*t~~~~~~~~~~~~~~~~~~n-~~~V--' 'Y PI~i( r '1~~~~~~~~~~~~~~~~~~~~~~~~~~~1 ~~ .~w-pw -p,4~~c~~.g~-%3 -At CM L~~~~~~~~~~~~~~~~~~~~~~~~~~~~- mggi ~ ~ ~ ~ v,M'vp 4. v ' ' '? .¼4$ &',~~VV3t-~~~4Y~- 4,. .tt.,v:?. Cj$J"?.)t.&tJ~~~~~~~~?3 - r%t,it-.~~~~fl.~~1 ~~. A.±.s4 '-J~~".r- .- NO ~~~~ -, 21:721 2~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~7 g ~~~44*~~~~~f9$~~~~~r.~~~~.A Public Hospitals in Developing Countries A World Bank Book Public Hospitals in Developing Countries Resource Use, Cost, Financing Howard Barnum and Joseph Kutzin Published for the World Bank The Johns Hopkins University Press Baltimore and London ( 1993 The International Bank for Reconstruction and Development/The World Bank 1818 H Street, N.W., Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America First printing April 1993 The Johns Hopkins University Press Baltimore, Maryland 21211, U.S.A. The findings, interpretations, and conclusions expressed in this publication are those of the authors and do not necessarily represent the views and poli- cies of the World Bank or its Board of Executive Directors or the countries they represent. The material in this publication is copyrighted. Requests for permission to re- produce portions of it should be sent to the Office of the Publisher at the ad- dress shown in the copyright notice above. The World Bank encourages dissemination of its work and will normally give permission promptly and, when the reproduction is for noncommercial purposes, without asking a fee. Permission to copy portions for classroom use is granted through the Copy- right Clearance Center, 27 Congress Street, Salem, Massachusetts 01970, U.S.A. The complete backlist of publications from the World Bank is shown in the annual Index of Publications, which contains an alphabetical title list and in- dexes of subjects, authors, and countries and regions. The latest edition is available free of charge from Distribution Unit, Office of the Publisher, The World Bank, 1818 H Street, N.W., Washington, D.C. 20433, U.S.A., or from Publications, The World Bank, 66 avenue d'Ina, 75116 Paris, France. Library of Congress Cataloging-in-Publication Data Barnum, Howard. Public hospitals in developing countries: resource use, cost, financing / Howard Barnum and Joseph Kutzin. p. cm. "Published for the World Bank." Indudes bibliographical references and index. ISBN 0-8018-4532-7 1. Hospitals-Developing countries-Finance. 2. Hospitals- Developing countries-Cost control. I. Kutzin, Joseph. II. International Bank for Reconstruction and Development. [II. Title. [DNLM: 1. Developing Countries. 2. Financial Management, Hospital. 3. Hospitals, Public-economics. 4. Hospitals, Public- utilization. WX 157 B263p] RA971.3.B37 1993 338.4'336211 '091724-dc2O DNLM/DLC 9249341 for Library of Congress CIP Contents Foreword vii Adetokunbo 0. Lucas Acknowledgments viii 1. Introduction I The Scope of the Book 2 The Issues 3 Organization of the Book and Overview of the Findings 4 2. Patterns of Hospital Resource Use 11 Distribution of Resources within the Government Health Sector 13 Distribution of Resources within Hospitals 27 Distributional Equity of Hospital Use 33 Cost-Effectiveness of Hospital Services 48 Summary 61 Appendix 2A. Resource Use 67 3. Hospital Costs and Efficiency 79 Accounting-Based Cost Studies 82 Service Statistics, Efficiency, and the Demand for Hospital Care 90 Efficiency of Input Use 108 Statistical Cost Studies 114 Summary 133 Appendix 3A. Step Down or Cost Center Analysis of Hospital Costs 135 Appendix 3B. Use of Service Indicators for Rapid Assessment of Relative Performance of Indonesian Type C Hospitals 137 v vi Contents 4. Hospital Financing Alternatives 143 Health Financing Objectives 144 Altemative Policies 152 Summary 165 Appendix 4A. Optimal Prices for Hospital Services 167 5. Hospital Financing Experience 179 Revenue Performance 180 Fee Policy in Selected Countries 194 Insurance Financing of Hospitals 229 The Effect of Alternative Hospital Financing Programs on Efficiency and Equity 239 Summary and Recommendations 254 6. Reallocating Hospital Resources to Improve Health Services 259 Changing Hospital Referral Pattems 262 Using Hospital Resources to Support Primary Health Care 274 Altematives to Inpatient Hospital Care 284 Summary 294 Appendix 6A. Quality and the Market for Health Services 295 7. Conclusions 299 Hospital Resource Use 299 Hospital Costs 301 Hospital Financing 303 Suggestions for Future Research 305 References 311 Index 327 Foreword Hospitals consume the largest share of government health resources, yet until recently, they have not been a focus of health policy research in developing countries. The attention paid to community-based and other nonhospital primary care interventions was appropriate, of course, be- cause of the cost-effectiveness of these measures compared with those delivered in hospitals. But this should not have led to analytical neglect of hospitals. It is precisely because hospitals account for such a large share of health expenditure that improvements in their efficiency can yield tremendous benefits for the entire sector. Moreover, strong hospi- tals at the first referral level are vital inputs to the success of an integrated primary health care strategy. For example, the strengthening of surgical and special obstetric services at the referral hospital is a key component of the strategy for reducing maternal deaths in developing countries. Most of the studies that are discussed in this book date back only to 1987, and data and analyses presented herein will prove useful to both researchers and policymakers. The cases reviewed in this book illustrate analyses that hospital managers in developing countries should carry out in their own institutions. In addition, the authors highlight a number of unanswered questions that can form the basis for further studies, either national or cross-national. By improving the efficiency of hospitals in developing countries, a greater number of persons can be served with available resources. By ensuring that hospitals provide appropriate support to the primary care level and that they are well integrated into national health systems, their effectiveness can be much enhanced. This is of vital importance in this era of worldwide fiscal constraints. I commend this book to all engaged in improving the health of people living in developing countries. Adetokunbo 0. Lucas Director, International Health Programs Department of Population and International Health Harvard School of Public Health vii Acknowledgments This book owes much to the interaction, help, and encouragement of many colleagues during a period of years. At an early stage in our formulation of the research effort, when hospitals were a comparatively neglected topic of research in the health sectors of developing countries, the support and interest of Nancy Birdsall, Anthony Measham, and Dean Jamison were invaluable. In order to identify the issues and establish a manageable focus, we began our research with a set of background papers prepared in 1986 and 1987 by Kenneth Chomitz (1986), David Dunlop (1986), Julio Frenk, Enrique Ruelas, and Avedis Donabedian (1989), Andrew Green (1987), DominiqueJolly (1987), Anne Mills (1990a and 1990b), and Godfrey Walker (1986). As indicated in the References, some of these studies have since been published and are generally available. Our examination of these broad overviews was followed by studies in specific countries: China (Barnum 1989), Ethiopia (Bitran- Dicowsky and Dunlop 1989), Indonesia (Djuhari and others 1988; Bar- num 1987b), Malawi (Mills 1991), and the Dominican Republic, Honduras, and Jamaica (Lewis 1990). The section on statistical cost functions in chapter 3 is based partly on ongoing research we are conducting with Adam Wagstaff on the characteristics of hospital cost fumctions. In addition to the activities commissioned specifically for this book, the findings and recommendations are based on a detailed survey of the relevant literature, both published and unpublished. Sector work con- ducted under World Bank auspices has contributed greatly to our re- search. Although we have drawn information from many World Bank reports, the work managed by John Briscoe and William McGreevey in Brazil, Richard Bumgarner in China, Nicholas Prescott in Indonesia, and Hazel Denton in Nigeria deserves specific mention. Many non-Bank studies were also used. Of particular note is the work carried out under two activities funded by the United States Agency for International Development: the Health Care Financing in Latin America and the Caribbean (HCF/LAC) project, managed by Dieter Zschock; and the Re- sources for Child Health (REACH) Project, managed by Gerald Rosenthal. viii Acknovledgments ix In May 1989, a workshop entitled "Hospital Resource Use and Cost Containment" was held at the World Bank. Presentations on hospital costs, financing, and health service alternatives were made by Ricardo Bitran-Dicowsky, David Dunlop, Charles Griffin, Maureen Lewis, Anne Mills, Germano Mwabu, Julia Walsh, Annemarie Wouters, and Mary Young. Jose-Luis Bobadilla, Richard Bumgarner, Philip Musgrove, Mead Over, and Donald Shepard added greatly to the discussion of these issues. Following the workshop, we began work on this book. Many people reviewed parts or all of earlier drafts of the manuscript, some on several occasions. We are most grateful for their written com- ments as well as contributions made during meetings held to discuss the first draft in August 1990 at the World Bank. Special acknowledgment is due to Martha Ainsworth, Ricardo Bitran-Dicowsky, Jos&Luis Bobadilla, John Briscoe, Richard Bumgarner, Guy Carrin, Andrew Creese, Carlos Cruz-Rivero, Hazel Denton, David Dunlop, Guy Ellena, Julio Frenk, Willy de Geyndt, Fred Golladay, Charles Griffin, Davidson Gwatkin, Salim Habayeb, Jeffrey Hammer, Jean-Louis Lamboray, Mau- reen Lewis, Patricio Marquez, Jo Martins, Anthony Measham, William Newbrander, Mead Over, Donald Shepard, Anne Tinker, Louis Vas- siliou, Adam Wagstaff, Julia Walsh, Marcia Weaver, Vivian Wong, and Annemarie Wouters. The exacting readings given the manuscript by four anonymous referees commnissioned by the World Bank's Office of the Publisher were also invaluable and resulted in imnportant revisions. Any remaining omissions and errors are our responsibility. We are grateful to Joanne S. Ainsworth for a splendid job of editing this book and to Leonila Jose, who provided excellent secretarial assis- tance throughout all stages of preparation of the manuscript. 1. Introduction This book examines economic and financial issues of hospital resource allocation with the objective of contributing to policies that will improve the use of public sector funds by hospitals. More specifically, the book concerns (a) the allocation of health sector resources between hospitals and nonhospital alternatives, (b) the internal efficiency of hospital oper- ations, and (c) effective and equitable cost-recovery policies for hospitals. Hospital operating expenses are at the core of the growing gap be- tween required and available resources in the health sector of many countries. Hospitals receive the lion's share of public recurrent resources in health. Although the actual percentage varies from country to country, it is common for 50 to 80 percent of public sector health resources, in money and trained personnel, to be used in hospitals. The remaining resources are used for preventive care, infectious and parasitic disease control programs, nonhospital maternal and child health programs, and other services that have been identified as generally more cost-effective than hospital care in countries with limited health resources. Reviews of the health sector in many countries suggest that these large recurrent expenditures on hospitals involve a great waste of resources because of the inappropriate allocation of funds within the health sector and the technical and managerial inefficiency within hospitals. Notwithstanding the large share of government health resources allo- cated to hospitals, much of health research has focused on maternal and child health programs and first-line services at the periphery rather than on hospitals and higher-level curative care. The emphasis on nonhospi- tal interventions reflects the limited results produced by capital-inten- sive hospital projects. The avoidance of capital expenditure on hospitals in health sector projects has been, and generally continues to be, desir- able, but the neglect of hospitals in sector research has been unfortunate, because research can be influential in changing the role of hospitals and the flow of resources in the sector. In addition, the emphasis on non- hospital interventions may have overshadowed the important potential contribution of hospitals to the integral health system. 1 2 Public Hospitals in Developing Countries Just five or six years ago there was relatively little information on hospital economics in developing countries. Recent studies, however, sponsored by ministries of health in collaboration with such donors as the World Health Organization (WHo), the United States Agency for International Development (UsALD), and the World Bank, have begun to fill the void in hospital research. Although our knowledge of hospital costs, finance, and resource use remains incomplete, the new research provides important data that are needed to develop improved hospital policy. This book draws widely on this research to support the analysis presented. The Scope of the Book The overarching viewpoint of this book is economic. Thus we examine the main questions of hospital efficiency within the framework of the economics of production and costs. Full answers to these questions must also involve the fields of hospital management, organization, and per- sonnel planning, but to keep the book manageable we have restricted the scope to economic and financial issues. Economic analysis can help to identify the areas in which changes are needed even though im- plementation of the changes ultimately requires noneconomic analysis. Also, some important policy areas, especially those related to incentives and pricing, directly involve economics and finance. The economic view of hospitals provides an important perspective, but it is not all-encompassing. This perspective is subsumed by a broader public health view of hospitals, which considers hospitals in relation to the epidemiology of the population and the structure of other health services. Public health policy will reflect the way in which health author- ities attempt to meet health needs, as determined by a population's epidemiology and demographic structure. The level of preventive activ- ities, the emphasis on basic care provided in health centers and through outreach programs, referral policies, and the choices made by consumers of health services all act to determine the case mix faced by a given type of hospital. Treatment norms within the hospital (and in the medical community in general), administrative policies, and staff composition and incentives determine the procedures used for cases of a given type. Although we discuss many aspects of public health policy in our eco- nomic approach to hospitals, we do not attempt an in-depth analysis of all aspects of the health system. The policy recommendations reached from the economic analysis of hospitals, however, do have implications for the entire health sector. The scope of this book is limited primarily to public hospitals and to public programs that finance hospital care, irrespective of whether such Introduction 3 care is delivered by governmental or private providers. This limitation is consistent with the objective of the book, which is to contribute to public sector efficiency. It also has kept the book within manageable bounds by restricting the information needed to the relatively more accessible public hospital data. Information on private sector hospitals is difficult to obtain because very few countries have any system of regular reporting to a central authority on the details of private sector hospital operations. Nevertheless, nonprofit hospitals operated by non- governmental organizations (NGos) as well as profit-seeking hospitals play an important role in the provision of services in most developing countries. Public and private sector hospitals inevitably interact in the market for services, and this interaction is taken into account in the policy discussion. The Issues Correcting the problems with hospital resource use does not simply mean allocating a greater proportion of development funds at the mar- gin to nonhospital projects. In order to control hospital expenditures and improve the efficiency, management, and role of hospitals in the health sector, we need a better understanding of the issues underlying the allocation of health sector resources to hospitals. Research on questions of hospital economics and finance is needed to gain this understanding and help formulate strategies for more effective hospital resource use. Some of the questions that must be answered follow: & What share offinancial and personnel resources do hospitals absorb? Who uses hospitals? Given an appropriate criterion, such as maximum health benefit, is the share of public health sector resources absorbed by hospitals optimum? The actual share of resources going to hospitals is not clearly known in many countries, but aggregate government data, together with special detailed studies in some cases, do suggest substantial differ- ences among countries in the allocation of public sector health re- sources. The amount of health sector resources absorbed by hospitals in contrast to that taken in by nonhospital programs can have im- plications for the success of primary health care and the overall effectiveness of the public health sector. The provision of hospital services also has implications for the distribution of benefits from public resources. * What are the levels of hospital average and marginal costs in selected regions and countries? How can these unit costs be measured? What are the relations among unit costs, occupancy rates, length of stay, and quality of 4 Public Hospitals in Developing Countries services? What are the determinants of hospital unit costs? To what extent can hospital unit costs be reduced and efficiency increased through adjust- ments to factor mix and changes in length of stay? How does efficiency vary with type and size of hospital? Knowledge of unit costs is needed to assist planning for recurrent budgets, as an indicator of efficiency, and to inform pricing of services. Costs are not known with any accuracy for most hospitals in develop- ing countries. Recent studies have demonstrated the feasibility, how- ever, of reconstructing hospital expenditures to obtain estimates of unit costs. This information can be used to suggest means of increasing hospital efficiency through adjustments to factor mix, changes in length of stay, and improved use of the referral system. * Is cost recoveryfeasibleanddesirable in hospitals? Whatprinciples should determine the level of hospital fees? Can practical guidelines be developed? Should risk-sharing mechanisms be used to finance hospitals? What institu- tional features should be included in insurance programs? Decreased availability of government recurrent revenues has put new pressure on government health officials to look for alternative sources of funds. Cost recovery has been promoted by donors and by govern- ment ministries but with little practical guidance on how to set fees and for what services. Cost recovery in health could start most readily with the use of fees for hospital services, where, because of adminis- trative capacity and concentration on curative care, the use of fees is more practical than it would be for lower-level services. The appropri- ate level of these fees, however, is open for discussion. * Can thefunctioning of the referral system be improved in order to increase access to care and the economic efficiency of the health sector? Are there practical, lower-cost alternatives to the delivery of health care now delivered through secondary and tertiary hospitals? The high cost of technically complex hospital services has motivated a search for less complex alternatives that could provide acceptable care at lower cost. Some of the possibilities considered have included redefining the role of district-level hospitals, use of home care, and greater reliance on short-term stays and outpatient care. Organization of the Book and Overview of the Findings The organization of the book flows directly from the issues set out above. The magnitudes of hospital use of public sector health resources and hospital unit costs are identified, and then possible financing and health sector service alternatives are considered. A brief overview of the book is given below. More detailed summaries are provided at the end of chapters 2-6. Chapter 7 recapitulates our principal findings and Introduction 5 recommendations and concludes with suggested topics for future re- search. Hospital Resource Use The magnitude and patterns of hospital resource use are examined in chapter 2. As expected, we found that in almost all countries hospitals receive the largest share of government health resources and that, within the hospital subsector of many countries, tertiary hospitals receive a large share of both financial and skilled personnel resources in relation to that received by district hospitals. Furthermore, we found that hospi- tal services are used disproportionately by urban populations, and there is evidence (although less clear) to suggest that hospital services are used less by the poor. Another finding is that not only do adults and the elderly use hospitals more than their share of the population would indicate, but they are also more expensive to treat. Thus as the popula- tion ages, pressure on hospital resources will increase if services continue to be delivered in the current manner. Finally, the chapter includes a survey of available studies on the cost-effectiveness of individual health interventions. These studies confirm that, in low-income economies, nonhospital interventions are more efficient in dealing with the preva- lent health conditions. Although not surprising, these findings are im- portant because they emphasize the continuing potential problem of extensive use of health sector resources in hospitals. In addition to these more evident results are some findings that lead to less evident conclusions. One surprising result is the lack of a clear inverse correlation between the hospital share of resources and success in achieving primary health care objectives. In countries that have achieved important gains in reducing infant mortality during the last twenty years, such as Sri Lanka and China, both a high absolute level of support for the health sector and the involvement of district-level hos- pitals in a broad sectoral strategy have contributed to the success of primary health care programs even though hospitals absorb more than 60 percent of public recurrent health spending. More important than the share of public health resources absorbed by hospitals is the use of the resources within the hospital subsector and the overall composition of health programs. In countries with successful primary health care programs, there is an emphasis on district-level hospitals rather than large tertiary facilities. Additionally, as the level of income increases, the relative importance of hospital services increases. Thus, the optimum proportion of health sector resources absorbed by hospitals must be determined by assessment of the epidemiological needs and the available personnel skills, the financial resources, and the structure of the hospital subsector of individual countries. 6 Public Hospitals in Developing Countries Hospital Costs Hospital unit costs and their implications for policy are discussed in chapter 3. This section relies both on accounting-based studies and on statistical estimation as complementary tools to examine internal effi- ciency issues and to generate a cost basis for planning and cost-recovery policies. The book does not give full answers to all the questions raised above. Information is provided on the magnitude of costs, the method- ologies for measuring costs, and the relationships between the various service indicators. But answers to the questions of the relation of hospital cost to efficiency remain tentative, primarily because of the difficulty in measuring the quality of hospital services. Quality issues have not received satisfactory treatment in the literature on hospital cost and are an important topic for future research. With the caveat concerning the issue of quality, we can summarize some of the chapter's findings. An important fact demonstrated by the survey of accounting studies of average costs is that such studies are feasible in low-income countries. Collection and analysis of the data required to calculate average costs can be made a routine hospital activity with the objective of improved planning, management, and budgeting. The estimated average costs demonstrate that tertiary-level hospitals have substantially higher costs per patient than do more basic hospitals. This situation is to be expected, but the treatment by tertiary facilities of cases not requiring specialized care reflects a waste and misuse of resources. The difficulty, of course, is in providing credible services at the district level that can support a more rational use of referral facilities. In chapter 3 (and chapter 6 as well) we discuss the importance of improving the quality of district facilities. There have been very few statistical studies of developing-country hospital costs, and it is thus premature to generalize from their findings. The cost functions discussed in chapter 3 do not show evidence of long-run economies of scale or of economies of scope.' These findings are consistent with similar studies of hospital costs in industrial coun- tries. Clearly, however, more examples from developing countries are needed to improve our understanding of the hospital production process in this context. Many relations between service statistics and efficiency have been identified. A cross-country survey reveals that occupancy rates, bed turnover rates, and average lengths of stay vary considerably across countries and among levels of hospitals.2 Low turnover and occupancy rates, especially among lower-level facilities, are often attributable to the poor quality of services caused by insufficient drugs and other supplies and a lack of skilled staff. Analysis suggests that, with plausible demand response, higher occupancy and possibly lowered average cost can be compatible with greater expenditures on facilities to improve quality. Introduction 7 Another finding is that low turnover and long average lengths of stay contribute to high costs per case treated. Much of this problem is mana- gerial and requires careful studies if solutions are to be proposed in individual cases. Some potential causes are poor scheduling of diagnos- tic services or surgery, outmoded treatment protocols, and misuse of secondary and tertiary facilities for extended care or convalescence. The discussion in chapter 3, as well as in chapters 4 and 5, also suggests that poor hospital performance, as revealed by service statistics, can relate to adverse incentives provided by the structure of hospital financing. Financing Hospital cost recovery is considered in chapters 4 and 5. In these chapters we build on the unit cost discussion and examine efficiency, equity, and revenue implications of the use of fees and insurance for hospital ser- vices. Chapter 4 provides a general framework for the analysis; here we review the deficiencies in the private market that lead to public interven- tion and set out the efficiency, equity, and revenue objectives of financing policy. In chapter 5 we review current cost-recovery policies in a variety of country settings. Chapter 4 contains some principles for setting hospital prices: (a) fees should be consistent with ability to pay, (b) fees should provide signals that promote economic efficiency, (c) fees and the quality of services should be linked, (d) fees should be subsidized for services that have externalities, are public goods, have associated informational deficien- cies, or are merit goods.3 A full answer to the question of how to set fees requires information on both sides of the market-household demand and the hospital cost of supplying services. Studies of hospital unit costs provide an important supply-side component to help determine appro- priate fees, whereas studies of service use response to income and price levels provide required demand-side information. Application of recent public enterprise economics integrates demand- and supply-side infor- mation to allow the development of guidelines for fee determination that make explicit the roles of costs and demand response. Briefly, prices should reflect price and income elasticities as well as costs. Services that are more income elastic (that is, a given percentage increase in income is accompanied by an even greater percentage change in the quantity of service demanded) should have a greater proportion of costs reflected in their price than services that are income inelastic. The benefits of applying this rule are improved equity and efficiency. The application of pricing rules must be tempered according to the institutional setting in which financing takes place. The use of fees, risk-sharing arrangements, management and organization, and the overall structure of the health system all greatly affect client and pro- 8 Public Hospitals in Developing Countries ducer incentives and performance in individual countries. In chapter 5 we review the systems used to finance health care in several countries and set forth some practical additions to the pricing principles outlined above. These additions can be abbreviated here: timely fee adjustments should be an integral part of the system, fee structures should be simple, and exemptions should be limited to those granted on equity grounds or for services that provide benefits well beyond the patient. The chapter also includes some suggested guidelines for risk sharing: cost contain- ment requires active involvement by the insuring institution to encour- age providers to behave in a cost-effective manner; cost sharing is needed in all risk-sharing schemes to provide consumer incentives for efficiency (but these appear to be less effective than provider incentives); prepaid capitation schemes (for example, health maintenance organizations) are practical in circumscribed communities. Publicly sponsored provision of insured hospital services for only a small proportion of the population should be done with great caution to avoid inequitable distributional consequences. Health Sector Alternatives Lower-cost alternatives to hospitals are considered in chapter 6. In this chapter we examine the role of hospitals in the wider network of health service providers and suggest that there is ample scope for improving the overall cost-effectiveness of the sector through a reallocation of resources in favor of relatively low-cost providers. The suggested ap- proach is to begin with an epidemiological picture of the population, delineate several viable packages of interventions for meeting health needs, and cost the alternative packages. The alternative that would yield the maximum health benefit for the available resources should be chosen. It is expected that a choice made in this manner would result in a network of providers and an allocation of resources that would encour- age treatment of frequently occurring, low-cost conditions in the least expensive, yet capable, setting. Alternative treatment settings would indude care in the home and community and noncomplex facilities such as health centers, which are often considered to be important compo- nents of a health referral system. Innovative alternatives not typically thought of in this manner, such as outpatient surgery facilities, could also be included if they are found to be cost-effective. As a result of the use of these alternatives, the case mix of hospitals of varying complexity would change so that, for example, tertiary hospitals would primarily treat patients requiring highly specialized care rather than using sub- stantial resources for patients who could be cared for in less expensive facilities. Introduction 9 Practical ways in which the development of alternatives to hospital care can be supported are discussed in the chapter. It is most important to improve quality at lower-level facilities so that the population will not bypass them in favor of hospitals. To do this would require reallocating resources in favor of noncomplex facilities, perhaps away from hospitals. Another suggested policy is to coordinate the management of all provid- ers at the district level, including the district hospital. The district hospi- tal can also support primary care provided in lower-cost settings with training programs, assistance with logistics, and provision of credible diagnostic backup. Finally, we discuss alternatives to expensive hospital inpatient care that may be useful in many developing countries. These include treatments that can substitute for hospitalization and the use of lower-level facilities for treatment and palliation of chronic diseases. Notes 1. See chapter 3 for definitions of these and other characteristics that can be derived from statistical cost functions. 2. See chapter 3 for definitions of these measures of service performance. 3. These terms are all aspects of market failure. A discussion of market failure and definitions of these terms are given in chapter 4. 2. Patterns of Hospital Resource Use A change in health policy can affect the balance between services pro- vided by hospitals and those provided by nonhospital facilities, and it can alter the share of government health sector resources used by hospi- tals. Interest in the share of health sector resources used by hospitals is motivated by the need to determine whether the allocation of resources within the health sector is economically efficient. An allocation of health sector resources among alternative activities is economically efficient if there is no possible reallocation that will improve health status. Unfor- tunately, appropriate measures of health status are not easily defined for the health sector in general and are especially hard to define for hospitals with their complex multiple outputs. Although hospitals are an essential part of any health system, the optimum allocation of resources is not obvious but is open to debate. Considerable analysis and discussion during the last twenty years have led to a rough consensus about the cost-effectiveness of primary curative care and preventive services com- pared with that of hospital services. Health planners generally agree that in countries with low levels of per capita gross national product (GNP) and high rates of mortality, the most effective allocation of resources would result in a lower share of public resources committed to hospitals than in countries with high GNP and low mortality. Accordingly, the share of government health resources going to hos- pitals is a rough indicator of the structure and emphasis within the health sector. A relatively low share suggests an emphasis on primary health care and concern with reaching rural populations, and a high share suggests an emphasis on curative care and concern with urban health. This interpretation of the hospital share, however, should be applied cautiously to any specific country because the types of facilities defined as hospitals vary across countries and, apart from definitional issues, the absolute level of resources flowing to the health sector and the quality 11 12 Public Hospitals in Developing Countries of health services are also determinants of the structure and effectiveness of the health sector. Despite the consensus that a greater share of public sector health resources should be allocated to nonhospital activities than is currently the case in many countries, the level of the appropriate percentage is not obvious. The provision of better quality and increased quantity of pri- mary health care services need not conflict directly with the existing allocation of resources. Primary curative care is one type of service currently provided in many hospitals, and as discussed in chapter 6, hospitals can also provide a variety of preventive, palliative, and reha- bilitative services and be structured so that they integrate with and provide support for nonhospital services. Hospitals are relatively high- cost institutions, especially with regard to investment, but primary health care outreach activities, although relatively low cost in terms of investment, can entail large recurrent costs. What is important is to obtain a balance of services throughout the health sector, with each investment being examined on the margin for its cost-effectiveness and contribution to sectoral efficiency. Interest in the share of public sector health resources going to hospitals is also motivated by a concern with equity. There is no universal defini- tion of equity because it is a normative concept that is defined in the context of prevailing social and moral values. But again based on a consensus view, equity might be defined as equal access across popula- tion and income subgroups to health services covering basic health needs. Using this minimum definition, equity could be enhanced with a progressive distribution of the burden of payment across income groups. Just what is included in basic health needs is only roughly defined, but at a minimum it encompasses safe childbirth, prevention and treatment of the main child health problems, protection from common tropical diseases, and access to first aid for accidents. Hospitals are not the most effective way to deal with many of these needs. In addition, hospital facilities are commonly distributed unequally across geographic areas and used unequally by different income groups. Thus, a relatively inequitable allocation of health sector resources is suggested by a very high percentage of recurrent health expenditures being absorbed by hospitals. This chapter is largely descriptive of observed patterns of government resource allocation in the health and hospital sectors. The data are too variable in definition across countries to support strong policy prescrip- tions. Also, policies to reallocate sectoral resources in a particular coun- try should be supported by detailed studies addressing the specific institutional issues of the setting. For these reasons we confine ourselves to identification of broad characteristics of hospital resource use and the potential policy implications of these patterns. Patterns of Hospital Resource Use 13 In the first two sections of this chapter we review cross-country information on the use of resources by public hospitals, first, with regard to the distribution of resources within the health sector and, second, with regard to the distribution of resources within hospitals. In the third section we assess data on the use of hospital services across income groups, geographic location, age, and disease types and consider the implications of existing use patterns for equity and future resource costs. In the fourth section of the chapter we review information related to the consensus view on the cost-effectiveness of hospital services. The chap- ter thus describes the resources used by hospitals and then turns to a normative evaluation of hospital resource use. The final section summa- rizes the principal findings of the chapter. Distribution of Resources within the Government Health Sector Hospitals compete with other health facilities and programs for both recurrent and capital resources. The amount allocated to hospitals must be further redistributed among different levels and types of hospitals. These allocational issues are described below. Recurrent Resources In most developing countries the distribution of recurrent resources within the government health sector strongly favors hospitals. Quanti- tative comparisons across countries are difficult because of variation in the definition of hospitals and a lack of comprehensive information available from the private sector and from different levels of government within the public sector. Broadly, the available data do reveal that a high percentage of government health resources are assigned to hospitals. Among the twenty-nine developing countries listed in table 2-1, over half spend more than 60 percent of recurrent health budgets on hospitals, and about two-thirds spend 50 percent or more. Only four countries distribute less than 40 percent of public sector health resources to hospi- tals. However, the implications of the hospital share for the success of primary health care (PHC) programs and for national health indicators are not clear and must be analyzed at the country level. Nepal, which has the lowest hospital share of any country in table 2-1, has a low level of absolute support for health services and notable problems in im- plementing quality primary care programs (World Bank 1989). It also has a low level of health status, reflected in an infant mortality rate (IMR) of more than 125 per 1,000 live births in 1987. Indonesia, also a low- income country, has been more successful at implementing primary care programs (Prescott 1991) and has a higher overall level of support for 14 Public Hospitals in Developing Countries Table 2-1. Share of Hospitals in Total Public Recurrent Health Expenditure (percent) Country Hospital share Year Source Bangladesha 61 1987 Griffin 1989 Botswana 49 1984 Brazil 68 1984 Burundi 66 1986 Chinab 61 1987 Colombia 67 1984 Cote d'IvoireC 46 1984 Vogel 1988 El Salvadord 62 1985 Fiedler 1987 Ethiopia 49 1983-84 The Gambia 45 1985-86 Indonesia 37 1985-86 Jamaica' 72 1986-87 Kutzin 1989 Jordan' 75 1987 Kenya 73 1985-86 Lesotho 74 1986-87 Malawig 81 1985-86 Mexicoh 58 1986 Mozambique! 36 1987 Nepali 25 1987-88 Niger 30 1988 Budd 1989 Papua New Guinea 45 1986 Newbrander 1987 Philippinesf 71 1985 Senegal 50 1982 Vogel 1988 Somalia 70 1989 Sri Lanka 70 1986 ADB 1987 Swaziland 52 1983-84 Turkeyk 63 1987 Ugandal 43 1982-83 Zimbabwe'l 54 1987 OECD meann 54 1980s OECD 1987 a. Includes Upazila Health Complexes (primary-level hospitals). b. Includes all expenditures by central and provincial governments on hospitals, hos- pitals of traditional Chinese medicine, and township hospitals. c. Percentage reflects mean of 40-51 percent range reported by Vogel. d. Percentage may reflect hospitals as a share of all faclity-related health expendi- tures, rather than of total public sector health expenditure. e. Includes mental health hospital and the share of drugs and other medical supplies allocated to hospitals from central administration. f. Excludes health expenditures by public sector other than Ministry of Health. g. Percentage reflects share of expenditure devoted to "curative" services, which over- states the hospital share to some extent. h. Includes capital and recurrent public health expenditures. i. Recurrent budget estimates. j. Total (capital and recurrent) budget devoted to secondary and tertiary care and to hospitals with fewer than fifty beds; excludes amounts budgeted for military and mis- sion hospitals. Patterns of Hospital Resource Use 15 k. Includes expenditures by the social insurance organization. 1. MOH expenditures (uncertain if capital included) with medical stores proportionally reallocated to faclities (may understate hospital share); some health center services in- cluded. m. Includes health expenditures by MOH and municipal or local government health authorities; excludes central government health expenditures by other ministries. n. Institutional health care, includes nursing homes, does not include ambulatory care. Source: World Bank sector reviews and appraisal reports, except as noted above. health services. Indonesia's imR was 70 in 1987. At the other extreme, eight countries in table 2-1 devote 70 percent or more of public health resources to hospitals; among them Lesotho, Malawi, and Somalia have IMRs of more than 100, and Jamaica, Jordan, the Philippines, and Sri Lanka have IMiRS below 50 (see the appendix table at the end of the chapter for IAR data). The use of beds per capita as a measure of the intensity of hospital resource use is common but is fraught with difficulties. The definition of a hospital bed varies across countries. Most developing countries do not distinguish between acute beds (used for short-term stays of less than, say, thirty days) and long-term beds, although some countries do distin- guish long-term specialty hospital beds, such as those for tuberculosis, leprosy, and psychiatric care. Even more troublesome are occasional large differences between the number of beds officially installed and the number of operational beds. A breakdown between acute and chronic beds, and confirmation that the beds given are operational, would add greatly to the interpretation of the comparative bed ratios, but this information is not available for many countries. Thus, bed ratios give only a gross measure of the availability of hospital services. Generally, there is a strong relation between the availability of hospital beds per capita and the level of per capita GNP (figure 2-1). The number of beds per capita increases by about 4 percent for every 10 percent increase in per capita income.' The structure of the health sector in terms of the balance of resources between hospitals and other health service alternatives, however, is not as easily explained as the overall availability of hospital beds. Despite the consensus that poorer countries should devote a larger share of public sector health resources to nonhospital interventions, available evidence suggests that there is no clear relation between the hospital resource share and the level of per capita national income. This lack of relationship is dear in the scatter of points relating the hospital share of government health expenditure on the Y axis and GNP per capita on the X axis in figure 2-2, which indicates that there are a number of poor countries with relatively high hospital shares. The high hospital shares of some poor countries may be explained (in part) by the nature 16 Public Hospitals in Developing Countries Figure 2-1. Hospital Beds per 1,000 Population Beds per 1,000 9 High income (38) S 8 7 6 * Upper-mniddle income (26) 5 4 3 * Lower-middle income (52) 2 * Low income (47) 1 0 2 4 6 8 10 12 14 16 Per capita GNP (thousands of 1989 US dollars) Source: World Bank data, 163 countries; most recent year, 1970-89. of hospitals as high-cost institutions. A considerable share of the total health resources of a small country may be taken up by one large referral hospital because of that hospital's high cost and the country's limited resources. In Lesotho, for example, one hospital, the national referral facility, absorbed 42 percent of total Ministry of Health (MOH) expendi- ture in 1986-87 (United Medical Enterprises 1988). Reducing that share without significantly increasing the absolute level of resources in the health sector may imply drastically changing hospital functions or re- ducing government responsibility for its financing. Both of these actions are the subject of later chapters. Some of the cross-country variation in hospital share may be attribut- able to differences in the definition of a hospital. In China, for example, if small township hospitals (which largely provide primary care) are not included with secondary and tertiary care facilities, the 1987 hospital share drops from 61 to 44 percent. The figures reported in table 2-1 reflect varied definitions of what constitutes a hospital in individual countries. This variation can mask the potential inequity and inefficiency resulting Patterns of Hospital Resource Use 17 Figure 2-2. Relation between the Share of Public Health Spending Devoted to Hospitals and per capita GNP, 1982-89 Percent 90 80 - * Malawvi Kenya . @ Lesotho * Jamaica * Jordan 70 - Somalia 0 0 Philippines Burundi * Sri Lanka * Colombia * Brazil 60 - * * China * El Salvador 0 Turkey Bangladesh Senegal * Zimbabwe Mexico 50 Etipia Sngl0 Swaziland * TheGambia Cote d'lvoire * Botswana 4 * gandia * Papua New Guinea 40 * Uganda * Mozambique 0 Indonesia 30 - Niger * Nepal 20 10 0 I 0 500 1,000 1,500 2,000 Per capita GNP (1980 US dollars) Source: World Bank sector reviews, project reports; see also table 2-1. from a large proportion of resources being allocated to large urban tertiary hospitals as opposed to smaller district and rural hospitals. Cross-country differences in the definitions of hospitals and in the share of recurrent hospital expenditures devoted to a few large tertiary hospitals compared with many smaller hospitals limit the usefulness of the overall public sector hospital share as an indicator of the priority given to various health programs. Comparing hospital with nonhospital expenditures is not equivalent to a comparison of primary care with secondary and tertiary care because hospitals provide a considerable amount of primary curative care. Therefore, as suggested above, the data for hospital share alone do not support definitive conclusions regarding 18 Public Hospitals in Developing Countries a country's level of support for primary health care or the effectiveness of primary, secondary, or tertiary care programs. Country-specific anal- ysis is required for such conclusions. Figure 2-3 indicates that there is a tendency for higher hospital shares to be associated with a greater number of doctors per capita. This relation is strongest for middle- and high-income countries. An examination of GNP per capita and doctors per capita as multiple determinants of the hospital share for twenty-nine low- to middle-income developing coun- tries produces results that are not very strong but are consistent with this observation.2 The statistical pattern in a cross-section of countries sug- gests that a 1.6 percent increase in the hospital share accompanies a 10 percent increase in the number of doctors per capita. Interestingly, the Figure 2-3. Relation between the Share of Public Health Spending Devoted to Hospitals and Physicians per 1,000 Population Percent 90 Malawi 80 - Lesotho Kenya Jordan * / Philippines Jamaica O 70 * * Sri Lanka Colombia Brazil Somalia 5 60 - 0 Bangladesh * El Salvador Turkeyo China Swaziland S I * Zimbabwe Mexico 50 ** - Senegal 's0 * * *0 Botswana **The Gambia-Ethiopia * * Papua New Guinea 40 Uganda c C^ote d'Ivoire S * Indonesia Mozambique 30 - Niger * Nepal 20 10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Physicians per 1,000 population Source: See figure 2-2. Patterns of Hospital Resource Use 19 coefficient on per capita GNP becomes insignificant when per capita physician supply is included in the regression. A country's income level would still seem to be an important factor, however, because high-in- come countries tend to have greater availability of physicians per capita (see the appendix table A2-1). In any case, the relation between the hospital share of health sector resources, the number of doctors per capita, and GNP per capita suggests that policies to limit the future supply of doctors (for example, redirect- ing subsidies for medical education to the training of other types of health professionals) might have an effect in the medium to long term on reducing the share of health resources absorbed by hospitals. Also, the correlation between physicians per capita and hospital share may indicate that the medical curriculum is oriented too much toward the practice of high-technology Western medicine and does not correspond to the health needs of the population. To a certain extent the transition in disease mix-away from the diseases of childhood and toward chronic disease and the diseases of adulthood-that accompanies growth in per capita income can require more hospital resources. A conclusion that is reached from a consideration of the cost-effectiveness of health sector alternatives later in the chapter is that primary health care services continue to have the greatest effect on health status per unit of expenditure even for countries that have advanced to middle-income levels. Despite national and international calls for an emphasis on primary health care in the years since the 1978 World Health Conference held in Alma-Ata, the share of recurrent resources going to hospitals increased in many countries but remained about constant or decreased in other countries during recent years (table 2-2). The support that the data in the table lend to conclusions regarding a country's ability to reallocate resources is limited by the varying lengths of time between observations of the hospital share, but the data do suggest clear trends in some countries. The rate of annual increase of recurrent expenditures on hospitals compared with total health sector expenditures was 2.8 percent in Brazil (including nonhospital curative care), 6.6 percent in the Philip- pines, and 1.5 percent in Kenya, to pick some representative countries. In several other countries, the hospital share remained relatively con- stant, which may reflect the inertia of a historically based budgeting process (see Fiedler 1987 for a discussion of such a process in El Salva- dor). In some other countries, the hospital share decreased, usually as a result of deliberate govemment policies to control hospital expenditures and protect PHC programs. Mexico, The Gambia, and Zambia exemplify countries in which such policies have been translated into resource allocation decisions. In The Gambia, for example, the health plan adopted for the period 1981 to 1986 emphasized a village-based PHC strategy and described the shift in resources implied by this policy. 20 Public Hospitals in Developing Countries Table 2-2. Changes in the Share of Public Recurrent Health Expenditures Allocated to Hospitals, Selected Countries and Years (percent) Hospital Average annual Country Year shlare clhange Source Botswana 1979 42 1984 49 3.0 Brazil' 1975 70 1982 85 2.8 Chinab 1980 38 1987 44 2.2 Colombia 1974 65 1984 67 0.3 El Salvador 1981 62 1985 62 -0.2 Fiedler 1987 The Gambia 1979-80 67 1985-86 45 -6.2 Jamaica 1983-84 73 1986-87 72 -0.5 Kutzin 1989 Jordan 1982 80 1987 75 -1.2 Kenya 1978-79 66 1985-86 73 1.5 Lesotho 1970-71 64 1981-82 71 0.9 Malawi 1976-77 73 1985-86 81 1.2 Mexico 1982 64 1986 58 -2.6 Papua New Guinea 1975 46 1987 45 -0.3 Newbrander 1987 Philippines 1981 55 1985 71 6.6 Swaziland 1976-77 69 1983-84 52 -4.0 Turkey 1981 68 1987 63 -1.4 Zambiac 1975 33 1981 30 -0.5 a. Percentages reflect the share absorbed by "curative" services (includes hospital and nonhospital). b. Includes only upper-level hospitals and hospitals of traditional Chinese medicine. c. Percentages reflect the share absorbed by the four largest hospitals only. Source: World Bank sector reviews and appraisal reports, except as noted. Although there were severe cuts in the real level of governrment health expenditure during the final three years of the plan period, the policy was adhered to, as evidenced by the data reported in table 2-2. Patterns of Hospital Resource Use 21 The recurrent resource needs of a country's health sector can interact dynamically with the decline of real public revenues that results, for example, from a fall in the prices of primary products coupled with softening economies in industrial countries and thus a drop in demand for these products. Nonhospital programs can therefore be placed in special jeopardy during times of economic difficulty, and a gap can be created between maintenance requirements and actual expenditures for hospitals. Declining public revenues have affected the quality and quan- tity of social sector outputs, although there has been no clear trend across countries as to whether these sectors have been affected dis- proportionately. Grosh (1990) found that, for eight countries from Latin America and the Caribbean between 1980 and 1988, declines in public spending on health exceeded declines in total government recurrent expenditures by an unweighted average of 25 percent. Public sector health expenditure, however, fell by less than total public spending in five of these eight countries, and total social sector spending (including education and social security) declined at only half the rate of total public spending. The way the burden of the reduction in health revenue is shared across programs differs greatly among countries. The fear of many public health specialists is that, contrary to the example of The Gambia reported above, as the real value of revenue allocations for health services de- clines, health ministries, faced with a painful choice in allocation be- tween hospital and nonhospital services, will favor hospitals, which have more visible and larger capital stock, over nonhospital programs, in which capital expenditures are less obvious and smaller. In the Phil- ippines, for example, during a decline in real government expenditure between 1983 and 1985, the allocation to the Department of Health (DOH) was reduced to a greater extent than the overall decline in public spend- ing. Within DOH expenditures, however, the share spent on hospitals increased, whereas expenditures on preventive programs declined rap- idly (Intercare 1987). But some countries have reacted to decreasing health revenues by protecting or even increasing primary health care programs while reducing hospital recurrent expenditures. In Zambia, for example, a decline in copper prices put pressure on public resources available in the health sector in the late 1970s, and the share of tertiary hospital recurrent expenditures was reduced by 3 percent from 1975 to 1981 as the real value of health expenditures declined by 13 percent (World Bank 1984). Investment Hospital investment is by its nature lumpy and irregular, and the hospi- tal capital stock produced by the investment has a life of two or more 22 Public Hospitals in Developing Countries decades. For this reason hospital investment expenditure for any given year can be a misleading indicator of the proportion of development funds absorbed by hospitals; the current size of the hospital sector results from much earlier health policies. Mills (1990a) contrasts selected data on the share of hospitals in investment in the 1960s with expenditures in the 1980s and notes the considerably higher share of capital expenditures spent on hospitals in the earlier period. For example, in Tanzania the hospital share of total health sector investment was reduced from 80 percent in 196142 to less than 32 percent in 1978-79. The high early figures result from large-scale hospital construction programs started in the 1960s, whereas the recent lower figures reflect the more limited donor funds now available for hospital construction and the greater priority given to primary health care. One cause of imbalance between recurrent resource availability and capital expenditures is that, in many countries, health ministries are largely responsible for recurrent budgets, whereas donors fund a large proportion of capital expenditures. Somalia provides an example of a very poor country in which both the recurrent cost and the investment cost of primary health care programs have been heavily supported by donors, but hospital recurrent costs are left to be supported almost strictly from government revenues (World Bank 1985). There is evidence that in recent years some health ministries have favored hospitals in their marginal recurrent resource allocation decisions and have focused on primary care in their capital investment decisions, which are largely donor-determined. Underfunding of the recurrent resource require- ments of these projects may in part be the result of the unwillingness of health ministries to adjust their marginal recurrent resource allocation decisions from an earlier hospital bias. Kenya provides an example of public sector health authorities' unwillingness or inability to shift recur- rent resources in favor of primary care investments. A comparison of Ministry of Health recurrent budgets with actual expenditures from 1982 to 1989 reveals that for curative services (mostly hospitals), annual budgets were overspent by an average of 7 percent and actual allocations to preventive and promotive programs and rural health services were 17 percent and 19 percent less, respectively, than the amounts budgeted. This evidence of actual resource allocation runs counter to the ministry's stated policy, as reflected in MOH budgets, of giving priority to rural, preventive, and promotive health services (World Bank 1991). A similar -pattern existed in Jamaica from 1983-84 to 1986-87; expenditures on hospital services were consistently above the amount budgeted (by an average of 16 percent for these four years), whereas primary health care expenditure averaged slightly less than the amount budgeted (Kutzin 1989). Hospital investment and recurrent expenditures are linked closely but with substantial lags. Present recurrent expenditures on hospitals result Patterns of Hospital Resource Use 23 from the cumulative hospital capital stock produced in the 1960s and 1970s and in many cases, especially in Africa, during an earlier colonial era. Consistent procedures are needed to assist planners in projecting the recurrent costs of project proposals and then incorporating these projec- tions into alternative budget scenarios. Ultimately, a careful project analysis by component and expenditure category needs to be made. Recurrent-capital cost (RCC) ratios provide a convenient alternative, however, prior to a more detailed analysis. Projections of recurrent expenditures based on RCC ratios are not reliable for a detailed analysis of the recurrent cost implications of specific investments because the relation between variable costs, fixed costs, and outputs that underlies the ratios may not be stable across countries, projects, and levels of output. Nevertheless, RCC ratios can be a valuable method of making quick first estimates of the recurrent costs associated with a contem- plated investment strategy. It is instructive to compare the recurrent cost ratios reconstructed from a number of more detailed sources. Using accounting-based analyses of existing hospitals or analyses of new pro- jects to estimate recurrent costs, the studies summarized in table 2-3 show RCC ratios varying from 0.10 to 0.40 and averaging roughly 0.20. Some of the cross-country variation results from methodological differ- ences described in the "Comments" column of the table (for example, inclusion of depreciation in recurrent costs), and some probably results from differences in budgeting conventions. Despite what may be char- acterized as these artificial sources of cost variation, most of the esti- mated RCC ratios in table 2-3 are dustered around the mean value of 0.20. These ratios are substantially lower than for projects that are less capital intensive, such as primary health care outreach programs, which can entail ratios in excess of 0.5 (Heller 1979; Over 1981). Allocation of Resources by Level of Hospital The scope and complexity of hospital services vary greatly within coun- tries. To sharpen the discussion it is necessary to categorize hospitals by the level of services offered. Definition of hospital levels. Differences in case mix, technical capacity, and skills differentiate hospital levels. These differences may also imply different sizes of facilities, roughly measured by the number of opera- tional beds, upper-level facilities often having a larger size. Bed size is not the definitive difference between levels of facilities, however, and there are many examples of facilities in which bed sizes run counter to the expectation of size based on level. It is difficult to distinguish between levels of facilities when one set of criteria is used for all countries. For convenience, three levels of general hospital facilities are distinguished here, not on the basis of bed size but on the basis of service characteristics. 24 Public Hospitals in Developing Countries Table 2-3. Recurrent Cost Ratios for Selected Hospital Projects Level of RCC Country hospital Year ratio Source Comments China Central 1986 0.22 Barnum Comparison of Provincial 0.22 1989 reported operating District 0.40 costs and replacement cost. Indonesia Central 1986 0.16 Barnum Based on Provincial 0.12 1987 accounting District 0.12 analyses for existing hospitals compared with new project costs. Jamaica District 1989 0.16-0.18 Kutzin Based on 1989 accounting analysis for new project. Kenya Unspecified c. 1970 0.10-0.30 Heller 1979 Malawi District 1988 0.14 Mills Based on 1991 accounting analyses for existing hospital compared with replacement cost. Malaysia Provincial 1971 0.18 Heller District 0.11-0.30 1975 Papua Central 1988 0.27 jsI 1990 Comparison of New Base 0.27-0.40 recurrent costs Guinea Provincial 0.22-0.33 (including depreciation) and estimated capital replacement cost. Rwanda Central 1987 0.18 Based on studies Provincial 0.13 of similar facilities District 0.14 plus data gathered locally. Uganda Unspecified 1970 0.25-0.30 Dunlop 1973 Zambia Central 1981 0.30-0.40 MOH accounts Provincial 0.16 compared with estimated capital replacement cost. Source: World Bank sector reviews and appraisal reports, except as noted in table. Patternis of Hospital Resource Use 25 The classification fits a pyramidal conception of a health service referral system, ranging from level I, the most technically complex, to level III, the most basic.3 The levels can be described as follows: Level 1: These hospitals have the most specialized staff and technical equipment. The clinical services are highly differentiated by function, including, for example, cardiology and specialized imaging units. More than ten clinical specialties are not unusual in large referral centers. The staff-to-bed ratios are the highest in relation to other hospitals. Hospitals at this level are often, but not exclusively, associ- ated with a medical school. As befitting their location at the top of the referral pyramid, level I hospitals are also called "central"- or "terti- ary"-level hospitals. Bed size, as suggested above, varies; depending on the size of the population to be covered, it ranges from 300 to more than 1,500 beds. Level II: Although lacking the most technical services available in level I hospitals, level H hospitals are also highly differentiated by function, with five to ten clinical specialties. Staff-to-bed ratios are intermediate between the highest and lowest level of hospitals. Level II hospitals are occasionally associated with medical schools. Bed size ranges from 200 to 800 beds. Level HI hospitals are often referred to as "provincial" hospitals. Level III: Level m hospitals have many fewer specialists than the higher-level hospitals. The availability of skill levels within this cate- gory is varied. Some entry-level hospitals may have specialists in internal medicine, obstetrics-gynecology, pediatrics, surgery, or radi- ology, whereas others may have only general practitioners. Limited laboratory services are available for general but not specialized patho- logical analysis. Staff-to-bed ratios are low in relation to the upper- level hospitals. Level Ill hospitals are often referred to as "district" or "first-level referral" hospitals. These definitions are by necessity very approximate and meant only to be suggestive. The actual differences between referral levels depend greatly on the resources available, the extent of health sector develop- ment, and the standards of service within a given health system. Addi- tional levels could conceivably be distinguished. Level III is particularly large and, for example, includes both class "C" and class "D" hospitals in Indonesia, "county" and "township" hospitals in China, and district hospitals in Malawi. Hospital level and resource use. Higher-level (level I and II) hospitals absorb a large share of total public expenditure on hospitals, despite the small number of these facilities in comparison with the number of level 26 Public Hospitals in Devoelopinig Coun1tries III hospitals in most countries. The greater case mix complexity in higher-level hospitals and their more intensive input use commonly translate into much higher operating costs per unit (see chapter 3) than at lower levels. Hospitals that have a teaching function also tend to have higher operating costs. This difference is expected on the basis of func- tion, and thus it is expected that level I and II hospitals will absorb a greater proportion of total hospital resources than simply their numbers would indicate. The extent to which resources are concentrated in these facilities, however, often goes beyond that which might be justified in order to fulfill their tertiary functions. In Zambia, for instance, the three large central hospitals use 30 percent of Ministry of Health resources and an estimated 45 percent of total MOH hospital resources, leaving the remaining 55 percent to cover thirty-nine lower-level hospitals (World Bank 1984). Similarly in Indonesia (Barnum 1987), the nineteen largest public hospitals (out of more than three hundred) used 53 percent of total recurrent hospital expenditure in 1985. In Kenya (World Bank 1991), Kenyatta National Hospital used almost 20 percent of recurrent MOH expenditure for curative services in 1986-87. Belize City Hospital in Belize (Raymond and others 1987) used 69 percent of recurrent MOH hospital expenditure in 1985. Finally, in Zimbabwe (Hecht 1992), 45 percent of MOH recurrent hospital expenditure in 1987 was for four central hospitals. This concentration of resources may have undesirable implications for overall sectoral resource use and equity. In each of these cases a relatively modest proportion of total hospital expenditures remained to finance the services of district hospitals, which are intended to serve as first-level referral facilities for most of the population. Excessive centralization of resources in a small number of tertiary hospitals is not consistent with the performance of crucial first-line functions by district hospitals, be- cause these smaller hospitals are relatively starved of inputs. Such centralization may result in poor-quality services at these first-level referral centers, possibly causing patients to bypass them in favor of the tertiary hospital. The end result is inefficiency in the overall network of health services. The importance of maintaining a reliable supply of services with an acceptable level of quality to enable district hospitals to perform their first-level referral role is examined in greater detail in chapter 6. Distribution of Resources within Hospitals The distribution of hospital expenditures by line item category reflects the input mix used in the production of services. Depending on manage- ment incentives, hospital input mix may be determined by many influ- ences in addition to input prices, including centrally determined Patterns of Hospital Resource Use 27 personnel staffing quotas and nationally acceptable medical practices. In any case, relative prices will strongly influence the share of expendi- ture on different inputs. In more industrialized countries, labor inputs (wages and salaries) commonly absorb a high proportion of operating expenditures, because wages and salaries tend to be high in relation to nonlabor inputs (Mills 1987). In contrast, in developing countries salaries and wages tend to be lower in relation to nonlabor inputs, and thus it might be expected that labor cost would be a smaller proportion of total expenditures. This expectation could be disappointed to the extent that economic incentives lead to substitution of equipment and other non- labor inputs for labor in high-wage countries and the reverse in low- wage countries, but accepted standards of care and medical procedures provide limits on this substitution. Data on labor shares for a selection of hospitals in a cross-section of countries support the expectation that the share of labor increases with income, at least up to the middle-income levels indicated in figure 2-4, in which the percentage of labor in hospital recurrent costs is compared with GNP per capita.4 Limits on the substitution of nonlabor inputs are also created by rigidities in employment policies and the setting of wage rates. The share of total hospital expenditure devoted to personnel has increased in many countries as a result of a decreased availability of public sector revenues and an inability or unwillingness to reduce the size of hospital staff, possibly because of civil service constraints on the hiring and firing of staff. Even though in low-income countries the share of total expenditures absorbed by labor tends to be less, labor remains an important input. The data in table 2-4 show that in African countries labor costs are 33 to 79 percent of total expenditures. In Indonesian hospitals, personnel costs are about 40 percent of total expenditures, and in Jamaica and Belize, personnel costs range from about 50 percent to 74 percent of total hospital expenditures. In China labor costs are 23 to 26 percent of total expenditures. In part, this unusually low personnel share may be caused by the exclusion of the value of nonmonetary or budgeted benefits from hospital accounts. Still, the low personnel share also results from a deliberate policy to emphasize nonlabor inputs, especially pharmaceuticals, in inpatient and outpatient hospital services. This latter factor is related to China's method of financing hospitals through user charges; drugs are priced at profitable levels, thus encouraging their use by providers. Largely be- cause of this situation, drug fees are the leading source of hospital revenues. Drugs and medical supplies vary greatly as a percentage of total hospital expenditures. These are often largely acquired in intemational markets with foreign exchange and at intemationally determined prices. 28 Public Hospitals in Developing Countries Figure 2-4. Relation between the Share of Labor in Hospital Recurrent Costs and per Capita GNP, 1978-90 Percentage of labor in hospital recurrent costs 80 * Nigeria 70 - 0 Colombia Papua New Guinea * St. Lucia Botswana * * Tunisia * Belize * Zambia 60 - * Ethiopia Morocco * * Jamaica 50 - * Turkey * Tanzania Benin So Rwanda 40 - * Indonesia * Niger S Malawi 30 - * China 20 l 0 500 1,000 1,500 2,000 Per capita GNP (constant 1980 U.S. dollars) Source: See table 2-4. Thus the expenditure on drugs and medical supplies does not reflect local prices to the same extent as labor costs. If the relative quantities of drugs and labor used per bed-day in countries with low wages were similar to those used in higher-wage countries, the share of expenditures on drugs would be higher. Central ministry pressures (in addition to relative prices) may dictate, however, that a lower quantity of medical supplies be used per unit of labor in low-wage countries. This may be a rational response to prevailing relative prices, providing allocational efficiency (as at point A in figure 2-5, where total output is 1,000 bed- days), or it may be a suboptimal response, forcing an economically Table 2-4. Hospital Recurrent Costs by Line Item Category (percent) Number Level of Total recurrent cost of hospitals Country Year hospital Personniel Drugs Other Total in study Source Belizea 1985 II 56 - 44 100 1 Raymond and others 1987 III 74 13 13 100 6 Beninb 1986 II 43 13 44 100 1 Pangu and others 1987 Botswana 1978 All 64 17 19 100 All MOH BMOH 1979 Colombiab 1978 I-II 71 15 14 100 8 PRIDES 1980 China 1986 I 23 38 40 100 8 Barnum 1989 II 26 43 32 100 11 III 24 50 26 100 7 1 Ethiopia 1983-84 Urban 56 17 27 100 9 Donaldson and Dunlop 1987 Rural 59 17 24 100 11 Indonesia 1985 1 33 - 67 100 2 Barnum 1987 II 42 - 58 100 15 III 39 - 61 100 296 Jamaica 1985-86 1 55 10 36 100 2 Kutzin 1989 II 49 12 39 100 2 III 55 11 34 100 1 MalawiC 1987-88 III 33 30 37 100 6 Mills 1991 Morocco 1987 All 54 24 2 100 All public Bennis and others 1990 Niger 1986 I 38 34 28 100 1 Wong 1989 Nigeria 1986 III 79 16 5 100 2 Papua New Guinea 1988 I 73 9 18 100 1 js 1990 11 63 9 28 100 4 III 64 9 27 100 8 (Table continues on thefollowing page.) Table 2-4 (continued) Number Level of Total recurrent cost of hospitals Country Year hospital Personnel Drugs Other Total in study Source Rwanda 1987 I 43 - 57 100 1 Shepard 1988 St. Lucia 1986 11 69 8 23 100 1 Russell, Gwynne, and Trisolini 1988 Tanzania 1979 All 46 35 19 100 All MOiH Dunlop 1984 Tunisiad 1990 1 65 12 23 100 22 Turkeyb 1987 I 50 30 20 100 3 Uganda 1968 All 63 14 23 100 All MOH Dunlop 1973 Zambia 1981 I 60 - 40 100 3 11, III 63 - 37 100 All MOH a. "Drugs" includes medical and nonmedical supplies. b. "Drugs" includes all medical supplies. c. Total amount includes some nonhospital district health expenditures. d. Figures reflect amount budgeted for a combination of teaching hospitals and specialized institutes. - Not available. Source: World Bank sector reviews and appraisal reports, except as noted in table. Patterns of Hospital Resource Use 31 inefficient solution (as at point B in figure 2-5, where output is 800 bed-days even though the budget remains constant). In poorer countries, drug shortages are common and absolute expen- ditures on drugs are limited, so that even though wages are low, drugs as a percentage of total costs may not be high. Many analysts have argued that the shortage of drugs forces technical inefficiency on the hospital sector. This is particularly true if rigid personnel policies make substitu- tion of supplies for personnel difficult. It is the combination of drug and supply shortages (and other manifestations of recurrent cost constraints, such as insufficient maintenance) and rigid personnel policies that cre- ates the technical inefficiency. Improved hospital information systems coupled with management policies and training could lead to greater flexibility by central authorities, hospital administrators, and physician- managers with respect to input substitution. Such reforms would be positive steps toward reducing this source of suboptimal hospital per- formance. The potential gains are limited, however, by operational constraints on managerial choice of inputs that may be unaffected by these reforms (lack of drugs or equipment, for example). One possibility for improving the internal efficiency of hospitals may be to change the mix of labor inputs in the production of hospital Figure 2-5. Underfunding of Supplies Supplies Budget constraint /~~~ 1,000 bed-days B 800 bed-days Labor hours 32 Public Hospitals in Developinig Countries services. Many hospitals in developing countries are organized and staffed based on industrialized country models. The different resource endowments and epidemiological profile of low-income countries sug- gest that such a model may not be appropriate here. Very little informa- fion is available on hospital staff mix in developing countries, and microeconomic-level studies are needed to examine the scope for im- proving the cost-effectiveness of hospital services through the realloca- tion of staff and the creation of new staff categories. This is a sensitive issue because medical (and in some cases, nursing and paramedical) professionals may perceive such reallocation as a threat to the positions they have worked hard to attain. Also, the internal organization of large hospitals (Harris 1977) leads to a bifurcation of labor between hospital support services (for example, laundry, pharmacy, blood bank) and direct medical care. Input mix and labor allocation on the medical side may be viewed as determined by scientific standards. These standards, however, are often determined in industrial countries. The medical input mix should not be viewed as fixed (Harris 1977). Training and opera- tional research can improve staff flexibility and increase medical decisionmakers' awareness of substitution possibilities. Consideration of alternative staffing patterns is particularly critical for those countries that are suffering severe shortages of skilled staff. Some substitution examples follow: * Substitution of nonphysician- for physician-managers. Physicians in hospitals commonly perform management functions that could be more effectively and cheaply filled by trained managers who are not physicians. The use of professional hospital managers is common in industrial countries but rare in rniddle- and low-income countries. The principal reason for this is the strong tradition in the medical profes- sion of physician control of hospitals. In many countries the lack of appropriate training and internship programs is a severe constraint on the development of a cadre of career hospital managers. * Substitution of nursing for physician time. Wider use of nursing staff could reduce the use of physician time in the hospital. Monitoring of patient condition, basic diagnosis, obstetrics, and simple curative intervention for trauma are possible areas in which nursing training and activities could be extended. * Substitution of clerical stafffor registered nurse time. In some countries, registered nurses (RNs) are widely used for clerical tasks and in some cases (primarily in Africa) for nontechnical nursing activities. Many of the tasks performed by RNs there could be adequately carried out by less-well-trained persons. Clerks could perform some clerical func- tions, and assistant nurses could perform some patient care and other Patterns of Hospital Resource Use 33 ward-related activities. The use of alternative staff could free RNs to focus their efforts on activities that make the best use of their skills. Distributional Equity of Hospital Use With a cross-country mean of about 60 percent, hospitals absorb a substantial percentage of public recurrent health expenditures. The share of health resources devoted to hospital services is believed to be inversely related to the equity of overall health service provision; the greater the share devoted to hospitals, the less is left for primary care programs and facilities that have the potential to provide relatively low-cost basic curative and preventive services to a broadly dispersed population. Thus, within the health sector, the share of resources ab- sorbed by hospitals has implications for the equity of the service delivery system. The characteristics of hospital services themselves may have a pro- found effect on equity in many poorer countries in addition to the effect the share of government health resources devoted to hospitals may have on the availability of primary care services. Hospitals use a large part of the total public sector health budget but can provide benefits to relatively few. Even with existing large expenditures only a limited number of beds can be provided, and inpatient service remains necessarily restricted to a small number of people. Hospital bed ratios per 1,000 population in low-income countries are commonly less than two, in contrast to means of more than four beds per 1,000 in middle-income and more than eight beds per 1,000 in high-income countries (see figure 2-6). In this section we describe the distribution of hospital expenditure and use across income groups, geographic location, age, and disease categories. In addition, we assess implications of projected demographic and epidemi- ological changes for the distribution of hospital services among popula- tion subgroups. Hospital Use across Income Groups Little recent research has been conducted on the income characteristics of hospital users. However, the analyses of individual hospitals in sev- eral Asian countries suggest that hospitals are used unequally by differ- ent socioeconomic groups within designated catchment areas. A survey of inpatient income from a sample of seventeen hospitals in Malaysia showed that only 21 percent of hospital inpatients were from households below a specified poverty criterion compared with 39 percent of the general population (Heller 1975). Upper-income groups used a substan- tially greater share of total inpatient services than would be expected from their proportion of the population. Using 1978 data from the Bicol 34 Public Hospitals in Developing Countries Figure 2-6. Distribution of Hospital Beds per 1,000 Population in 163 Countries, Most Recent Year, 1970-89 Percentage of countries from each group 100 80 60 40 20 . . .... 0 Less than 2 2-4 4-6 6-8 More than 8 Hospital beds per 1,000 population Low income (47) l.Middle income (78) High income (38) Mean = 1.45 Mean = 4.12 Mean = 8.60 Source: World Bank data. region of the Philippines to examine the distribution of public health subsidies across income groups, Ching (1986) found that, for public hospitals, the per capita subsidies to the poorest three quintiles of the population were below the regional mean, whereas those to the wealth- iest two quintiles were above the regional mean. In fact, per capita subsidies were greatest to families in the richest quintile. A household survey in Indonesia revealed that households in the lower 40 percent of per capita income distribution accounted for only 16 percent of hospital service use, whereas households in the upper 30 percent accounted for 63 percent (Meesok 1984). Hospital-based surveys conducted in 1985 in the Nusa Tenggara Barat province of Indonesia had similar findings. Patterns of Hospital Resource Use 35 Patients from the richest 9 percent of the province's income distribution accounted for 32 percent of inpatients and 56 percent of outpatients. Conversely, the poorest 55 percent of the population accounted for only 32 percent of inpatients and 18 percent of hospital outpatients (Gish, Malik, and Sudharto 1988). Thus, persons in the highest income groups benefited disproportionately from Indonesia's subsidies to public hos- pitals, with a greater degree of regressivity involved in the use of hospital outpatient care. A study of the characteristics of patients at one African tertiary hospi- tal had results similar to these Asian studies. Surveys of patients at Niamey National Hospital in Niger showed that inpatients had a median income that was comparable to or slightly higher than other urban residents (who had, in turn, higher incomes than rural residents), and outpatients had a higher median income than inpatients (Weaver, Handou, and Mohamed 1990b). The pattern suggested by this and the above studies is that government subsidies to public hospitals are not well targeted to the poor and that this regressivity is magnified for hospital outpatient services, perhaps because of their more discretionary nature. Not all the available information (particularly studies from Latin American countries), however, confirms inequitable use across income groups. Selowsky (1979) used household survey data in conjunction with public sector and social security expenditures on health services to analyze the distribution of public subsidies for health across income groups, facilities, and regions in Colombia in 1974. He found that, in public sector hospitals, the lowest-income quintile received the greatest subsidy for inpatient care and that this share declined monotorncally for higher-income groups. McGreevey (1990) cites information for five South American countries (Argentina [1980], Costa Rica [19821, Chile [1982], the Dominican Republic [1980], Uruguay [1982]) that demonstrates that the benefits from social security expenditures on health, much of which is used for public hospital services, are inversely related to income. In those few countries in which social security covers most of the population, as in Brazil, the overall distributional effect of public hospital services is to improve equity. One possible explanation for the better equity effects of public hospital subsidies in Latin American countries is the existence of a strong modern private sector serving the wealthiest population groups, leaving lower-income groups to use the public (ministry of health or social security) hospitals. Similar studies in other countries are needed, especially ones distin- guishing between the use of public and the use of private hospital services by socioeconomic group. When there are perceived quality differences between public and private services, either material differ- ences related to health outcome or differences in amenities, high-income 36 Public Hospitals in Developing Countries groups will favor private services. In Sri Lanka, for example, the value of per capita use of public services by low-income households far ex- ceeds the expenditure by such households on private care, whereas the value of government services received by the highest-income group is about one-half that group's expenditure on private medical care (Al- ailima and Mohideen 1984). Geographical Distribution Hospitals are located primarily in urban areas, and even when they are intended to provide a referral service for a broad geographical popula- tion base they serve in actuality a disproportionately urban clientele. Because urban populations generally have higher incomes than those of rural areas, the urban location of most hospitals affects income equity as well as geographic equity. The urban bias provided by hospital services extends even to China, which has decreased mortality in rural areas during the last four decades and is widely perceived to have focused on equity. Other factors, such as improved nutrition and education, have probably been at the root of China's improved health status in this period. In a study of the distribution of health care resources, Prescott and Jamison (1985) noted that the fall in mortality has not been achieved through a more equitable distribution of health resources. There is a wide variation in the availability and distribution of hospital beds, health workers, and health expenditure among Chinese provinces. They found that the distribution of health resources is strongly related to the degree of urbanization in the province and to urban income. In their study, a 1.0 percent increase in urban income is associated with a 1.5 percent increase in the ratio of medical doctors to population, a 2.5 percent increase in the general hospital bed ratio, and a 3.0 percent increase in recurrent expen- diture. These increases are also likely correlated with the extent of health insurance coverage, which is associated with greater health care use and expenditure, as well as with greater income. Similar inequalities in the availability of hospital beds and in the distribution of hospital expenditures are documented in health sector assessments of other countries. Evidence from a few representative studies can be cited. The number of hospital beds per thousand popula- tion in Indonesia varies from 0.2 to 1.2 across provinces, and the variation again is related to levels of urbanization and income (Barnum 1987); the income elasticity of hospital beds is 1.0 (Prescott 1991) and, given that hospital beds located in more heavily urbanized and high-income areas tend to be staffed and equipped more intensively, the income elasticity of hospital expenditure is even greater.5 In Malaysia (Heller 1975) and Papua New Guinea (Baker 1977), hospital expenditure per capita varied more than twentyfold across provinces. In Brazil, expenditure per capita Patterns of Hospital Resource Use 37 by the social security health system in the lower-income northeastern region in 1986 was less than half of those of the higher-income states in the southeast (World Bank 1988), and the availability per capita of hospital beds in the north and northeast was about half of that in the south and southeast in 1984 (Briscoe 1990). In Colombia the average government health subsidy per urban household in 1974 was more than twice as large as that received by rural households (Selowsky 1979). In many African countries probably at least equal variation exists because of the predominantly urban location of hospitals and the largely rural, dispersed populations. The above examples are not intended to constitute an argument for establishing hospitals in rural areas to meet equity concerns but to highlight the importance of improving the reliability of lower-level facilities that serve more widely dispersed populations. The examples also draw attention to the importance of improving transport and, more generally, the referral network to widen access to hospital services. The "bias" of hospital services toward urban dwellers is to be expected from the nature of hospitals as institutions with high fixed costs. It would not make sense to establish large hospitals in sparsely populated areas or rural areas with a widely dispersed population; a network of low-cost primary care that included smaller district hospitals would seem to be more appropriate for these areas. Other risks are associated with using hospital investments in poor geographic areas as a means of improving equity, such as the creation of a two-tiered medical system, in which poor quality and stigma are associated with the facilities serving poor com- munities. It is logical for hospitals to be located in areas with a concen- tration of population high enough to keep the relatively large amount of resources (staff, equipment, supplies) occupied. Ensuring that those not living near these areas have access to hospital care when needed is the function of the referral system. In concept, referral within the health system is supposed to provide a wider distribution and appropriate use of health services, entry being gained at the bottom of the referral pyramid through primary care facilities and other providers. In practice, self-referral-often resulting from the perception or fact of poor quality primary care services because of a lack of drugs or personnel-and poor administrative, transport, and communications linkages between levels of care, frequently defeat the intended purpose of the referral system. Many referral studies have revealed that hospital entry is a matter of proximity, rather than the appropriateness of the classification, and that most inpatients in upper- level hospitals in the most urbanized areas are admitted directly. This situation is to be expected in the absence of barriers to easy admission, when, for many urban dwellers, the nearby hospital is the closest source of primary care. A sample of hospitals in Indonesia revealed that more 38 Public Hospitals in Developing Countries than 50 percent of inpatients come from less than 5 kilometers away from the hospital and more than 80 percent of outpatients come from less than 1 kilometer away (Barnum 1987). Cumper, WaIlker, and MacCormack (1985) concluded that much of the variation in hospital admission rates across census constituencies in Jamaica (from 13 to 92 per 1,000) could be explained by hospital location. The referral system is discussed in greater detail in chapter 6. Distribution of Hospital Use by Age Both the overall level of demand for hospital services and the specific services required are affected by the age and disease profiles of the population. As the economic development of a country proceeds, both profiles change in what is commonly characterized as demographic and epidemiological transitions. The demographic transition refers to the change in the population age structure that accompanies the fall in fertility and mortality as economic and social development occur. The epidemiological transition refers to the changes in age-specific death and illness rates that result in reduced infant and child mortality and longer expected life span. Interaction of these two transitions results in an aging population that is characterized by a lower prevalence of infectious disease and a higher prevalence of chronic disease (Jamison and Mosley 1990). The implications of this transition for hospital resource use can be derived from the experience of specific countries. Hospital inpatient services are used primarily by adults, both because they comprise the largest proportion of the population and because children, after they survive their first year of life, are much less likely to need hospitalization (for epidemiological reasons) than they will in later life. For example, according to samples from hospitals in the countries concerned, adults comprise 88 percent of hospital admissions in county hospitals in China (Over and others, 1992); 87 percent in Papua New Guinea (Jsi, 1990); 84 percent in Belize (Raymond and others 1987); and approximately 70 percent in Malawi (Mills 1991), Niger (Wong 1989), and Uganda (Over and others, 1992). The lower percentages of adult use of hospitals in the African countries reflect the smaller percentage of population that is more than fifteen years of age. Significantly, however, it also reflects the reduction in child mortality and the increase in the relative importance of chronic diseases in China and Papua New Guinea during the last twenty years. This pattern of use is apparent in more detail in figures 2-7a and 2-7b, which present comparisons of the age distribution of the population with the age distribution of hospital admissions in Jamaica and the Republic of Korea.6 The figures indicate that, in both countries, the distribution of hospital admissions by age corresponds closely with the Patterns of Hospital Resource Use 39 Figure 2-7a. Distribution of Population and Admissions by Age Group, Jamaica, 1985 Percentage of total 60 56 50- 40- 30- 24 20- 2< n 110 12lW l i i1 X 10~~~~~~~~~~~~~21 6 98 60 36 0 0 1-4 5-14 15-44 45-64 65+ Age Population Admnissions Source: GOJMOH 1987. Figure 2-7b. Distribution of Population and Admissions by Age Group, Korea, 1986 Percentage of total 60 54 50 40 30 20 1 7 16 8 10 7 6 7 0 0 1-4 5-14 15-44 45-64 65+ Age Population Admnissions Source: FKMIS 1987. 40 Public Hospitals in Developing Countries age distribution. Exceptions are the group age five through fourteen, which makes least use of hospitals compared with that group's share of the population, and infants (less than one year), who make most use of hospitals in relation to their population share. The elderly also tend to use hospitals disproportionately, as would be expected. The wealth of data available on insured patients from Korea provides a better look at inpatient resource use by age group. In figure 2-8 the population distribution of the insured is compared with the age distri- bution of inpatient expenditure. In Korea, most inpatient care is pro- vided by private hospitals that are financed through a combination of insurance reimbursement, copayments, and user fees. Therefore, expen- diture data provide a good proxy for measuring resource use. Conclu- Figure 2-8. Population and Inpatient Cost for Insured Patients by Age Group, Korea, 1986 Percentage of total 60 50 4 40 30 2 20 17 10 9 8 4 5 ..... 0 0 1-4 5-14 15-44 45-64 65+ Age Population Z Cost Source: FKMIS 1987. Patterns of Hospital Resource Use 41 sions drawn from this figure are similar to those from figures 2-7a and 2-7b, with the exception of the group age fifteen through forty-four, which uses slightly less than its population share of inpatient resources, though admissions for this group exceed its population share. This suggests that the average cost per case for young adults is low in relation to the mean across all age groups. This conclusion is supported by figures 2-9a and 2-9b, which depict mean expenditures per admission and per patient-day for each age group in relation to their respective overall means. The results confirm what would be expected: older adults and the elderly are the most expensive to treat in the hospital. The implication is that as a country's population ages, if there is no significant change in the delivery of health care services, hospital costs will rise. Diseases and Conditions Treated in Hospitals Data on hospital admissions from six countries are summarized in table 2-5 by the primary headings of the International Classification of Dis- eases (ICD). Conditions relating to pregnancy and childbirth were the leading cause of admissions in five of the six countries. This is as expected. A country's policy or the population's preferences regarding hospital deliveries, however, are also important factors. Trauma care related to accidents, injuries, and poisonings was among the top five causes of admission in each country, reflecting both the relatively fre- quent nature of serious accidents across countries at various levels of development and the important role of hospitals in providing trauma care. Other leading causes of admission are infectious and parasitic diseases, respiratory diseases, and diseases of the digestive and genito- urinary systems. The causes of admission to hospitals can be reorganized into groups to assess the broad epidemiological composition of inpatient use. Table 2-6 presents such a reorganization into five inclusive groups. The table provides a snapshot of several countries at different stages with respect to the epidemiological transition. The relative sizes of the communicable disease and the chronic and noncommunicable disease categories are of particular interest. The relative share of communicable diseases is great- est in those countries (Nigeria and Malawi) with the highest infant mortality rates and the shortest average life span (see table A2-1). Con- versely, the countries with the lowest fMR and longest average life span (Jamaica and Korea) have the greatest share of chronic and noncommu- nicable admissions. These statistics are not simply a function of a country's level of development, as measured by per capita GNP; Oman has the highest per capita GNP, followed by Korea, Belize, Jamaica, Nigeria, and Malawi. Communicable diseases are, in general, more amenable to broad-based primary prevention efforts than are noncom- 42 Public Hospitals in Developing Countries Figure 2-9a. Cost per Admission in Relation to the Population Mean, Korea, 1986 Index value 0.6 0.4 - 0.2 - 0 -0.2- -0.4- 0 1-4 5-14 1544 45-64 65+ Age Source: FKMIS 1987. Figure 2-9b. Cost per Patient-Day in Relation to the Population Mean, Korea, 1986 Index value 0.6 0.4 0.2 0 -0.2 - -.4 0 1-4 5-14 15-44 45-64 65+ Age Source: FKvI4S 1987. Table 2-5. Ten Leading Causes of Admission to Hospitals, Selected Countries Rank Belize (1985)a Jamaica (1985)b Korea (1986)c Malawi (1986)d Nigeria (1984) Oman (1984)e I Pregnancy/ Pregnancy/ Pregnancy/ Pregnancy/ Parasitic/ Pregnancy/ childbirth (40.8%) childbirth (42.9%) childbirth (25.9%) childbirth (29.8%) infectious (31.3%) childbirth (30.2%) 2 Accidents/injuries Accidents/injuries Digestive (15.3%) Parasitic/ Pregnancy/ Parasitic/ (9.8%) (11.4%) infectious (26.9%) childbirth (23.1%) infectious (18.7%) 3 Parasitic/infectious Digestive (6.0%) Respiratory (9.8%) Respiratory (9.1%) Respiratory (9.8%) Respiratory (13.0%) (8.6%) 4 Respiratory (6.8%) Genitourinary Parasitic/infectious Anemias (6.6%) Genitourinary Accidents/injuries (5.9%) (7.0%) (5.8%) (8.1%) 5 Digestive (6.2%) Heart/circulatory Accidents/injuries Accidents/injuries Accidents/injuries Genitourinary (5.9%) (6.9%) (5.0%) (5.3%) (5.8%) 6 Heart/circulatory Respiratory (5.2%) Neoplasms (6.6%) Endocrine/ Digestive (5.0%) Senility/ill-defined (3.7%) metabolic (4.1 %) (4.7%) 7 Perinatal (3.0%) Parasitic/ Genitourinary Genitourinary Nervous/sensory Digestive (4.6%) LQ infectious (4.8%) (5.8%) (3.3%) (3.3%) 8 Genitourinary Neoplasms (3.3%) Heart/circulatory Skin (2.9%) Anemias (3.0%) Heart/circulatory (2.3%) (5.8%) (4.1%) 9 Endocrine/ Endocrine/ Nervous/sensory Nervous/sensory Endocrine/ Nervous/sensory metabolic (2.1 %) metabolic (2.5%) (3.2%) (2.3%) metabolic (2.8%) (2.7%) 10 Skin (2.0%) Skin (1.6%) Perinatal (2.7%) Digestive (2.2%) Skin (2.4%) Skin (1.8%) a. All public hospitals. b. All public acute care hospitals except 200-bed children's hospital and 500-bed university hospital. c. Data on those covered by Industrial Employment Medical Insurance, Medical Insurance for Government Employees and Private School Teachers, and Compulsory Regional and Occupational Medical Insurance (99 percent of the insured and 46 percent of the total population in 1986). d. All public hospital inpatients. e. All MOI I facilities. Source: Raymond and others 1987 (Belize); GOJMOH 1987 (Jamaica); FKMIS 1987 and KMIC 1987 (Korea); Mills 1989 (Malawi); World Bank data (Nigeria and Oman). 44 Public Hospitals in Developing Countries Table 2-6. Distribution of Causes of Hospital Admissions across Major Categories of Conditions, Selected Countries (percent) Chronic and non- Pregnancy Commun- Commun- and icable icable Accidents, Country Year perinatal diseasesa diseasesb injuries Other c Belize 1985 44 15 22 10 9 Jamaica 1985 44 10 30 11 5 Korea 1986 29 17 45 7 3 Malawi 1986 31 36 26 5 3 Nigeria 1984 23 41 22 5 8 Oman 1984 31 32 24 8 5 a. Infectious and parasitic diseases and respiratory diseases. b. Noncommunicable diseases indude neoplasms; endocrine, nutritional, and meta- bolic diseases; anemias; mental disorders; and diseases of the nervous system and sense organs, circulatory system, digestive system, the skin and musculoskeletal system, and the genitourinary system. c. Congenital anomalies and ill-defined conditions. Source: Raymond and others 1987 (Belize); GOJMOH 1987 (Jamaica); FKMIS 1987 and KMIC 1987 (Korea); Mills 1989 (Malawi); World Bank data (Nigeria and Oman). municable diseases, and thus the relative share of these diseases in total hospital admissions in part reflects the effectiveness of a country's primary health care (and specifically, communicable disease control) policies, in addition to a country's underlying demographic and epide- miological characteristics. Countries that have been successful in controlling the spread of com- municable diseases and that have experienced increases in the average length of life of their citizens are faced with a new problem: a greater percentage of persons with chronic and noncommunicable diseases than existed previously. In figures 2-8, 2-9a, and 2-9b we showed that, in Korea, the cost of treating a person over age forty-five was expensive compared with the cost of treating a younger person. In figure 2-10 we again use Korean insurance data to compare the percentage of total admissions by category of condition with the percentage of total costs for the same categories.7 The figure indicates that communicable dis- eases and pregnancy and perinatal conditions are relatively inexpensive admnissions, whereas admissions for accidents and noncommunicable diseases are relatively expensive. The cost data suggest that more than 70 percent of inpatient care resources are used for patients in these latter two categories. The distribution of admissions can be used to portray relative resource use across disease categories if the percentages are adjusted for case mix Patterns of Hospital Resource Use 45 Figure 2-10. Distribution of Admissions and Inpatient Costs by Category of Principal Condition, Korea, 1986 Percentage of total 80 70 63 60- 50 -4.8 40- 30 25 20 17 13 :10 8 1 3 4 0 -] 8 1~ F-- Preg/ Communicable Accidents Noncommun- Other Perinatal icable Category of disease or condition Admissions Costs Source: KMIC 1987. (that is, resource use per case of a particular type). The Korean data on admissions and cost by disease categories were used to develop case mix weights. These weights were applied to the distribution of admissions by disease in selected countries to approximate relative resource use by condition. The estimates are only indicative because they apply results from one country to other countries. The weights were generated by dividing the percentage of total costs for a disease category by the percentage of total admissions for that same category. The percentages of admissions for a country were then multiplied by the appropriate weight for a disease or condition to generate a distribution of inpatient resource use (corrected to sum to 100 percent). The results of this exercise for six countries are presented in table 2-7. 46 Public Hospitals in Developing Countries Although the results in table 2-7 are only indicative, they suggest two important conclusions: (1) in all countries, a greater share of hospital resources is devoted to treating patients with noncommunicable condi- tions and for trauma care than is indicated by data on total admissions, and (2) as the demographic and epidemiological transitions proceed, the demand on hospital resources will increase as the mix of patients be- comes more expensive to treat. The implications of the epidemiological transition for the health sys- tems of Mexico, Brazil, and China have recently been analyzed (Frenk and others 1989; Briscoe 1990; and Bumgarner 1992, respectively), and the findings are consistent with the conclusions reached above. In the Mexico study, Frenk and his colleagues call for a transition in the health system that involves proactive, innovative primary health care measures to meet the changing pattern of service needs. In the Brazil study, Briscoe warns that the demand for individual treatment of "post-transition" diseases will escalate and may divert resources from preventive efforts aimed at these as well as at communicable diseases, which will remain Table 2-7. Distribution of Hospital Costs across Major Categories of Conditions, Based on Case Mix-Adjusted Admissions, Selected Countries (percent) Chronic and non- Pregnancy Commun- commun- and icable icable Accidents, Country Year perinatal diseasesa diseasesb injuries Other c Belize 1985 24 14 34 15 13 Jamaica 1985 23 9 44 17 7 Korea 1986 13 13 61 9 4 Malawi 1986 17 32 39 8 4 Nigeria 1984 12 35 34 8 11 Oman 1984 16 27 37 12 8 Case mix weights 0.46 0.74 1.31 1.30 1.25 Note The percentage distribution of admissions for each country was first multiplied by the appropriate case mix weight. The resulting percentages were then normalized to sum to 100 percent for each country. a. Infectious and parasitic diseases and respiratory diseases. b. Noncommunicable diseases include neoplasms; endocrine, nutritional, and meta- bolic diseases; anemias; mental disorders; and diseases of the nervous system and sense organs, circulatory system, digestive system, the skin and musculoskeletal system, and the genitourinary system. c. Congenital anomalies and ill-defined conditions. Source Raymond and others 1987 (Belize); GOJMOH 1987 (Jamaica); FKMIS 1987 and KMIC 1987 (Korea); Mills 1989 (Malawi); World Bank data (Nigeria and Oman). Patterns ofHospital Resource Use 47 important in many parts of the country. In his study of the health system in China, Bumgarner conservatively estimated that, solely because of the epidemiological transition, the annual rate of growth of per capita health care costs will be 2 percent higher than that of per capita GNP. These studies emphasize the need to allocate resources to chronic disease prevention and control programs immediately as a means to avoid future preventable loss of life and keep hospital treatment costs from reaching completely unmanageable levels. Cost-Effectiveness of Hospital Services This section contains a brief review of the relative cost-effectiveness of hospital and nonhospital health care services. Both hospitals and PHC programs cover multiple and often overlapping activities, and it is difficult to assess the effectiveness of all interventions collectively. "Pri- mary health care" has become a particularly amorphous term, pervasive in the literature but difficult to define in operational terms. The term is nevertheless useful because its connotation of community-level delivery programs contrasts with the connotation of hospital services delivered through large facilities socially detached from the community. Primary health care also connotes prevention rather than cure, although the term does include simple curative care such as oral rehydration for diarrhea and first-level curative contact for other health problems. In chapter 6 we argue that hospitals and PHC should be more integrated, and the services provided by the health sector should be balanced and inter- linked, from lower-level preventive and curative outreach programs to upper-level facilities. The question remains, however, of the appropriate balance of services within the integrated system. In order to provide one dimension of an answer to this question, we discuss below the individual activities carried out by hospitals and nonhospital programs. Itis difficult to discuss the cost-effectiveness of health services without a knowledge of the epidemiological and resource environment because the relative effectiveness of services changes with the context. The dis- cussion below distinguishes loosely between low-resource countries with comparatively high mortality rates (say, lower-middle-income countries and below that have infant mortality rates above fifty) and high-resource countries with low mortality rates. Our interest encom- passes mortality and morbidity for all age groups, but health status has been found to correlate roughly with infant mortality. There are a few exceptions, but generally, high-mortality countries have fewer health resources, and low-mortality countries have relatively greater resources. The exceptions-countries that have fewer resources but have neverthe- less achieved lower levels of infant mortality-are instructive. For exam- ple, China, Sri Lanka, and Costa Rica are low- and lower-middle-income 48 Public Hospitals in Developing Countries countries that have achieved relatively low infant mortality rates through innovative and encompassing PHC programs plus investments in education, nutrition, clean water, and sanitation (Halstead, Walsh, and Warren 1985). If health planners working in poorer countries were asked to rank health interventions by their efficiency in achieving decreased morbidity and mortality, they would almost universally place primary health care, especially basic services delivered through outreach or rural health centers or health posts, near the top and large, urban-based institutional facilities near the bottom, with regard to both cost-effectiveness and equity. This consensus view derives from the fact that health problems targeted by primary health care programs are epidemiologically the most important, especially to low-income groups, in low- and middle- income countries. Table 2-8 contrasts the epidemiological pattern of diseases in Ghana, China, Mexico, and the United States as examples of high- and low-mor- tality countries. Ghana (1979) provides an example of an epidemiological environment typical of many low-income and high-mortality countries. Table 2-8. Total Days of Life Lost by Major Category of Disease in Selected Countries (percent) Ghana China Mexico United States Category (1979) (1985) (1985) (1988) Infections, respiratory 58 25 22 and digestive diseases, malnutrition Chronic and 21 34 19a 39 cardiovascular diseases, malignancies, psychiatric disorders Newborn, pregnancy, 15 9 14 6 gynecological complications Injuries, accidents, 5 27 23 31 homicide, suicide Other 1 5 22 24 Total 100 100 100 100 - Not available. a. Malignancies and cardiovascular only. b. Includes infectious diseases, which represent approximately 5 to 10 percent of total days lost. Source GHAT 1981 (Ghana); People's Republic of China 1986 (China); Hijar-Medina 1990 and Cavazos-Ortega and others 1989 (Mexico); Centers for Disease Control 1989 (United States). Patterns of Hospital Resource Use 49 The leading causes of morbidity in Ghana and other high-mortality, low-resource countries are upper respiratory illness, diarrhea, parasitic diseases, and accidents. The leading causes of mortality are vaccine- preventable diseases, respiratory diseases, malnutrition, diarrhea, and accidents. With the exception of accidents, hospitals do not play a dominant role in reducing lost years of life from these causes. Accumu- lating studies (see, for example, Barnum, Tarantola, and Setiady 1980; Feachem 1986; and Shepard, Brenzel, and Nemeth 1986) demonstrate that preventive measures, such as immunization and prenatal care, and simple curative measures, such as oral rehydration, can be delivered efficiently through rural health post and outreach programs and are only a fraction of the cost of the alternative inpatient care where it is available. The epidemiological picture in China in 1985 illustrates the pattern of diseases in a low-mortality country. After earlier success at reducing mortality from infectious diseases and lowering fertility, chronic dis- eases have emerged as a significant health problem. Accidents and injuries have also become important sources of morbidity and mortality. Increasingly, with a lowering of overall mortality, the pattern of diseases can be expected to resemble that in the industrial countries (the 1988 pattern in the United States of years of life lost is given for reference in the last column). Hospitals play a somewhat larger role in addressing the problems of chronic disease than in treating infectious diseases. Prevention and primary health care programs, however, still have a central role in determining the disease pattern of low-mortality coun- tries. Hospitals do play an essential role in the delivery of a program of coordinated health services and provide an essential backup and credi- bility for primary health care programs in both low- and high-mortality countries. In particular, as we will argue in more detail in chapter 6, more effort to integrate lower-level hospitals could greatly increase the effec- tiveness of outreach and community-based programs. Central-level hos- pitals can also provide technical support for lower-level services and a focus for training of skilled manpower. Nevertheless, as routinely ap- plied, hospital services, especially in upper-level hospitals, are less cost- effective in reducing mortality or morbidity than many alternative uses of health sector resources, as shown below. The cost-effectiveness of an array of alternative health interventions for primary and secondary prevention and treatment is summarized in tables 2-9a and 2-9b. The effects are measured, depending on the study and availability of data, by years of life gained (YLG) from prevention of premature mortality or, if the required additional morbidity data are available, by healthy years of life gained (HYLG) or, if an index of the quality of health status has been constructed, by quality-adjusted life years (QALY). Table 2-9a. Approximate Cost-Effectiveness of Selected Primary and Secondary Prevention Activities in Health (percent of GNP per capita) Low-income, hig&l infant mor-tality countriesa Higli-iticome, low infanit mortality countriesb Cost per Cost per Cost per Cost per ° Intervention discounted YLG discounted HYLG discounted YLG discounted QALY Primary prevention activities Immunization EPI packageC 3 2 Measles (alone) 5 5 Polio (alone) 120 21 Hepatitis B 20 Cholera 16 14 Diarrhea prevention Weaning education 9 8 Breast-feeding 7 6 Prevention of new smoking starts 2 2 Tropical disease vector control Malaria 22 12 Schistosomiasis 46 26 Onchocerciasis 85 Secondary prevention activities Prenatal screening and high-risk delivery (maternal death only) 19 Cervical cancer screening 25 26 Breast cancer screening Physical exam 12 Mammography added 150 Hypertension screening cr Mild hypertension (90-110 mm Hg) 140 Moderate hypertension ( 110 mm Hg) 70 Hypercholesterolemia screening (drugs) 500 Note: All notes appear at the end of table 2-9b. Table 2-9b. Approximate Cost-Effectiveness of Selected Treatment Activities in Health (percent of GNP per capita) Estimated foreign Low-income, high infant mortality countriesa High-inicome, low infatt mortality countriesb cotntent in lower- Cost per Cost per Cost per Cost per middle income Intervention discounted YLG discounted HYLG discounted YLG discounted QALY countryC (proportion) Diarrhea treatment Oral rehydration 5 5 Intravenous therapy 37 33 Tuberculosis treatment > Outpatient (rifampicin) 6 6 Inpatient-outpatient 16 15 Hospital treatment 40 Neonatal intensive care: 1,000-1,499 grams 400d 70 0.3 500-599 grams 3,000d 500 0.5 Cancer treatment Cervix 40 10 0.2 Breast 40 11 0.2 Colon and rectum 140 50 0.2 Lung 1,300 220 0.5 Stomach 2,700 460 0.5 Liver 4,000 660 0.5 Hip replacement 50d 12 0.5 Hemodialysis Hospital 1,000d 300 0.7 Home l,oood 200 0.7 Coronary treatment Pacemaker 60d 12 0.7 Valve replacement 80d 15 0.7 Coronary bypass Severe angina (left vent.) 90d 25 0.7 Moderate angina (2 vessel) 300d 60 0.7 Note: The estimates should be regarded as approximate. The intention is to allow an order of magnitude comparison among broad categories of interven- tions. Local conditions may cause great variance in actual cost-effectiveness across countries. Cost is expressed as a percentage of GNP per capita. The pur- U pose of using the percentage of per capita GNP rather than monetary units is to reduce program costs across countries to roughly comparable units. The U measure is deficient in that it primarily adjusts for labor cost differences among countries but does not account well for differences in foreign supply costs or productivity. The deficiences are offset, however, by the convenience of the measure. Sources for the GNP and exchange rates are various years of the World Bank, World Development Report, and IMF, International Financial Statistics. Many of the sources are reviews of cost-effectiveness rather than primary sources. If several studies gave the cost-effectiveness of an intervention, as was the case for oral rehydration therapy or immunization, an average was taken and outliers were excluded. Basic information in the literature is reported variously as cost per undiscounted or discounted years of life lost, per healthy years of life lost, or per death prevented. Several procedures were used to convert the information to the comparable measures used in the table. Undiscounted results or results reported using a different discount rate were con- verted to discounted units using a 3 percent discount rate. Deaths prevented were converted to years lost using life expectancies for the original time and place of the study and information on average age of death from the Ghana or China data sets summarized in table 2-8. Ratios of years of life lost from death to total healthy years lost in the Ghana study were used to convert years of life lost to healthy years in other studies. a. IMR >50. b. IMR <50. c. The foreign exchange proportion is approximate. The foreign exchange requirements of hospital services are expected to vary with the size and level of development of the country. No studies have specifically examined the foreign exchange content of hospital services by function in low-income coun- (Table continues on thefollowing page.) Table 2-9b (continued) tries. In a study of hospitals in Tanzania, Dunlop (1984) estimated that 40 percent of total government hospital costs entailed foreign exchange expenditures in 1979. This is an average, however; basic services would have a smaller foreign exchange content and more technically complex services would far exceed 40 per- cent because of the need for special skills and training. u, d. Estimated cost per discounted QALY based on results for high-income countries. Source: Barnum, Tarantola, and Setiadi 1980; Robertson 1985; Shepard, Sanoh, and Coffi 1986 (immunization). Feachem 1986; Horton and Claquin 1983; Shepard Brenzel, and Nemeth 1986 (diarrhea). Barnum and Greenberg 1991 (smoking). Barlow and Grobar 1986 (malaria, schistosomiasis). Prost and Prescott 1984 (onchocerciasis). Herz and Measham 1987 (prenatal screening). Barnum and Greenberg 1991 (cervical cancer screening, breast cancer screening). Torrance 1986 (hy pertension screening, coronary treatment, neonatal care, dialysis, hip replacement). Williams 1985 (coronary treatment, neonatal care, dialysis, hip replacement). Barlow 1976 (general hospital treatment). Mills 1985; Mills and Drummond 1987; Drummond 1985 (additional sources). Patterns of Hospital Resource Use 55 The three measures are not strictly comparable. The HYLG measure primarily has been applied to prevention, whereas the QALY measure has been applied to treatment. For any given preventive intervention, an ordering of the magnitude of the measures will, by definition, give HYLG 2 YLG.8 In contrast, for a given treatment, an ordering of the magnitudes will, by definition, give, YLG 2 QALY.9 Thus if the cost per QALY or YLG for a first intervention is less than the cost per HYLG for a second, the cost-effectiveness of the first is evidently greater even though different measures have been used in reporting the effects. In any case, if the cost per unit measure of a first intervention is an order of magni- tude greater than the cost per unit measure of a second (that is, the difference is sufficiently great that it cannot be due to differences in the technical definition of the measures), then the cost-effectiveness of the first can be accepted as a practical conclusion. Costs are measured as a percentage of the GNP per capita (percent GNPN) at the time and in the country of the study. The purpose of using the percentage of per capita GNP rather than monetary units is to facilitate comparisons across countries. Cost as a percentage of GNP gives an intuitively clear measure of the cost of the intervention in terms of the resources used in relation to the productive capacity of the country. The measure adjusts primarily for differences in labor costs and local supply costs between countries, however, and does not account well for differ- ences in foreign supply costs or productivity. Thus, the degree of com- parability of the cost-effectiveness estimates expressed in percent GNPN is limited; differences in technical capacity and in productivity are substantial between countries, and there are important technical inter- ventions such as complex surgery in high-income countries that can be replicated in poorer countries only at a substantially greater cost in termns of per capita GNP because of the need to use imported materials and technical training. The use of common monetary units, such as dollars, is even more problematic because differences in resulting cost estimates may reflect differences in exchange rates and wage rates more than the content of services. Conceptually, the separate components of interven- tion costs could be corrected for price differences across countries, but it would be difficult to do so because of the lack of appropriate indexes across the range of countries and dates in the cost-effectiveness studies surveyed. In recognition of this problem in comparability, tables 2-9a and 2-9b are divided into two sections; the first two columns give the results of studies from low-income countries and the third and fourth columns show the results of studies carried out in high-income countries. The results in table 2-9a, for primary and secondary prevention, respectively, are not greatly affected by this distinction because most of the studies on which the table is based were carried out in developing countries, and 56 Public Hospitals in Developing Countries the foreign exchange costs for these interventions is low. For neonatal intensive care, cancer, hip replacement, hemodialysis, and heart treat- ments, the original studies on which table 2-9b is based were carried out in industrial countries. For these interventions an estimated cost for hospital procedures in percent GNPN for an average lower-middle-in- come country has been computed based on the estimated foreign ex- change component in the last column and calculation of a weighted sum of the foreign and local costs.10 The general order of magnitude of the difference in cost-effectiveness between hospital services and primary health care programs can be established, and the dominance of lower-level interventions, especially prenatal care, diarrhea control, and immunizations is clear from the tables. Looking first at diseases that are of primary importance in high- mortality countries, we see that the cost per year of life gained from the expanded program of immunization (EPI) package or measles vaccina- tion varies from 3 to 5 percent GNIPN, and for diarrhea control using weaning, breast-feeding, or oral rehydration therapy, the cost per year of life gained varies from 5 to 9 percent GNIPN. Obstetrics and neonatal care are of particular importance because of the relatively large share of total hospital resources used for delivery in high-fertility countries. In low-resource countries, routine delivery in hospitals is not cost-effective compared with health center or attended home delivery (see, for instance, the analysis in Barnum and others 1980). Neonatal intensive care at 400 percent GNPN for births of 1,000-1,500 grams is very expensive, and for births of less than 1,000 grams the cost of wide coverage would be prohibitive. Much of the need for neonatal care could be prevented by prenatal care programs, especially with early detection of pregnancy (see the discussion in Rao 1990). A potentially cost-effective use of hospital services is for high-risk deliveries, with an estimated cost per year of life gained of about 20 percent GNPN. Few countries, however, currently have an effective screening program for high-risk deliveries or adequate transportation to provide access, and the ability to mount such a program depends crucially on the develop- ment of the supporting primary health care infrastructure. Cesarean deliveries are not mentioned in the tables but have great implications for the misuse of hospital resources. When performed in conjunction with high-risk screening, cesarean deliveries can be an important component of programs to reduce maternal mortality. Rapid growth in the number of cesarean deliveries in some middle-income developing countries (Bobadilla and Walker 1991), and in industrial countries (OECD 1987), has raised the question of inappropriate use. In selected countries in which medical training and the financial reimbur- sement system provide adverse incentives, the proportion of total deliv- eries that are cesarean has become a potential hazard to maternal health Patterns of Hospital Resource Use 57 and an unwarranted financial burden on the health system. Brazil pro- vides a dramatic example. The cesarean rate (1981) is 17 percent for low-income households and climbs to 58 percent for upper-income households, with an overall average of 31 percent for the country (Saxenian forthcoming). Clinically, less than 15 percent of all deliveries, on average, can benefit from a cesarean section. For routine, normal deliveries, which comprise more than 85 percent of births, cesarean sections are of greater mortality and morbidity risk for the mother than normal vaginal deliv- ery. A World Bank country study (Saxenian forthcoming) estimates the cost of unwarranted cesareans in Brazil to be US$53 million per year. In countries that have experienced the epidemiological transition from high to low infant and child mortality, the cost-effectiveness of hospital services increases, but not dramatically. As noted earlier, increased life expectancy greatly shifts the mix of diseases with which the health system must contend, and problems of adult ill health, especially acci- dents and chronic diseases, become even more important than they already are. Columns three and four of tables 2-9a and 2-9b summarize the cost-effectiveness of a number of health interventions in high- resource countries. Selected hospital services are relatively cost-effective, but preventive measures remain crucially important. Primary prevention programs for smoking are clearly dominant among interventions for noncommunicable diseases, and hepatitis con- trol is also a priority area compared with the application of hospital resources to lung, stomach, or liver cancers. At an estimated cost per year of life saved of 2 percent GNPN, antismoking programs are much more cost-effective than the use of hospital resources to treat the associated lung cancer or heart ailments over the lifetime of the disease. Treatment of lung cancer or heart disease costs 200 percent GNPN per year of life saved and 25 to 60 percent GNPN per QALY in a high-income country and 1,300 percent GNPN per year of life saved and 90 to 300 percent GNPN per QALY in a low- or middle-income country. Similarly, hepatitis B im- munization in a high-prevalence country at 20 percent GNPN per year of life saved is more cost-effective than the 660 percent GNPN per year of life saved from treating the associated primary liver cancer. Secondary prevention programs, such as those for cervical and breast cancer and tuberculosis, can also be cost-effective if the incidence of the targeted diseases is high and sufficient hospital infrastructure is avail- able for follow-up. Several important secondary prevention programs (for example, prenatal care, screening for high-risk delivery, and screen- ing for breast and cervical cancer) require the use of hospital resources for follow-up treatment of those who test positive. In fact, the cost and effects of screening and treatment are intertwined, and in some cases it is difficult to isolate screening and treatment for separate cost-effective- ness evaluation. 58 Public Hospitals in Developing Countries Cervical cancer provides a good example. The low cost per year of life saved from cervical cancer treatment is an effect of the long existence of screening programs in industrial countries. These programs have re- sulted in earlier detection and, if treatment is undertaken, improved survival. Thus, the estimated cost-effectiveness of treatment, which is based on improvement in survival rates, is increased for cervical cancer. As a complement, the secondary prevention program is cost-effective partly because treatment for cervical cancer involves simpler and less costly procedures when the cancer is detected early. The relation of secondary prevention and the appropriate design of a referral system is considered further in chapter 6. Unfortunately, most studies on the cost-effectiveness of hospital ser- vices have been of technical procedures, such as dialysis, that have generated controversy in industrial countries. Most of these procedures are clearly not cost-effective in developing countries and are performed, if they are performed at all, primarily in larger provincial or central facilities. Only a few studies, such as that carried out by Barlow (1976) in Morocco, have addressed the general cost-effectiveness of relatively basic services provided in hospital facilities. Recast in percent GNPN, the cost per year of life gained for hospital services in the Morocco study was 40. This cost, although higher than many nonhospital interventions, is an average of high- and low-cost hospital activities and suggests that more detailed analysis of less technically complex hospital interventions in low-income countries could demonstrate their relative cost-effective- ness. Marginal Cost-Effectiveness of Hospital Resources The marginal cost-effectiveness of a given activity can be defined as the change in the number of discounted healthy life years gained with a change in the expenditure on the activity. The studies cited in the discussion above do not distinguish between marginal and average cost- effectiveness. Allocational efficiency requires that the marginal cost-ef- fectiveness of interventions be equal, and the use of average cost-effectiveness introduces bias into the comparisons. This bias can be especially important in comparing interventions in environments of low mortality and high resources, because the marginal cost-effectiveness of any intervention falls as the incidence of its related disease falls and the level of coverage by the given intervention as well as other interventions increases. In evaluating health care interventions in countries with low levels of resources and high disease rates, the difference between resource allocations based on average and marginal effects is not great. The distinction may be important, however, in resource environments in which the coverage with more basic interventions may be high. Patterns of Hospital Resource Use 59 A study of the optimum use of resources to improve child survival demonstrated that marginal cost-effectiveness can change rapidly as the coverage and use of interventions increases, with the result that the optimum intervention mix changes at alternative resource levels (Bar- num and others 1980). At low-resource levels and with an imR greater than 100, the activities with the highest marginal cost-effectiveness in improving child survival are outreach programs promoting nutrition, breast-feeding, antidiarrheal measures, prenatal care, and attended home delivery. At middle- to upper-resource levels, hospital outpatient programs and inpatient delivery become increasingly cost-effective. Only at upper levels of resources, as the IMR falls well below fifty, does inpatient care become cost-effective, and then only for selected uses. The principle remains sound and can be extended to resources used by adults as well as children and to morbidity as well as mortality. Preventive and primary curative programs should be maintained at all resource levels, but hospital inpatient care becomes increasingly cost-effective as the general mortality level falls and greater health resources become available. If a country were following an optimum allocation strategy as the health sector developed, the marginal cost-effectiveness of any given primary health care intervention, even ones with as high an average cost-effectiveness as the diarrhea prevention activities listed in table 2-9a, would decrease with each additional unit of resources until at some point it fell below inpatient hospital care. The principle is illustrated in figure 2-11. If we follow the two lines depicting the marginal cost-effec- tiveness of diarrheal disease prevention and inpatient care for the case in which a country has initially invested only a limited amount of resources in hospitals, we can see how the required investment in hospitals might change as increasing resources become available. The change in the number of years of life gained with, say, an additional expenditure on diarrhea prevention or hospitals is measured on the y-axis and the level of resources used is given on the x-axis. A low level of resources used for diarrhea prevention results in great marginal cost-effectiveness compared with an equivalent expenditure on inpa- tients. As the resources used for diarrhea prevention activities increase, the marginal cost-effectiveness of the activities decreases until at some point (at high resource levels) diarrhea prevention becomes less cost-ef- fective than an additional unit of inpatient care. In most resource-poor countries, however, the health sector invest- ment program has not followed an optimum path, and the countries have already invested heavily in hospitals even though the level of sectoral development is low. Investments in nonhospital health care programs are vastly more cost-effective in these countries. This situation is depicted in figure 2-12. As diarrhea prevention activities increase, the relative cost-effectiveness of hospitals nevertheless remains low. In these 60 Public Hospitals in Developing Countries Figure 2-11. Change in Marginal Cost-Effectiveness with Low Initial Investment in Hospital Services Marginal cost-effectiveness Diarrhea prevention Health resource level countries it becomes especially important to release public resources for more cost-effective programs by finding ways of diminishing the government's financial responsibility for hospitals. In practice it is often politically or technically difficult actually to reduce recurrent expendi- tures on hospitals and transfer the funds to nonhospital programs. Changes must be made at the margin by altering investment patterns; this strategy can have significant effects in time. An alternative method would be to alter the function of existinghospitals to integrate them more fully into primary care programs. Summary In this chapter we have surveyed the use of hospital resources in devel- oping countries with respect to their distribution within the government health sector, their distribution within and across hospitals, their dis- Patterns of Hospital Resource Use 61 Figure 2-12. Change in Marginal Cost-Effectiveness with High Initial Investment in Hospital Services Marginal cost-effectiveness Diarrhea Inpatient care Health resource level tributional equity, and their cost-effectiveness. We have arrived at the following conclusions: * In nearly all countries, the largest share of public sector health expenditure is for hospitals, regardless of a country's health status and income level. * Within the hospital subsector in many countries, tertiary hospitals use a very large share of public resources in relation to district- level hospitals. The relatively limited resources available for district-level hospitals can be detrimental to service quality and im- pedes the function of district hospitals as the institution of choice for first referral. * The benefits of hospital services are not distributed equitably throughout the population but instead are received disproportionately by residents of urban areas. There is also limited evidence to suggest that hospital services are used less by the poor, with the possible 62 Public Hospitals in Developing Countries exception of hospital services covered by social security in those countries in Latin America where such coverage is high. * Nonhospital interventions are both more cost-effective and more equitable as a means of improving health status for most of the prevalent health conditions in low-income countries. These conclusions are not surprising and are consistent with widely held views of hospital resource use. Nevertheless, the discussion leading to the conclusions is important because it marshals available evidence to sustain the prevailing consensus and underlines the importance of find- ing solutions to the problem of heavy hospital resource use. Within these broad and obvious conclusions the chapter reveals a number of less ob- vious details that have implications for health sector resource planning. Distribution of Resources within the Health Sector The large share of recurrent resources going to hospitals is only indica- tive of the actual priority placed on other health programs or sector strategies. The hospital share does not, in itself, demonstrate a country's level of support for primary health care or the effectiveness of its primary health care strategy. Although exceptionally large expenditures for hos- pitals are competitive with PHC programs, lower-level hospitals can provide substantial primary health care services directly as well as support to nonhospital aspects of PHC programs, as will be brought out in chapter 6. For these reasons a comparison of hospital share with health status indicators, such as the infant mortality rate, related to the goals of primary health care reveals only a modest inverse pattern with many significant exceptions. Thus, an examination of tradeoffs between hos- pital resource use and PHC requires country-specific analysis. The hospital share of recurrent resources can be changed by deliberate government policies affecting either the short-term or long-term alloca- tion. In the short term, the share may change during periods of fiscal difficulties when fewer public resources are available. Some countries have acted to protect primary care programs in the face of declining availability of recurrent resources for the entire sector, whereas others have concentrated a greater share on hospitals. Policymakers should be especially careful during periods of acute fiscal adjustment to protect the recurrent resource needs of basic health care programs. In the long term, although the number of physicians in relation to the population appears to affect the hospital share, a more worthwhile activity than simply limiting physician supply may be to alter the medical curriculum to place a greater emphasis on primary health care and less emphasis on technically complex curative techniques. Perhaps physicians could spend more training time at district hospitals and in Patterns of Hospital Resource Use 63 primary care outreach programs. A by-product of such a change may, in the medium to long term, be a reduction in the hospital share of public sector health expenditure. The long-term balance of recurrent resource use between hospitals and other health programs is obviously affected by past and current investments, but health sector plans often fail to project the implications of recurrent costs. Recurrent-capital cost ratios can be used to generate a rough order-of-magnitude estimate of the recurrent cost implications of current capital expenditures until project-specific financial analysis is available for more accurate projections. Whatever projection methodol- ogy is used, governments should act on the projections to coordinate government and donor investment in the sector to achieve the intended long-term balance of recurrent resource use. Allocation of Resources within the Hospital Subsector Tertiary-level hospitals absorb a large share of total public expenditure on hospitals. Tertiary hospitals are intended for patients with complex conditions, who are more costly to treat. It is likely that many, if not most, of the patients treated in large urban tertiary hospitals could be treated in less costly facilities but are not because the referral system is not functioning effectively and because viable alternative urban treatment centers are not available for the population living relatively close to the hospital. If concentration of funds on tertiary hospitals leads to un- derfunding of first-level referral hospitals, the quality of these low-level facilities is likely to deteriorate, and people who are able will tend to bypass them and seek out the nearest tertiary center. Therefore, excessive concentration of funding on a few facilities will feed on itself and result in an increasingly inefficient and inequitable allocation of hospital re- sources. There appears to be a positive correlation between the share of labor in hospital recurrent expenditures and a country's level of income, though there are several exceptions. This correlation is expected on economic grounds as long as the technical possibilities for substitution between personnel and other inputs, especially drugs and supplies with relatively fixed real prices across countries, are relatively limited. Al- though there may be a greater relative quantity of labor used per unit of output in low-wage countries, it is not enough to cancel out the price effects of imported inputs on the relative expenditure shares. As a result, drugs and medical supplies absorb a larger share of total recurrent expenditures than in high-wage countries. This high-cost, low-produc- tivity environment leads to a recommendation that hospitals in low-in- come countries should not pattern themselves after those in high-income countries; rather, emphasis needs to be placed on research to develop 64 Public Hospitals in Developing Countries productive, alternative, labor-using technologies for delivering hospital services in developing countries. In contrast to the majority of low-income countries that have, as expected, relatively low (40 to 60 percent) personnel shares, some low- income countries have very high labor shares. Examination of the hos- pitals in these countries reveals the cause of this discrepancy: in the face of declining resource availability in the public sector and civil service constraints on reducing the number of employees, there are very few funds available for nonpersonnel inputs after the staff has been paid. The result is extremely low productivity, because the staff then has very few nonlabor inputs available with which to produce hospital services. Greater health sector funding and more flexibility in assigning staff to nonhospital programs within the health sector are needed to increase productivity. Distributional Equity of Hospital Use The use of hospital resources is important for the relative welfare of different population groups distinguished by income, geographical lo- cation, and age. Given that hospitals absorb a large share of health sector resources but can provide services to relatively few persons, these dis- tributional implications are of great importance in setting policy. Sur- prisingly little information is available on this question, and there is a great need for additional research, possibly based on household surveys or comparisons of the characteristics of patients discharged from hospi- tals with wider demographic and income information. The limited in- formation on hospital use by income groups indicates, but not conclusively, that bias in favor of high-income groups exists. Stronger evidence exists of geographical bias and of greater use of hospital outpatient and, to a somewhat lesser extent, inpatient services by urban populations than rural. This evidence suggests that there are distributional biases that favor high-income regions and urban populations. Hospitals are necessary as part of the overall health system, and economic logic dictates that they be located in areas with the highest population. The key equity issue raised by the geographical bias is whether the health needs of rural persons are being met by the services available to them, and whether the existing referral system provides them with access to more complex services when needed. Consideration of the distribution of hospital services by age and disease or condition raises issues of future resource use related to the epidemiological and demographic transitions. Limited data from devel- oping countries indicate that infants and the elderly comprise a greater share of admissions than they do of the population, whereas the five- through-fourteen age group comprises fewer admissions than its popu- Patterns of Hospital Resource Use 65 lation share. The available empirical evidence supports the hypothesis that greater resources are used per admission and patient-day for adults over age forty-five than for younger persons. This suggests that older people use more services and are more expensive to treat. An implication of the demographic transition is that as populations age, if services are organized and delivered as they are currently, hospital costs will rise. This implication is made stronger by the fact that specific disease categories require greater hospital resources. The current leading causes of admission relate to pregnancy and communicable (infectious-para- sitic and respiratory) diseases. These cases, however, have a relatively low average cost per admission, and noncommunicable conditions and accidents consume a far larger share of hospital resources than is indi- cated by the admissions data. It is difficult to disentangle the separate effects of the demographic and epidemiological transitions, but it is clear that as they occur their combined effects will lead to a more costly hospital case mix, further straining available resources, assuming health services continue to be financed and delivered as they are today. World Bank studies from Brazil and China stress the need to reallocate re- sources immediately in favor of chronic disease prevention and control programs in order to avert future high hospital costs that will arise when patients with these conditions present themselves at hospitals. This condusion is applicable (to varying degrees) to many developing coun- tries. Cost-Effectiveness of Hospital Services Understanding the relative cost-effectiveness of hospital services helps to determine the appropriate levels of hospital and nonhospital services that should be provided in an integrated referral system. Because of different resource endowments and relative prices across countries, a ranking of interventions by their relative cost-effectiveness may be dif- ferent in different countries. Despite these differences, however, it is likely that the leading causes of morbidity and mortality in low-income, high-mortality countries are conditions that can be treated or prevented most cost-effectively through nonhospital interventions. In a low-mor- tality country, in which chronic disease is more important, hospital services become relatively more cost-effective, although primary care and prevention remain very important. To recapitulate the conclusions of the chapter: hospital services have a role in providing referral services to support and complement non- hospital health programs, but in resource-poor countries the magnitude and diversity of hospital services to be provided should be limited. In most countries the resources going to hospitals appear to exceed the 66 Public Hospitals in Developing Countries amount required for allocational efficiency. Thus, to reduce the use by hospitals of public sector resources and to make additional resources available for nonhospital programs, it is important to find ways of increasing internal efficiency in hospitals, to find mechanisms to reduce the dependence of hospitals on public finance, and to develop low-cost alternatives to hospitals. The next four chapters address these problems. Patterns of Hospital Resource Use 67 Appendix 2A. Resource Use The table on following pages provides indicators of health resource availability and health status for a cross-section of countries. Information is also provided for total health spending and government spending on hospitals. This latter information is available for only a limited number of countries. Table 2A-1. Selected Health Indicators, Most Recent Estimates Infant Health Hospitals as a Per capita Life mortality spending as a percentage of GNP a expectancy (per 1,000 Health inputs per 1,000 population percentage public health Country (U.S. $) at birtha live births) a Physiciansb Nursesc Hospital bedsd of GNP spendinge Low-income countries Mozambique 81 49 137 0.03 0.17 1.09 4.4 36 Tanzania 120 49 112 0.04 0.18 1.38 3.2 Ethiopia 122 48 133 0.01 0.19 0.30 3.6 49 Somalia 170 48 128 0.06 0.65 1.43 - 70 Malawi 173 47 147 0.09 0.32 1.54 2.7 81 Nepal 175 52 124 0.03 0.21 0.17 1.4 25 LaoP.D.R. 175 49 105 0.73 1.88 0.93 - - Bangladesh 182 51 106 0.15 0.11 0.28 1.7 61 Guinea-Bissau 183 40 147 0.14 0.89 1.86 - - Chad 188 46 127 0.03 0.29 1.31 - - Bhutan 195 48 125 0.10 0.33 0.71 2.0 - Sierra Leone 209 42 149 0.07 0.92 1.21 3.0 - Burundi 218 49 70 0.05 0.23 0.68 - 66 Madagascar 220 51 117 0.10 0.58 2.50 9.2 - The Gambia 236 44 138 0.09 0.46 1.67 - 45 Zaire 244 53 94 0.08 0.56 1.65 - - Nigeria 249 51 100 0.16 1.18 0.73 7.8 - Uganda 250 49 99 0.05 0.49 1.52 2.1 43 Mali 260 48 167 0.04 0.74 0.71 0.8 - Niger 292 45 130 0.03 2.18 0.50 - 30 Burkina Faso 313 48 135 0.02 0.59 0.57 - - Rwanda 320 49 118 0.01 0.23 1.67 3.5 India 347 59 95 0.40 0.59 0.77 4.3 71 China 356 70 30 0.99 0.71 1.98 4.0 61 Haiti 359 55 94 0.14 0.44 0.72 - - Equatorial Guinea 359 46 122 - 0.80 - - - Kenya 364 59 68 0.10 1.55 1.65 2.3 73 Pakistan 365 55 106 0.34 0.20 0.59 3.5 52 Sao Tome and Principe 373 66 71 0.50 3.55 - - - Central African Republic 379 51 100 0.04 0.45 1.55 - - Benin 382 51 112 0.06 0.57 1.13 4.1 Ghana 383 55 86 0.05 0.60 1.57 2.4 Togo 394 54 90 0.11 0.90 1.43 - - Guyana 407 64 53 0.16 1.13 3.33 4.4 Zambia 412 54 76 0.14 1.34 3.53 5.6 Maldives 417 61 73 0.07 1.63 - - - Sudan 428 50 104 0.10 0.79 0.88 6.0 Sri Lanka 433 71 20 0.18 0.78 2.94 2.3 70 Lesotho 437 56 96 0.05 0.26 1.67 2.0 74 Guinea 442 43 140 0.02 0.19 1.53 - - Comoros 456 55 94 0.08 0.45 2.10 - - Indonesia 503 61 64 0.11 0.79 0.55 2.4 37 Mauritania 505 46 123 0.08 0.85 0.78 - - Solomon Islands 577 64 49 0.13 1.69 5.68 Low-income countries with missing GNP data Afghanistan - 42 170 0.16 0.11 0.27 - - Cambodia - 50 121 - 0.73 1.08 - - Liberia - 54 137 0.11 0.73 1.67 - - Myanmar - 61 66 0.27 1.18 0.85 3.2 33 Viet Nam - 66 43 1.06 1.68 3.70 - - (Talile continues on thefollowing page.) Table 2A-1 (continued) Infant Health Hospitals as a Per capita Life mortality spending as a percentage of GNP a expectancy (per 1,000 Health inputs per 1,000 population percentage public health Country (U.S. $) at birtha live births)a P,lysiciattsb Nursesc Hospital beds" of GNP spendinge Middle-income countries Angola 604 45 132 0.06 0.99 2.72 - - Bolivia 628 54 106 0.65 0.44 2.00 2.3 Arab Republic of Egypt 639 60 68 1.30 1.28 2.09 - - Republic of Yemen 650 48 125 0.15 0.58 0.45 6.0 Senegal 654 48 82 0.08 0.49 1.25 3.1 50 Zimbabwe 654 64 45 0.14 1.39 2.01 4.2 54 Kiribati 702 55 59 0.51 4.40 4.80 - - Philippines 714 64 42 0.15 0.37 1.70 2.4 71 Western Samoa 722 66 48 0.28 2.45 4.36 - - Cape Verde 778 66 41 0.19 1.40 2.19 Dominican Republic 781 67 61 0.57 0.83 2.50 - - C6te d'lvoire 788 53 92 - 0.49 1.16 5.4 46 Morocco 878 61 69 0.21 0.95 1.24 1.2 Papua New Guinea 892 54 59 0.16 1.14 4.81 3.8 45 Swaziland 896 56 114 0.05 0.95 3.37 5.5 52 Tonga 907 67 24 0.60 1.83 3.56 - - Honduras 909 65 66 0.66 1.49 1.25 Guatemala 916 63 55 0.46 1.17 1.67 - People's Republic of the Congo 943 54 115 0.12 1.73 4.59 9.4 Syrian Arab Republic 963 66 44 0.77 1.12 1.13 - Vanuatu 981 64 71 0.19 2.18 6.08 - Cameroon 995 57 90 - 0.51 2.50 7.1 Ecuador 1,023 66 61 1.22 1.64 2.50 3.4 Namibia 1,027 57 101 - - - 4.4 Paraguay 1,032 67 32 0.69 1.20 1.67 - Peru 1,061 62 79 0.96 0.99 1.69 - - El Salvador 1,063 63 55 0.35 1.75 2.00 - 62 Colombia 1,212 69 38 0.80 1.52 1.65 4.9 67 Thailand 1,224 66 28 0.16 1.42 1.54 3.8 58 Tunisia 1,261 66 46 0.46 2.73 2.13 4.5 70 Turkey 1,368 66 61 0.73 0.98 2.08 3.0 63 Jamaica 1,400 73 16 0.49 2.50 3.33 4.6 72 Botswana 1,603 67 39 0.14 1.43 2.37 3.3 49 Jordan 1,640 67 52 0.90 0.79 0.94 6.8 75 Fiji 1,675 67 19 0.49 2.36 2.75 3.8 - Belize 1,720 68 46 0.45 2.17 3.32 - - Panama 1,764 72 22 1.00 2.57 3.33 5.6 - Dominica 1,764 75 17 0.32 1.88 4.44 5.3 - Chile 1,771 72 19 0.81 2.70 3.41 6.1 - Costa Rica 1,773 75 17 1.04 2.22 3.41 - Poland 1,774 71 16 2.05 5.35 7.64 3.7 - St. Lucia 1,810 71 20 0.26 1.90 5.07 5.0 - Mongolia 1,824 62 64 9.81 4.72 11.35 - - Grenada 1,891 69 32 0.47 - 5.95 7.0 - Mauritius 2,017 70 21 0.53 1.71 3.33 9.0 - Mexico 2,079 69 40 0.80 1.14 1.25 3.4 58 Argentina 2,103 71 30 2.68 1.19 5.59 7.1 - Malaysia 2,136 70 22 0.52 0.99 2.50 3.5 59 Algeria 2,287 65 69 0.43 3.30 2.63 5.4 - Uruguay 2,448 73 22 1.95 5.29 3.25 6.4 (Table continues on the following page.) Table 2A-1 (continued) Infant Health Hospitals as a Per capita Life mortality spending as a percentage of GNP a expectancy (per 1,000 Health inputs per 1,000 population percentage public health Country (U.S. $) at birtha live births)a Physiciansb Nursesc Hospital bedsd of GNP spendinge Venezuela 2,449 70 35 1.43 2.71 2.60 South Africa 2,470 61 68 - 2.43 Brazil 2,496 66 59 0.93 0.83 5.00 5.6 68 Hungary 2,585 71 16 3.26 5.76 9.17 5.4 - Bulgaria 2,662 72 14 3.63 6.44 11.14 - - Yugoslavia 2,937 72 24 1.82 3.93 5.98 - - Gabon 2,992 53 98 0.36 3.67 1.25 - - Suriname 3,006 67 40 0.79 3.62 8.90 - - Islamic Republic of Iran 3,007 63 90 0.34 0.87 1.52 - - Trinidad and Tobago 3,338 71 15 1.05 3.98 5.00 - - Czechoslovakia 3,453 71 12 3.60 6.91 12.45 - - Seychelles 4,226 70 18 0.46 5.03 4.98 - - Portugal 4,249 75 14 2.42 1.59 5.00 6.4 44 Republic of Korea 4,400 70 23 0.87 1.72 1.68 5.1 33 Oman 5,217 65 36 0.91 2.55 1.81 3.4 Libya 5,308 62 77 1.44 2.85 4.83 - Greece 5,346 77 12 2.85 2.24 6.16 5.3 33 Malta 5,832 73 10 1.14 8.84 10.00 - - Countries believed to be middle income, but with missing GNP data Antigua and Barbuda - 74 20 - 3.07 6.53 Djibouti - 48 117 0.24 1.98 3.60 French Guiana - 73 - 1.40 16.65 12.15 Gibraltar - - 5 - - - Iraq - 63 67 0.55 0.58 1.85 Lebanon - 65 47 1.49 1.59 4.34 Macao - 72 11 1.80 2.35 5.49 Martinique - 76 10 1.42 - 10.06 Montserrat - 71 30 0.42 - - New Caledonia - 69 33 0.66 2.41 - Nicaragua - 64 57 0.67 1.87 2.50 Pacific Islands Trust Territories - 72 20 - - Reunion - 72 13 - 3.62 8.66 Romania - 71 24 1.76 3.65 8.77 3.0 53 St. Kitts and Nevis - 69 38 0.46 7.94 8.65 - - St. Vincent and the Grenadines - 70 25 0.23 1.48 4.99 Highi-incomie countries Saudi Arabia 6,020 64 67 1.35 2.90 1.46 - - Barbados 6,346 75 13 0.89 4.49 8.53 - - Bahrain 6,380 69 33 1.22 2.71 3.31 - - Cyprus 7,040 76 11 1.34 3.67 5.55 - - Ireland 8,713 74 8 1.47 7.16 9.71 7.5 73 Spain 9,329 77 8 3.17 3.87 5.20 6.0 48 Israel 9,787 76 11 2.90 9.37 5.00 - - Singapore 10,352 74 7 0.76 2.96 4.05 - - Hong Kong 10,372 77 7 0.93 4.15 4.89 - - The Bahamas 11,119 68 24 0.94 4.85 4.40 - - New Zealand 12,067 75 10 1.74 12.39 10.13 6.8 66 Australia 14,357 77 8 2.29 8.73 12.00 7.6 56 (Table continues on the following page.) Table 2A-1 (continued) Infant Health Hospitals as a Per capita Life mortality spending as a percentage of GNP a expectancy (per 1,000 Health it1puts per 1,000 population percentage public health Country (U.S. $) at birtha live births)a Physiciansb NursesC Hospital bedsd of GNP spendinge United Kingdom 14,612 76 9 1.64 8.34 9.33 6.0 57 Italy 15,118 77 9 4.28 4.00 10.59 7.3 52 Qatar 15,833 70 29 1.74 4.48 2.54 - - Netherlands 15,923 77 7 2.22 5.96 12.53 8.4 65 Kuwait 16,153 74 15 1.57 4.94 4.14 - - Belgium 16,223 76 9 3.02 9.26 9.38 7.3 23 Austria 17,303 76 9 2.57 5.43 11.14 8.4 18 France 17,821 77 7 3.13 9.51 7.22 8.6 55 N United Arab Emirates 18,414 71 24 0.98 2.56 2.96 - - Canada 19,032 77 7 1.96 8.25 9.88 8.6 59 Iceland 20,411 78 6 2.30 11.49 - 7.9 70 Germany 20,442 75 8 2.65 4.42 11.50 8.1 43 Denmark 20,453 75 8 2.51 16.43 9.68 5.9 76 United States 20,910 76 10 2.12 13.58 5.85 11.1 59 Sweden 21,574 77 6 2.59 1.48 14.81 9.1 75 Finland 22,121 75 6 2.26 16.92 15.54 7.5 51 Norway 22,294 77 8 2.22 17.50 15.00 7.4 74 Japan 23,811 79 5 1.51 5.43 11.58 6.9 42 Luxembourg 26,217 75 9 1.81 4.14 11.84 7.2 29 Switzerland 29,883 78 6 1.44 7.74 11.16 7.7 56 Countries believed to be high income, but with missing GNP data American Samoa - - - 0.83 0.51 Bermuda - - - 0.76 - 9.31 Brunei Darussalam - 75 9 0.55 3.78 3.11 Channel Islands - 77 8 1.52 - 10.60 Faeroe Islands - 74 23 - - 8.13 French Polynesia - 72 21 1.25 2.35 Greenland - 63 29 1.14 - 14.35 Guadeloupe - 74 15 1.39 1.90 11.14 Guam - 73 11 1.18 4.75 Isle of Man - - 16 - - Netherlands Antilles - 77 13 - - 9.12 Puerto Rico - 75 13 - - Virgin Islands (U.S.) - 74 18 - - @ Other economies Albania - 72 26 - 5.25 7.10 Cuba - 76 12 1.89 3.51 4.62 Democratic People's - 70 27 2.38 - Republic of Korea Former U.S.S.R. - 70 24 3.70 5.90 10.97 -Not available. a. Data are for 1989. b. Most estimates are for 1985 or later. c. From about 1985. d. From about 1980. e. For OECD countries, measured as the share of public inpatient expenditures in total public sector health expenditure. Source: OECD 1990 and World Bank data. 76 Public Hospitals in Developing Countries Notes 1. The relationship of the number of beds per capita (Beds/Pop) to GNP per capita (GNPN) using the data in figure 2-1 is Log (Beds/Pop) = -9.416 + 0.477 [Log (GNPN)I (-35.47) (13.45) R2= 0.57 n = 139 (t-statistics are given in parentheses). 2. The log-linear relationship explaining the hospital share (HOSP%oEX) is Log (HOSP%EX) = 1.230 - 0.068 [Log(GNPN)] + 0.159 [Log(MD/Pop) ] (1.23) (-0.76) (2.71) R = 0.29 n = 29 (developing countries) (t-statistics are given in parentheses.) 3. A pyramidal conceptualization of a health system is discussed in chapter 6. 4. The relationship of the personnel share in total hospital cost to GNP per capita (using the data from table 2-4), controlling for the level of hospital (D = 1 for district-level hospitals, 0 for other levels) is Personnel Cost Share (%) = 36.5 + 0.020 GNPN + 7.30 D (7.31) (3.48) (1.57) R2 =033 n= 31 (t-statistics are given in parentheses). 5. The income elasticity of hospital beds gives the percentage change in hospital beds in response to a percentage change in per capita income. 6. The Jamaican data include admissions to all acute care public hospitals. The Korean data relate to those covered by Industrial Establishment Medical Insur- ance, which represented 60 percent of the insured population and 32 percent of the total population in 1986. 7. These data are for insured government employees and private school teachers and staff (and all dependents), who made up 10.4 percent of the population in 1986. 8. The number of healthy years of life gained (HYLG) is the sum of years of life gained from prevention of mortality (YLG) and years gained from avoidance of disability. Polio provides an example. Using the data for Ghana provided by the Ghana Health Assessment Team (GHAT 1981), we find that a total of 3.3 HYLG per thousand people per year are gained from elimination of polio. Of this total 0.6 is gained from elimination of mortality (YLG) and the remainder, 2.7, is gained from elimination of disability. Thus, in this example, the number of YLc is substantially less than the number of HYLG. 9. A QALY is, literally, a quality adjustment to a year of life gained (YLG). The adjustment yields a fraction of a year of life with full health. In the example given by Williams (1985), coronary artery bypass surgery on a patient with severe Patterns of Hospital Resource Use 77 angina and disease of the left main vessel gives a gain of 6 years compared with no surgical intervention. Adjusting these years for the loss of quality of life from the treatment and from continued chronic disability gives a gain, measured in QALYS, of 3.5 years. 10. Given an original cost expressed in U.S. dollars of C$ in a high-income country and foreign exchange costs as a proportion, f, of total costs in a lower- middle-income country, the estimated cost as a proportion of GNPN in a low- income country is: C GNPN H f*C H + (I-f)CH L ~~~$ $ GNPN L GNPN H = FOREIGN + LOCAL COMPONENT COMPONENT The superscript indicates whether the cost is measured in $ or GNPN, and the subscript indicates a high-income (H) or low-income (L) country. 3. Hospital Costs and Efficiency In this chapter we provide estimates of recurrent hospital costs and consider the relation of costs to services. Our immediate objective is to provide information on the determinants of hospital costs that can be used to formulate policy recommendations concerning efficiency and financing. Related benefits that can arise from an understanding of costs include estimates of the potential savings from improved referral pat- terns, the relative cost-effectiveness of alternative programs, and projec- tions of future recurrent resource requirements of current and proposed facilities. The methodology and findings of studies that use two very different methods of cost analysis are reviewed in this chapter. The first method makes use of accounting information and reanalysis of hospital service records to examine hospital costs and performance. The second one makes use of statistical procedures to infer the relation of hospital costs to services provided. The accounting method can be applied usefully to a single hospital and can involve a labor-intensive, detailed examination of hospital accounts, staffing patterns, and admissions. It is also possible, although somewhat less accurate, to derive hospital accounting costs by using aggregate government budgets or expenditure data. Less detailed data are needed in the statistical method, but it requires observations of costs and service use for many hospitals. Statistical studies provide insights into cost issues-the relation between marginal and average cost, and the degree to which hospitals exhibit economies of scale and scope-that accounting studies do not reveal as readily. Ideally, the information used for the statistical analyses would be derived from a large number of detailed and well-documented observations. In actual- ity this is not often possible, and the lesser quality of data in a statistical analysis must be compensated for by inferring a general pattern of costs from a large number of observations. Thus, the accounting and statistical methods yield different but complementary views of costs. 79 80 Public Hospitals in Developing Countries Whichever method of costing is used, the unit of analysis (such as patient-days, admissions, or outpatient visits) and consideration of both average and marginal costs are important, because the use of any single output or cost measure may produce misleading results. The reasons for this can be seen in figure 3-1, which provides a schematic description of the nature of inpatient hospital costs. As depicted in the figure, recurrent inpatient costs are considered to have three components: * Overhead costs. These costs remain essentially constant regardless of whether a bed is occupied. Typically, they include items such as heating and maintenance, but for many public hospitals in developing countries, personnel may be a large component of overhead costs, because it may not be possible for staff to be reduced in the short run during periods of low occupancy. The magnitude of overhead costs is related to hospital size. * "Hotel" costs. These are costs, such as catering, laundry, and linen, that are incurred for each patient-day in the hospital. They tend to be constant for each day of a patient stay, though there will be some variation related to diagnosis and patient characteristics. * Treatment costs. These are case-dependent costs associated with the particular diagnostic, therapeutic, and other treatment services pro- vided to the patient. As depicted in figure 3-1, these costs tend to peak in the first few days of a patient stay, when, for example, there might be an operation, and then diminish thereafter. The actual pattern of treatment costs in any hospital will vary depending on the clinical management of inpatients. Figure 3-1 illustrates the importance of using various units of analysis when interpreting cost data. The figure posits that marginal costs per day equal average costs per day for overhead costs and per patient-day for hotel costs but differ for treatment costs during the period of the patient stay. If the case were the only unit of analysis, the changing nature of marginal treatment costs during a stay would not be observed. Cost comparisons between two hospitals based solely on the number of inpatient days, however, would not account for the differences in aver- age costs per case arising from differences in the average length of stay (ALOS). The hospital in which the length of stay of patients was longer, other things equal, would tend to have a lower average cost per day because the treatment costs for the additional days would be likely to be far below the average for the case. The extra day's stay in the hospital would probably contribute little to the improvement of the patient's condition, and thus the lower average cost would actually mask ineffi- cient hospital performance. The figure illustrates the potential use of hospital service statistics in understanding the efficiency implications of unit cost estimates. Knowl- Hospital Costs and Efficiency 81 Figure 3-1. Inpatient Cost Profile Cost per day Length of stay Length of stay Turnover Hotel costs interval Overhead costs Days Source: Adapted from Forte, 1985. edge of the average length of patient stay, the bed occupancy rate (percentage of beds occupied by patients), and the annual bed turnover rate (average number of inpatients per bed during one year) can help in explaining variation in inpatient unit cost measures. Assuming that the treatment cost profile is similar, high occupancy rates tend to result in lower average costs per patient-day because overhead costs are spread over beds that are usually filled. If high occupancy results from relatively few admissions but very long stays, however, hotel costs will be high in relation to the number of patients and average cost per admission will be high. The expected marginal cost per bed-day will be low because the treatment costs at the end of a long hospital stay tend to be minimal. Alternatively, if the bed turnover rate is high, average cost per admnission is apt to be lower because hotel costs are spread over a larger number of patients, whereas the marginal cost per day will be relatively high. Increasing the bed occupancy rate through a greater number of admis- 82 Public Hospitals in Developing Countries sions per bed rather than longer stays will allow more patients to be served and thus improve hospital productivity. The service units to which costs are compared in this chapter-admis- sion, patient-day, and outpatient visit-are intermediate output or pro- cess measures. Ideally, efficiency in the use of health sector inputs should be assessed in relation to health outcomes (for example, quality adjusted years of life gained, as discussed in chapter 2) rather than process measures. Unfortunately, a massive level of resources would be needed to conduct empirical studies of hospital costs per QALY or other outcome measure for all hospital services, and we are unaware of any such studies having been done. Therefore, we are left to focus on these process measures for analysis of hospital efficiency. The analysis of hospital costs and efficiency in developing countries is a relatively new phenomenon. Although a few accounting studies of average costs were performed in the 1970s and early 1980s (see, for example, the work on Malaysia by Heller [19751 and reviews of earlier studies in Robertson 1985, and Mills 1987), hospital costing has received increased attention in more recent years. Of the studies reviewed in this chapter, only one precedes 1987. The relative newness and dearth of hospital cost studies circumscribes the contribution of this chapter. On the one hand, the chapter pulls previously widely dispersed information together for the first time and insight is thus gained into economic functioning of hospitals in developing countries; on the other hand the data base for the chapter is relatively small compared with what is available in industrial countries, which underlines the need for further cost studies. Accounting-Based Cost Studies In this section we review cost estimates derived from a selection of accounting-based studies. Such studies are often termed "unit cost" studies to indicate that they provide estimates of the average cost of a unit of service. Unfortunately, the term "unit cost" has evolved in the language of health planners to refer ambiguously to both average and marginal costs per unit. To a certain extent the ambiguity in the use of the term "unit" has practical roots. By separating costs into relatively fixed components (utilities and some categories of staff, for example) and variable components (examples are drugs, medical supplies, and food), we can approximate marginal costs by average variable cost using accounting methods. Also, if it is reasonable to assume that average cost is invariant for the relevant scale of production, then marginal and average cost will be equal and estimates of change in cost with projected output can be based on the "unit" cost estimates. Thus, average costs can Hospital Costs and Efficiency 83 be used with care to approximate the marginal cost needed to project the recurrent costs of existing and planned hospitals. There are a number of additional reasons for measuring average cost. Perhaps the foremost is that an examination of the levels and determi- nants of costs holds out some possibility of providing useful insight into the relative efficiency of hospital operations. Comparisons of average costs between hospitals with similar roles in a country's health system may be useful for assessing individual hospital performance and iden- tifying hospitals whose average costs are far from the norm. Compari- sons of average costs of performing the same activity among different levels and categories of hospitals may help to determine hospital devel- opment policy and give an approximation of the potential saving (or increased availability of services) to be derived from improving the referral system. A closely related use of average costs in planning is to provide information that can be used to compare the cost-effectiveness of alternative health sector interventions or to provide information needed to calculate the cost of treating particular diseases and the expenditure avoided by prevention. Average costs also provide infor- mation that can be used in justifying budgets when the level of govern- ment subsidies is set for public hospitals or when reimbursement is made to nongovernmental organizations. Finally, knowledge of recurrent op- erating costs is needed for the formulation of economically efficient hospital cost-recovery policies. Methodology of Accounting-Based Studies We have divided accounting-based average cost studies into two cate- gories. The first develops detailed cost information for individual hos- pitals using allocational assumptions in what is sometimes designated a "step down procedure" to distribute aggregate costs across departments and functions. The second makes less detailed estimates of hospital average costs based on aggregate central ministry information or aggre- gate reported hospital statistics and accounting records. Step down analyses. Until recently, detailed average cost studies were not available for hospitals in developing countries. Such analyses were seen as too difficultbecause of the lack of general accounting information and not useful because of the probable imprecision of the results. This view was not correct; a number of studies, many employing variations on the step down costing methodology, have recently demonstrated the feasibility and usefulness of doing cost studies in a variety of economic and developmental environments. To date, this costing method has been applied only as an exception. The experience from these applications 84 Public Hospitals in Developinzg Countries suggests, however, that it would be possible to institutionalize this type of accounting methodology at the hospital, regional, or ministerial level. Step down cost accounting is a disaggregated method of analyzing the costs associated with specific hospital outputs. It is based on scrutiny of the hospital production process to enable the best assignment of costs to the outputs to which they are related. All hospital expenditures are attributed to specific departments (cost centers), and then allocational criteria, such as time use, are employed to distribute all costs (including overhead and the cost of intermediate outputs) to final service categories. A summary of the step down procedure is given in appendix 3-1. Our focus is on the final cost estimates derived from step down cost analyses, but it is worth noting that the analytical process can be as important as the final estimate. In itself the process of attributing costs to cost centers gives hospital managers considerably more useful infor- mation than the line item accounting information with which they typically operate. The cost center method enables managers at the hos- pital and departmental level to know the level of resources they have available to produce services, and this information can be used with measures of departmental output (in the dietary department, for exam- ple, a comparison can be made between the number of meals served and expenditures in this cost center) to develop benchmarks against which performance can be measured through time and across facilities. The use of cost centers, thus, also promotes financial accountability of depart- mnental managers. The process of allocating costs across intermediate and overhead departments to the final patient service departments yields much greater understanding of resource flows within the hospital. Finally, a comparison of fully allocated costs with service statistics produces average cost estimates of important performance indicators, such as cost per patient-day and cost per outpatient visit. Average cost estimates derived from hospital step down cost studies are included in the top half of table 3-1. Other accounting cost studies. An accounting alternative to step down analysis is the use of aggregate data, either for individual hospitals or for groups of hospitals. The calculation of meaningful average costs from aggregate data requires that cost and service information be related with respect to time and institutional and geographic coverage. Although studies based on aggregate data require much less analysis than do step down studies, the importance of data coverage, accuracy, and complete- ness remains the same. Making an informed assumption of the resource use of other outputs (such as an outpatient visit) from a single output measure (such as an inpatient-day) simplifies the process of relating cost and service information to produce average cost estimates. For example, such a study for Rwandan hospitals (Shepard 1988) assumed that the Hospital Costs and Efficiency 85 resources consumed in four outpatient visits equaled that of one patient- day. A single measure of hospital output, the day-equivalent, was cre- ated as the sum of patient-days and one-fourth of outpatient visits. Unit cost per patient-day was calculated as the total cost divided by day- equivalents, and cost per outpatient visit was one-quarter of this. Cost studies based on aggregate hospital information can be per- formed in less time but provide fewer details and insights than do the step down studies that allow functional analyses. Nevertheless, the information provided from more aggregate studies can be useful in making comparisons among similar hospitals, or different levels of hospitals, and can serve as a basis for making budgetary allocations to hospitals. Aggregate studies should be supplemented by step down studies that give more details on the relative average cost of inpatient and outpatient services to inform the choice of a day-equivalent mea- sure. The aggregate studies can then be used for a quick approximation of average costs for a comparatively large number of hospitals. Average costs based on aggregate accounting studies are cited in the lower half of table 3-1. Caveats on the Interpretation of Average Costs Information on average costs provides one useful input needed for assessing hospital performance. Average cost data alone, however, are not sufficient for reaching definitive conclusions regarding hospital efficiency within a country, and even greater caution is warranted in interpreting results from cross-country studies. Differences in the com- pleteness of the data used in each study and in the health, institutional, and economic environment underlying the estimates place limitations on comparisons. Under ideal circumstances, a study comparing the cost per unit of output for several hospitals would tell us which one provided services with the greatest efficiency (technical or economic). The follow- ing conditions would have to be met, at a minimum, however, for the results to be unequivocal: * The quality of services provided in each facility would have to be the same (or adjusted for) so that costs per an equivalent unit of output were being compared. * The clinical composition of the patients (the case mix) at each of the facilities would have to be the same (or adjusted for). * For economic efficiency, the cost information would have to mea- sure the social opportunity cost of resources used, not merely the amounts reported to have been spent.1 Without an understanding of differences in quality and case mix across hospitals, the efficiency implications of variation in average costs Table 3-1. Hospital Recurrent Average Cost Estimates from Accounting Cost Studies, Selected Countries, Selected Hospitals (1988 U.S. dollars) Inpatient cost Number Level Per Per Outpatient of hospitals Country and year of data of hospital patient-day admission Per bed cost per visit in study Source Step down studies Belize, 1985a If 60.4 370 15,075 1 Raymond and others 1987 III 42.3 126 4,714 6 China, 1986 1 9.1 260 3,223 1 Chen 1988; Chen 1987 11 4.4 87 1,446 1 c Indonesia, 1987 II 15.5 6.2 2 Djuhari and others 1988 III 8.1 2.3 30 Jamaica, 1985-86 1 30.7 327 9,419 12.6 2 Kutzin 1989 11 20.9 140 6,646 8.7 2 III 23.0 176 6,699 9.0 1 Malawi, 1987-88b III 3.0 27 1,264 0.6 6 Mills 1991 Niger, 1986-87c I 6.6 93 2,090 15.9 1 Wong 1989 Papua New Guinea, 1988 1 28.7 286 8,384 6.2 1 js1 1990 11 25.8 276 7,549 2.5 4 III 26.7 364 5,822 5.3 8 St. Lucia, 1986-87d if 43.9 311 11,907 19.4 1 Russell, Gwynne, and Trisolini 1988 Other accounting studies China, 1986e 1 8.7 218 2,993 2.2 8 Barnum 1989 II 5.7 103 1,807 1.4 11 III 4.3 57 1,489 1.1 7 Colombia, 1978 I, II 56.9 421 16,526 14.2 8 PRIDES 1980 Indonesia, 1985f 1 18.6 174 5,102 4.6 2 Barnum 1987 II 14.6 127 3,647 3.7 15 III 7.0 41 1,384 1.8 296 Rwanda, 1984k 1 16.1 5,146 4.0 2 Shepard 1988 II 13.8 4,218 3.5 1 III 7.6 1,572 1.9 17 Turkey, 1987g 1 39.8 372 10,649 3 World Bank 1990 Zimbabwe, 1987h 1 28.0 219 9,123 10.6 4 Hecht 1992 II, III 17.9 109 4,366 1.9 90 United States, 1988' All acute 581.1 4,194 5,579 AHA 1989 Note: Current estimates in local currency were converted to U.S. dollars by exchange rate in that year. The U.S. consumer price index was used to adjust to 1988 terms. a. For level Im hospitals, attribution of inpatient and outpatient costs was based on relative amounts estimated for Belize City Hospital. >, b. Outpatient department unit cost relates to new outpatients (unit cost across all outpatients would be lower). c. Intermediate service (for example, laboratory) costs were attributed to inpatient and outpatient services for comparability with other studies. d. Radiology, laboratory, operating theater, and physiotherapy costs were attributed to inpatient and outpatient services for comparability with other studies. e. Cost data identified more than 50 percent of expenses as either inpatient or outpatient. For the remainder, expenditures were attributed on the assump- tion that the cost of four outpatient visits was equal to that of one patient-day. f. Day equivalents assume that the cost of four outpatient visits equal that of one patient-day. g. Total hospital recurrent costs were divided by inpatient statistics; no attempt was made to apportion costs between inpatient and outpatient. h. Estimates were based on Hecht's assumption that 80 percent of total hospital costs were attributable to inpatient services. The category "I, 1Hi" in- cludes provincial, district, and small rural hospitals, plus health centers. i. Estimates exclude the value of staff time of physicians who are not salaried employees of hospitals. Source: As noted in table. 88 Public Hospitals in Developing Coun tries cannot be properly interpreted. For example, high average costs may reflect high quality, poor efficiency, or the characteristics of the patients at one institution in relation to another. Low average costs may be a result of an inadequate provision of drugs, and thus would represent poor quality, not greater efficiency. If information on the quality of services and the case mix of patients is added to cost data, the efficiency implications of average cost information become clearer. Conclusions from Accounting Studies With these caveats in mind, it is still possible to glean useful information from table 3-1. Perhaps the most striking aspect of the table is the magnitude of cross-country variation in the average cost estimates. The level of average costs is associated with per capita GNP, because the richer countries in the table (Belize, St. Lucia, Turkey, and Jamaica) have the highest unit costs, and Malawi and Niger, the poorest nations, have the lowest unit costs. Given that personnel usually comprises the largest component of hospital costs (see discussion in chapter 2) and that wage rates are associated with per capita income, this finding is not surprising. It suggests that the appropriate technological mix of inputs in a poor country's hospitals is probably different, that is, would use relatively more personnel, from that which is appropriate in a richer country's hospitals. The extent of the difference depends on the feasibility of substituting labor for nonlabor inputs. Cross-country variation in recurrent average costs may also reflect differences in the quantity of recurrent inputs used, which may suggest qualitative differences in hospital outputs across countries. These differ- ences between outputs across countries may mean that, for example, a hospital day in St. Lucia bears little resemblance to a day in a Chinese hospital of the same level. Therefore, comparison of the cost of produc- ing a patient-day in each country is difficult to interpret because the outputs may be qualitatively different. Detailed information on the prices and quantities of inputs is needed to interpret more meaningfully the relative cost-effectiveness of hospital services across countries. A study by Lewis, Sulvetta, and LaForgia (1990) of one tertiary hospi- tal in the Dominican Republic sheds some light on the relation between quality and average costs. The authors collected information on specific clinical practices in the hospital and compared them with norms of clinical treatment established by Dominican physicians that described the inputs needed to provide adequate quality of care for selected diagnoses. Price information gathered during the study was used to estimate the cost of achieving these norms. To reach the norms, current expenditures for the appropriate diagnostic tests and drugs would each have to increase tenfold. This suggests that current estimates of average Hospital Costs and Efficiency 89 costs at this hospital are measures of the resources used to provide care of low quality. Although this finding is only suggestive because it assesses quality as a process rather than as an outcome, it has important implications for the interpretation of average cost estimates. The devel- opment and costing of country-specific norms and their comparison with estimated average costs would go a long way toward enabling a hospital's average costs to be adjusted for quality so that the cost per equivalent unit of output could be compared across facilities. Data in table 3-1 show that, within a country, tertiary hospitals tend to have the highest average costs and that the less technically complex district-level hospitals have the lowest. There are several possible expla- nations for this. First, the teaching role of most tertiary hospitals contrib- utes to higher costs. Second, tertiary hospitals are intended for treatment of patients with the most complex and severe conditions. Therefore, higher costs per patient might indicate different case mixes across facil- ities. Such differences reflect appropriate use of tertiary facilities and are not, in themselves, cause for concern. Third, higher average costs in tertiary hospitals might also result from their having more equipment and other resources available than lower-level facilities have. This may lead to the use by tertiary hospitals of a more expensive mnix of inputs to treat cases of similar complexity and severity to those in provincial or district hospitals. Such use of resources might be inappropriate and require corrective action with respect to referral policy. Treatment norms and analyses of hospital case mix are needed to inform such policy decisions. Each of these factors-teaching function, case mix complexity, and costlier input mix-contribute to the higher average costs found in tertiary hospitals, with the relative contribution of each varying by hospital. A common finding of the step down studies was that official govern- ment budgetary information and even the financial reports of individual hospitals understate expenditures (and, to a greater degree, costs) by and on behalf of public hospitals. Although this finding has implications for future cost studies (particularly for more aggregate-level analyses), per- haps a more important finding is that those within the hospital who are responsible for allocating resources do not have a complete picture of their available revenues. For example, if the hospital's budget does not indude services provided by a regional maintenance unit, hospital ad- ministrators will find it difficult to manage maintenance services. These detailed studies point to the need for greater transparency in public hospital accounting systems, so that managers have an accurate depic- tion of the level of resources with which they are working. The data in table 3-1 allow an indirect comparison of the step down studies with those that use aggregate data. Although the studies were performed on a different set of hospitals, the estimates of cost per day 90 Public Hospitals in Developing Countries for Chinese hospitals and for Indonesian hospitals were similar under the two methods, which suggests that reasonably accurate results can be achieved through use of aggregate costs and day-equivalents when available resources do not allow for step down studies. In the case of the Indonesian studies, however, there are greater differences with regard to outpatient costs between the step down and accounting results. The three studies that used the day-equivalent method-China (Bar- num 1989), Indonesia (Barnum 1987), and Rwanda (Shepard 1988)- each assumed that, for line item categories for which the direct cost was unknown, the cost of one inpatient day was equal to the cost of four outpatient visits. For China, a separate linear regression of total cost on inpatient bed-days and outpatient visits gave a rough confirmation of the 1:4 ratio. For Indonesia, Djuhari and others (1988) gave a rough confirmation of the same ratio for lower-level hospitals but suggested a ratio closer to 1:3 for upper-level hospitals. With few exceptions, how- ever, the step down studies found the cost of an outpatient visit to be greater than one-fourth the cost of an inpatient day. This finding sug- gests that it may be worthwhile for countries with large numbers of hospitals to perform a few step down studies on hospitals of each type to generate the appropriate day-equivalent measure. Table 3-2 presents the average costs from table 3-1 as a percentage of each country's per capita GNP to show hospital costs in relation to the level of income in each country. The percent GNPN measure is a good indicator of the burden of hospital costs on the overall economy of a country. For example, this table indicates that, compared with other countries, the average cost of a hospital day and the annual cost associ- ated with a hospital bed in Rwanda and Zimbabwe are high in relation to the overall level of resources. Alternatively, it is evident that in the United States, where the absolute level of hospital costs is high (table 3-1), costs per discharge in relation to this country's level of resources are lower than in many poorer countries. Service Statistics, Efficiency, and the Demand for Hospital Care The above discussion suggests that although average cost studies yield useful results, accurate interpretation of the implications of the calcu- lated level of average costs requires data on service indicators. In this section we examine data from a number of countries on three interrelated hospital service indicators-bed occupancy rate, average length of stay, and the bed turnover rate-and describe a methodology developed by Pab6n Lasso (1986) for assessing hospital performance based on the simultaneous analysis of these statistics. Table 3-3 presents data on these indicators (plus an additional indicator showing the relative use of Hospital Costs and Efficiency 91 hospitals for outpatient services) from hospitals in a number of develop- ing and industrial countries. The bed occupancy rate measures the percentage of total available beds that are occupied by patients.2 The average length of stay ALOS is the mean number of days from admission to discharge for each inpa- tient.3 The bed turnover rate is the average number of inpatient admis- sions or discharges per bed.4 Each of these indicators is usually (but not necessarily) defined on an annual basis and can refer to a particular ward, inpatient department, entire hospital, or group of hospitals. Any one of these indicators provides useful information that can help de- scribe the performance of a hospital's inpatient services, but their explan- atory power is multiplied when they are used together. Because they are interrelated, knowledge of any two indicators defines the third. In the following subsections we consider each of the service statistics individ- ually and then present a graphical technique using the indicators simul- taneously to examine the relative efficiency of hospital performance. Occupancy and Turnover Rates The data in table 3-3 indicate that occupancy and turnover rates vary greatly from country to country and between levels of hospitals. For the most part, occupancy rates decrease as the level of the hospital decreases. There are exceptions, such as China and Malawi, where occupancy rates are high at the district level, but in most developing countries low occupancy rates at the district level are an important reflection of eco- nomic inefficiency in the hospital sector. Annual bed turnover rates do not show a consistent trend according to the level of hospital. In many countries, turnover is higher in middle- and low-level hospitals than in tertiary hospitals. In these countries, low- and middle-level hospitals are serving a greater number of patients per bed than tertiary facilities. Even if this is the case, however, there may still be room for improving sectoral efficiency by encouraging more patients to use lower-level rather than tertiary hospitals. Low occupancy rates are a commonly observed problem in many countries, especially in lower-level facilities. Individual facilities have a level of services, usually somewhere in the neighborhood of 85-90 percent occupancy, at which they have been designed to operate most efficiently. In the short run, a relatively small percentage of hospital costs can be varied; most costs are fixed and determined by the scale of the facility and the personnel establishment (the overhead costs shown in figure 3-1). In chapter 2 it was reported that personnel costs, which make up the bulk of fixed cost, represent a range of about 35 to 75 percent of total recurrent costs. The effect of low occupancy is to spread the cost of personnel and other fixed inputs over a smaller number of service units Table 3-2. Hospital Recurrent Average Cost Estimates fromn Accounting Cost Studies, Measured as a Percentage of per Capita GNP Inpatient cost Number Level Per Per Outpatient of hospitals Country and year of data of hospital patient-day admission Per bed cost per visit in study Source Step down studies Belize, 1985a II 4.9 30 1,231 1 Raymond and others 1987 ill 3.5 10 385 6 China, 1986 I 3.2 90 1,119 I Chen 1988; II 1.5 30 502 1 Chen 1987 Indonesia, 1987 II 3.6 1.4 2 Djuhari and others 1988 ill 1.9 0.5 30 Jamaica, 1985-86 1 3.7 40 1,148 1.5 2 Kutzin 1989 11 2.6 17 810 1.1 2 ill 2.8 21 817 1.1 1 Malawi, 1987-88b III 1.9 17 806 0.4 6 Mills 1991 Niger, 1986-87c I 2.2 32 710 5.4 1 Wong 1989 Papua New Guinea, 1988 1 3.3 33 962 0.7 1 jsi 1990 II 3.0 32 866 0.3 4 III 3.1 42 668 0.6 8 St. Lucia, 1986-87d II 3.0 21 808 1.3 1 Russell, Gywnne, and Trisolini 1988 Other accounting studies China, 1986e 1 3.0 76 1,039 0.8 8 Barnum 1989 11 2.0 36 627 0.5 11 III 1.5 20 517 0.4 7 Colombia, 1978 I-II 3.4 25 985 0.8 8 PFRDES 1980 Indonesia, 1985f 1 2.8 26 756 0.7 2 Barnum 1987 11 2.2 19 540 0.5 15 III 1.0 6 205 0.3 296 Rwanda, 1984f 1 5.2 1,667 1.3 2 Shepard 1988 11 4.5 1,366 1.1 1 III 2.5 509 0.6 17 Turkey, 1987g I 3.1 28 816 3 World Bank 1990 Zimbabwe, 1987h 1 4.3 33 1,393 1.6 4 Hecht 1992 II, III 2.7 17 667 0.3 90 United States, 19881 All acute 2.0 15 5,579 AHA 1989 Note: Percent GNPN cost figures derived by dividing current year average cost estimates by the country's per capita GNP for that year, both measured in local currency. a. For Level III hospitals, attribution of inpatient and outpatient costs was based on relative amounts estimated for Belize City Hospital. 4 b. Outpatient department unit cost relates to new outpatients (unit cost across all outpatients would be lower). c. Intermediate service (for example, laboratory) costs were attributed to inpatient and outpatient services for comparability with other studies. d. Radiology, laboratory, operating theater, and physiotherapy costs were attributed to inpatient and outpatient services for comparability with other studies. e. Cost data identified more than 50 percent of expenses as either inpatient or outpatient. For the remainder, expenditures were attributed on the assump- tion that the cost of four outpatient visits was equal to that of one patient-day. f. Day equivalents assume that the cost of four outpatient visits equal that of one patient-day. g. Total hospital recurrent costs were divided by inpatient statistics; no attempt was made to apportion costs between inpatient and outpatient. h. Estimates were based on Hecht's assumption that 80 percent of total hospital costs were attributable to inpatient services. The category "ll, 1I" in- cludes provincial, district, and small rural hospitals, plus health centers. i. Estimates exclude the value of staff time of physicians who are not salaried employees of hospitals. Source: As noted in table. Table 3-3. Hospital Service Statistics, Selected Countries, Selected Hospitals Occupancy Bed turn- Mean Outpatient Number Level rate over rate length visits per of hospitals Country and year ofdata ofhospital (%) per year ofstay bed-day in study Source Argentina, 1980 Acute public 66 20.0 12.0 - All Belize, 1985 II 68 40.7 6.1 1.3 1 Raymond and others III 31 37.3 3.0 - 6 1987 China, 1986 I 94 13.7 25.1 2.1 8 Barnum 1989 II 86 17.6 17.9 3.6 11 III 95 26.1 13.3 3.3 7 Colombia, 1980 I 73 37.8 7.2 - 9 Pab6n Lasso 1986 II 61 38.7 6.0 - 20 III 55 42.8 5.2 - 44 t Ethiopia, 1983-85 Urban 47 14.7 11.8 1.7 6 Donaldson and Dunlop Rural 59 29.7 7.2 2.4 13 1987 Fiji, 1987 1 83 42.5 7.2 2.7 3 III 46 47.9 3.5 4.2 19 Indonesia, 1985 I 75 29.2 9.4 2.5 2 Barnum 1987 II 68 28.7 8.7 1.8 15 III 54 33.6 5.9 2.2 296 Jamaica, 1985 1 79 35.2 8.2 1.0 5 GOJMOH 1986 II 84 43.2 7.1 0.6 4 III 61 28.6 7.8 0.6 13 Jordan, 1986 All MOH 71 66.0 3.9 1.7 - Korea, 1986 All 60 17.8 12.3 2.3 546 FKMIS 1987 Lesotho, 1985 I 125 50.7 9.0 0.7 1 III 129 54.9 8.6 1.1 7 Mission 56 19.2 10.7 1.0 9 Malawi, 1987-88 III 116 47.4 9.0 1.3 6 Mills 1991 Morocco, 1987 All Public 57 20.2 10.3 - - Bennis and others 1990 Niger, 1986-87 I 87 22.5 14.1 0.3 1 Wong 1989 Papua New Guinea, 1 80 29.4 9.9 1.4 1 jsi 1990 1988 II 80 28.1 10.4 2.2 4 III 60 16.9 12.9 1.8 8 Rwanda, 1984 1 88 - - 0.9 2 Shepard 1988 II 83 - - 0.5 1 III 57 - - 2.0 17 St. Lucia, 1986-87 II 74 38.8 7.0 0.6 1 Russell, Gwynne, and Trisolini 1988 Tanzania, 1989 Mission 81 33.3 8.9 1.5 43 CMBT 1991 Tunisia, 1989 1 76 27.6 10.1 1.0 9 Turkey, 1987 I 73 28.7 9.3 - 3 All MOH 46 25.3 6.6 -- All hospitals 50 25.5 7.2 - Zirnbabwe, 1987 MOH I 89 41.7 7.8 0.7 4 Hecht 1992 MOH II 91 54.5 6.1 1.4 8 MOH III 76 40.8 6.8 1.9 31 OECD mean, 1980-83 All 81 16.4 17.9 - - OECD 1987 Finland, 1982 All 85 13.9 22.2 - - OECD 1987 France, 1983 All 73 18.9 14.1 - - OECD 1987 Ireland, 1982 All 80 32.5 9.0 - - OECD 1987 Spain, 1981 All 75 18.6 14.6 - - OECD 1987 United Kingdom, 1981 All 81 16.0 18.6 - - OECD 1987 United States, 1981 All 79 29.0 9.9 - - OECD 1987 United States, 1988 Acute 65 33.2 7.2 1.2 5,579 AHA 1989 - Not available. Source: World Bank sector reviews and appraisal reports, except as noted in table. 96 Public Hospitals in Developing Countries and raise the average cost of services. Even if hospital inputs are being used with technical efficiency, low occupancy implies economic ineffi- ciency. A high bed occupancy rate does not necessarily indicate better hospital performance. Indeed, bed occupancy rates can be too high, in the sense that the volume of services is above the design level of the facility. The implications of high occupancy for average costs and hospital efficiency are ambiguous without information on the other service indicators. The reason for this is that a high occupancy rate may reflect a relatively efficient situation, as when many patients with modest lengths of stay are served (that is, the hospital has a high bed turnover rate), or an inefficient situation, as when the high proportion of filled beds largely results from long lengths of stay. The latter situation is signaled by a low average cost per day but a relatively high average cost per admission. Consider, for example, the level I hospitals (from the non-step down studies in table 3-1) in China and Indonesia. The Chinese hospitals have a lower cost per day but a higher cost per discharge. From table 3-3 it is clear that the Chinese hospitals have a higher occupancy rate, yet the mean length of stay figures suggest that this high occupancy rate does not reflect an efficient situation. A comparison of the turnover rates shows that, at each level, Indonesian hospitals are serving more patients per bed than Chinese hospitals. There are other reasons why a high occupancy rate does not by itself imply a relatively efficient hospital. For example, with high occupancy rates, scheduling of individual service activities, maintenance, and man- agement becomes more difficult and more costly. The measured average costs from accounting-based studies do not provide sufficient evidence to support or reject the hypothesis that average costs are regularly lower for hospitals with extraordinarily high occupancy rates. More likely is that the quality of services is compromised as staff attention and labora- tory and ancillary services are divided among a greater number of admissions that exceeds the hospital's design capacity. In addition, very high occupancy rates may reflect overcrowding, which can facilitate the spread of hospital-acquired infections. In general, conditions treated in lower-level facilities require simpler interventions and may require shorter lengths of stay than more complex cases seen in tertiary facilities. In those countries where turnover rates are highest in tertiary hospitals, the likelihood is that a large percentage of the cases treated in these hospitals are basic cases not requiring tertiary care. Although the hypothesis needs to be tested in future research, there is a presumption that turnover at tertiary facilities that is high in relation to turnover at lower-level hospitals may be an indicator of sectoral inefficiency. Hospital Costs and Efficiency 97 Quality and Hospital Use The determinants of low hospital turnover and occupancy rates are not well understood and need careful empirical research to establish statis- tically confirmed causes. Field experience and qualitative analysis do, however, suggest some likely hypotheses. Possible causes include de- mand-side factors affecting use that are not within the control of central health authorities, for example, underlying morbidity patterns, reflected in a high proportion of communicable diseases that require less hospi- talization than the mix of diseases in higher-income areas, and education and cultural factors, specifically, the lower inclination of poorer and less-educated people to look to hospitals for care. Supply-side factors that interact with demand, however, are probably at least as important and are more amenable to policy intervention. These include cash prices, in the form of user fees, for services and drug charges; nonmonetary prices of access, for example, the value of time spent in gaining access, which is inversely related to the proxirnity of the hospital to patients in the catchment area, plus the availability and monetary cost of transport; and the quality of services with respect to the adequacy of drugs and other medical supplies, staffing, and the availability of critical special- ties. Quality has both supply- and demand-side characteristics. The critical demand issue is perceived quality: the consumer's assessment of the relative quality of different health care providers. Differences in per- ceived quality, with a basis in fact, provide an important explanation of why some people bypass district facilities and refer themselves directly to provincial or central tertiary facilities despite the greater price in time and money that use of these facilities often entails. Adequate staff and supplies are obvious supply-side factors affecting actual quality of ser- vices that are important in affecting perceived quality. Thus, demand for services can be responsive to policies on the supply side to improve the availability of key inputs at lower-level facilities. Availability of drugs and supplies. In low-income countries, a sporadic supply of drugs and supplies, especially in lower-level facilities, is commonplace. Foreign exchange constraints severely limit the purchase of drugs by central ministries, and poor distribution systems further restrict the regular availability of drugs at lower-level facilities in outly- ing areas. The correlation between drug availability and use of services is quite clear to local hospital managers and is revealed by the variation in occupancy and outpatient visits during the course of a year-when drugs are available service use is high and when drug supplies are limited service use is low. This expected effect of drug availability on perceived quality and thus on demand is supported by a World Bank 98 Public Hospitals in Developing Countries study of health facility demand in Nigeria's Ogun State, which found that facility use increased with the percentage of time during the year that drugs were available (Akin and others 1991). Important to note is that the shortchanging of drug and supply expen- ditures in hospitals may not result in lower average costs but quite the opposite. The elasticity of service demand with respect to drug availabil- ity is probably well above, say, 0.3 at low levels of drug provision, so that a 10 percent fall in drug availability is accompanied by a more than 3 percent fall in service demand. In this case a simple calculation will show that, if drugs make up 30 percent of total hospital unit costs, average costs will increase as the availability of drugs decreases even though drug costs are reduced.5 Lack of skilled staff. A similar deficiency in demand is associated with a lack of trained personnel and skilled functions in lower-level hospitals. Because of greater opportunities for private practice, better opportuni- ties for advancement, better amenities, and training that emphasizes Western medical practices and the use of modern technology, medical doctors often resist assignments to rural areas, preferring to be posted in large urban areas. The lack of skilled staff for basic services such as high-risk obstetrics, radiology and laboratory diagnostics, surgery, and pediatrics undermines patient confidence in lower-level facilities. A lack of skilled staff may also indicate an overall insufficiency of other inputs, such as equipment, and an overall inferior quality of services. Indonesia offers an appropriate context to examine the relation be- tween low occupancy rates and the quality of hospital services. There is wide variation in the occupancy rates in the 296 lower-level hospitals (designated C and D in the Indonesian classification system). Among all class C and D hospitals, half have bed occupancy rates under 50 percent, and one out of six has rates lower than 25 percent. Some insight into the cause of this variation is given by an analysis (reported in the study edited by Prescott [19911) of the relation between inpatient use and a number of variables, including staffing characteristics. Inpatient use is weakly related to district population size, is higher for urban hospitals, lower outside Java, and is higher when there is a large proportion of the population covered by the compulsory insurance system for govern- ment employees (ASKES). But the most important result is the extremely strong effect of staffing, used as a proxy for service quality. Hospitals that offer surgical services had 42 percent higher inpatient use than similar hospitals that do not offer surgery. Adding a specialist doctor (not necessarily a surgeon) would boost use by 83 percent. These results suggest a strong conclusion: perceived service quality as measured by physician availability is a key determinant of the level of hospital use in Indonesia. Adding a specialist doctor and providing basic Hospital Costs and Efficiency 99 surgical facilities in the lower-level hospitals could increase service quality and quantity dramatically. The total costs in the hospitals would rise, but average costs could fall if use increased. Health system costs may not increase much if physician staff is transferred from urban to rural facilities. Clearly, the problem in implementing this policy is that it is difficult to attract specialist doctors to small outlying hospitals: the financial opportunity cost to specialist doctors at present levels of public sector salaries levels is too great because of the lack of effective demand for supplementary private practice in poor, remote areas. But this anal- ysis strongly suggests that it might be worthwhile to offer a salary supplement to attract specialists to these hospitals. Without a physician and basic facilities the hospital is little more than an empty shell, and the fixed operating costs and sunk investment costs are substantially wasted. Paying market wages for specialists and providing necessary facilities might yield a high payoff. Alternatively, smaller hospitals with very low occupancy rates could be converted into health centers or clinics that provide primary care. Any excess staff or supplies that may result from this conversion could be reallocated to other facilities. Average Length of Stay Average length of stay is an important indicator of the efficiency of hospital resource use. There is no reason to conclude that longer stays contribute to higher-quality care; in fact the steady decline during the last thirty years in lengths of stay in most countries belonging to the Organisation for Economic Co-operation and Development (OECD) has occurred at the same time that the technical quality of care in hospitals has improved. Without information about case mix and severity it is difficult to use length of stay as a direct indicator of efficiency for individual hospitals, but stays that are unusually long raise questions regarding efficiency and should provoke a closer search for an explana- tion of the cause. Differences in the average length of stay among a large number of hospitals of comparable type imply differences in prevailing treatment practices across countries, but again, case mix must be taken into account. Although average length of stay varies across countries, it appears that stays are generally shorter in developing countries than in Europe and are roughly comparable to the average lengths of stay in the United States (see table 3-3). High fertility, however, may bias the average length of stay in developing countries toward short-term con- finement, whereas the higher chronic disease rates in the United States and Europe impart the opposite bias. Thus, average lengths of stay adjusted for case mix may actually be higher in many developing coun- tries in relation to industrial countries than is indicated by the aggregate data of table 3-3. The problems arising from cross-country differences in 100 Public Hospitals in Developing Countries the definition of a hospital must be noted in comparisons of average lengths of stay. Whether hospitals include long-term as well as acute care beds is particularly relevant to this issue. China, with extremely long average stays at all hospital levels, is an important example of instances in which a long ALOS reflects technical inefficiency in hospital resource use. In two general hospitals in Shang- hai, average stays in 1986-87 for many conditions (for example, hyper- tension, bronchitis, malignant tumors) were about two to three times as great as in the average OECD hospital in 1980 (Chen 1987 and 1988; OECD 1985). There are many possible reasons for the longer confinements in China, including a lack of alternatives for long-term care, poor schedul- ing for diagnostics or surgery, and a poor recovery environment in posthospital home care. The means by which inpatient services are financed can have a direct effect on average length of stay. If patients do not face a monetary price for each day that they stay in the hospital, they have no financial incentive to minimize their length of stay. From the hospital's perspec- tive the financial incentives can be powerful. If a hospital is reimbursed for the costs of an inpatient stay on the basis of a constant per diem fee, they have an incentive to keep patients for lengthy periods. This incen- tive can be strong; it was shown in figure 3-1 that the last days of a patient stay have low marginal costs, thus making these days the most profitable under this method of hospital reimbursement. Korean and Chinese hospitals are financed in this way. In Korea, the average length of stay was 10.9 days in 1986 (FKMIs 1987) and in China the average length of stay was 19.1 days in 1989 (from a sample of twenty-six hospitals; Barnum 1989). The lower average length of stay is expected in Korean hospitals because the Korean health insurance system includes substan- tial per diem cost sharing by beneficiaries, whereas most Chinese cov- ered by health insurance have no copayment requirements. A study (Yu 1983) comparing insured with uninsured patients in a Korean hospital from 1978 to 1980 found that the average length of stay was significantly higher for insured patients. Alternatively, hospital reimbursement policies can serve to limit length of stay. This is the most likely cause of the short average lengths of stay in the United States, where hospitals are increasingly being reimbursed a case-related prospective price for inpatient stays (by pri- vate as well as public payers). Under this system, which was im- plemented in 1984 for the publicly funded Medicare program (serving the acute care needs of the population age sixty-five and older), patients are categorized-according to their diagnosis, procedures received, and other characteristics-into Diagnosis Related Groups (DRGs), and hospi- tals are reimbursed a fixed price based on the DRG. If the costs of a patient stay are less than this price, the hospital profits. If they are not, the Hospital Costs and Efficiency 101 hospital loses money on that patient. Clearly, under this system hospitals have had an incentive to reduce length of stay (and, some have argued, other resource inputs) per case and to maximiuze the number of profitable admissions. The average length of stay in nonfederal acute care hospitals for Medicare beneficiaries fell from 10.7 days in 1980 to 8.6 days in 1987 (NCHS 1989). This 20 percent decline bears out the strength of these incentives. Developing countries have had little experience with case basis pric- ing, and it is premature to evaluate or recommend policies of this type for low- or middle-income countries. Case-based pricing as it operates in the United States (the DRG system) requires continual updating and monitoring of payment rates and a massive amount of data collection and reporting. These features of the system limit its use in low-income countries at present. Nevertheless, because of the growing need to control costs, less administratively complex versions of case-based pric- ing may have application. Simplified adaptations of case-based pricing should be evaluated on a pilot basis, particularly in large university or central hospitals. Brazil's social health insurance system has recently begun to reimburse hospitals on a procedure-oriented prospective price basis, and thus one would expect its average length of stay to have fallen since the implementation of this system. Unfortunately, such informa- tion is not available (Rodrigues 1989a). In Zaire's Bwamanda Health Zone, the reference hospital is reimbursed a fixed amount per case, with different payment rates for each of sixteen case categories (Shepard, Vian, and Kleinau 1990). As in Brazil, however, no data are yet available to assess the effect of this payment system on ALOS. Apart from differences in case mix and methods of financing, cross- national variation in average lengths of hospital stay may result from differences in the role of hospitals across countries. In the United States, for example, acute care hospitals rarely provide extended care to those in need of such care. Separate facilities exist to provide long-term care; therefore, the reported mean length of stay of hospitals is not biased by the long stays of those treated in extended care facilities. In many developing countries, separate extended care facilities do not exist, and hospitals must often be a source of long-term as well as acute care (see discussion of extended care, below). In some countries, hospitals also serve other social functions, such as orphanages and as nursing homes for the elderly. These factors must be considered before drawing conclu- sions regarding the efficiency implications in comparisons of cross-na- tional mean length of stay. Reducing the average length of stay in upper-level facilities with high occupancy rates would enable turnover rates to increase and thus allow hospital benefits to be extended to a greater number of people, dimin- ishing the pressure for capital investment in new hospital capacity. In 102 Public Hospitals in Developing Countries both upper- and lower-level facilities, even where occupancy rates are low, the reduction of excessive ALOS would increase the cost-effective- ness of services by reducing the average cost per admission of specific treatments, although the cost per day may rise from the reduction in relatively inexpensive days. Many factors subject to management inter- vention have been identified as contributing to long ALOS and are dis- cussed below. Scheduling. Poor scheduling of diagnostic and therapeutic care contrib- utes to longer stays, especially in upper-level hospitals. In many hospi- tals it is common for patients to be admitted to inpatient care for diagnostic tests and then confined until results are received. Similarly, patients receiving therapy are often kept in the hospital between treat- ments rather than treated in short stays or as outpatients. Dramatic examples are the use of scarce high-technology therapies, such as kidney dialysis and radiation treatment, but elective surgery and antibiotic therapy and basic laboratory tests are also poorly scheduled. Blanpain (1987) notes that a selection of Chinese hospitals had waiting periods of three to eleven days from the conclusion of preoperative diagnostics to the day of actual surgery. In the context of figure 3-1 the hospital incurs hotel costs prior to as well as after the principal treatment activity. Insufficient equipment does not fully explain the scheduling bottlenecks; often the same hospitals that report long diagnostic waiting periods have uneven use of special facilities, excess capacity being observed during particular times of the day or week. Physicians who divide their time between public hospital service and private practice are another source of scheduling problems, although the efficiency of outpatient services may suffer more than that of inpatient services. Many hospitals provide scheduled specialist or nonspecialist services in outpatient clinics. Physicians who maintain a private practice in addition to public employment may experience conflicts in schedul- ing, which result in their lateness for or absence from scheduled public clinic appointments. It is critical for hospitals to work out arrangements with their physicians to ensure that such conflicts are minimized. Problems in diagnostic services. These problems often manifest them- selves as scheduling problems but result specifically from equipment failures or shortages of supplies or staff needed to conduct diagnostic tests. Lack of recurrent resources to fund maintenance, supplies, and staff adequately is at the root of these problems, which then spill over into delayed testing and, further, into extended lengths of stay. A short- age of reagents often presents itself as a cause of delays in performing and analyzing laboratory tests. Parallel problems arise in radiological services. For example, lack of staff and equipment failure were cited as Hospital Costs and Efficiency 103 the primary reasons behind scheduling delays for radiological exams in Jamaica's public hospitals (GOJMOH 1987). Extended care. The provision of extended care services in acute care hospitals is sometimes considered an inefficient use of hospital re- sources. In Europe and the United States, patients convalescing or suf- fering from long-term degenerative or chronic diseases are often kept in nonhospital facilities (nursing homes, rehabilitation facilities, convales- cent homes, or their own homes), where less resource-intensive but more appropriate care is available. Transportation and logistics problems, however, plus shortages of trained staff and other resources, make it difficult to replicate this array of providers in developing countries. Given shortages of recurrent resources for hospitals, many countries probably do not consider the creation of separate facilities for extended care to be feasible. Even though some staff can be reallocated, the additional overhead costs that must be incurred may be prohibitive. Home care, provided reliable outreach support is made available, may be a feasible lower-cost alternative. The unsatisfactory health environment, poor sanitation, and inade- quate housing into which convalescing patients are discharged is often cited by local hospital administrators as a reason to keep patients beyond their direct need for hospital care. Patients or their families may also be a source of demand for longer hospital stays for convalescence or chronic diseases.The extent to which hospitals are used for these purposes probably varies greatly between urban and rural areas and by income level, the poorer and more rural patients using hospitals for extended care to a much lesser extent. Nonhospital alternatives for extended care are discussed further in chapter 6. Standard treatment practices. Treatment protocols for the same causes of admission vary among countries and through time. After considerable professional controversy concerning the effect of early discharge and the use of ambulatory care on the quality of outcomes, physicians in the industrialized countries concluded that dramatic reductions in length of stay were possible without compromising, and perhaps increasing, the quality of outcome. From 1960 to 1980, the average length of stay was reduced from twenty-one to ten days in the United States (all hospitals; OECD 1987). The reduction was achieved, in part, by changing the stan- dard practice for specific causes of admission. Physicians exert ultimate control over the way in which the treatment plan involves hospitals, including diagnostic and treatment procedures used and lengths of stay. Medical school training and commonly accepted professional practice are determinants of physicians' choices and affect the rate at which new standards are adopted. 104 Public Hospitals in Developing Countries Older standards of care and medical protocols may contribute to the longer hospital stays in some countries. To the extent that shorter stays require the use of high-cost modern technology, such change might not be desirable; however, much of the reduced length of stay in OECD countries has been achieved with changed standards based on a better understanding of disease and recovery, changed financial incentives, and widely available modern pharmaceuticals. Through physician re- training programs, government information and management guid- ance, and appropriate financial incentives (see chapters 4 and 5), practices contributing to shorter stays could be encouraged by govern- ment policy choice. Hospital-acquired complications. Hospital infections and accidents add to extended lengths of stay, but reliable data on rates of hospital infection and their effect on average lengths of stay in developing countries are lacking. Overcrowding, poor sanitation, limited use of disposable items, inadequate use of pressurized steam sterilizers, the misuse of antibiotics, and the use of contaminated blood products all contribute to hospital- acquired medical problems and consequent longer hospital stays. Wide- spread inadequate maintenance is an important cause of conditions leading to accidents and infection. Policy intervention is also needed to provide explicit guidelines on infection control and to enforce standards of care. A survey of twenty-one hospitals in China measured an overall rate of nosocomnial infection of 8.4 percent, with rates as high as 13.2 percent in surgery and 10.3 percent in orthopedics (Blanpain 1987). Despite these seemingly high rates, China's hospital environment may be better than in many other countries. Matching Facilities to Patient Care Needs The large average cost differences between high- and low-level hospi- tals, the inverse relationship between hospital level and occupancy rates, and the scope for further increases of the turnover rate at low-level hospitals indicates that there are economic gains to be obtained from improving the match between patient care needs and the type of facility at which patients seek treatment. An indication of the potential gain from successfully encouraging patients to use low-level rather than middle- level facilities is suggested in table 3-1. Using the averages for the sample of hospitals in Indonesia, and assuming that all cases are of the same complexity and that marginal cost equals average cost, we calculate that the saving (measured in 1988 U.S. dollars) per inpatient bed-day and outpatient visit would be $7.60 and $1.80, respectively. Similarly in Rwanda, the savings per inpatient bed-day and outpatient visit would be $6.20 and $1.60. There are two reasons why these figures provide only Hospital Costs and Efficiency 105 a rough estimate of the potential saving to be gained by such shifts in facility use. First, the average treatment cost of the patients shifted to lower-level facilities is likely to be somewhat less than the average for all patients at mid-level hospitals because the goal is to shift the patients needing basic care toward lower levels, and the average treatment cost of these patients is likely to be less than the average for all patients at mid-level hospitals. The second reason, which perhaps offsets the first, is that marginal costs are likely to be greater than the average cost in mid-level hospitals with high occupancy rates, whereas marginal costs are likely to be lower than average costs in low-level hospitals, which often have lower occupancy rates. Despite these provisos, if patients needing basic care are shifted and appropriate treatment can be pro- vided, they will be treated in facilities with lower overhead costs, and it is likely that a less costly combination of inputs will be used for treat- ment. This systemwide efficiency gain would provide savings that could be translated into the provision of services to a greater number of clients. If the patients who have less severe conditions are moved out of tertiary facilities, the remaining case mix in these hospitals will be more costly, on average, than previously, so average costs will probably rise. This is desirable, however, because these hospitals will then be able to concentrate their efforts on the complicated cases for which they were intended. It is expected that average costs in lower-level hospitals will fall, as considerable underused capacity (as evidenced by low occupancy rates) is activated. The magnitude of some of the calculated average cost differences between types of facilities in a country is probably a function of the level of resources available at each level. Successfully encouraging consumers to use lower-level facilities would probably entail increasing the supply of drugs and staff to these hospitals, which might increase their average costs, though they would still likely be less than those in large urban facilities.6 Nevertheless, a shift of less severely ill patients from crowded urban tertiary hospitals into lesser-used secondary facil- ities would represent an improvement in resource allocation in the sector. There is some evidence that self-referral becomes an increasing prob- lem as economic conditions, education, transportation, and communica- tions improve, because these improvements increase the accessibility of higher-level facilities. This phenomenon will occur as long as nothing is done to alter the perception of the quality of services available in basic facilities. Mechanisms for improving the referral system are related to the quality of services at lower levels and to the financing of services pro- vided at each level. Most important, an effective referral system requires the availability of lower-level services in which consumers have confi- dence. The means by which services at each level are financed can serve 106 Public Hospitals in Developing Countries to encourage or discourage appropriate use of the referral system. Achieving a balance between demand and service availability at each hospital level depends on a system of relative prices, fee penalties for nonreferred entry at upper levels, and enforcement of referral that is in balance with the quality of services. Implementation of imnproved refer- ral policies is facilitated by a district and regional health management structure that includes both primary and secondary care under a com- mon administrative and financial unit responsible for all health facilities and programs in the region. Simultaneous Use of Hospital Service Indicators to Assess Efficiency Pab6n Lasso (1986) devised a graphical technique to summarize the three inpatient service indicators-occupancy rates, turnover rates, and ALOS-for similar levels of hospitals within a country in order rapidly to assess their relative performance. In figure 3-2 we adapt his graphical technique to describe performance measures across countries. The data represent either the average of a sample of hospitals or the average for all hospitals in the selected countries. The x-axis is the average bed occupancy rate, and the y-axis represents the annual bed turnover rate. Because of the mathematical relationship among these three indicators of hospital performance, a ray drawn from the origin that passes through any point on the graph represents a constant average length of stay, and this measure increases monotonically from left to right across the top and down the right-hand side of the graph.7 The graph is divided into four regions by two intersecting lines drawn from the mean values of the bed occupancy and turnover rates (which in turn identify the mean value of the average length of stay). Alterna- tively, normative values of at least two of these three indicators of hospital performance could be used as the basis for subdividing the graph. Within countries, the division of the graph into four quadrants is useful to identify hospitals that are outliers and demand specific atten- tion. Among countries it is useful to give some idea of the relative performance of the hospital subsector. (The cross-country interpretation needs to be made cautiously because country averages disguise large internal variations.) Pab6n Lasso (1986) describes some of the character- istics that could be expected to be found in hospitals falling in each region. As they apply to figure 3-2, hospitals in the countries of region I may be characterized by excess bed availability, low demand for hospi- talization in relation to installed capacity, and possibly demand that has been reduced by patients being diverted to other institutions. Hospitals in the countries of region II may have some or all of the following characteristics: excess bed availability, unnecessary hospitalizations, many beds used for patient observation, and a predominance of normal Hospital Costs and Efficiency 107 Figure 3-2. Indicators of Hospital Performance, Selected Countries Bed turnover rate Average length of stay (days) 60 4.5 5.8 6.6 7.1 50 II Jordan Sr 7. -7.8 * Fiji * Zimbabwe 409. 4 Belize . ; 190. ,Jamaica 10.3 Indonesiag U.-. 11.5 30 1? 3 Tunisia * Turkey,* 14.1 . TurkeyX * , - ,apua New Gunea 20 .thiopia a * - ,iger . . ' . '. .- * -'. -Argentima 19.1 Korea China IV 10 . 0 ,, . - l l I 0 10 20 30 40 50 60 70 80 90 100 Average bed occupancy (percent) Source: See table 3-3. (as opposed to complicated) deliveries. Region III countries have hospi- tals that are performing relatively well, on average (though perhaps not compared with normative concepts of what occupancy rates and aver- age lengths of stay should be). They are characterized by a relatively small proportion of unused beds. Hospitals in countries in region IV tend to exhibit some or all of the following: a high proportion of severely ill patients, a predominance of chronic cases, and unnecessarily long inpa- tient stays. For international comparisons, this graphical technique is more useful for descriptive than for policy purposes. Figure 3-2 merely describes countrywide averages; it does not suggest specific problems to be ad- 108 Public Hospitals in Developing Countries dressed. As an operational tool, it can be applied on a national level as a quick identification guide to hospitals that seem to be performing poorly or very well (see appendix 3-2 for an application to Indonesia). Some of the caveats that apply to the interpretation of unit cost information, however, are also relevant here, especially the need to consider case mix. For example, it would be expected that a tertiary hospital would be clinically composed of more severe cases than would a district hospital and that a greater average length of stay in the tertiary hospital would be one manifestation of this greater severity. The effect of differences in case mix across hospitals can be mitigated somewhat by grouping hos- pitals according to type and creating one graph per hospital type, as in appendix 3-2. Outliers could then be identified and investigated as to whether they deviate from the norm because of differences in case mix, quality of care, or other "legitimate" reasons, or because of differences in technical or economic efficiency. Efficiency of Input Use Inefficiency of input use can lead to high costs per unit of service delivered to patients. In chapter 2 we defined two important sources of inefficiency. The first is technical inefficiency, in which output is less than is technically possible with the mix of inputs used by the hospital (that is, if the institution is operating on the "frontier" of its production possibilities). The second is economic inefficiency, in which the institu- tion may be technically efficient but is not economically efficient in the sense that it is not using the least expensive combination of inputs for its given production of services. Technical Inefficiency Technical inefficiency in the use of staff, supplies, and equipment has been cited in many studies of hospital operations. Defining staff assign- ments too narrowly can restrict the full use of staff time in district-level facilities that have insufficient demand for specialized skills. For exam- ple, in a field study of six district hospitals in Malawi, X-ray technicians were used only about half time in their specialty, yet their excess time could not be used in other departments because they lacked qualifica- tions or were unwilling to be reassigned on a part-time basis (Mills, Njoloma, and Chisimbi 1989). This example comes from hospitals with high occupancy, where the effect of technical inefficiency on the quality of health services is likely to be greatest. Nevertheless, the problem of staff inflexibility is more apparent in hospitals with low occupancy rates and high ratios of staff per bed. Broadening the role of lower-level staff to include more support for primary health care activities could add to Hospital Costs and Efficiency 109 the productivity of both hospital and PHC operations. Extending the role of lower-level hospital staff is discussed further in chapter 6. Theft, bribery, and fraud are serious problems in the management of medical and nonmedical supplies, such as pharmaceuticals and food, in some countries. Security problems can occur at any point in the distri- bution chain, but hospitals are particularly vulnerable to losses due to petty theft by staff of items for personal use, diversion to the black market, and private medical practices. Poor storage is also a source of pharmaceutical loss in hospitals. Use of careful inventory and account- ing systems and well-maintained and secure storage areas is a prerequi- site for efficient drug and supplies management.8 A large number of sources of technical inefficiency in drug use can be identified. They indude extravagant prescribing (a less expensive drug or generic could be used with comparable effect), overprescribing (the drug is not needed or is taken in too large a dose or for too long a period), incorrect prescribing (the wrong drug is prescribed for the diagnosis or it is improperly prepared), multiple prescribing (two or more drugs are used when one would have the same effect), and underprescribing (the dosage or length of treatment is inadequate). The effect of these practices is to increase unnecessarily the cost of effective treatment, as well as to create deleterious health effects. Training, appropriate guidelines, and management surveillance are needed to reduce technical inefficiency in drug use. Low levels of use and incorrect use of equipment are also sources of technical (and economic) inefficiency at all levels of hospitals. In some cases, especially in low- and middle-level hospitals, the equipment is little used or misused because it is inappropriate. For example, in China, radiological equipment at district and provincial hospitals and compu- terized axial tomography (CAT) scanners, neonatal intensive care, and dialysis machines at provincial and central hospitals are often unused or inappropriately used (Banta 1987). Fluoroscopes are often used for rou- tine examinations, although they should be reserved for specific prob- lems not resolved by X ray. Expensive 500-milliampere X-ray machines are used rather than the safer and cheaper WHO Basic Radiological System (BRS) (Palmer 1985), which should be all that is needed at the lowest level. In other cases there is a lack of trained personnel to make adequate use of equipment. Ultrasound, for example, is inexpensive and powerful and thus a tempting device to have at low- and middle-level hospitals. Yet reading the ultrasound image requires experience and more training (six months) than often is available. Another source of technical inefficiency associated with equipment is a lack of uniformity and consistency in supply. A country may receive medical equipment from donations or under multilateral or bilateral assistance programs. Frequently the result is a variety of items from all 110 Public Hospitals in Developing Countries over the world and with considerably different technical specifications. Acquiring spare parts and managing the maintenance of, for example, ten different types of X-ray machines from different countries can be extremely difficult. This is an area in which a health ministry must actively manage its donors. Because all these sources of inefficiency involve a technical waste of resources, it is conceptually possible to obtain more hospital services without any increase in input. Unfortunately, achieving this conceptual possibility can provide a significant challenge; in most cases the increase in output cannot be obtained without deliberate and potentially difficult institutional changes. The inefficiencies that exist have arisen because of the incentives inherent in the existing system of financing health ser- vices, a lack of incentives for management or workers, a lack of manage- ment skills, a lack of information, or a lack of forethought in investment. Economic Inefficiency The production of services, including those of hospitals, involves a choice among alternative techniques and combinations of inputs. Even if the hospital is operating with technical efficiency, hospital managers and service providers are faced with a range of means of achieving a given service level. Ideally, the choice will be economically efficient, that is, the combination chosen to produce a certain quantity and quality of output will have the lowest cost among the range of alternatives. The problem of input choice can be stated as substituting one input for another while keeping the output (quantity and quality of services) the same. Input substitution will occur until the incremental value of ser- vices from another unit of expenditure on an input is the same for all inputs. Unfortunately, the lowest cost combination of inputs is often not achieved. The failure to use the economically efficient combination relates to many of the same reasons as the failure to achieve technical efficiency. These include deficiencies in incentives, lack of necessary training or skills, institutional constraints beyond the control of hospital managers (for example, rigid budgetary policies that expressly disallow input substitution), shortages in the flow of real resources to the hospital, and failures in information or investment planning. Overarching institutional constraints, such as civil service or budget- ary regulations, and failures of information and planning must be ad- dressed through coordinated and consistent national policies. Problems of economic inefficiency are generally outside the control of individual hospital managers. It is often at the level of the health ministry that the issues of the appropriate types and quantities of facilities, staff, and equipment must be answered. For example, the least-cost way to im- prove service coverage may be to convert underused district hospitals Hospital Costs and Efficiency 111 to health centers and reallocate staff. Alternatively, in rural facilities it may be possible to install basic diagnostic equipment that can be oper- ated by lesser-trained technicians and to reallocate more specialized equipment and staff to larger facilities. Analysis of which alternative would achieve the desired population coverage at least cost is not a job for individual hospitals. Regional or national issues of economic effi- ciency in the health sector involve decisions regarding many issues that affect the performance of individual hospitals but are not within their power to control. Common problems with economic efficiency can be illustrated by an examination of some substitution choices among important input com- binations. In the next few paragraphs we consider the possibilities of substitution between different categories of labor, between equipment or pharmaceuticals and labor, and between maintenance and other expenditure categories. Substitution among categories of labor. Professional standards of care and accepted practices provide restrictions on service tasks assigned to dif- ferent categories of medical care staff in a hospital. Most affected are the range of activities undertaken by nurses and paramedical staff compared with the accepted role of medical doctors, as well as the roles within a field or discipline (such as the role of registered nurses compared with that of assistant nurses, or the role of professionally trained radio- graphers compared with that of X-ray technicians). Possibilities for substitution between and within these categories of staff exist, but they are often limited by roles defined by professional associations. The quantity of each type of staff person can be related to hospital size and translated into ratios of staff to beds. There are no internationally accepted norms for staffing ratios, and there probably should not be any because staffing choices must be made in the context of local constraints and wage levels. An examination of staffing ratios (table 34) reveals cross-country variation in the total staff per bed and, for those countries for which information is available, in the composition of staff. As ex- pected, the limited data support the conclusion that higher-level hospi- tals tend to employ more staff per bed than lower-level hospitals. In particular, the table supports the observation that physicians are concen- trated in large, tertiary hospitals. Nurses and paramedical personnel make up the largest component of hospital staff, although in Indonesia, Jamaica, and Niger the number of other (nonmedical) workers is sub- stantial. Staffing ratios per bed or bed-day are not an infallible proxy for quality of services. Training and skill level, supporting technology, team work, and the organization of services are all essential complementary co-de- terminants of quality. In addition, expected differences in case mix Table 3-4. Hospital Staffing Characteristics and Related Data, Selected Countries, Selected Hospitals Staff per bed Occupancy Nurses Nurmber Level rate Bed-days and of hospitals Coumtry and year of data of hospital ( per staff Physicians Parameds Other Total in study Source Belize, 1985 11 68 218 0.1 0.9 0.1 1.1 1 Raymond and others 1987 III 31 225 0.1 0.4 0.0 0.5 6 Colombia, 1979 1,11 68 100 0.2 1.4 1.0 2.6 8 PRIDES 1980 China, 1986 1 94 177 - - - 1.9 8 Barnum 1989 11 86 172 - - - 1.8 11 III 95 232 - - - 1.5 7 Dominican Republic, 1989 1 - - 0.9 0.8 0.4 2.1 1 Lewis, Sulvetta, and LaForgia 1990 Fiji, 1987a 1 83 225 0.2 1.0 0.1 1.4 3 III 46 176 0.1 0.8 0.1 1.0 19 Indonesia, 1985 1 75 97 0.6 1.0 1.2 2.8 2 Barnum 1987 II 68 118 0.4 0.9 0.9 2.1 15 III 54 197 0.1 0.6 0.4 1.0 297 Jamaica, 1985-86b I 84 160 0.2 1.4 0.4 1.9 2 Kutzin 1989 II 87 238 0.1 0.7 0.5 1.3 2 III 80 158 0.1 0.9 0.9 1.8 1 Niger, 1986-87 1 87 476 0.1 0.3 0.3 0.7 1 Wong 1989 Papua New Guinea, 1988 I 80 328 0.1 0.6 0.2 0.9 1 JSI 1990 II 80 276 0.1 0.6 0.4 1.1 4 III 60 287 0.0 0.4 0.3 0.8 8 - Not available. a. Staffing data refer to number of established posts. b. Staffing data refer to number of established posts as of August 1988. Source: World bank sector reviews, except as noted in table. Hospital Costs and Efficiency 113 between hospitals suggest that tertiary hospitals require greater staffing intensity than district hospitals. Staffing ratios are, however, an import- ant indicator of hospital performance, and low staffing ratios in district hospitals remain an impediment to adequate service provision in rural areas in many countries. Substitution of equipment and pharmaceuticals for labor. One reason that ratios of staff to beds vary considerably from country to country, even if quality is roughly comparable, may be the flexibility in the choice of input mix made possible by some forms of new hospital technology. Monitoring equipment can be used in intensive care departments and in wards to replace some nurses, freeing them for other patient care activ- ities. Automated laboratory equipment, such as that for urinalysis and hemoanalysis, can increase the effectiveness and productivity of labora- tory workers. The kitchen and laundry can also involve capital-labor substitution possibilities with the introduction of electric, gas, or steam- driven appliances to replace wood fires and hand labor. The potential for efficiency-improving input substitution exists, but it should not be overstated. Although some technological breakthroughs can be charac- terized as labor-saving, new technology often requires as much labor as older equipment does, and often of a higher level of skill. Country- specific conditions would determine the desirability of such substitu- tion. Pharmaceuticals can also reduce required staff or increase staff effi- ciency. Psychiatric drugs render patients more controllable with fewer workers and less-secure facilities. Drugs to reduce infection and increase resistance after surgery can reduce length of stay. The use of drugs, for example, rifampicin or ethambutol in tuberculosis treatment, sometimes makes outpatient care possible for illnesses that previously required long inpatient stays (Barnum 1986). Substitution between maintenance and other expenditures. Poor mainte- nance of buildings, equipment, and vehicles is a ubiquitous problem. Maintenance as a share of total recurrent expenditures is often budgeted at less than 4 percent of recurrent costs (for example, see Mills 1991). The level of required maintenance depends on the operating environment and the complexity of the facility, but an estimate of the desirable level of maintenance expenditure for a secondary referral hospital is between 10 and 15 percent of annual recurrent cost.9 Even if maintenance is budgeted at this level, the actual expenditure is often less, as mainte- nance is typically the first item to suffer cutbacks in the face of budget shortfalls. Although technically classified as economic inefficiency, un- derallocation to maintenance has important implications for the overall technical efficiency of the hospital. For example, insufficient mainte- 114 Public Hospitals in Developing Countries nance can lead to broken steam units or inoperable X-ray equipment or vehicles, which contribute to the inefficiency of the remaining 96 to 97 percent of recurrent expenditures. Conclusion. The examples of input substitution given above do not constitute specific recommendations for change. No prejudgment should be made of the appropriate factor mix because what is appropri- ate will vary across countries and hospitals. For some hospital services, substitution of equipment for labor may reduce diagnostic or therapeutic errors; however, such substitution may also increase errors in other services or in the same services in other hospitals. The choice should be made on the basis of economic efficiency. Careful country-specific anal- ysis needs to be carried out to determine the relative cost-effectiveness of alternative mixes of factors. Statistical Cost Studies The minimum cost of providing a specified set of services when input prices are given can be represented by a point on a cost function. The cost function summarizes the economics of production and can be used to determine the cost of both an additional unit of output (marginal cost) and an average unit and to describe the possibilities of economies or diseconomies of scale or scope. Although hospital managers do not, of course, know the exact nature of their cost function or even of its existence, by their choices through time they necessarily operate within its confines and, if their incentives are sufficiently strong, they operate on or near it. Knowledge of the cost function or at least the rough magnitudes of some of its parameters is especially important for central policymakers because it promotes the setting of policies that are consis- tent with economic reality. By examining data for a large number of hospitals we can hope to discover some of the characteristics of cost functions. This section summarizes the findings from some statistical cost functions for developing countries. Modeling Hospital Cost Accounting studies of costs, such as those reviewed in the first section in this chapter, have an implicit underlying cost function, represented by the sum of the products of the quantity of each input multiplied by its respective price.10 In fact, such an accounting cost function represents the cost of production, at one point in time, on a cost function. The accounting view of the cost function-sometimes referred to as a "pas- sive" cost function-is rigid and does not allow management or techni- cal responses to changes in input prices or quantities. Figure 3-3 depicts Hospital Costs and Efficiency 115 Figure 3-3. Accounting Cost Function in Contrast to Actual Cost Function Cost Actual cost function (one of many possibilities) g Accounting cost function Output the relation between an accounting cost function and one possibility for the actual underlying cost function. An accounting study generates a point estimate of total costs at the observed output, as at point A in the figure, but does not tell much about what is likely to happen when the price or quantity of an input changes. The only point we are certain that the actual and accounting cost functions have in common, if the hospital is minimizing costs, is the cost and quantity combination depicted at point A. The implicit assumption of an accounting cost function is that the underlying cost curve is linear and, thus, that marginal costs are constant. This assumption may or may not be true and must be tested. Econometric models of cost and production can be specified suffi- ciently flexibly that they allow the assumption of constant marginal cost and similar rigid assumptions inherent in the accounting model to be tested. Econometric or statistical models also provide a better depiction of how total costs change in response to differences in service mix, inputs, input prices, and scale of operations. For example, the accounting model would show a direct and simple relation between an increase in a factor price and an increase in total costs; that is, total costs would rise 116 Public Hospitals in Developing Countries by the increase in price multiplied by the quantity of the factor. The econometric model would let total costs rise at a rate less than that of the price increase, to allow for substitution of other factors and a reduced quantity of the input for which the price rose. This cost function is curved, not linear as in the accounting model. In spite of the potential usefulness of econometric cost and production studies, very few such studies of hospitals in developing countries have been carried out thus far. Data deficiencies have been the primary constraint. Few hospitals regularly collect the information needed for such studies, and inconsistencies among hospitals limit the usefulness of the data that are gathered. To a certain extent, statistical methods are robust enough to allow data problems to be absorbed in error terms or imprecision of estimates, but limits on this absorption of poor data quality are soon reached, and the statistical confidence that can be placed in most of the studies to date is low. Another important limitation in econometric studies is that they are built largely on an economic model that assumes the hospital is minimizing costs at some point along its production frontier. Econometric studies also assume that it is fairly easy for managers to change their factor input mix or their capital stock in response to the economic environmnent. Furthermore, the means by which hospitals are financed have specific implications for the behavior of hospital managers and the characteristics of hospital costs, but the econometric models formulated so far have not been able fully to incor- porate the influence of finance. In reality, hospital managers may be insufficiently motivated to operate on the production or cost frontier and instead may be operating comfortably and placidly inside. Or, even if sufficiently motivated, managers may be severely constrained by poli- cies set outside their influence. Finally, as with accounting-based studies, econometric studies suffer from the difficulty of controlling for the quality of output. Thus, the results of the studies to be discussed below are interpreted with caution. Estimation of hospital cost functions is made difficult by the lack of a clear model of hospital behavior that can be used to interpret the relation between cost and output (service) data. The difficulty of developing an economic view of the hospital lies in the inadequacy of the conventional profit maximization model of the competitive market to explain the incentives, and thus the objectives and behavior, of hospital managers. Many alternative views of hospitals as economic entities have been proposed, ranging from the appealing simplicity of the model suggested by Newhouse (1970)-which proposes that hospitals achieve least-cost production as they strive to maximize a combination of quantity and quality of output constrained by a budget-to a complex, but institution- ally apt, model suggested by Harris. Harris (1977) suggests that many hospitals can be characterized as actually composed of two firms-the Hospital Costs and Efficiency 117 first consisting of the medical staff, which delivers the services, and the second consisting of the administrative staff. The input mix, determined in advance by the administrative staff, defines the capabilities of the hospital to meet the demand identified by the medical staff. This view of a hospital may not be appropriate for many hospitals in developing countries, in which the administrative staff and service staff often over- lap considerably. It is useful, however, because it recognizes that the demand for services is controlled by the service staff, especially physi- cians, with resource constraints imposed by managers (centrally or locally). A not uncommon institutional arrangement is that medical staff income is dependent, at least partly, on the volume and type of services offered by the hospital. Thus, this model replaces as a hospital objective the profit maximization of the conventional economic model with the income maximization of physician cum manager. Still other models have been proposed to explain the incentives of managers in nonprofit organizations in general or hospitals in particular. Pauly (1987) reviews many possible specifications of hospital objectives and notes that, with the current econometric formulations now in use, these models cannot be distinguished from each other. He condudes that the differences in the models may be important for policy decisions involving incentives and hospital pricing behavior, but the models have similar behavioral implications with regard to cost. A complete review of possible models is potentially large. Indeed, there is no one model that would be appropriate across the tremendous diversity of institutional settings that exist in developing (or industrial) countries. To indicate the variety, a partial list of alternative behavioral models can be given: * Maximization of output (patient admissions), given a fixed budget * Maximization of some function of output and quality of care (as- suming a tradeoff between the two) e Minimization of cost, given an exogenous demand for admissions e Maximiization of some function of profit and output e Maximization of institutional prestige, which is a function of hos- pital size, facilities, and the prestige of associated physicians * Nonmaximization-so-called "satisficing" models of behavior-in which managers and staff only hope to achieve some level of output and quality within a fixed budget that will satisfy their own expecta- tions and those of higher-level managers. Extending Pauly's observations, all these alternatives, with the possi- ble exception of the last two, are consistent with a close empirical relation between total costs and hospital output. The last two models can lead to behavior that obscures the cost-output relationship but, more likely, they 118 Public Hospitals in Developing Countries coexist with budget constrained behavior circumscribed by a cost func- tion. Our expectation is, therefore, that even in the absence of an under- lying maximization objective that is universally applicable, a functional relation exists between observed hospital costs and output. The rela- tively close fits of the econometric functions estimated below sustain that expectation. During the last twenty years, estimation of cost functions for hospitals in industrial countries has become commonplace. Reviews by Cowing, Holtmann, and Powers (1983); Wagstaff (1989a, 1989b), and Breyer (1987) document the progress. The earliest attempts to estimate cost functions using data from hospitals in industrial countries employed specifications of the regression equation that used composite measures of hospital output (for example, Cohen 1967), used average or unit costs of inpatient-day or admission as the dependent variable (Feldstein 1968), and included a variety of interrelated explanatory variables such as occupancy rates, patient flow, length of stay, and capacity as explanatory variables (Mann and Yett 1968). The primary reason the authors of these studies adopted the average cost formulation was to avoid the potential problem of error terms with a nonuniform variance (heteroscedasticity) in the estimated regression. (When total rather than average cost is used as the dependent variable, there is some potential that the error term may be correlated with the size of the hospital.) In these estimates, the functional form used in the specification is not derived from a structure based in theory but is generally defined for convenience of estimation as either log-linear or additive-linear. In more recent attempts, authors have specified a functional form and included variables that are consistent with a theoretical production structure (for example, Cowing and Holtmann 1983; Grannemann, Brown, and Pauly 1986; and Vita 1990). They have generally estimated total cost (rather than average cost) functions with multiple outputs and have employed flexible functional forms. Although the total cost speci- fication can present the problem of heteroscedasticity, this can be dealt with if it arises through the proper selection of an estimation procedure. Some advantages of the more recent work compared with the earlier work are (a) the multiproduct nature of the hospital is explicitly recog- nized, (b) an economic interpretation of functional form and induded variables is provided, (c) econometric problems created by using output (inpatient bed-days) on both sides of the estimating equation are avoided. Authors of recent cost studies have also been more careful to distin- guish between short- and long-run cost functions. This distinction is important for the identification of economies of scale, which is a long-run concept. In empirical applications the distinction between long- and short-run cost functions is related to the specification of the time period Hospital Costs and Efficiency 119 and the inclusion of a scale or capital proxy as an independent variable. If the hospital cannot adjust the capital stock within the period of time defined by the data, as would commonly be the case if the cost and output data refer to a one-year accounting period, then a short-run variable cost function is appropriate. It is possible to examine the behav- ior of the variable cost function with changes in the capital stock (in this case the number of beds) and derive a long-run function (see Vita 1990). The importance of this is that it is then possible to derive an index of long-run economies of scale from an estimated short-run variable cost function. For heuristic purposes (though it may not be descriptive of the actual shape of a given estimated function), we show in figure 3-4 a general graphical description of the relation between long- and short-run cost curves, presented in average cost format, for a single output. With reference to the figure, we can distinguish between the short- and long-run behavior of cost with changes in output by differentiating between returns to scale (a long-run concept) and returns to the variable factor (a short-run concept). Looking first at different locations along the long-run average cost function (LRAC), we see that the hospital with the short-run average cost (SRAC 1) curve is at an optimum plant size and can be characterized as having constant returns to scale when producing the long-run minimum cost output at A. To the left of A on the LRAC, average cost falls with increases in hospital size (increasing returns to scale). To the right of A, average cost increases with hospital size (decreasing returns to scale). Looking at locations on a short-run average cost function, for example, SRAC 2, we see that, given the fixed scale, the hospital is at its most efficient level of output at B and is above this short-run optimum at, say, C. At C, the cost of an additional unit-the marginal cost (MC)-is above the average cost. At point B the hospital is at a point of constant returns to the variable factor. To the left of B, short-run average cost decreases with greater output, yielding increasing returns to the variable factor, and to the right of B, say at C, there are decreasing returns to the variable factor. Thus we distinguish between movements along the short-run cost curve by discussing returns to the variable factor and movements along the long-run cost curve by discussing returns to scale. This distinction is not always carefully made in the literature on hospital cost functions. Vita (1990) has recently pointed out that failure to make this distinction has led to errors in estimating economies of scale and that such failure is caused by applying long-run formulas to short-run empirical cost functions. To avoid this problem we explicitly define short- and long-run formulas. Our central interest in estimated cost functions is in obtaining infor- mation on the magnitude of average and marginal cost, on returns to the 120 Public Hospitals in Developing Countries Figure 3-4. Relation of Short- and Long-Run Average Cost Functions Cost LRAC \ ~~~~~ ~~SRAC2/ Output variable factor, and on the importance of economies of scale and scope. These concepts can be defined in terms of a general specification of a short-run cost function, C = ftBeds,Yl, Y2,.. . Yn), where Beds identifies the scale of the short-run function, C is the total cost of hospital opera- tions, and the Yi are the hospital's n outputs. We then have the following definitions. For the short run: 1. The marginal cost of producing an additional unit of the ith output: MC, = aC/aYi. 2. The average incremental cost of the ith output: AIC = [C - C (Yn -i) ]/Yi where Yn-i is the total cost of production with the exclusion of the ith product. The average incremental cost is the average added cost per unit Hospital Costs and Efficiency 121 of producing the ith product in comparison with producing all products except the ith. 3. Short-run returns to the variable factor: SRVF = C/1YiciYi. The index of short-run returns to the variable factor measures the effect on costs of a general increase in output when the output mix and bed size remain fixed. If the SRVF is greater than one, the level of output is below optimum efficiency. If the SRVF is less than one the level of output is above maximum efficiency. 4. Short-run product-specific returns to the variable factor: SPRVF = A1C/MC,. The indexes of product-specific returns to the variable factor measure the effect on costs of a proportional increase in all inputs on the output of the ith product while the level of output of all other products remains constant. Product-specific returns to the variable factor are said to exist if the SPRVF is greater than one. 5. Economies of scope: SCOPEs = [C(YJ) + C(Yn _ -C(Y) ]/C(Y). Economies of scope exist when it is cheaper to produce selected outputs jointly than to do so separately. Economies of scope between a subset of outputs (Y,) and all other outputs (Yn-s) exist when SCOPE S is greater than zero. In addition to these definitions, we have for the long run: 6. Economies of scale: With beds included as a proxy for scale, the measure of long-run economies of scale (EOS) is EOS = (1 -aCBeds)/ GCyi where Gab indicates the elasticity of a with respect to b. The index of economies of scale measures the effect on cost of a general increase in output when the output mix remains unchanged and all inputs are allowed to vary.1' If the EOS is greater than one, economies of scale are said to exist, if less than one, diseconomies exist. We now turn to estimated cost functions for developing countries and, where possible, apply these formulas to derive estimates of marginal cost, returns to the variable factor, and economies of scale and scope. 122 Public Hospitals in Developing Countries Five Studies of Developing Countries The literature on developing countries follows much the same evolution- ary pattern as that for industrial countries, but in this case there are only a handful of studies. In chronological order the studies known to us cover hospitals in Kenya (Anderson 1980), Ethiopia (Bitran-Dicowsky and Dunlop 1989), Nigeria (Wouters forthcoming). In addition to these we add our own analyses of data for Colombia and China. It is evidence of the relatively recent emergence of this topic that only one of the studies (Anderson's) was obtained from the published literature.12 Individually, several of these studies suffer from data inadequacies that cloud an interpretation of results; collectively, however, the studies indicate the variation in the magnitude of some critical cost parameters across differ- ent countries and different health system environments. We first review the five studies and then briefly discuss the general implications of the findings. Kenya. The study of hospital costs in Kenya is in the tradition of the earliest studies done in industrial countries and uses an average cost specification: I = aO [365.4' [-16 j [jAT B4 ZFSZ where C/I = Costs (C) per inpatient bed-day (1), I/(365-B) = Occupancy Rate, B represents beds, 0/1 = Outpatient visits (0) per inpatient bed-day, I/A = Length of stay, A represents admissions, ZF = Weighted index of the number of associated subhospital facilities, ZL = Level of hospital, 1 = province, 0 = district. The results of the Kenyan study are clouded by the fact that hospital costs include the costs of related subcenters (health centers and health posts) as well as those of the specified hospital, and it is not possible to match the estimated coefficients clearly with the hospital services per se. An attempt was made to correct for this problem by including a weighted average (ZF) of the number of subcenters as an explanatory variable. The data set included fifty-one provincial and district public hospitals. Natural logs were taken of all variables before regression so that the estimated coefficients could be interpreted directly as elasticities. The ordinary least-squares estimates of the coefficients are given in table 3-5. Anderson does not use the estimated equation to calculate the mar- ginal cost of services, and the mean values for the variables employed in the analysis are not reported. He interprets the significant negative coefficient on beds (B) to imply long-run economies of scale with regard Hospital Costs and Efficiency 123 to hospital size, but the specified equation does not yield a direct estimate of economies of scale because of the unit cost formulation and the several places in the specified cost function that inpatient bed-days (I) occur. Additionally, Anderson's interpretation of the coefficient on beds is not strictly correct; the estimated function is a short-run cost function, and allowing bed size to change will account for movements from one short-run cost curve to another, keeping output constant, but will not give movements along the long-run cost curve. By multiplying through by I and rearranging terms, however, we can put the estimated equation in a total cost form. The rearranged equation can be used to derive the elasticities with respect to output and beds that are needed to obtain a measure of long-run economies of scale (EOS above). Following this method, we arrive at 1.5 as the estimate of the index of economies of scale. The policy implications are that, if additional capacity is to be constructed, moderate cost savings could be obtained by expanding existing facilities instead of building new small-scale hospitals. This is, of course, only one of several factors, such as accessi- bility, that planners would take into account in planning hospital con- struction. The function rearranged in total cost form yields an estimate of the index of general short-run returns to the variable factor of 2.0 as well as an estimate of the elasticity of unit cost with respect to inpatient bed-days of -0.8. This would seemingly indicate that Kenyan hospitals can be characterized by decreasing cost with greater output. It would be diffi- cult to reach the conclusion, however, that greater efficiency would be achieved by increasing hospital use and occupancy rates. Anderson points out that many Kenyan hospitals at the time of the study already had occupancy rates in excess of 100 percent and increased use would further compromise quality. Perhaps greater throughput could be achieved by shortening the length of stay; the coefficient on length of Table 3-5. Cost Function for Selected Hospitals in Kenya Variable Coefficient t-statistic Constant 6.57 I/B -0.44 -4.17 0/i 0.29 2.45 I/A -0.07 -0.69 B -0.20 -2.97 ZF 0.19 8.23 ZL 0.29 2.37 2 =0.75 n=51 Source: Anderson 1980. 124 Public Hospitals in Developing Countries stay (I/A) is negative but not statistically significant. It was not possible to calculate economies of scope. A further finding, however, was the significantly greater unit cost for provincial hospitals than for district hospitals. This probably reflects a mix of more severely ill patients in provincial hospitals or a greater allocation of resources per patient to provincial hospitals than to district hospitals. Ethiopia. As do recent studies of industrial countries, the study of Ethiopian hospitals uses a flexible functional form and estimates a short-run variable total cost function. The specification used is quadratic in outputs: C = e(ao +aB) eYX) where f(Y,X) = bjI + b212 + b30 + b402 + b,5I 0 + b6Del + b7Surg + b8Lab. All variables remain as defined previously with the addition of the number of deliveries, laboratory tests, and surgical operations repre- sented by Del, Lab, and Surg, respectively. The data set included a pooled cross-section-time-series data set of thirty-eight observations for fifteen public hospitals. Ordinary least-squares estimates of the coeffi- cients are given in table 3-6. The coefficients on outpatients and the square of bed-days are not statistically significant. Taken together, the high R2 and large standard errors of the estimated coefficients suggest that multicollinearity is a problem. The sign on the square of bed-days is incorrect and implies an inverted U-shaped average cost curve for bed-days. The peak of the implied average cost curve, however, is well outside the range of the data; throughout the range of the data, average cost increases slowly with increases in output. The main findings are that, at the sample average, marginal costs are only slightly above average costs for inpatient bed-days and for deliver- ies and laboratory services. Given the statistical variation in the sample it cannot be clearly established that marginal and average costs differ for inpatient services, and the sample of hospitals is best characterized by nearly constant (slightly decreasing) short-run returns to the variable factor. The study also identifies mild economies of scope between inpa- tient and outpatient services. Bitran-Dicowsky and Dunlop (1989) do not report estimated results for hospitals of different sizes but note that the output structure varies between large and small hospitals. They suggest that with additional data it would be important to disaggregate the sample by hospital size. Equally important would be a disaggregation by functional level. Hospital Costs and Efficiency 125 Table 3-6. Cost Function for Selected Hospitals in Ethiopia Variable Coefficient t-statistic Constant 5.45 22.51 B 4.71 E-3 8.89 I 2.18 E-5 3.44 12 -1.65 E-12 -0.02 0 1.91 E-6 0.08 02 1.42 E-10 0.26 I 0 -7.50 E-10 -2.42 Deliv 1.68 E4 5.39 Surg 3.21 E-5 0.11 Lab 7.63 E-6 7.97 i2 =0.96 n =38 Source: Bitran-Dicowsky and Dunlop 1989. Nigeria. The analysis in Nigeria covers twenty-four health institutions in Ogun State, of which eight are health centers with ten to twenty inpatient beds each, seven are maternities (about seven beds each), and nine are dispensaries. Three of these facilities (one health center and two maternities) are privately owned. For our purposes the lack of facilities classified as hospitals limits the applicability of the results. Interpretation of the results is also clouded by the small sample size. In spite of these limitations the study is of interest because it gives some information on the behavior of costs of inpatient care in small facilities and because it uses an econometric specification that explicitly recognizes that facilities may not be operating efficiently. In this technique the production func- tion, designated a frontier production function, is estimated as an enve- lope of observed points. Wouters (forthcoming) estimates a production function for outpatient visits to compute the marginal products of health workers and non-health workers and compares the ratio of marginal products with the ratio of wages to derive an efficiency index (a value of one represents optimum economic efficiency). The efficiency index is then used in a multiproduct cost function as a technically neutral shift variable. The estimated equation is log-linear. The specified cost equa- tion is, C = aO 0a1 zaa i, W a ' expfA,B) where f(A,B) = bo + b1A* + b2B* + b3D and all variables remain as defined previously with the addition of 126 Public Hospitals in Developing Countries Z = an index of drug availability as a measure of quality, WH = wage of health workers, WN = wage of a non-health worker, E = Efficiency index based on first stage estimate of a production function, D = dummy variable set to 0 if facility has no beds, * indicates a Box-Cox transformation (2 = 0.1) of the variable. The results of the cost function estimation are given in table 3-7. The main finding from the production function estimate (not reported here) is that the Ogun State facilities use substantially greater numbers of non-health workers than is economically efficient; the ratio of the mar- ginal product of non-health workers to that of health workers is about two-thirds of their ratio of wages. The main findings from the estimated cost function are short-run returns to the variable factor for admissions and approximately constant returns with respect to outpatient visits for the group of small facilities included in the analysis. That is, the marginal cost of admissions is substantially less than the average cost, whereas marginal and average costs for outpatient services are approximately equal. The study also indicates slightly decreasing short-run returns to the variable factor with respect to a neutral expansion of facility produc- tion. Long-run economies of scale were not calculated. Colombia. The Colombian cost function is derived from data given in a special survey (PRIDES 1980) evaluating the financial management of eight class I and II hospitals (hospitales de referencia) distributed through- out the country. The costs, given in 1975 constant pesos, and service data cover the period 1975 to 1978 and provide a pooled cross-section and time series data set of thirty-five observations. The function estimated is a short-run variable cost function and includes beds as a measure of Table 3-7. Cost Function for Selected Health Facilities in Ogun State, Nigeria Variable Coefficient t-statistic Constant 1.628 0.58 B 0.093 0.44 A' 0.011 0.25 0 0.597 5.28 z -1.361 -2.06 WH 0.586 2.46 WN 0.387 2.08 E -0.160 -0.88 D 0.221 0.62 R2=0.91 n=24 Source: Wouters forthcoming. Hospital Costs and Efficiency 127 scale. A functional form similar to that used by Grannemann, Brown, and Pauly (1986) was adopted for the analysis. The specified function is cubic in outputs: C=e e(a° + a,B) .,P(Yx) where f(Y,X) = b,I + b212 +b313 + b40 + b502 + b603 + b71. 0 and all variables remain as defined above. A preliminary estimation of this equation gave coefficients on the squared and cubed number of outpatient visits that were not statistically significant and of the wrong sign. Cubic equations often provide diffi- culty in estimation because of multicollinearity and the magnified effect of outliers on the squared and cubed terms. Reestimation of the cost function omitting the power terms on outpatients gives the results presented in table 3-8. The significance of the coefficient on outpatients remains low, but the sign and magnitude are plausible. With regard to inpatient costs, the findings are similar to those for the Ethiopian study. At the sample average, marginal costs for inpatient services are approximately equal to average costs, and the sample of hospitals exhibits constant short-run returns to the variable factor. With regard to outpatient services, marginal costs are again very close to average costs and there appear to be constant returns to the variable factor, although in this case the statistical weakness of the coefficient on outpatients reduces the confidence that can be placed in this result. The estimated equation also does not confirm the existence of economies of scope between inpatient and outpatient services. Calculated at the Table 3-8. Cost Function for Selected Hospitals in Colombia Variable Coefficient t-statistic Constant 2.76 10.34 B 3.96 E-4 1.16 1 1.73 E-5 3.27 12 -6.42 E-1 1 -1.74 13 8.78 E-17 1.44 0 3.34 E-06 1.28 I 0 -7.01 E-12 -0.39 R2 = 0.91 n = 35 Source Authors. 128 Public Hospitals in Developing Countries means of the data, the estimated value of EOS is close to 1 and suggests constant returns to scale. China. As a final study, a short-run variable cost function was esti- mated for the sample of thirty Chinese hospitals summarized in the discussion of accounting unit cost. The sample includes cost and service data for three years, 1984 through 1986. Cost data were converted to constant 1986 yuan. After exclusion for missing observations, the total sample size was seventy-two. Again, a flexible cost function similar to that used by Grannemann, Brown, and Pauly (1986) and cubic in outputs was used for the estimating equation, C e(ao+a,B) AjYAX) where f(Y,X) = bil + b20 + b3I2 + b413 + b502 + b6O3 + b71- 0 + bgDg4 + b9D85. The variables remain as previously defined with the addition of dummy variables, D84 and D85, set to 1 for years 1984 and 1985. After an analysis of the residuals, a maximum likelihood estimate of a frontier production function specification was rejected in favor of an ordinary least-squares estimate (see Aigner, Lovell, and Schmidt 1977). Also, the residuals were not found to be heteroscedastic with respect to bed size (Goldfield-Quandt test). The ordinary least-squares estimates are sum- marized in table 3-9. The main findings are diseconomies of scale and only mild economies of scope and short-run inefficiencies in the level of operation with respect to bed-days and outpatients. It is of interest to examine the returns to the variable factor and economies of scale and scope for different levels of hospitals. Table 3-10 gives the marginal cost, short-run returns to the variable factor, and economies of scope and long-run economies of scale for low-, middle-, and high-level Chinese hospitals. The estimates were obtained by sub- stituting the means for the groups in the estimated cost function. The short-run returns to the variable factor of 0.9 for lower-level hospitals and 0.7 for upper-level hospitals indicate that low- and middle-level hospitals are operating at or slightly above efficient volumes of output, whereas larger hospitals are clearly above an efficient volume of output. The index of long-run scale economies indicates that increases in the size of hospitals is not warranted and that, particularly in the upper level, a smaller scale would be more efficient. These results are consistent with the high levels of occupancy rates in most of the hospitals included in the sample (and in Chinese hospitals in general). Occupancy rates in China are very high, so output takes place Hospital Costs and Efficiency 129 Table 3.9. Cost Function for Selected Hospitals in China Variable Coefficient t-statistic Constant 12.93 71.54 B 6.12 E-4 0.72 1 2.03 E-5 4.63 12 -6.04 E-11 -1.90 13 1.16 E-16 1.56 O 2.01 E-6 1.49 02 -9.65 E-13 -0.35 03 14.72 E-18 0.83 1 0 -11.51 E-11 -2.71 D84 -0.166 -2.23 D85 -0.111 -1.53 2 =0.89 n = 72 Source Authors. well to the right of the optimum output on the cost function, in a region where marginal cost is greater than average cost. A mechanical conclu- sion would be that unit costs could be reduced by using a larger number of institutions and reducing the average occupancy slightly. This inter- pretation, however, fails to acknowledge the inordinately long stays in Chinese hospitals. The long stays are largely responsible for high occu- pancy rates, which would be much more modest if length of stay were managed more efficiently. In fact, the productivity of hospital beds in regard to the number of patients treated (reflected in the turnover rate) in Chinese hospitals is lower than in the hospitals of many other coun- Table 3-10. Marginal Cost and Economies of Scale and Scope by Level of Hospital from Cost Function for a Sample of Chinese Hospitals (1986 yuan) Lower level Middle level Upper level Marginal cost Bed-day 16.2 20.3 40.1 Outpatient visit 1.5 3.2 1.8 Indices of scale and output Short-run SRVF 0.9 0.9 0.7 Long-run EOS 0.8 0.7 0.5 Economies of scope Bed-days and outpatient visits 0.4 0.3 0.2 Note: Marginal costs are calculated from the estimated cost function in table 3-9 with all variables set at the means of the indicated group. 130 Public Hospitals in Developing Countries tries, despite generally much higher occupancy rates in China (see table 3-3). Given the long stays, reducing the cost per day through expanded bed capacity should not be the focus of policy. Reducing length of stay would be likely to increase costs per day slightly (because relatively less expensive days toward the end of the stay would be reduced), yet there would be overall efficiency gains because more patients would be ac- commodated and the cost per admission would be sharply reduced. The shortened length of stay could also result in a more efficient level of inpatient use. Discussion of Statistical Cost Functions The small number of cost function studies for developing countries that are available for the survey above precludes any clear generalization of results. The great variation in results demonstrates that the conclusions must remain country-specific. In addition, it is important to avoid a mechanical interpretation of the statistical results, such as the finding of inefficiency in high Chinese occupancy rates implied by marginal cost that exceeds average cost. The statistical functions are useful only if one has a knowledge of the context. The scope of the survey could be expanded to include results from industrial countries, but such an expansion should only be done with caution. The greatly restricted budgets of hospitals in developing coun- tries and the limited capacity for training skilled personnel in medical and nursing schools constrain the production technology choices to a much smaller range of activities than in industrial countries, and the underlying production functions are probably very different, especially in low- and middle-level hospitals. Furthermore, hospital managers will respond differently according to the manner in which hospital services are financed. The results from analyses of industrial countries might be of more relevance in the higher-income South American economies and in the emerging countries of the Pacific rim. Some similarities in the findings between industrial and developing countries do, however, appear indicative. An important debate running throughout the literature of industrial countries concerns the extent of economies of scale. Significantly, the earlier literature, using the single product, unit cost specification, commonly identified significant econo- mies of scale or, strictly speaking, of size. But these findings were criticized, as we noted earlier, because of the poor theoretical specifica- tion of the models upon which they were based. The view of economies of scale changed substantially with the introduction of the more flexible estimation forms and the use of total cost, multiproduct specifications. In particular, some of the later studies found constant or mildly dimin- ishing returns to scale, especially for larger hospitals. The general con- Hospital Costs and Efficiency 131 clusion of Cowing, Holtmann, and Powers (1983, p. 276) in their survey of industrial country studies is that "economies of scale may exist for small hospitals but that moderate and large size hospitals can generally be characterized by constant returns to scale." The results from the five studies of developing countries surveyed here are consistent with the literature concerning industrial countries. With regard to long-run economies of scale, Anderson, using a unit-cost specification in his study of Kenyan hospitals, found economies of scale. However, we used flexible cost functions in our analyses for Colombia and China and found either diseconomies of scale or constant returns to scale at the sample averages. Table 3-11 summarizes the results from the four flexible function cost estimates. With regard to short-run returns to the variable factor, with the exception of the returns to admnissions in the small facilities with low occupancy rates included in the Nigerian study, the four studies found either decreasing or constant returns at the sample averages. In two of the three studies in which the functional form allowed an index to be computed, there was no clear evidence of economies of scope. The Colombian and Chinese data yielded an index value of 0.2 or less between outpatient and inpatient services. The Ethiopian data suggested that some economies of scope may exist between bed-days and outpa- tient visits. Computation of the scope index for separate levels of hospi- tals in the Chinese data set (see table 3-10) suggests that the index increases inversely to the level of hospital and that low- and middle-level hospitals with an index value of 0.3 or more may have slight economies of scope in providing both outpatient and inpatient care. Despite the slight variation found in the Ethiopian data and the Chinese middle-level hospitals, the overall results indicate that economies of scope are not an important factor in planning hospital activities. These results are also consistent with the literature on industrial countries, although in medium to large hospitals some diseconomies of scope have been identified. In their analysis of U.S. hospitals, Grannemann, Brown, and Pauly (1986) found that there were some diseconomies of scope between outpatient visits and inpatient care- hospitals with larger numbers of inpatients also had a higher unit cost of outpatient visits. Their explanation for this phenomenon may be useful in understanding the interaction of inpatient and outpatient costs in large hospitals in developing countries. They state that the difficulty of coordinating a greater range of services may contribute to higher costs. Also, larger hospitals may have outpatient visits of greater complexity that give rise to longer or more costly inpatient stays. They did not suggest, but it may also be, that larger hospitals have more trained personnel and available techniques that are applied to outpatient ser- vices independently of case complexity. In other words, the "quality" of 132 Public Hospitals in Developing Countries Table 3-11. Marginal Cost and Economies of Scale and Scope from Statistical Cost Functions Study Item Ethiopia Nigeria China Colombia Marginal cost In 1988 U.S. dollars Bed-day 1.4 7.6 22.0 Outpatient visit 2.9 0.5 9.9 Admnission 4.2 Delivery 89.8 As a % of per capita GNP a Bed-day 0.9 1.8 2.7 Outpatient visit 0.3 0.1 1.2 Admission 0.4 Delivery 55.9 Indexes of scale and output Short-run SRVF 0.8 0.9 1.1 Long-run Eos 0.7 0.9 Product returns Bed-day 1.0 0.7 1.0 Outpatient visit 0.8 0.5 0.8 Admission 2.6 Deliveries 0.9 Economies of scope Bed-days and outpatient visits 0.4 0.2 0.1 a. In year of data. Source: Bitran-Dicowsky and Dunlop 1989 (Ethiopia); Wouters forthcoming (Nigeria); authors (China and Colombia). outpatient care in larger hospitals in terms of resources used per patient may be better (although it is not clear that the health outcome for noncomplex cases is superior; that is, qualitatively different inputs are used, which may or may not lead to qualitatively different health out- comes). The general lack of marked economies of scope for all hospitals and possible diseconomies of scope for larger hospitals removes an economic argument for expanded outpatient departments in large hospitals. Taken at face value, the estimated results imply that limiting, or perhaps eliminating, the outpatient department of large hospitals and shifting the burden of outpatient care to lower levels of facilities would improve the efficiency of the hospital subsector. Much more detailed study is needed before such a far-reaching policy conclusion is recommended, however. Hospital outpatient services are quite diverse, ranging from emergency trauma care to scheduled specialty treatment and consulta- Hospital Costs and Efficiency 133 tion services. It is unlikely that large hospitals could divest themselves of their casualty and emergency departments, nor is it likely that highly specialized outpatient care requiring expensive equipment and person- nel could be moved from a hospital setting, as both the capital and recurrent costs of opening separate outpatient facilities are likely to be prohibitive. A much more feasible policy conclusion is that lower-level facilities should be upgraded to provide basic outpatient services to general outpatients requiring nonspecialized care and to provide follow- up ambulatory services to patients who have received specialized inpa- tient or outpatient care at a tertiary facility. In other words, first-level referral hospitals should be capable of playing a dual role: that of referring patients to more specialized care and that of providing pallia- tive and follow-up care to these patients after they receive specialized services and are referred back down the provider pyramid to their local institutions. Summary The growing number of accounting analyses of hospital costs demonstrates both the feasibility of such studies, even in situations in which information on cost and use is ostensibly lacking, and the useful- ness of such analyses. Within countries, the studies can be used to develop performance standards, assist with projections of future resource requirements as demand for hospital services increases, and identify hospitals that require special management attention to improve efficiency. Cost studies can be supplemented by examination of service statistics through use of the technique suggested by Pab6n Lasso (1986) to provide further identification of hospitals with substandard perfor- mance. Once such hospitals are identified, the analysis of specific effi- ciency issues can quickly move beyond the realm of economics and into management, organization, and personnel planning. Problems with low turnover and occupancy rates, inappropriate length of stay, and ineffi- cient use of personnel and technical inputs may be solvable if direct changes are made in administrative rules or management decisions. But if inefficiency is widespread, it is probable that corrective policies will involve changing the incentives that guide the behavior of management, physician, and client. The average cost estimates reviewed in the first section of the chapter show the remarkable similarity across countries in the average cost of a bed-day when measured as a percentage of per capita GNP. With few exceptions the cost per bed-day in a middle-level hospital varies from about 1.3 to 3.0 percent GNPN. The actual constant dollar cost varies by considerably more, however, and this variation is undoubtedly accom- panied by broad differences in quality. The variations in quality impede 134 Public Hospitals in Developing Countries cross-country comparisons in efficiency of service delivery. Two facts do stand out, however. First, the average cost of lower-level hospitals is nearly always substantially less than that in higher-level referral institu- tions. The waste of resources resulting from the inappropriate use of provincial and central facilities to treat cases that do not require more specialized technical care can be large. The difference between upper- and lower-level facilities in cost of services gives a rough measure of the resource benefits from improved referral policies, although one must keep in mind that some of the cost variation is likely to be attributable to case mix and quality differences. Second, lower-level hospitals are characterized by low occupancy and turnover rates in many countries. These low rates are commonly related to services that are perceived to be and actually are of low quality. Increased use cannot be achieved merely by new written guidelines for a patient referral system; rather, an adequate supply of inputs at lower levels to improve quality is needed to stimulate demand. The results of statistical studies performed to date are too limited to provide definitive guidance for policy. Most of the studies have been carried out on small or poorly specified data sets that obscure the conclusions. The potential usefulness of such studies is established, however, by the illumination they could shed on the relation between marginal and average cost and between economies of scale and econo- mies of scope. The setting of hospital prices can be done more efficiently with a knowledge of marginal costs, and economies of scale and scope should be included in hospital design and organization. As experience grows with statistical analysis of hospitals in low-resource environments and with larger hospital data bases and improved specifications, the studies will be of increased usefulness for setting prices and planning hospitals. The individual studies surveyed in this chapter demonstrate the variability that can be expected across countries and the need for doing country-specific studies before assuming constant cost. The pas- sive accounting view of the cost function can be taken as a good first approximation of the relation between output and cost and is certainly preferable to basing planning decisions on rough assumptions. The accounting and true cost functions, however, can be expected to diverge for policies that carry output far from prevailing levels. The findings in this chapter complement the policy discussion in the subsequent three chapters on pricing and hospital service alternatives. Pricing should be intimately related to costs, which constitute half of the information required for determining optimum pricing levels. The next chapter will add demand and income distribution information to cost information to build suggested guidelines for hospital pricing. With regard to hospital alternatives, the discussion of cost and efficiency in this chapter has suggested that improving referral patterns by directing patients to lower-level hospitals and nonhospital alternatives might pro- Hospital Costs and Efficiency 135 vide savings without sacrificing health status. Chapter 6 outlines inno- vative and practicable lower-level hospital and nonhospital alternatives. Appendix 3A. Step Down or Cost Center Analysis of Hospital Costs Anyone who does detailed step down recurrent cost studies faces two significant problems. The first is a problem that all authors of cost studies must address: cost data may not be directly available for individual hospitals. Multiple sources of budgeting (for example, central as op- posed to district-level expenditures) and assorted means of making payment for different line items (for example, the salaries of physicians and nurses may be paid by the central health ministry, other salaries may be paid at the district level or directly by the hospital) make the recon- struction of actual expenditures laborious. The second problem, of par- ticular importance for step down studies, is that cost information may be available only on an aggregate basis for the hospital. The need to reconstruct hospital cost data from multiple sources provides some insight into the problems of resource allocation that confront hospital managers. Often, because of institutional constraints and multiple budg- etary sources, they do not have a complete picture of their costs or the level of resources that will be available to them over a period of time. The first requirement of the estimation process is to get as complete a picture of total recurrent hospital costs as is possible. This means sup- plementing the hospital line item expenditure data with information on resources used that do not appear in the hospital's budget or financial statement. For example, it is common that the financial statements of public hospitals do not include most expenditures on drugs and medical supplies or on maintenance services provided to the hospital. It is necessary to get such information from central medical stores or other central or regional distributional agencies. It is also necessary to estimate nonfinancial costs, such as depreciation and the value of donated goods and services. Next, because most hospital or health ministry budgets are in a line item format (typical line items are salaries, drugs, other supplies, public utilities, and so on), line item costs must be attributed to cost centers, which reflect specific hospital departments. The specific cost centers vary from study to study, but typically three categories of cost centers are used: * Overhead. These cost centers produce only those services that are consumed by other departments (cost centers) of the hospital, not by patients. Examples include Administration, Housekeeping, Mainte- nance, and Utilities. 136 Public Hospitals in Developing Coun tries * Intermediate. These cost centers produce services that are used by other departments but also provide services directly to patients. Ex- amples include Laboratory, X ray, Operating Theater, and Physiother- apy. * Final. These cost centers provide services directly to patients, not to other departments. Examples are Inpatient and Outpatient, with some studies disaggregating these broad categories into specific de- partments, such as Medicine, Surgery, Obstetrics-Gynecology, and Pediatrics. Step down costing can be depicted algebraically as follows: Let Cio = direct costs in Overhead cost center i; Cj, = direct costs in Intermediate cost center j; and CkF = direct costs in Final cost center k. The direct costs are the costs attributed to each cost center prior to their allocation to the cost centers associated with hospital outputs. After the direct costs of each cost center (that is, all of the Cio, Cji, and CkF) are identified, the step down method is applied to allocate all costs to final cost centers. The basis for allocating specific proportions of each cost center's costs to other departments should reflect the consumption of the source department resources by the receiving department (as an example, the distribution of dietary [source] costs among inpatient departments [receiving] would typically be based on the proportion of total patient- days in each inpatient department). First, direct overhead costs are allocated to all other departments. The bases of allocation of costs from each overhead cost center to the other cost centers are proportions that can be represented as: aij = the proportion a of Overhead cost center i's costs "used" by Intermediate cost center j; cxik = the proportion a of Overhead cost center i's costs "used" by Final cost center k; pjk = the proportion p of Intermediate cost center j's costs "used" by Final cost center k; where p= 1. In the first step the overhead costs are allocated to intermediate and final cost centers using the allocational proportions cic and xik, resulting in the first step allocated costs, C 'jI and C 'kF.13 Explicitly, Hospital Costs and Efficiency 137 C 'I, = C,, + -a,i C,0 C 'kF = CkF + Yak* CiO where C 'J = fully allocated costs of Intermediate cost center j; and C 'kF = partially allocated costs of Final cost center k. Then, in the second step, the allocated ("indirect") costs from the intermediate cost centers in the first step are allocated among the final cost centers using the proportions pfk. Explicitly, the fully allocated costs are C "k = C 'k + CPk /1 where C "kF = fully allocated costs of Final cost center k. After all costs are fully allocated to each of the final cost centers, average costs are calculated by comparing fully allocated costs to the relevant use statistics. For example, if the only final cost centers are simply Inpatient and Outpatient, statistics on total patient-days and discharges can be compared with fully allocated inpatient costs to gen- erate measures of the average cost per day and per discharge. In a similar manner, the average outpatient cost per visit can be calculated. Average costs of intermediate services can also be calculated if service statistics are available. Appendix 3B. Use of Service Indicators for Rapid Assessment of Relative Performance of Indonesian Type C Hospitals Each point in figure 3A-1 (in accord with the method described by Pab6n Lasso [1986]) represents one of seventy-eight Indonesian type C hospi- tals (eighteen on Sumatra, forty-three on Java, and seventeen on the other islands). The points are defined by the rates of bed occupancy and turnover for each hospital; these rates in turn define the average length of stay (shown for its mean and extreme values across the top and down the right side of the graph). The dotted lines are one standard deviation (plus and minus) from the means of the occupancy and turnover rates. Hospitals that lie outside the central rectangle formed by the intersecting dotted lines (with the intersecting solid lines in its center) are considered outliers and merit further investigation to understand their deviation from the norm. 138 Public Hospitals in Developing Countries The most striking aspect of the figure is the apparently poor perfor- mance of Sumatran hospitals compared with those of Java and the other islands. Most of the outliers in region I are Sumatran. Hospitals in this region have the least desirable characteristics-low use and poor pro- ductivity. This region is characterized by low demand for hospital beds in relation to installed capacity, either because of a generalized low demand for inpatient care or because alternatives to type C beds are preferred by the population. The data in the figure suggest that an Figure 3A-1. Indicators of Hospital Performance in Indonesia, Type C Hospitals, 1985 Turnover rate Average length of stay (days) 3.5 70 - 60 _ - 6.2 5 .' * 0 50 II 61*11 II ~~~~~~~~~~~~~~~~~~~~~~~1 2. 40 30 0 - 3 E+ Averag bedoccupancy(perce eIV )14.2 20 + t v .a 10 + 1+97 0 10 20 30 40 50 60 70 80 90 100 Average bed occupancy (percentage) + Sumatra 0 Java Other islands Source: Bamnum 1987. Hospital Costs and Efficiency 139 investigation should be made to explain this situation and develop possible remedies: * It should be determined whether poor performance is a function of low medical staffing ratios, low budgets per bed, management defi- ciencies, or overbedding in Sumatra. * Since demand for type C inpatient services in Sumatra is low, it may be cost-effective to consolidate inpatient services in a smaller number of facilities and to convert some facilities into strictly ambu- latory centers. These changes would enable reallocation of staff and allow remaining type C facilities to become more productive. * If it is determined that demand for inpatient care at a type C hospital is low because the population bypasses this hospital to reach a higher-level (type B) hospital, it may be possible to improve the quality of services visibly at the type C hospital (for example, by transferring resources, such as one or several physicians, from higher levels to the type C) in order to change the population's preferences and improve the referral system. Alternatively, if this is not feasible, it may be better to curtail the inpatient services of the type C hospital. * The reasons for the relatively good performance of the Javanese hospitals that are outliers in region IJI should be assessed. Because of expected variation in case mix, staffing, and possibly other characteristics across hospitals of different types, for example, tertiary in contrast to district, this methodology for assessing hospital performance is most appropriate when applied to facilities that are similar in nature (as in this example). Upon detailed investigation, however, one may find that, even among a group of similar hospitals, deviations from the norm are caused by case mix or other factors that do not necessarily imply differences in relative efficiency across hospitals. Nevertheless, this graphical means of identifying performance outliers among a group of similar hospitals is both quick and effective. Once identified, these outliers can be required to explain their deviations from the norm. By itself, then, this graphical technique does not answer the question of whether a given hospital is performing efficiently. It does, however, enable one to focus attention on those specific hospitals whose perfor- mance varies considerably from the norm. Notes 1. Existing prices may not reflect the true scarcity of certain items in the economy. For example, public sector wage rates may understate the value of physician or nursing services, or an overvalued exchange rate may result in the domestic prices of imported pharmaceuticals being artificially low. Cost esti- mates should adjust for these distortions, or, at a minimum, note their existence and their approximate magnitude. 140 Public Hospitals in Developing Countries 2. The occupancy rate (occ) is calculated as: occ = I / (365-B), where I = Annual number of inpatient days, and B = Average number of available hospital beds during a year. 3. The average length of stay is defined as: ALOS = I / A, where I = Annual number of inpatient days, and A = Annual number of inpatients (admissions or discharges). 4. The turnover rate (C) is calculated as: T= A / B, where A = Annual number of inpatients (admissions or discharges), and B = Average number of available hospital beds during a year. 5. Let drug availability be measured as drug expenditure (prices are constant) and write total costs as C = F + D, where D is drug expenditure, and F is all other costs (assumed fixed in the short run). Define the elasticity of service use with respect to drug availability (expenditure) as EQD = (aQ/aD)-(D/Q) and the elas- ticity of costs with respect to drug availability as ECD = (dC/aD)-(D/C). Average costs can be written c = C/ Q. Then the change in average cost with a change in drug expenditure is &/aD = ac CaQ]/D2. Multiplyingby D/D, C/C, Q/Qand rearranging gives theelasticityof average cost with respect to drug availability, Tcd = (ac/aD) (D/C) = (Q/D2)- [(aC/aD)- (D/C) - (aQQ/aD)- (D/Q)] or TlcD = (Q/ID2) . [ECD - EQD] = (Q/D2) [DIC - EQDI where the two terms in the brackets on the right-hand side are the elasticities of total cost and quantity with respect to drugs. If the expression in brackets has a negative absolute value the effect of a reduction in drug availability will be to increase average costs. This highlights D/C and EQD as the appropriate param- eters to estimate for an empirical study of the effect of drug availability on demand and average costs. Typically D/C is about 0.3 so that EQD merely has to be greater than 0.3 for a drug shortage to result in increased average costs. 6. The increase would depend on the values of the elasticities in the previous note. 7. On the graph, national mean values of thebed occupancy and turnover rates define a point. Because the average length of stay is defined as ALOS = I/A, Hospital Costs and Efficiency 141 knowledge of the occupancy and turnover rates identifies the average length of stay: I = (occ)(B)(365), and A = (T(B), therefore ALOS = (occ)(365) / T. 8. A careful analysis of inefficiencies and remedies for drug supply is given in Quick 1982. 9. In a modestly equipped referral facility, equipment may be 30 percent of total construction cost (about 40 percent of the cost of the building construction). Using a rule of thumb that annual maintenance expenditure should be 1.2 to 2 percent of the capital cost of the building and 7 percent of equipment cost, maintenance should be about 3 percent of total capital cost (0.012 x 0.7 + 0.07 x 0.3 = 0.029). If the recurrent cost ratio (see chapter 2) is between 0.2 and 0.25, maintenance will be between 12 and 15 percent of the recurrent cost (0.03/0.2 = 0.15). 10. The accounting view of the cost function can be written as C = F + lWi Li + lq BDj, i i where F is fixed cost, Wand L are the wage and quantity of the ith type of labor, and q and D are the price and quantity of the jth type of other nonlabor inputs. 11. The index of short-run returns to the variable factor, SRVF, set out above could be appropriately interpreted as an index of long-run general economies of scale if applied to a long-run cost function in which costs include the cost of capital and the data are defined for a time period that is sufficiently great so that all inputs, including capital stock, can vary. The estimated cost functions pre- sented later in the chapter are all short-run cost functions (as, indeed, are most econometric hospital cost functions). When applied to a short-run variable cost function, the SRVF provides a measure of short-run returns to the variable factor, as we note above, but does not measure economies of scale because scale is fixed. If the function estimated is a short-run variable cost function the appropriate index for economies of scale is the EOS as set out above (Vita 1990). 12. Four of these studies were done to provide background for recent World Bank research. The Ethiopia, Colombia, and China research efforts were done as background studies for this chapter, and the Nigeria study was carried out to support the Nigeria health sector analysis reported in Akin and others (1991). 13. It is also possible to allocate the costs from one overhead cost center (for example, Administration) among other overhead cost centers (for example, Housekeeping). The step down procedure is closely related to input-output analysis of the firm, and it could be formulated in terms of matrix algebra. For an example of step down hospital cost study with a detailed description of methodology, see Russell, Gwynne, and Trisolini 1988. 4. Hospital Financing Alternatives A growing number of countries have implemented, or are considering, alternatives to government budget allocations for financing health ser- vices. Because of the individual nature of the services provided and the importance of hospitals in the total health sector budget, hospitals are a focus for the practical application of these alternatives. Although donors and ministry of finance officials have been supportive and, in some cases promotive, of the use of fees and insurance schemes in the health sector (see, for example, Akin, Birdsall, and de Ferranti 1987), the support has in most cases fallen short of specific guidelines for pricing and revenue collection. In this chapter we summarize the potential role of alternative, nonpublic sources for hospital financing; present a brief outline of the most important issues in financing alternatives; and suggest some gen- eral criteria for setting hospital prices. The impetus for adopting alternative financing policies in hospitals comes from the difficulty of mobilizing sufficient funds for the health sector from public general revenues alone and the inefficiency and inequity of using public funds to support institutions or programs that do not have wide benefits. In addition, problems in achieving or main- taining acceptable quality within government budgets dictate a search for alternative financing sources. Rapid hospital cost escalation has also distorted the allocation of health sector resources between facilities and programs. Financing mechanisms are intrinsically related to potential solutions to these problems because, in addition to augmenting reve- nues, they affect demand and supply decisions and the allocation of resources. Broadly, financing alternatives that respond to these problems in- volve, either separately or in combination, public or private insurance and direct charges to users of hospital services. User fees and insurance are, in themselves, large topics that go beyond the scope of this study.1 In this chapter and the next, we attempt to keep the focus narrowly on 143 144 Public Hospitals in Developing Countries the financing mode and its effects on the use and provision of public hospital services, although much of the discussion is applicable to other health services as well. In order to clarify criteria for setting health financing policy, we first set out the rationale for public sector involve- ment in hospitals and identify the objectives of public sector health financing, particularly as these objectives relate to hospitals. Then, in the second section, we discuss alternative financing policies, provide broad principles for designing cost-recovery programs, and briefly outline the institutional characteristics of user fees and insurance as they relate to hospitals. Health Financing Objectives Efficiency, equity, and revenue collection are the objectives of a health financing policy. Problems of efficiency and equity in the delivery of hospital services and insufficient funds for recurrent operating costs have been emphasized in the preceding chapters. Hospital financing policy can contribute to improved equity and efficiency of service use and improved funding or, if ill-considered, can contribute to a worsening of these problems. Efficiency, equity, and revenue collection, thus, are criteria by which the performance of a financing policy can be assessed. Efficiency The need for public intervention in specific markets, either through provision of goods and services directly by the government or through market regulation, arises from market defects or failures that prevent the achievement of an economically efficient outcome or that lead to ineq- uities that are unacceptable. Not only does the existence of market failures affect the choice of public services to provide, but it can affect the appropriate choice of public financing. In this subsection we briefly review the market failures that provide a rationale for government intervention in the market for hospital services, and in the following subsection we review failures related to equity. Market failures that lead to allocational inefficiency and that require government financing or provision of services are of great importance for health markets in general and of moderate importance for hospitals. Five general sources of market failure are commonly recognized-public goods, externalities, economies of scale, inadequate consumer informa- tion, and incomplete markets. Public goods are those goods whose con- sumption by one person does not diminish the consumption by another and from whose benefits it is not possible to exclude nonpaying consum- ers if the good is provided at all. Externalities are benefits or costs that accrue to persons other than the direct market participants. Economies of Hospital Financing Alternatives 145 scale occur when the unit costs of production decline throughout the range of production. Information failure exists when consumers or pro- ducers (or both) lack information about the benefits or costs of consump- tion or production that is needed for rational market participation. Incomplete markets occur if uncertainty, the lack of contingent markets (for example, consumer loans), or the lack of future markets prevents consumers and producers from forming a market. Classification of some services as Merit goods, is also given as an argument for public sector activities (though this classification cannot strictly be defined as a market failure). The argument for public support of merit goods is not based on economic analysis but is a normative, extramarket, argument based on a subjective evaluation by society (planners or voters) of the desirability of providing and consuming specific goods or services. Table 4-1 summarizes one view of the strength of each of these sources of market failure as a rationale for government intervention in the provision of selected broad categories of hospital services. The contents of the table are not rigorously determined. The existence or absence of market failure is a question of positive economics and is conceptually Table 4-1. Strength of Rationale for Public Sector Involvement in Hospital Services Rationale Econo- Informa- Incom- Public Exter- mies of tion plete Merit Type of service goods nalities scale failure markets goods Inpatient Communicable diseases .. ** .. ** Chronic diseases * *** Accidents Mental health * * ** Surgery * * *** *** ** Depends Diagnostics * ** ** ** on social Obstetrics and Normal * * political High risk .. * ** *** values Outpatient Communicable diseases .. ** ** * Chronic diseases .. * *** * Emergency and trauma * ** * Preventive care * *** ** *** Strong. ** Moderate. * Applicable but weak. Of negligible importance. 146 Public Hospitals in Developing Countries open to deductive reasoning and empirical validation. Still, there is actually very little related empirical analysis on which to construct such a table, and the importance of the various market failures could also change in different settings. In addition, there is a wide variation of services subsumed within each category. For all these reasons, the con- tents of the table are open to debate. By citing specific service categories, however, the table provides a concrete frame of reference to allow a judgment of the overall importance of market failure in the case of hospitals. The first three types of market failures-public goods, externalities, and, possibly, economies of scale-are of importance only in special categories of hospital care. The last two-inadequate consumer informa- tion and incomplete markets-are of greater importance. In addition, the merit goods argument, though often implicit, is perhaps the strongest rationale for the public sector provision of hospital services. With the possible exception of the security provided by the existence of emergency or trauma units, very few hospital services have the characteristics of true public goods. Many hospital services have the weak externality of greater labor productivity of individuals whose health is improved by the services. In addition, a few services, especially those involving communicable diseases, such as tuberculosis, have the important externality of the improved health of others. In general, however, hospital services are characterized by excludability in con- sumption and the fact that the bulk of benefits accrue to the individual (or household) receiving care. Thus, externalities and public goods are not important in arguments for public provision of services or regulation of the prices of services. As found in the last chapter, economies of scale are not pronounced in hospitals. When occupancy rates are low, the marginal cost of produc- tion may be below average cost, but in hospitals operating at full capac- ity, marginal cost is likely to be equal to or greater than average cost. It cannot be convincingly argued that hospitals are a naturally declining cost industry that universally require government production, but in smaller communities, especially those in which transportation is limited, an efficient scale of facility may have declining cost in the range of production relevant to the local level of demand. Subsidized production may be required in such communities if the service is to be provided. In addition, some researchers in industrial countries (Grannemann, Brown, and Pauly 1986) have found important economies of scale for emergency care, and the high fixed cost of some diagnostic techniques can also involve economies of scale. From an efficiency point of view the most obvious reasons for public provision or regulation of hospital care are the inadequacy of consumer information and the incomplete market for hospital services in many Hospital Financing Alternatives 147 settings, especially in rural and low-income areas. The inadequacy of consumer information restricts the appropriate use of hospital services and creates the opportunity for market exploitation through supplier-in- duced demand. The consumer has little basis on which to judge the appropriateness of producer-requested surgery, laboratory tests, or other interventions. Public provision of services or private market regu- lation, although not completely eliminating this problem, can subject producers to greater control. Lack of consumer information may also be an especially large problem in a setting in which the level of education is low, because a failure to recognize the need for care and to seek services at the appropriate time can restrict the demand for services. On the consumer side of the market, the uncertainty of when diseases will occur, information failure, and the long-term time horizon involved in individual consumption decisions in health lead to incomplete mar- kets for hospital services.2 Incomplete markets are manifest in the failure of contingency or futures markets to arise to finance private purchases of hospital services. They are especially important in the failure of the market to provide for less frequent and higher-cost medical services even when consumers, if adequate financing were available, would pay to protect against the risk of incurring the full cost of these services. On the supply side of the market, capital market imperfections limit funds for construction of hospitals in rural areas, and skilled labor shortages restrict the operation of rural hospitals. The market failures cited above prevent the private market from financing an economically efficient allocation of resources directly. Pub- lic sector involvement in the market for hospital services does not automatically eliminate the efficiency problems posed by market failures but instead introduces the new problem of selecting an appropriate financing system. The financing system creates incentives for consumers and hospital managers that affect the efficiency of resource allocation. In general, efficiency is served if the price paid by the user reflects the additional benefits to society from consumption of the service and the additional cost to society of producing the service. The appropriate pricing and financing system to achieve this is, however, not obvious. All financing systems have potentially undesirable allocational effects. Regulation or public sector provision of services may lead to nonmarket inefficiency. Hospital services provided without charge may be con- sumed beyond the point of economic optimality, that is, the marginal cost to society may exceed the marginal benefit of the excess consump- tion. Regulated prices, as set, for example, by a price control board, may inadvertently provide socially adverse profit incentives for hospital managers to produce particular services, such as CAT scans, and neglect others, such as general wards for inpatients. Finally, unregulated private provision may lead to a service mix that is not socially optimal because 148 Public Hospitals in Developintg Counitries of a lack of consumer information and poorly functioning markets for hospital services. Thus, among the range of choices, from free public provision of the good, through regulated pricing of services provided either by public or private institutions, to a laissez-faire private system of service provision, there are associated potential welfare losses. This is a sobering fact, for there are no pat answers to the hospital financing and allocational efficiency problem. The best that health planners can do is to remain aware of the importance of allocational efficiency as a criterion and to attempt to achieve, not necessarily an optimal system, but prac- tical improvements to existing arrangements that are clearly suboptimal, as revealed in the preceding two chapters. Equity Evaluating the equity effects of a potential financing mechanism requires identifying who pays, who benefits, and how much (Hoare and Mills 1986). Market failures contributing to a worsening of social equity are of great importance for hospitals. Whereas most publicly provided health care is intended to improve social welfare and involves a redistribution of government revenues collected from a narrow base to provide health services having broad-based benefits, hospitals are, in many countries, an exception. The primarily urban location and the pattern of hospital use by type and cost of service across income classes have equity effects that are too often either neutral or even negative (see discussion of these effects in chapter 2). As with efficiency, the financing system creates incentives for consum- ers and hospital managers that affect equity. Financing affects the inter- action of supply and demand and thus the distribution of benefits from resources devoted to hospital services. There are also direct implications for equity that arise from government, NGO planning, and entrepreneurial investment decisions. In general, equity is served if the financing system promotes wider accessibility and use of services across income and risk groups. Equity is further improved if the burden of payment is distrib- uted progressively across income. Equity thus involves an interaction of the risks of illness across different social groups, the availability and use of services for the illnesses, and the ability of different groups to pay. From an equity point of view, incomplete markets for services and the inadequacy of consumer information provide even stronger rationales for the public provision or regulation of hospital services than efficiency does. Information failure is apt to be correlated with low education or rural location, and incomplete markets attributable to limited access to contingency financing are a greater problem for low-income households. By locating in relatively poor geographic areas, using financing methods Hospital Financing Alternatives 149 that promote the use of services by low-income groups, and providing a referral network that is tied to outreach programs intended to increase accessibility, public hospitals are meant to promote social goals. In incomplete markets, hospital services are not made available, even if demand would be sufficient to cover costs, because of a structural imperfection that prevents a market from forming. Merit goods are a somewhat different case in that a market with a socially acceptable price and quantity of services cannot be established because of deficient demand (perhaps a result of low household incomes) or the high cost of production. In the absence of government intervention it is unlikely that hospital services would be available and used in many rural areas and among low-income groups in urban areas. Most countries recognize health services as merit goods and provide some services through the public sector, but the scope of the services included varies widely. In some countries, free health services, including hospitals, are specified as a social right. Other countries provide only basic hospital care to indi- gents as a merit good. Economics does not directly determine what is regarded as a merit good but does determine what is feasible to provide given available resources. Many countries must face the conflict between all that they would like to provide as a merit good and what they can actually afford to provide. Again, as was true for efficiency, the appropriate financing system to achieve equity goals is not obvious, and all financing systems-whether services are provided free or for fees by the public sector or are provided through the private sector-have potentially deleterious (as well as beneficial) equity effects. The provision of free hospital services by the public sector is potentially the least deleterious, but does have the possible adverse effect of using the public budget to pay for services that do not address the greatest health needs of the poor (either locationally or epidemiologically). Also, if the tax system is regressive and the demand for hospital services is income elastic, the provision of free hospital services may decrease equity. The introduction of fees carries with it the obviously significant risks of damaging the accessibility of health services to the poor. These risks can be reduced if targeting, fee exclusion criteria, and differential pricing can be put into effect. Finally, the equity risk of relying on the private (for profit) provision of services is that such services will cater to the needs of higher-income groups and be offered only where sufficient monetary incentives exist. Conversely, however, each system also has potentially beneficial effects. It is more illuminating to view alternative financing mechanisms not as purely competing but as potential components in an eclectic system of balanced public and private care that combines elements of free service provision, the use of fees, and some form of risk coverage. 150 Public Hospitals in Developing Countries Revenue If we distinguish three levels of budgeting-central finance, the health sector, and the hospital-then we can identify three competing revenue goals that motivate administering agencies to use nongovernmental sources of hospital financing. Conceptually, the three goals result from the desire of the administrative agency at each level to maximize reve- nues under its control and retain flexibility in allocation across budget items within its jurisdiction. The three competing goals are: * to supplement a hospital's resources derived from the government budget, that is, to add to the hospital's budget; * to substitute for the hospital's allocation from the health sector budget and provide supplemental funds for other health activities, that is, to add to the health sector budget and reallocate part of the hospital budget to other programs; and * to substitute for governmental sources of health revenues, that is, to add to the central government budget. Outcomes between supplementation and substitution are also possi- ble. For example, a part of alternative collected revenues could be used to supplement the hospital's budgeted resources, a second part could be used to supplement other health activities, and the remainder could provide a partial substitute for the government subsidy to health at either the local or central level. Such a remainder would be analogous to tax revenues and would yield an increase in government general reve- nues. These revenue goals and the efficiency and equity goals of cost recov- ery are related. For analytical darity, the theory of public finance sepa- rates the effects of revenue and expenditure policies, and to the extent practical we attempt to keep this distinction. Yet in execution, revenue and expenditure policies are often linked. The distinction between the three goals above implies that such links are seen by government admin- istrators and motivate revenue policy. If revenues are kept by the hospi- tal and do not supplant existing subsidies, they make improved quality of services or greater quantity of services possible, and the effects on efficiency and equity are then determined by hospital policy. If revenues are kept within the health sector at the local level, they may make additional nonhospital services possible; depending on local health sector policies there may then be cross-subsidization between hospital and nonhospital services and attendant changes in efficiency and equity. The accrual of proceeds of hospital financing alternatives at the central level holds other possible benefits. In this case, the government can use the resources to bolster the central-level program that is deemed to have the greatest social benefit. Hospital Financing Alternatives 152 Strictly interpreted, the theory of public finance leads to the conclusion that the central level is the best locus for accrued revenues. If geographic redistribution of the funds is needed to improve efficiency and equity, the central government may be a more effective end user of new reve- nues. Practically, however, accrual of revenues at the local level appears to provide a good compromise between the flexibility of central govern- ment accrual and the increase in quality made possible by hospital accrual. In many practical situations, the retention of revenues at the local level would provide the greatest possibility for effective use of the funds. Local governments have the most immediate knowledge of alter- native effective and equitable uses of funds and are most flexible in response. Local officials also are more acutely aware of hospital funding requirements. An unconstrained choice among program alternatives, rather than restriction of choice to hospitals or a given sector, can result in the greatest benefit. Revenue stability. Achieving greater revenue stability for the health sector can also be an objective that motivates administering agencies to use nonbudgetary financing altematives. During economic recession, brought on, for example, by a decline in the price of an important export commodity such as oil in Indonesia or copper in Zambia, recurrent financing in the health sector falls more or less in tandem with a decine in government revenues. By providing an alternative source of funds that is less sensitive to fluctuations in the government budget, the use of alternative financing for hospitals can reduce the effect of economic recession on health programs. TIhe efficiency of revenue collection. There is a further relation between these goals, the kind of nongovernmental financing, and the degree of efficiency of collection of nongovernmental revenues (where efficiency is measured as the proportion of potential revenues, given the defined revenue policy, that are actually collected). Depending on the form of financing, the collection of revenues may be more efficient at the central, intermediate governmental, or institutional level. The revenue goals can involve either partial or full recovery of hospital costs (or even the generation of hospital profits). There is a link between the extent to which actual revenue collection can achieve the intended proportion of cost recovered and the kind of financing and institutional form and level of collection. As an illustration we consider the relation between the efficiency of fee collection and government accrual policy. For any given level of prices, potential levels of demand and revenues are determined. The revenues actually collected from hospital user charges are generally less than those that potentially could be collected if all users of hospital 152 Public Hospitals in Developing Countries services paid the designated amount. The principal reason for this is that lax enforcement of fee policy allows nonexempt patients to receive hospital services and never pay for them. Comparing actual with poten- tial revenues allows an assessment of the performance of alternative financing mechanisms, given prices. Anecdotal evidence (for example, see Collins 1990; Overholt and others 1990; Vogel 1988) indicates that there is an inverse relation be- tween the governmental level at which fees accrue and the efficiency of fee collection. The general recommendation is to allow fees to accrue as closely as possible to the collecting level in order to provide an incentive for managerial surveillance and enforcement of fee policies. There is, of course, no conflict between this recommendation and the revenue goals if the intention is to allow supplementation at the institutional or local level (the first two revenue goals listed earlier). Critically, even if the goal of government fee policy is substitution at the central level (that is, the third revenue goal), and, therefore, a reduction of the net government subsidy after fee revenues returned to the central level are subtracted, some retention of fees at the collecting level is needed to provide a collection incentive.3 Alternative Policies Public sector hospitals in developing countries receive revenues from a large variety of budgeted sources. Financing from government budgets can occur at the central, local, or intermediate level, and at each of these levels more than one agency budget may be involved. Sources of the government budgeted revenues are in themselves based on a diverse tax and financial base, perhaps including earmarked taxes and donor trans- fers. Collectively, the hospital revenues derived from these sources are a subsidy from the governmental budget allocation. In the next two subsections we discuss fees and risk sharing as alternatives to subsidies from governmental budget allocations that would shift a greater part of the financing burden more directly to household or community sources. Fees Fees, if they are to be substantial enough to achieve revenue objectives and cover a significant proportion of hospital costs, must be set with a recognition of their effects on demand and user welfare. Conceptually, fees can improve efficiency of resource use by reducing use of hospital services for care with negligible benefits, removing demand in excess of existing supply capacity, and providing appropriate allocational incen- tives to both producers and consumers. Fees can also have adverse effects on equity by impeding the access of the poor to needed services. Hospital Financing Alternatives 153 The few studies that have been done of the demand elasticity of user choice of services (table 4-2) suggest that users are not highly responsive to changes in the price of health care. The low elasticities in these studies, however, do not of themselves indicate either that the efficiency gain from fees or the adverse equity effects of fees is low. First, all the econometric studies on user fees cited in table 4-2 have been carried out where fees are low or nearly negligible. It does not follow that one can extrapolate from the results of these studies to estimate the quantity response to substantial fees. Second, even if the price elasticity is truly low, the welfare effect of fees can be significant because, in households that have fixed incomes, increases in fees imply that the consumption of other goods or services, possibly food or education, could be reduced. Thus, the use of fees can have implications not only for revenue but also for efficiency and equity objectives. Pricing principles. Guided by the efficiency, equity, and revenue objec- tives, we can elaborate a normative set of practical principles for a system of health sector fees in government institutions.4 These principles in- clude considerations of (a) ability to pay, (b) fees as resource allocation signals, (c) the relation between fees and quality, and (d) market failures. Table 4-2. Price Elasticities of Demand for Health Services Price range Year of data Country Service (U.S. dollars) Price elasticity 1985 CUted'Ivoirea Clinic Free-$0.11 -0.32 $0.11-$0.22 -0.62 Hospital Free-$0.11 -0.38 $0.11-$0.22 -0.83 1985 Perua Clinic Free-$1.56 -0.46 $1.56-$3.12 -0.68 Hospital Free-$1.56 -0.41 $1.56-$3.12 -0.64 1984 Kenya Outpatient Free-$0.13 -0.05-0.20 1975 Malaysia Public outpatient - -0.15 Public inpatient - -0.00 1981 Philippines Prenatal care - -0.01 1985 Ethiopia Outpatient - -0.05-0.50 1990 Nigeria Outpatient - -0.04 1986 Sudan Outpatient care - -0.37 - Not available. a. Arc elasticity was calculated in price ranges given and for middle-income group. Sources: Gertler and van der Gaag 1988,1990 (C6te d'lvoire and Peru); Heller 1982 (Malaysia); Akin and others 1986 (Philippines); Donaldson and Dunlop 1987 (Ethiopia); Mwabu and Mwangi 1986 (Kenya); Akin and others 1992 (Nigeria); Schwabe, n.d., as quoted in Jimenez 1989 (Sudan). 154 Public Hospitals in Developing Countries * Fees should be consistent with ability to pay and should not prevent essential access to health care. A direct interpretation of this principle would sup- port price discrimination among users on the basis of income. Some price discrimination may be practical between urban and rural locations or across geographical regions insulated by high travel costs, but within a given geographic location or institution, price discrimination can be difficult to enforce. In the appendix to this chapter, we show that indirect price discrimination in which fees with differing profit or loss margins for services consumed by different income groups can achieve consider- able equity gains and can substitute for direct price discrimination. Detailed price discrimination in which several price tiers are used may be impractical, but some provision should be made to recognize and exempt the very poor from the burden of fees. It may be feasible in smaller institutions to base such exemptions on the judgment of local officials. For larger hospitals, however, a system of formal identification of income status is needed. Still, local control of the identification of the indigent may be practical. In setting equitable fees, it should be recognized that the actual fee paid to the hospital or clinic is only a part of the true price to the user. Transportation and out-of-pocket drug expenses can be large. In some health systems patients or their relatives supply meals and bed linen. Surveys of the cost of services to patients should inform the design of a fee system. The system of fees needs to provide some limitation on the out-of-pocket expenses of patients. Such a limitation is particularly important for long stays in hospitals or extended outpatient treatment of chronic diseases. - Fees should provide correct signals for the direction of the use of health care and health sector resources. One of the most commonly cited reasons for imposing user fees is to provide signals that discourage unwarranted use of services that have high costs but comparatively low benefits. The system of fees is an essential component in establishing an efficient referral network that avoids the loss of welfare that accompanies inap- propriate use of services at upper referral levels by patients who ignore the referral requirements. Patients who use upper-level services directly can be viewed as using luxury services and be charged as such. One possibility is to forgive part of upper-level charges for patients who are appropriately referred. * Fees and the quality of services should be linked. The introduction of fees will result in consumer dissatisfaction and possibly unacceptable de- creases in demand if the quality of services is not perceived to be sufficiently high. Quality that is perceived as low at entry levels of referral is the primary reason that clients attempt to avoid established referral chains and go directly to more costly institutions. Several studies have advocated tying increases in fees to increases in quality of services Hospital Financing Alternatives 155 by using the revenues obtained to improve the availability of pharma- ceuticals and provide other improvements in quality. A crucial empirical assumption underlying this recommendation is that the decrease in demand for services in response to the higher fee will be more than offset by the increase in demand in response to the higher quality. A simulation of the equity and efficiency effects of fee increases with accompanying quality increases based on data from several hundred health units in Kenya demonstrated a net increase in welfare (Mwabu and Mwangi 1986). Similarly, simulations based on estimated demand functions demonstrate that quality increases more than offset the minimal reduc- tion in demand that accompanied fee increases in Ogun State in Nigeria (Akin and others 1991). These findings are particularly important with regard to public hospital care at the district level; low usage at this level, even when services are free, is often explained by the poor quality of services and lack of availability of essential complementary inputs. * Fees should be subsidized for services that have important externalities, are primarily publicgoods, have low consumption due to informationaldeficiencies, or are merit goods. As noted earlier, goods with externalities have benefits that accrue not only to the individual receiving the service but to others as well. The externalities involved in the consumption of many preven- tive health services can be substantial and justify government interven- tion, including subsidization, to achieve an economically efficient level of use. It can be argued that obstetrics and some curative services, such as tuberculosis treatment, have external benefits, but generally hospital services do not have significant externalities. Also, as previously indi- cated, public goods are characterized by the impossibility of excluding nonpaying users from the benefits of consumption. Examples in health are vector control and some of the community monitoring services of public health laboratories. Hospital services have negligible public good attributes. Inadequate consumer information, in contrast, is an import- ant problem that affects the appropriate use of health care, including hospital services. Without a good understanding of the benefits of a potential service, the consumer cannot determine if or when consump- tion is warranted. Subsidies are warranted if the lack of information results in underuse of services. Although merit goods do allow excludability in consumption, it is considered necessary to subsidize them or provide them free to the public because their consumption is deemed to have social merit or be a social right. Public provision of merit goods is related to equity values and political processes. Many countries implicitly consider hospital services merit goods as evidenced by their heavy subsidization, given that they do not qualify for broad-based subsidies on other grounds. Merit goods need continued and careful justification for their public provision, favorable pricing, or subsidization. 156 Public Hospitals in Developing Countries Administrative feasibility of fee collection. The introduction of fees in a hospital that has not previously charged for services, or in which the charges have previously been minimal or sporadically collected, can present potentially daunting problems of administration. The manage- ment system and physical plant arrangements must provide checks against theft, fraud, and uneven enforcement of exemption rules and allow for accountability and monitoring of the flow of collected funds. Installation and maintenance of a practical system of collection require staff time, training, and even some minimal equipment, all of which have an associated cost. The collected fees (together with the value of any gains in economic or technical efficiency from the introduction of a price system) must exceed the cost of collection in order to justify implementa- tion of a user charge system. In a survey of fee collection systems in West Africa, Vogel (1988) gives some useful pointers for successful fee collection systems: * Well-defined entrance points for the hospital * The issuance of receipts, with duplicate copies, to serve as evidence of payment * A rigorously enforced system for determining those eligible for exemption * Training for all staff to confirm the importance of enforcing collec- tion * Periodic spot checks to establish that the above points are being carried out by all staff * Periodic audits of the financial transactions and flow of funds. These elements are needed for successful collection with even the most simply defined fee schedule. Daunting though this may seem, the prac- ticability of fees in diverse settings is demonstrated by the existence of active fee collection in nongovernmental, nonprofit hospitals, often in geographic areas where governmental hospitals provide services with- out charge. Optimal pricing. A standard result of economics is the optimality, in the sense of achieving the greatest welfare with a given set of resources, of prices equal to marginal cost. This optimality holds for a competitive equilibrium in markets that supply private goods (as opposed to public or merit goods) to well-informed consumers and in which there is free entry and exit of firms. Hospital services are not provided under these conditions. The failure of the market for hospital services requires regu- lation of provider behavior and may require subsidized provision of services. Under these conditions the greatest welfare within the con- straint of the public budget (or quasi-public institutional budget) can be Hospital Financing Alternatives 157 achieved by prices that reflect the demand and equity characteristics of the good as well as its marginal costs. The appendix to this chapter reviews some rules of optimal pricing as they relate to the problem of pricing hospital services. The rules provide an explicit specification of prices that are consistent with the general thrust of the principles out- lined above. Broadly, the rules contribute to equity goals by incorporat- ing the distribution of income and setting lower prices for services consumed disproportionately by the poor. The rules also contribute to efficiency goals by setting prices that interfere minimally with private preferences. Detailed exploration of optimal pricing would carry the discussion too far afield and into the realm of technical economics, but it is noted here because it holds out some promise of being useful as a guide in setting rational hospital prices that are consistent with planning objectives. Optimal pricing is suggested in this context, not with the thought that the principle should be applied rigidly, but that it can be used for guidance, together with a less quantitative interpretation of the pricing principles, to set prices that can achieve revenue and efficiency objectives without sacrificing equity. Price simulations for three broad categories of bundled services, which can be interpreted as successively higher amenity levels of inpatient care, are discussed as an example in the appendix to this chapter. The simu- lations apply optimal pricing rules, using plausible ranges of required information (on income distribution and distributional objectives, price and income elasticities, budget and subsidy levels) to derive pricing coefficients that are multiplied by the marginal costs of services to give the optimal prices. Some broad implications can be derived from the simulations. First, the optimal prices are small, but positive, for services that would be used by groups that have the lowest income. Given the plausible sizes of income elasticities used in the simulations, the proportion of marginal costs recovered is very small for the category with the lowest income elasticity (this would correspond, say, to ward care) but substantially higher for the category with the highest income elasticity (this would correspond, say, to a private room with special amenities). In hospitals with 80 percent of total cost subsidized from public revenues, the ratio of the pricing coefficients of the highest to the lowest categories of care is approximately 30:1. Second, as the subsidy decreases, that is, as the proportion of total cost to be recovered from patient fees increases, the ratio of the high- to low- category pricing coefficients decreases markedly. Going from an 80 to 40 and then to 0 percent subsidy, the ratio falls from 30:1 to 12:1 and finally to 4:1. Thus, as the subsidy is reduced the latitude for cross-subsidization 158 Public Hospitals in Developing Countries is also reduced, with the lowest elasticity category taking on an increas- ingly greater burden of the cost of services. Even at a zero subsidy, however, the optimal pricing rules generate some cross-subsidization, with the pricing coefficient remaining less than one for the lowest service category but rising to above one for the highest categories. Third, the pricing coefficients are moderately sensitive to the income elasticities. A lower elasticity for a given service produces a lower pricing coefficient on that service and a higher coefficient on other services. At a 40 percent subsidy and with income elasticities of 0, 0.3, and 1.5 for the three categories of services, the pricing coefficients are 0.12, 0.68, and 1.42, respectively. If the income elasticity for the highest category falls to 1.0, the pricing coefficients for the two lowest categories rise to 0.15 and 0.75, whereas the coefficient for the highest category falls to 1.20. This sensitivity illustrates the importance of further empirical studies to derive the demand characteristics of hospital services. Fourth, the pricing coefficients are not highly sensitive to the income distribution through a range of realistic values from recent household surveys, but the coefficients are sensitive to the distributional objectives guiding government pricing policy. A standard normative principle in economics is that added income brings about increased welfare, but successive additions to income bring ever-smaller increments in welfare (this is the principle of diminishing marginal utility). The simulations are based on the subjective assumption that a given percentage increase in income is accompanied by an equal proportional decrease in the increment to welfare. The effect of this assumption is sufficiently human- itarian to lead to relatively aggressive redistribution objectives. If, in- stead, government planners believe that the increment to welfare falls faster (more slowly) than this assumption indicates, the extent of cross- subsidization implied by the pricing coefficients increases (decreases) markedly. The actual pricing schemes in selected public hospitals in which there is an attempt to mount more than a nominal cost-recovery program are not greatly at variance with the implications of these simulations. An examination of the pricing coefficients produced by the simulations and the coefficients derived from the actual costs and prices for a level II hospital in Indonesia serves to illustrate their similarity. Expressly stated goals of Indonesia are to provide accessible hospital services for groups with the lowest incomes and to institute a program of modest cost recovery. On average, public hospitals in Indonesia subsidized about 80 percent of total cost, recovering the remaining 20 percent from user charges. With an 80 percent subsidy the optimal-pricing coefficients from the simulations are 0.03, 0.18, and 0.60, respectively, for the low-, middle-, and high-amenity (low-, middle-, and high-income elasticity) services. Based on the 1984 pricing guidelines of the Ministry of Health Hospital Financing Alternatives 159 as implemented in a small level II hospital, the implicit price coefficients for low-, middle-, and high-amenity services are 0.04, 0.30, and 0.51.5 Thus, the Indonesian coefficients are taken as consistent with the opti- mum pricing simulation, and the pricing scheme used in the hospital is in line with the relatively egalitarian equity goals as well as a revenue objective of 20 percent cost recovery. Risk Sharing The high cost of hospital services, coupled with the randomness of many health needs, is the primary reason for the importance of insurance as a means of financing health services. The introduction of optimal pricing as a means of achieving the equitable distribution of care will not fully adjust for all the equity concerns that arise in the use of hospital services. If the total revenues raised through fees are to offset a substantial part of total costs, then the prices charged for all goods, even those demanded by low-income groups, must be a high proportion of unit costs. Hospital services are by far the most expensive health goods consumed and, depending on the illness, the cost of a hospitalization can easily amount to a multiple of per capita annual income. Modest fees for primary health care services at and below the level of the health center may be absorbed by the majority of the population without a risk of great financial loss, but the introduction of substantial fees for hospital services adds a risk of heavy financial costs to households and creates a need for insurance. Some health needs occur randomly throughout the life of all individ- uals and thus, within a narrow confidence band, are predictable. But savings arrangements or contingent asset and credit markets may be inadequate to finance these predictable costs. In addition, other, unusual, health needs are unpredictable from the point of view of the individual. The cost of adequate treatment for many unpredictable illnesses can easily prove to be a catastrophic burden substantially affecting the welfare of the household. Health insurance improves efficiency by pro- viding a form of earmarked savings for predictable risks and can also improve equity by spreading the risk of the cost of unpredictable illness among all households. The availability of health insurance varies among developing coun- tries. In those in which health services are heavily subsidized, govern- ment is implicitly covering individual risk, though this coverage is not actuarially based. Interpreted in this light, government provision of free services is a form of social insurance with no deductible and with zero coinsurance rates. Such subsidization limits the demand for more ex- plicit forms of health insurance. High administrative costs and a lack of an institutional mechanism for collection in rural areas may also impede the use of health insurance. Also, an actuarially adequate premium may 160 Public Hospitals in Developing Countries be beyond the capacity of many households. For these reasons private insurance markets that cover individuals are often not well developed, and those that do exist usually cover only a small fraction of the popu- lation. Informal insurance arrangements, in which the financial risk is shared among the members of a community or extended family, may exist in some areas, but such mechanisms are apt to function unevenly. Employer plans providing either direct services or third-party risk cov- erage are also not common and cover only selected parts of the popula- tion. Government social insurance programs with a medical care component are more prevalent in Latin America but are rare and gener- ally cover only a small proportion of the population in most African and Asian countries. There is growing interest in government-provided insurance in many emerging economies, but experience in Latin America and the industrial countries demonstrates that adoption of social insurance must be done circumspectly if the programs are not to have unintentionally adverse effects. In many low-income countries the best choice may not be an explicit program of government-sponsored health insurance. In spite of the theoretical advantages of health insurance, specific schemes must be formulated carefully if substantial positive equity and efficiency benefits are to be realized. Poorly devised health insurance schemes and those designed to ben- efit only specific population subgroups can result in a deterioration rather than an improvement in social welfare. For example, the use of health insurance to cover subsets of the population, such as civil servants or urban workers, in a partially monetized economy raises important equity issues. This is especially true if, as is generally the case, govern- ment revenues partially subsidize the superior quality and greater per capita quantity of services consumed by the insured population. Schemes can also create incentives to use resources inefficiently. Finally, the use of insurance introduces problems of administrative feasibility that are even greater than those brought about by the use of fees them- selves because of the need for more careful accounting and administra- tion, both in the collection of premiums and in the delivery of services. This difficulty in administration has been an important reason that insurance schemes have not been used more widely in poor economies. In the next chapter, we review experience with a selection of schemes and note potentially adverse effects in equity and efficiency. In the next few pages, we briefly review the principles of some risk-sharing alterna- tives and note salient aspects related to hospitals. The type of insurance plan and efficiency. There are several possible ways to design health insurance schemes, and within the general formulation of each scheme there are many parameters that must be set correctly if Hospital Financing Alternatives 161 the scheme is to have a positive effect on the overall welfare of the population (Akin 1987). Broadly, we can classify insurance schemes into two principal types: third-party retrospective reimbursement and pre- paid health care organizations. Among the many parameters defining the scheme, we focus our attention on the services to be covered, the magnitude of the health event to be covered, the population groups induded, and the size of the premium and copayment. The choice of scheme and parameters closely affects the functioning of hospitals. Critical to an appropriate insurance plan are the implications it contains for client and provider behavior. Different reimbursement arrangements can bring about marked differences in the kind and quantity of services demanded by patients and given by hospitals. In plans providing retrospective reimbursement on a fee-for-service basis, the individual pays a periodic premium to cover possible expen- ditures for specified services in the future. The premiums from different individuals are pooled and only used to cover services as needed at random by members. Hospitals are reimbursed retrospectively for the cost of each service provided. This payment system gives hospitals an incentive to add to revenues by maximizing the volume of services provided per patient and providing the most costly services possible. As is true of all insurance systems, patients have no incentive to be concerned with the cost of care unless there are cost-sharing provisions. The effect of this hospital reimbursement system is to increase health care costs. One possible means of countering these adverse incentives is a regulatory process whereby hospitals can be denied reimbursement for services determined to be unnecessary. Of course, such a process is itself costly. An alternative retrospective reimbursement plan reimburses the hos- pital a fixed amount per case or admission, which is to cover all services provided to the patient during his or her stay. The cost-containment incentives are superior to those of fee-for-service reimbursement, as the hospital is encouraged to minimize service inputs per admission. On the other hand, the hospital may try to maximize the number of admissions for which its margin of reimbursement above treatment costs is greatest (that is, hospitals may try to choose healthier patients). To mitigate this problem, diagnostic and other patient or procedural information may be used to group patients into categories, each with its own case payment rate (Fetter and others 1980). The difficulties of defining diagnostic categories, establishing and periodically adjusting case reimbursement rates, and policing and administering this type of plan can be formidable, however. Case-based reimbursement raises concerns about quality of care, given the incentives to minimize inputs per case. These concerns can be addressed through some combination of quality-based competi- tion among hospitals or insurers, and utilization review and other forms of regulation. 162 Public Hospitals in Developing Countries Finally, prepaid capitation provider plans (an alternative terminology is health maintenance organization, or HMo) remove provider incentives to increase the cost of care. In these plans, the provider is prepaid a fixed amount to cover health care needs during a specified time interval and then delivers services as needed without further reimbursement. This arrangement focuses on the population rather than on the providers to be reimbursed; therefore, a capitation health plan should involve all levels of personal health services so that the care of individual patients may be managed cost-effectively. One form of this model is direct insurance, in which the provider and insurer are the same institution and thus respond to the same incentives. In contrast to the retrospective reimbursement plans, the capitation plan introduces provider incentives to reduce both the total number of admissions and quantity of services provided per case. With these incentives providers may lower the qual- ity of care in order to reduce the cost of services. As in case-based reimbursement, the means of mitigating such effects would be either competition among prepaid plans or some form of regulation that would ensure quality. In all these schemes there is an incentive, termed "adverse selection," for high-risk patients, whose health needs will probably exceed the cost of the average claim, to join and the converse incentive for low-risk people. There is also an incentive, termed "selection bias," for insurers to exclude high-risk persons from their risk pools in an attempt to maximize the margin between premium income and claims paid. Ad- verse selection can greatly increase the cost of an insurance program if there are significant differences in relative risk across insurance groups, and selection bias can create equity problems if high-risk persons are unable to obtain insurance. The effects of adverse selection and selection bias can be reduced by requiring enrollment in insurance plans across broadly defined client groups or by organizing insurance plans around other broadly defined groups. Insurance can also create an incentive for clients to change their behavior, which can affect the cost and quantity of services demanded. Such changes occur because the clients are less concerned about possible financial loss. For example, high-risk pregnancies and the use of hospital services for delivery might increase if the cost of obstetrics is covered by insurance. This phenomenon is called "moral hazard." The effect of moral hazard can be reduced by limiting benefits or by requiring a copayment or deductible from clients for part of the cost of services. The argument for including cost sharing as part of an insurance scheme is similar to that for instituting user charges in formerly free care systems. Excess use of services resulting from moral hazard is equivalent to what is often termed "unnecessary utilization," which arises in free care systems. In both cases, the high demand is the rational response of consumers faced with a very low priced good. Hospital Financing Alternatives 163 Equity and government insurance. Government-sponsored health insur- ance for specific employment groups such as civil servants has been proposed as a means of transferring resources from urban to rural areas and from high-income to low-income health service users. The argument is that most high-cost hospital services are consumed in urban areas and by specific employment groups whose income is higher than average, and that the revenues obtained from insurance premiums and copayments in urban areas can be used to cross-subsidize rural services for poorer people. This is a valid argument as long as the revenues obtained are greater than the cost of services provided; in this case the cross-subsidization will go in the direction intended. Review of actual experience, however, reveals that revenues seldom exceed the cost of services. Instead, the government subsidy of urban hospital services covered by insurance is often substantial and, depending on the source of revenues, the equity effects are adverse. For example, the ASKES insurance program provides free hospital care for Indonesian civil ser- vants but reimburses hospitals only about 15-25 percent of the cost of an average inpatient stay. Furthermore, the hospitalization rate for ASKES beneficiaries is about five times the national average (Prescott 1991). Thus, moral hazard magnifies the negative impact on equity of provid- ing insurance coverage for subsidized hospital services to a part of the population that is relatively well off. Government health insurance plans, therefore, need to be established in conjunction with a careful choice of services covered, a knowledge of unit costs, and premium and copayment levels that will achieve the desired distributional objectives. One possible way to reduce the adverse equity problem is for the government-sponsored plan to include only a defined minimum benefit package covering amenity levels and services that provide adequate care for the average consumer, to charge fees for services excluded by the benefits package, and to leave the provision of additional coverage to private insurance. Prepaid plans and hospitals. Prospective payment plans can be organized on a modest scale and provide a means of increasing revenues for specific hospitals or groups of hospitals within a community. Through use of a prepayment plan, rural communities served by district hospitals can achieve greater financial autonomy and reduced dependence on central budget sources. Such plans can also yield improved efficiency in the use of providers if the prepayment amount covers all personal health ser- vices for plan members, not just hospitalization. In this case, providers have the incentive to steer enrollees in the plan to the least costly service delivery setting. The advantages of a prepaid plan are that they reduce the incentives for excessive consultations, excessive diagnostic tests, and higher drug use and surgical rates that are reported to exist with a retrospective 164 Public Hospitals in Developing Countries payment system (Shimmura 1988). The administrative costs of prepaid plans are less than those of reimbursement schemes because of the reduced need for billing information and records, and the requirements by management for information within the hospital are less than those in retrospective payment schemes. As is the case for retrospective plans, however, a copayment may be needed to reduce demand for less neces- sary treatment and to improve economic efficiency. Type of service covered. There is a conflict between the need to cover only catastrophic losses to reduce the cost of an insurance plan and the need to include broader, noncatastrophic coverage to avoid the introduction of spurious overuse of higher-cost hospital services. This conflict exists under both retrospective and prospective plans, although it is most relevant for retrospective plans because coverage under such plans is often narrow, whereas prepaid plans commonly provide broad coverage of services. Some authors (for example, Griffin 1989) have recommended that only catastrophic costs be covered. The difficulty lies in defining the basis and limits of the coverage. Perhaps the best possibility would be to set the limits of the coverage on the basis of individual annual health expenditures or (slightly less desirable) on the basis of an individual health event. Insurance would become applicable only above a certain absolute amount. This arrangement would place a burden on the con- sumer of keeping records and then filing claims as justified. The require- ments for literacy, numeracy, and organization may be too great, however, for low-income groups in many countries. A second possibility would be to define "catastrophic" in terms of specific services to be covered, such as inpatient care of more than a certain number of days, emergency services, or specific diseases. This arrangement, although more practical in terms of recordkeeping and administration, could greatly distort the use of hospital services by encouraging providers to use unwarranted services in order to claim insurance coverage. Control- ling such misuse of services would probably require regulation and monitoring. Limiting insurance to catastrophic coverage also results in a loss of the efficiency gain provided by the credit or savings function that comes from covering the cost of subcatastrophic random care. By extending coverage to selected services provided by health centers and hospital outpatient departments, copayment levels for the lower referral levels can be set so that patients are encouraged to use these less costly settings. The cost of the services can then be offset from the insurance fund composed of accumulated premiums. Hospital Financing Alternatives 165 Summary In this chapter we have outlined the principal cost-recovery alternatives to the financing of hospitals from public sector budgets. All the alterna- tives have potential effects on the use and provision of services that are both beneficial and detrimental to welfare. The planner's dilemma is to weigh the benefits and costs associated with the alternatives in order to design a scheme that is most suitable for a given environment. Many of the effects can be attributed to the incentives created by the financing mechanism for consumers and providers. Responses to these incentives may differ across institutional and cultural environments. For example, the behavior of providers depends on whether they are motivated by quantity, revenue, or profit maximization; cost minimization; quality objectives; or some combination of these. The behavior of consumers is affected by their education, income, and perception of the quality of facilities; the range of providers from which to choose; and the degree of practical control the consumer has over treatment. There are dozens of variations on the basic fee and risk-sharing alter- natives. Table 4-3 provides a very general summary of the incentives inherent in six broad financing options, ranging from high fees and no insurance through high fees and partial insurance to capitation pay- ments for managed health care. Some of the options, such as high user charges and no insurance, clearly create incentives that adversely affect equity. Other options, such as free public provision of services or, equivalently, full insurance coverage without cost sharing, create incen- tives that encourage inefficient use of resources. Practical systems that avoid or control these adverse effects through exemptions, partial cov- erage, capitation, or price discrimination are apt to vary and be tailored to the situation in a given country or geographic area. In the next chapter, we examine the actual experiences of hospitals in recovering costs with the financing options outlined above. 166 Public Hospitals in Developing Countries Table 4-3. Summary of Incentives Inherent in Alternative Financing Policies Incentives Policy Providers Consumers High user charges Maximize billable Incentive to reduce use, (fee-for-service) services, possibly but without insurance, constrained by access to expensive awareness that services is difficult for consumers' ability to many pay is limited Fee-for-service plus Maximize billable FoT the insured third-party insurance services; producer is population, quantity with no cost sharing virtually unconstrained demanded is the same as by consumers' ability when money price is to pay zero; no cost consciousness Third party fee-for- Maximize billable Some cost consciousness, service insurance with services but demnand still greater cost sharing than if consumers faced full prices Insurance coverage Minimize costs per Extent of cost using prospectively patient, maximize consciousness depends determined admissions of patients on existence and reimbursement whose treatment costs magnitude of cost- rates per case are less than sharing provisions reimbursement amount (that is, shift case mix); report diagnoses or procedures for ambiguous cases that maximize reimbursement High fees plus Maximize billable Inequitable access to care insurance coverage services; focus on the because price to the of only part of the insured from whom insured is much less population reimbursement is than to the uninsured guaranteed Insurance coverage Keep total health care Guided through the financed through costs for covered network of providers; capitation payments population below sum typically faced with high and administered as a of capitation payments; money prices if they managed care system; this may lead to cost- violate the structure of also direct insurance effective allocation of the managed care system resources or underprovision (especially if there is no competition among providers) Hospital Financing Alternatives 167 Appendix 4A. Optimal Prices for Hospital Services The crux of a cost-recovery policy lies in setting prices and insurance parameters that promote the equity, efficiency, and revenue goals that are the objectives of financing. Fortunately, the general problem of setting prices in the public sector has received growing attention by welfare theorists during the last twenty years, and a flexible theory of public sector enterprise pricing has arisen. The literature is capable of illuminating important pricing and insurance issues and, in particular, holds promise for application to hospitals. Fees A well-known result of welfare economics states that to achieve optimal (Paretian) efficiency, prices for all firms should be set equal to marginal cost: (4-1) P =MC. Were it not for market failure, and equity, the setting of fees for public hospitals could follow this simple rule to achieve economic efficiency. However, market failure and a concern for equity limit the application of marginal cost pricing for hospitals (Baumol and Bradford 1970). Optimal pricing principles for public enterprises provide an adjustment to marginal cost pricing that can be used to address questions of market failure and equity. Public enterprise pricing (Ramsey pricing). To achieve social and political objectives, hospital services are provided at less than cost and supported by government subsidies. Additionally, some hospitals may have declin- ing costs throughout the relevant range of production, and marginal cost will be below average cost. Thus, a price equal to marginal cost will perforce entail a loss, and a subsidy will be required. A pricing rule developed by Ramsey (1927) and later modified by Boiteux (1971) max- in-izes welfare constrained by a budget equal to cost-recovery revenue plus a fixed subsidy. This rule, which forms the basis for pricing in modern public enterprises, can be stated in terms of the price-cost margin (that is, the ratio between the price-cost difference and price) as (4-2) Pi [ i =1. . . n where Ei is the elasticity of demand for service i and x is the added benefit or shadow value of an additional unit of budget (subsidy). 168 Public Hospitals in Developing Countries Although X is not directly observed, the Ramsey rule nevertheless provides important implications for pricing (elaborated in Bos 1985). The rule states that the price-cost margin is inversely proportional to the own price elasticity of the good in question. The Ramsey rule does not provide for explicit cross-subsidization, because the deficit is spread over all services whether luxuries or not, but the rule does affect distributional objectives through the role that price elasticities play in setting prices. In effect, the inverse elasticity of each good is multiplied by the same value, (X - 1)/X, to obtain multipliers that can be used to mark down (or mark up) costs so that the budget deficit (or surplus) is distributed over all goods. Lower (absolute value) elasticities will have higher absolute price-cost margins. If the goods to be priced are produced at a loss (that is, the price-cost margin is negative, as is generally the case for public hospitals), then the rule leads to relatively lower prices of price inelastic goods and higher prices of elastic goods. If the goods purchased by lower-income households are price inelastic, the equity effects of the pricing rule will be favorable compared with, say, those of marginal cost pricing.6 The Ramsey rule is capable of practical application. Given a predeter- mined level of subsidy, costs and estimates of the price elasticities for n goods, the values of the prices, and the critical proportion can be imputed from the n sets of Ramsey relationships and the budget constraint. In practice all the elasticities required may not be available from economet- ric estimates, but subjective estimates together with the restrictions of demand theory may be sufficient to identify the required prices. Public enterprise pricing with distributional weights (Feldstein pricing). The equity benefits of applying the Ramsey rule to deficit budgets are not the result of directly including distributional considerations in the planning process but are an artifact, or side aspect, of the Ramsey result. The equity effects of Ramsey pricing, although a move in the right direction, fall short of the optimal effects that can result from directly including distri- bution. Recent literature that incorporates distributional goals as well as efficiency as a policy objective has grown as a result of seminal work by Feldstein (1972a, 1972b, 1972c). The Feldstein method derives a rule for pricing that maximizes social welfare defined as a weighted function of the consumption of individual households with varying incomes and consumption patterns affected by the level of their incomes. The weights explicitly recognize that the value of additional consumption decreases with rising household income. The Feldstein rule can be expressed as Pi - mci rX-Ril 1 (4-3) P K2E Z=1 . .. n Hospital Financing Alternatives 169 where Ri is the "distributional characteristic" of the good and is, in fact, the weighted function of household consumption. In general the speci- fication of R can be complex. Feldstein suggests an approximation that allows the calculation of the Ri in terms of the mean and variance of income, an estimate of the income elasticity of demand and a normative value for the elasticity of social marginal utility with respect to income.7 The implications of the Feldstein rule for pricing can be elaborated. Lower values of R yield higher prices.8 The value of Ri is inversely related to the income elasticity of demand (see Sherman 1989). Thus, goods with higher income elasticity of demand have relatively lower values of R and higher optimal prices. Ramsey prices can be used for reference. When the income elasticity of demand is 1, Ri will equal 1. Equation 4-3 then reverts to the Ramsey rule. If the income elasticity of demand exceeds 1, the value of R will be less than 1 and the Feldstein price will exceed the Ramsey price. Con- versely, if the income elasticity is less than 1, the Feldstein price will be less than the Ramsey price. Thus, the Feldstein rule increases the prices paid for services consumed by the rich and reduces the prices of services paid by the poor. Cross-subsidization will occur if the price elasticities of the goods consumed by the rich are sufficiently high (in absolute value) that P > MC. The Feldstein prices will also vary with the distribution of income and the rate at which marginal social valuation of income diminishes. The greater the inequality of income distribution or rate at which the mar- ginal valuation of income falls off as income rises, the lower will be the value of R for any given income elasticity. Using the Feldstein approxi- mation for Ri (given in note 7) and an estimated variance of household expenditure for C6te d'lvoire (Glewwe 1988), and choosing an elasticity of social marginal utility of -0.5, we find that the value of Ri is 1.3 for goods with an income elasticity of 0, 1.1 for goods with an income elasticity of 0.5, and 0.8 for goods with an income elasticity of 1.5. Two-part pricing. Optimal public enterprise rules have also been de- rived for two-part pricing, in which the first part is a flat fee for all consumers to have access to services and the second part is a price for services used. Optimal two-part pricing rules are relevant, for example, if an admission fee is charged for hospital inpatient services or an outpatient ward in addition to fees for specific services received. In general two-part pricing rules with distributional weights are similar to the Feldstein rule set out above. The principal difference is that fixed-fee receipts are added to the subsidy in the budget constraint with the result that the value of X changes. We do not pursue two-part pricing further here because it would carry the discussion too far afield, but we do note that the problem is tractable (Feldstein 1972b; Shernan 1989). 170 Public Hospitals in Developing Countries A suggested pricing procedure. In application, a hospital has too many services to allow the practical identification of all the information that would be required to implement Ramsey or Feldstein pricing fully. Instead, a procedure is suggested that would approach, but not fully achieve, the optimal pricing solution and promote increased efficiency and equity in comparison with either zero prices or arbitrary pricing not directly relating costs and demand. The procedure is as follows: * Classify hospital services into large groups of bundled services and differentiate the groups by subjective estimates of the income elasticity of demand into, say, three broad categories: low-income elasticity goods (including inferior goods), normal goods, and luxury goods. * Carry out a unit cost analysis using, say, step down disaggregation of cost as discussed in the previous chapter. Calculate the unit cost (ci) of the services to be priced and the estimated total recurrent cost of the principal service groups. A second subscript can be added to unit cost to indicate the low-, normal-, and high-income elasticity categories as ciL, ciN, and c1H, respectively. The total recurrent costs of the groups can be designated CL, CN, and CH, where, for instance CL = TXcL QiL. Total recur- rent costs are then C =CL + CN + CH * Identify the magnitude of the hospital revenue objective as the ex- pected total recurrent costs (C) less the government subsidy (S), D=C-S. Then, the problem to be addressed in the next steps (d and e) is the selection of markdown coefficients (6,) for unit costs, so that the prices determined, Pi = °i ci, will completely distribute the deficit among the categories, that is, Y-6iL CiL QiL + Y SiN CiN QiN + X OjH CiH QuH = D or, if 5A are the same within each group, 8L X CiL QiL + 6N Y CiN QiN + 8 CiH QiH D, which can be written Hospital Financing Alternatives 171 (4-4) &LCL+ NCN+6HCH=D. * Determine the distribution parameters, Ri. If information on income distribution and income elasticities is available, Ri can be calculated for a chosen elasticity of marginal social utility of income. If information is lacking, a practical approximation is to set the value of RH at 0.75 for the high-income elastic group, RN at 1 for the normal group, and RL at 1.25 for the low-income elastic group of services. * Finally, choose markdown coefficients that are consistent with Feldst- ein prices for each group. This is done through application of equation 4-3. In the absence of better information, use average unit costs (ci) as a substitute for marginal costs (MCi) in equation 4-3. Assuming that price elasticities are comparable within the groups, equation 4-3 yields expres- sions for the markdown coefficients,9 (4-5) 5, = -Ei i = L, N, H The three equations represented by 4-5, together with equation 4-4, can then be solved simultaneously for X, 8L, 8N, and 8H, which can in turn be used to set prices within the groups. As an example we take a hospital with total costs of C = 250, which can be broken down into three primary service categories with group total costs of CL = 100, CN = 100, and CH = 50. (To provide a frame of reference, the three service categories can be thought of as a bundle of services associated with inpatient care in wards, semiprivate care with moderate amenities, and private room care with substantial amenities.) The subsidy is 100, giving a deficit of D = 150 to be financed from revenues on the sale of services. Using RL = 1.25, RN = 1.00, and RH = 0.75 and assuming the price elasticities are -0.05, -0.3, and -1.00 for the income-elastic categories of low, normal, and high, respectively, one can solve the budget constraint (equation 4-4) and the markdown coefficient (equation 4-5) to obtain a markdown coefficient of 8L = 0.12 for low-in- come elastic services, 8N = 0.79 for the middle category of services, and 8H = 1.21 for high-income elastic services. The burden of the deficit then falls on the high-income elastic category, which generates net profits, whereas the lower-income elastic category receives a large share of the cross-subsidy profits and the government subsidy. Sensitivity of the pricing coefficients to the subsidy. The government sub- sidy together with profits is the primary determinant of the markdown coefficients. When the subsidy is very low the proportion of cost recov- ered must be high even for the low-income elasticity good. In contrast, when the subsidy is very high none of the goods is sold at a profit, and the markdown on the low elasticity good becomes nearly complete. The 172 Public Hospitals in Developing Countries data in table 4A-1 illustrate how the markdown coefficients vary with the level of subsidy. The situation described is similar to the preceding example with the exception that the Ri are calculated from the parame- ters specified in the table. When the subsidy is 100 the high-elasticity category is sold at a considerable markup above costs (8H = 1.42) and the lower two categories are cross-subsidized. If the subsidy should rise to 200 (80 percent of total costs), &L would fall to a negligible value (0.02) and even luxury services would be subsidized (FH = 0.59). In contrast, if the subsidy should fall to 0, 5L would rise to 0.4, 5N would go above unity to 1.2, and 6H would rise to 1.8 to provide additional profits to cover the services provided at a loss in the low-income elastic category. Sensitivity of pricing coefficients to the distributional parameters. The pric- ing coefficients are also sensitive to the distributional characteristics of the services, that is, the parameters determining Ri. Data in tables 4A-2 through 4A-4 illustrate the change of the markdown coefficients with alternative values for the relative variance of the income distribution, the income elasticity of marginal social welfare (n), and the income elastici- ties of the services to be priced. The markdown coefficients are not highly sensitive to the income distribution through a range of values based on recent household sur- veys in Ghana (Glewwe and Twum-Baah 1991) and C6te d'Ivoire (Glewwe 1988). Ghana represents a moderately equitable distribution of income with a normalized standard deviation of 0.70 (a variance ratio of 0.49), whereas C6te d'Ivoire represents a considerably less equitable distribution with a normalized standard deviation of 0.91 (a variance ratio of 0.84; in comparison, the variance ratio for the United States in 1970 was 0.55). The suggested distributional parameters to be used in the absence of specific information (see step [d] above) used a normalized standard deviation of 0.64 (a variance ratio of 0.80). The income elasticity of marginal social welfare (rj) is a subjective parameter. The more egalitarian the social objectives determining gov- ernment policy the higher the value of rl that should be chosen. A value of T = -1 implies that a 10 percent increase in income is associated with Table 4A-1. Sensitivity of Pricing Coefficients to the Subsidy Pricing Income Own price Subsidy (percent of total cost) coefficient elasticity elasticity 0 40 80 oL 0.0 -0.05 0.41 0.12 0.02 °N 0.3 -0.40 1.18 0.68 0.18 °11 1.5 -1.00 1.80 1.42 0.59 Note; Eta = -1.0; total cost = 250 (cost of L = 100, cost of N = 100, cost of H = 50). Hospital Financing Alternatives 173 Table 4A-2. Sensitivity of Pricing Coefficients to the Relative Variance of the Income Distribution Relative variance of income 0.84 Pricing Income Own price 0.49 (Cote coefficient elasticity elasticity (Ghana) 0.65 d'Ivoire) xL 0.0 -0.05 0.13 0.21 0.09 °N 0.3 -0.40 0.68 0.64 0.59 °H4 1.5 -1.00 1.42 1.48 1.63 Note: Eta = -1.0; total cost = 250 (cost of L = 1 00, cost of N = 100, cost of H = 50); sub- sidy = 100; net deficit = 150. Table 4A-3. Sensitivity of Pricing Coefficients to the Income Elasticity of Marginal Social Welfare OWtn Pricing Income price Elasticity of marginal social welfare (-Eta) coefficient elasticity elasticity -0.1 -0.5 -1.0 -1.5 -2.0 oZ, 0.0 -0.05 0.26 0.18 0.12 0.08 0.05 °N 0.3 -0.40 0.76 0.73 0.68 0.57 0.47 OH 1.5 -1.00 0.94 1.14 1.42 1.69 1.97 Note: Total cost = 250 (cost of L = 100, cost of N = 100, cost of H = 50). Table 4A4. Sensitivity of Pricing Coefficients to the Income Elasticity of Good 3 Pricing Income Own price Income elasticity of good 3 coefficient elasticity elasticity 0.5 1.0 1.5 OL 0.0 -0.05 0.17 0.15 0.12 5N 0.3 -0.40 0.82 0.75 0.68 °H (see at right) -1.00 1.00 1.20 1.42 Note: Eta = -1.0; total cost = 250 (cost of L = 100, cost of N = 100, cost of H = 50); sub- sidy = 100. 174 Public Hospitals in Developing Countries a 10 percent decrease in marginal social utility. The values in the table range from extremes of 71 = -0.1 to -2.0. Through this broad range the markdown coefficients vary considerably from no cross-subsidization from profits (on the high-elasticity services) at the lower value of r1 to considerable cross-subsidization from profits at the higher value of 11. The suggested distributional parameters to be used in default of other information employ a unit elastic value for 1. Because the services are gathered broadly into only three groups the income elasticities are highly approximate. Also, knowledge of income elasticities depends on household survey and econometric information that may not be available or affordable. The most critical category of goods to classify are those with the higher-income elasticities that pro- vide the profits for cross-subsidization. The sensitivity analysis suggests that the markdown coefficients on the categories that are less income elastic are not sensitive to a plausible range of elasticities for the luxury category of goods. The markdown coefficient for the high-income elastic good is only moderately sensitive, varying by less than 20 percent (from 8 = 1.2 to 1.4) as the elasticity goes from 1.0 to 1.5. The default values for Ri assume income elasticities of 0.0, 0.3, and 1.0. Marginal compared with average costs. The above procedure assumes constant costs, that is, that marginal costs equal average costs or, as will be made clear below, that the ratios of marginal and average costs are the same for all service groups. Commonly, this is taken to be a reason- able approximation of reality. The evidence reviewed in the previous chapter was inconclusive for developing countries, but the consensus, after many econometric studies in industrialized countries, appears to reject hospitals as a declining cost industry. If additional information on marginal costs is available, however, and the marginal cost and average cost ratios are not the same, the costing procedure can be modified to set prices with greater accuracy (in the sense of achieving the welfare and efficiency objectives). If marginal and average costs are not equal the markdown equations become (46) i-= +XE1 _ c1 i = L, N, H. Because the 6i are determined simultaneously they will be affected in comparison with the prices set by using equation 4-5 only if the ratios MCi/ci differ across the groups. The number of groups. The procedure outlined above is generalizable to more than three groups, say n. Additional categories only require the addition of further equations parallel to 4-5 and an appropriate modifi- cation to the budget constraint. Determination of the values of the markdown coefficients, Si, will then require the solution of n + 1 simul- Hospital Finiancing Alternatives 175 taneous equations, and this, as a practical matter, is easily executed using nonlinear iterative techniques. The purpose of the sparse choice of only three groups is to keep the problem tractable for small hospitals that do not have a well-established costing system. It is also probable that the additional welfare and efficiency diminishes rapidly with an expansion of the number of groups much beyond four or five, as long as the elasticity characteristics (both price and income) within the groups do not vary too greatly. Health Insurance and Optimal Pricing (Harris pricing) The introduction of insurance alters the pricing rules set out earlier for an environment without insurance. Harris (1979) has extended the Ramsey analysis to derive optimal prices for a hospital operating in an environment in which all members of the community are covered by insurance. The addition of insurance means that the optimal price will vary with the copayment rate, the price elasticity of demand, and the marginal utility of income. The Harris analysis is explicitly designed for retrospective reimbursement insurance that has a fixed premium equal to the policy's actuarial value and fixed coinsurance rates. The full derivation of Harris's results would take the discussion too far afield, but the results can be summarized in the following pricing rule:10 (4-7) P [ xj ] 5 where wiher = -cov [ x5, x] + (1 - ri) Xi and, in addition to the notation introduced in the previous section, ri is the coinsurance rate (determined exogenously) for service i, xi is the demand of the average consumer for service i, and cov [g5, xJ is the covariance of the marginal utility of income (i'J and the consumption of service i across states of health, s. This rule differs from the Ramsey optimal price (equation 4-3) with the addition of 1i/X,, which adjusts for risk effects across health states. The rule states that the optimal price will be increased by a decrease in the copayment, ri, or the covariance term, cov f[,s, xis]. The covariance term describes the extent to which the demand for specific health services is associated with costly health states and large sacrifices in consumer welfare. If the health service is covered by insurance, the covariance between the marginal utility of income and the quantity of service consumed will be reduced and, because the sign on the covariance term in equation 4-6 is negative, the optimal price is increased. The effect of the copayment rate can be regarded as a correction for the efficiency 176 Public Hospitals in Developing Countries distortion of subsidized services provided under insurance. The covari- ance term represents the equity benefits derived from spreading the risk of expensive health states across the pool of insured persons. Harris used this method to reexamine the prices for services in a United States hospital. Sufficient data were available to allow the pricing of a diverse set of surgical, diagnostic, intensive care, and ward services. To be applied in most hospitals in developing countries, the analysis would require fewer and more aggregated pricing categories. The infor- mation required for the analysis would be estimates of the unit cost of services, the price elasticities of demand, and a survey of the economic and health status of consuming households. Harris pricing is applicable when most or all of the population is covered by insurance; any of the population not covered would be served either by subsidized prices or at full prices as determined by income screening. Thus, Harris pricing may be appropriate for a com- prehensive system of national insurance or for smaller, circumscribed communities served by a given health plan. If insurance covers only a small proportion of the population, however, the Harris prices need to be modified. Adding distributional weights. There are several possible extensions of Harris prices that have not yet been discussed in the literature on optimal pricing but that could be useful for setting policy and should be ad- dressed by future research. First, distributional objectives could be intro- duced directly into the optimization objectives. The distributional effects of Harris pricing are attributable to the welfare gain from spreading the risk of exceptionally costly illness across the pool of insured households. As with Ramsey prices, the inequity of the distribution of income is not explicitly recognized, but the protection insurance provides against market prices at least partly mitigates this failure. Distributional objec- tives parallel to those specified by Feldstein could be combined with Harris's specification to obtain a more complex rule, introducing the distribution term, to set prices. Optimnal coinsurance rates (Arrow prices). As a second modification to Harris prices, it would be possible to reverse the optimization procedure and derive the optimal coinsurance rates, given specified prices. Harris prices are optimal, given a prespecified set of coinsurance rates. From the point of view of a single consumer, Arrow (1976) discussed the welfare implications of changes in coinsurance rates in the situation where prices for health services are determined by the market but did not actually derive optimal rates or consider the distributional im- plications. His findings could, however, be modified to encompass nonmarket prices and to yield optimal coinsurance rates. Optimal coin- Hospital Financing Alternatives 177 surance rates, given prices, are potentially useful when a large part of the population is not served by insurance. A two-step procedure to set prices could provide substantial welfare gain in situations in which the insured comprised only a minority of the population. Feldstein pricing, which does incorporate the distribution of income, could be used to set prices in the first step. The Arrow technique would then be used to determine the optimal coinsurance rates given the prices set in the first step. Notes 1. For nontechnical reviews of the issues, see Griffin 1987; Hoare and Mills 1986; de Ferranti 1985; and Akin, Birdsall, and deFerranti 1987. A more technical, yet accessible, discussion of pricing in the social sectors in general is in Jimenez 1987. Finally, excellent technical reviews of public sector pricing alternatives are Atkinson and Stiglitz 1980; Bos 1986; and Sherman 1989. This chapter draws substantially on all these studies. 2. A general discussion of incomplete markets in given in Laffont 1989. 3. This phenomenon can be made more explicit if we set out the algebraic relation between the net government subsidy and fee retention and examine the change in the net subsidy with the percentage of fees retained in order to identify an optimal retention policy. Given the actual level of autonomous revenues received (HA) and a proportion (p) of these revenues returned to the central treasury, the net government subsidy (S) can be defined as (1) S=S'-p HA O-Tlral[itv--*tTJtitm_ AB_ . :D -U_ POP 0 5 o [ X 0-801B-4532-7 ld