This is an open access article published under a Creative Commons Attribution (CC-BY) License, which permits unrestricted use, distribution and reproduction in any medium, provided the author and source are cited. Article Cite This: Environ. Sci. Technol. 2018, 52, 7928−7936 pubs.acs.org/est Fecal Indicator Bacteria along Multiple Environmental Transmission Pathways (Water, Hands, Food, Soil, Flies) and Subsequent Child Diarrhea in Rural Bangladesh Amy J. Pickering,*,†,‡,▽ Ayse Ercumen,§,∥,▽ Benjamin F. Arnold,§ Laura H. Kwong,†,⊥ Sarker Masud Parvez,# Mahfuja Alam,# Debashis Sen,# Sharmin Islam,# Craig Kullmann,¶ Claire Chase,¶ Rokeya Ahmed,□ Leanne Unicomb,# John M. Colford, Jr.,§ and Stephen P. Luby† † Woods Institute for the Environment, Stanford University, Stanford, California United States ‡ Civil and Environmental Engineering, Tufts University, Science and Engineering Complex, 200 College Avenue, Medford, See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles. Massachusetts United States § Division of Epidemiology, School of Public Health, University of California, Berkeley, California United States ∥ Department of Forestry and Environmental Resources, North Carolina State University, Raleigh, North Carolina United States ⊥ Downloaded via 138.220.206.73 on March 5, 2019 at 20:34:37 (UTC). Civil and Environmental Engineering, Stanford University, Stanford, California United States # Infectious Disease Division, icddr,b Dhaka 1000, Bangladesh ¶ Water Global Practice, World Bank, Washington, D.C. 20433, United States □ Water Global Practice, World Bank, Dhaka 1207, Bangladesh * S Supporting Information ABSTRACT: Enteric pathogens can be transmitted through multiple environmental pathways, yet little is known about the relative contribution of each pathway to diarrhea risk among children. We aimed to identify fecal transmission pathways in the household environment associated with prospectively measured child diarrhea in rural Bangladesh. We measured the presence and levels of Escherichia coli in tube wells, stored drinking water, pond water, child hand rinses, courtyard soil, flies, and food in 1843 households. Gastrointestinal symptoms among children ages 0−60 months were recorded con- currently at the time of environmental sample collection and again a median of 6 days later. Incident diarrhea (3 or more loose stools in a 24-h period) was positively associated with the concentration of E. coli on child hands measured on the first visit (incidence rate ratio [IRR] = 1.23, 95% CI 1.06, 1.43 for a log10 increase), while other pathways were not associated. In cross-sectional analysis, there were no associations between concurrently measured environmental contamination and diarrhea. Our findings suggest higher levels of E. coli on child hands are strongly associated with subsequent diarrheal illness rates among children in rural Bangladesh. ■ INTRODUCTION Diarrhea is a leading cause of global mortality, causing over 1 Additionally, most previous studies have used cross-sectional associations between levels of fecal contamination in the million deaths in the year 2016.1 The morbidity burden of environment and concurrent diarrhea prevalence. For example, diarrhea is also substantial: in 2010 there were an estimated 1.7 the association between fecal contamination levels in stored billion episodes of diarrhea.2 Diarrheal pathogens are trans- drinking water and concurrent diarrhea has been extensively mitted along multiple environmental pathways, traditionally studied, with equivocal results.9,10 One study in Tanzania conceptualized as the “five-Fs”: fluids (water), fingers (hands), found that levels of hand fecal contamination was a stronger food, fields (soil), and flies.3,4 Fecal indicator bacteria and predictor of concurrent diarrheal illness within a household some enteric pathogens have been measured in source water, than fecal contamination levels in stored drinking water.11 environmental waters, stored drinking water, on child and caregiver hands, in stored food, in soil, and on flies in low- Received: February 16, 2018 income countries.5,6 However, there is limited evidence on Revised: June 9, 2018 which of these pathways are the most important for Accepted: June 14, 2018 transmission of diarrhea among young children.4,7,8 Published: June 14, 2018 © 2018 American Chemical Society 7928 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936 Environmental Science & Technology Article Another study in Tanzania found that pathogenic E. coli in visit, as well as prospectively at a second household visit. The stored drinking water was associated with a decrease in the second visit was targeted to be 4−10 days later; however, due odds of concurrent caregiver-reported diarrhea.12 Cross- to field logistical constraints, the range of days between the two sectional associations are difficult to interpret because the visits was wider (2−20 days; median 6 days). The spacing fecal contamination measured in an exposure pathway could between visits was selected to capture symptoms using have caused the diarrhea, or been caused by the diarrhea, or respondent recall based on typical incubation periods for been influenced by human behaviors elicited in response to the gastrointestinal pathogens, such as enterotoxigenic E. coli (3−4 illness (e.g., treatment of water for sick individuals). days), Salmonella (2 days), Shigella (1−2 days), rotavirus (2 The few studies that have prospectively examined fecal days), and norovirus (1−2 days).16 contamination along an exposure pathway in a low-income Households that were absent at the time of the field team’s country and diarrhea have focused on drinking water. Luby et visit were revisited two more times before being marked as loss al. found that contaminated stored drinking water was to follow-up. In households where the pregnant women associated with diarrhea measured 3−100 days later in rural enrolled into WASH Benefits did not have a live birth or where Bangladesh.13 Also in rural Bangladesh, Ercumen et al. the study child died after live birth, the team proceeded with reported that prospective measurement yielded a stronger the interview and sampling if there was any other child <5 association between Escherichia coli in stored drinking water years living in the household. If a household had no other and diarrhea than cross-sectional measurement.14 Neither children <5 years, it was considered lost to follow-up. study measured other fecal transmission pathways. Environmental Sampling and Analysis. We collected Our objective for this study was to assess fecal source water (tube wells), stored drinking water, pond water, contamination along multiple transmission pathways (includ- child hand rinse, soil, food, and fly samples from each ing drinking water, ambient waters, hands, food, soil, and flies), household. At tube well water sources, field workers removed to better understand their contribution to incident child fabric or other materials attached to the tube well mouth and diarrhea in rural Bangladesh. Diarrheal pathogen transmission flushed the tube well by pumping five times.17 They collected pathways are likely heterogeneous across different settings; 250 mL of water directly from the water source using sterile however, studying the relative risk of diarrhea from exposure to Whirlpak bags (Nasco Modesto, Salida, CA). Field workers multiple fecal transmission pathways in this setting could asked the respondent to provide a glass of water (in the same provide useful insight into child diarrheal pathogen exposure in way they would fetch it for their <5 children) to obtain a similar settings. We also examined how a prospective analysis sample of stored drinking water. Field staff asked the with incident episodes measured after exposure compared with respondent to pour the water collected in the glass from the cross-sectional analysis of environmental fecal contamination storage container into a sterile Whirlpak bag. If a pond and concurrently measured diarrhea prevalence. (typically used for bathing or washing dishes or clothes) was ■ METHODS Study Design. The data collection for this study was present in the compound, a sample was collected from the area where the household reported most commonly accessing the pond by dipping a sterile Whirlpak bag into the pond and nested within the WASH Benefits trial in rural Bangladesh, a collecting 250 mL of pond water. One hand rinse sample was multiyear randomized controlled trial of water, sanitation, collected from the youngest child in the household. To sample hygiene and nutrition interventions.15 Compounds (extended hands, field workers rinsed both of the child’s hands in a family groups of 1 or more households sharing a courtyard) Whirlpak bag filled with sterile water.18 To collect soil, field were eligible for participation in the WASH Benefits trial if workers marked a 30 cm by 30 cm area with a metal stencil they had at least one pregnant woman in her first or second (sterilized with ethanol) where the youngest <5 child had most trimester who did not plan to move in the following 24 recently played or spent time. The top layer of soil within the months. There were seven arms in the WASH Benefits trial; stencil was scraped into a sterile Whirlpak bag using a sterile this substudy only included households from the control, scoop; the sample area was scraped once vertically and once sanitation, and combined water, sanitation, and hygiene horizontally to collect approximately 50 g of soil from the (WSH) arms. These arms were selected to allow for analysis ground surface. Stored food to be served to children <5 years of the e ff ect of the sanitation and combined WSH in the household was collected by asking the respondent to interventions on fecal contamination levels, which will be provide a small amount of food in the same manner she fed her reported in a separate manuscript. The WASH Benefits trial child. Food was scooped to fill a 50 mL sterile plastic tube targeted enrolling 720 households in each intervention arm. using a sterile spoon attached to the lid of the tube. To capture Study participants provided written informed consent. Human flies, field workers identified a suitable location in the food subjects approval was obtained from the International Centre preparation area (away from the stove and smoke, under a roof for Diarrheal Diseases Research, Bangladesh (icddr,b) (PR- or protected from rain if possible) and horizontally hung three 11063), University of California, Berkeley (2011−09−3652), 1.5-foot long strips of nonbaited, sticky fly tape. The fly tape and Stanford University (25863). was left in place for 3−6 h to capture flies. Field workers then Data collection for this substudy occurred during the first removed one fly from the center of the strip with the greatest year of the trial after interventions were delivered and extended number of flies using metal tweezers that were sterilized with from July 2013 through March 2014, spanning both the rainy ethanol and placed the fly into a sterile Whirlpak bag. season (Jul−Oct) and dry season (Nov−Mar). We collected All samples were preserved on ice and transported to the data through two successive visits to each study household. field laboratory to be processed on the same day, typically Samples from the household environment were collected within 12 h of collection. Upon arrival at the laboratory, during the first visit and analyzed for fecal indicator bacteria samples were kept on ice until they were processed. Food and levels. Caregiver-reported gastrointestinal illness symptoms in soil samples required a homogenization step before processing children <5 years were recorded cross-sectionally at this first for fecal indicator bacteria detection. These samples were 7929 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936 Environmental Science & Technology Article aliquoted upon arrival in the lab (10 g of food; 20 g of soil) contamination in the environment and thus could detect bias and placed into a sturdy blending bag (BagFilter P, 400 mL, in illness reporting.21 All household survey responses were Interscience, Saint Nom, France) with sterile distilled water recorded electronically using Open Data Kit (ODK) software (100 mL for food; 200 mL for soil). The contents of the installed on tablets. blender bag were then homogenized in a laboratory food Statistical Analysis. We imputed a concentration of zero processer (BagMixer C, Interscience, Saint Nom, France) for 1 MPN fecal contamination for fly samples in households where min at a specified mixing speed (speed 4 for food and speed 2 no flies were captured at the food preparation area and zero for soil). A sterile pipet was used to extract the appropriate MPN fecal contamination for households for pond water volume of the homogenized mixture and further dilute it with samples that did not have access to a pond in the compound distilled water as specified below before processing with the area (because the absence of flies or ponds would indicate no IDEXX Quanti-Tray system. An additional aliquot (5 g) from child exposures via these pathways). MPN counts of fecal the original unhomogenized food and soil samples was indicator bacteria were log10 transformed for the analysis; half weighed and placed in a drying oven overnight for determining of the lower detection limit was substituted for samples below the sample moisture content to calculate bacterial counts per the detection limit to calculate the logarithm. Relevant fecal dry weight of each sample. Fly samples (still encased in the contamination indicators were treated as missing data if the sterile collection bag) were ground with a pestle on a hard household did not have stored drinking water or food available surface to homogenize the fly parts. Distilled water (100 mL) for sampling, the tube well was temporarily not working or if it was added to the Whirlpak bag containing the pulverized fly. was not possible to rinse a child’s hand or collect soil (because The contents were well mixed and further diluted with distilled these types of missing samples would not indicate lack of child water (1 mL of fly mixture was added to 99 mL of distilled exposure). water). We did not prespecify a definition of diarrhea for this Stored water, tube well water, pond water, and hand rinse analysis, so we estimated all models for each of the following samples were diluted with distilled water as follows: no dilution caregiver-reported health outcomes we measured: WHO- for tube well and stored water samples (100 mL of sample defined diarrhea, caregiver-defined diarrhea, and blood in processed directly), 2-fold dilution for hand samples (50 mL of stool. For the prospective analysis, we estimated the incidence sample diluted with 50 mL of distilled water), and 100-fold rate of these outcomes by dividing the number of new cases dilution for pond samples (1 mL of sample diluted with 99 mL since the environmental sampling household visit by the child- of distilled water). All samples were analyzed using the IDEXX days at risk. If a child’s symptoms were not measured on the Quanti-Tray system with Colilert-18 media (IDEXX Labo- first household visit, we excluded them from the prospective ratories, Inc., Westbrook, ME) and incubated for 18 h at 44.5 analysis. For children who did not experience an incident °C.19 Ten percent of trays were recounted by the lab diarrhea episode, we calculated days at risk as the difference supervisor to detect and minimize intercounter variability. between visits. For children with incident diarrhea episodes, we One laboratory control (composed of dilution water) per lab estimated their days at risk assuming onset occurred at technician per day and 2% replicates were processed. Field midpoint between visits as our data collection instrument did workers also collected field blanks by pouring sterile water not record the day of symptom onset; assuming disease onset from a sterile bottle into a Whirlpak as if collecting a stored at the midpoint of the follow-up period is unbiased since the water sample and by opening and massaging a prefilled interval length in this context is independent of disease Whirlpak as if sampling hands. status.22 In the prospective analysis we quantified the Diarrhea Measurement. At each household visit, field association between E. coli levels and diarrhea incidence staff recorded the caregiver-reported gastrointestinal symptoms using the incidence rate ratio (IRR). We modeled binary among children <5 years living in the compound. We recorded gastrointestinal illness outcomes as a function of log10 MPN in symptoms for up to three children on the first visit (prioritized E. coli for each pathway using generalized linear models with a by child age, with the youngest sampled first), and we log link, a Poisson error structure, an offset for each child’s attempted to measure the same children at the second visit. days at risk, and robust standard errors that treated the study’s First, the field staff asked if the child had “diarrhea” using the geographic clusters as independent units. Each pathway was local Bengali term “patla paikana” (“loose stool”). Caregiver- estimated with a separate model since interactions between defined diarrhea may be a useful outcome measure since pathways could have affected estimates in a multivariable caregivers know the typical frequency and consistency of their model.23 The exponentiated coefficient on the E. coli levels children’s bowel movements, and can identify abnormal stool. estimated the incidence rate ratio (IRR) associated with a 1 − Field staff then asked if the child had three or more loose or log10 increase in E. coli MPN. Since rash and bruising were only watery stools in a 24-h period, following the World Health measured on the second household visit, we conducted the Organization definition of diarrhea, and if the child had any negative control prospective analysis by estimating prevalence blood in their stool (an indicator of more severe diarrhea). ratios. In a sensitivity analysis, we re-estimated results using Field staff asked illness questions using a recall period of 7- only data from the control group to remove any potential days.20 On the second household visit, field staff recorded confounding from intervention. caregiver-reported diarrhea (both WHO-defined and caregiver- To compare our prospective analysis with a cross-sectional defined) since the previous household visit, as well as in the analysis approach, we repeated the analyses described above past 7 days (in order to allow for reporting of the 7-day using 7-day prevalence of diarrhea measured at the first diarrhea prevalence at both visits). On the second household household visit as the dependent variable (measured visit, field staff also recorded if the caregiver reported the child simultaneously with the environmental sampling); this analysis had any rash or bruising in the past 7 days. Rash and bruising used illness prevalence as the outcome measure as incident were selected as negative control outcomes in the analysis, as diarrhea or blood in stool episodes could not be identified at they would not be expected to be influenced by levels of fecal the first household visit without prior knowledge of symptom 7930 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936 Environmental Science & Technology Article onset. We estimated the prevalence ratio using a modified visit, 7-day prevalence of rash was 12.2% and 7-day prevalence Poisson generalized linear model (log link) with robust of bruising was 4.7%. The average number of days between the standard errors.24,25 first and second household visits was 7.7 days (median 6; range For pathways that were significantly associated with diarrhea 2, 20; interquartile range [IQR] 6). in the prospective analysis, we conducted a subgroup analysis Among the enrolled households, a total of 9960 environ- by child age to explore if the relationship between fecal mental samples were collected along seven environmental contamination and diarrhea incidence was different for transmission pathways. Sample collection was evenly dis- children that could have different levels of mobility and tributed over the rainy season (43% of samples, Jul 2013−Oct contact with each pathway. The age categories were selected 2013), and dry season (57%, Nov 2013−Mar 2014). At least based on the US Environmental Protection Agency guide- one fly was captured from 34% of households; among lines26 for selecting age groups for monitoring and assessing households with flies, the median number of flies captured child exposure to environmental contaminants, as follows: 0−5 was 2 (range 1, 161). Most (80%) food samples were months, 6−23 months, and 24 months or older. Since this precooked rice, 17% were rice or wheat porridge, and 3% were subgroup analysis was not prespecified, we re-estimated the other types of food. Field staff observed covers on the stored results using different age cut points as a robustness check. The food 85% of the time. All (100%) of sampled food was alternative age categories were selected based on WHO reported to have been prepared in the home (not purchased windows of achievement for hands and knees crawling (5−14 outside). Half (49%) of households had access to a pond in months) and walking alone (8−18 months) and were as their compound. Fecal contamination was prevalent in all follows: immobile (0−4 months), rolling, crawling, and seven pathways (Table 1; see Ercumen et al. for additional learning to walk (5−18 months), and walking well (19−60 months).27 We also re-estimated models restricted only to Table 1. Fecal Indicator Bacteria Levels (Most Probable children that provided a hand rinse sample (considering an Number [MPN] E. coli) by Pathwaya individual child’s hand contamination might be more closely E. coli linked to that individual child’s health). We also conducted subgroup analyses by season (rainy vs dry) to explore if the geometric units N meanb IQR % positive relationship between environmental contamination in each pathway and subsequent diarrhea was modified by seasonality. tube well 100 mL 1676 1 0 24 All models controlled for season (rainy vs dry), household stored water 100 mL 1627 4 20 58 wealth (monthly income over or under 6000 BDT pond 100 mL 824 5393 14350 98 [∼$75USD]), and mother’s formal education (0 vs >0 years) child hands 2 hands 1772 7 8 40 as potential confounders of the relationship between environ- soil dry g 1799 117979 938650 94 mental contamination and diarrhea, as well as controlled for food dry g 1650 2 18 59 study arm (control, sanitation, or WSH). The unit of analysis flies fly 612 715 9670 54 for all models was at the child-level. We did not adjust p-values a Interquartile range (IQR) is the difference between the 25th and for multiple comparisons. 75th percentiles; IQR includes samples with zero E. coli detected. ■ b Geometric mean calculated by including value of 0.5 MPN for RESULTS samples under the detection limit. A total of 2098 households were eligible for enrollment into this substudy from the control, sanitation, and combined WSH details on fecal contamination levels and correlation between arms of the WASH Benefits trial. Of these, we successfully pathways). E. coli was detected in 1% of all lab and field blanks enrolled 1843 households, with 255 households (12%) lost to (8 out of 672 total blanks). The Pearson correlation coefficient follow-up due to stillbirth, miscarriage, abortion, or death of between replicates was 0.9 (n = 233). children in the target age range (7.4%), short-term or long- Prospective Analysis. Estimates of association between term relocation (3.4%), or refusal (1%). A total of 2430 environmental contamination and diarrhea were very similar children <5 years were enrolled into the substudy at the first for WHO-defined diarrhea and caregiver-defined diarrhea household visit and are included in the cross-sectional analysis. (Figure 1 and Tables S1 and S2). Higher E. coli levels on child On the second household visit, 2200 (90%) of these children hands were significantly associated with both incident WHO- were successfully measured and included in the prospective defined and caregiver-defined diarrhea. We estimated the analysis. WHO-defined diarrhea 7-day prevalence was 17.8% at WHO-defined diarrhea incidence rate was 23% higher with the first household visit and 16.9% at the second visit; 7-day each log10 increase in E. coli on child hands (n = 1875 children, prevalence of caregiver-defined diarrhea was 11.9% at the first IRR = 1.23, 95% CI 1.06, 1.43) and 51% higher if any E. coli visit and 10.9% at the second visit; and 7-day prevalence of was detected on child hands (IRR = 1.51, 95% CI 1.17, 1.95; blood in stool was 1.3% at the first visit and 1.5% at the second Table S1). Hand contamination was also the only pathway visit. WHO-defined diarrhea and caregiver-defined diarrhea associated with caregiver-defined diarrhea (Figure 1 and Table were in agreement for 92% of children at the first household S2). Restricting the model to include diarrhea data from only visit and 90% of children at the second household visit. The those children that had their hands sampled for fecal indicator incidence rate of WHO-defined diarrhea between visits was 18 bacterial levels (instead of all <5 children in the compound) episodes per 1000 child-days (250 episodes during 13882 total showed the same relationship between hand contamination child-days at risk); the incidence rate of caregiver-defined and incident WHO-defined diarrhea (n = 1381 children, IRR diarrhea was 13 episodes per 1000 child-days (192 episodes 1.23, 95% CI 1.05, 1.44). during 14877 total child-days at risk); and the incidence rate of Higher E. coli levels in food were significantly associated with blood in stool was 3 episodes per 1000 child-days (46 episodes subsequent blood in stool (Figure 1 and Table S3). We during 16604 total child-days at risk). At the second household estimated that the incidence rate of bloody stool was 34% 7931 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936 Environmental Science & Technology Article Figure 1. Estimates of World Health Organization (WHO) defined incident diarrhea (3 or more watery stools in 24 h), caregiver-defined incident diarrhea, and incident bloody stool associated with log10 MPN E. coli in each fecal transmission pathway, prospectively (left column) and concurrently (right column). Incidence rate ratio (IRR) and prevalence ratios (PR) estimated with generalized linear models with a log link, a Poisson error structure, an offset for each child’s days at risk (IRR models only), and robust standard errors; models were adjusted for season, monthly income, mother’s education, and study arm. The axis for bloody stool is on a different scale than the other outcomes to accommodate wider confidence intervals. higher with each log10 increase in E. coli in stored food (IRR = 40% for 24−60 months. Subgroup analyses by child age ranges 1.34, 95% CI 1.07, 1.68), (Table S3), and the bloody stool revealed that hand fecal contamination was a statistically incidence rate increased by more than 2-fold when E. coli was significant risk factor for diarrhea among children aged 0−5 present in food (IRR = 2.64, 95% CI 1.17, 5.98). E. coli months, but not for children aged 6−23 months or 24−60 presence on hands was also positively associated with bloody months (Table 2; results are similar if model is restricted to stool incidence (IRR = 1.91, 95% CI 1.04, 3.51), while each log10 increase of E. coli on hands were marginally associated only those children that provided hand samples (data not with bloody stool (IRR = 1.37, 95% CI 0.97, 1.94) (Table S3). shown). Using different age range cut-offs gave similar results; The mean prevalence of E. coli on hands across child age hand contamination was only a significant risk factor for ranges was: 37% for 0−5 months, 45% for 6−23 months, and children aged 0−4 months (Table 2). 7932 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936 Environmental Science & Technology Article Table 2. Prospective Associations between WHO-Defined (Left) and Caregiver-Defined (Right) Diarrhea and Fecal Indicator Bacteria (log10 MPN E. coli) Measurement on Child Hands, Stratified by Child Age Range in Monthsa WHO-defined diarrhea caregiver-defined diarrhea age (months) children (n) child-days at risk IRR 95% CI children (n) child-days at risk IRR 95% CI EPA categories 0−5 1007 7505 1.29 1.10 1.51 1096 8384 1.44 1.16 1.77 6−23 424 2672 0.99 0.70 1.40 432 2764 1.04 0.69 1.55 24−60 444 3700 1.26 0.85 1.85 443 3729 1.22 0.79 1.90 WHO categories 0−4 796 5874 1.37 1.13 1.67 871 6622 1.55 1.24 1.96 5−18 613 4097 1.10 0.87 1.39 635 4321 1.10 0.84 1.45 19−60 465 3906 1.23 0.84 1.79 465 3935 1.17 0.77 1.80 a Age categories based on Environmental Protection Agency (EPA) exposure guidelines and World Health Organization (WHO) mobility windows. Incidence rate ratio (IRR) estimated with generalized linear models, adjusted for season, monthly income, mothers’ education, study arm, and child days at risk. The incidence rate of WHO-defined diarrhea was 14 episodes per 1000 child-days at risk during the rainy season ■ DISCUSSION We examined E. coli contamination levels along seven different and 22 episodes per 1000 child-days at risk during the dry diarrheal illness transmission pathways in the household season; the incidence rate of caregiver-defined diarrhea was 12 environment (source water, stored water, pond water, child episodes per 1000 child-days in the rainy season and 14 episodes per 1000 child-days in the dry season; and the hands, courtyard soil, complementary food, and flies caught in incidence rate of blood in stool was 3 episodes per 1000 child- the compound) as risk factors for incident diarrhea among days in the rainy season and 2 episodes per 1000 child-days in children under five. Increased E. coli levels on child hands the dry season. During the dry season, the incidence rate of predicted incident child diarrhea episodes, while other WHO-defined diarrhea was 29% higher with each log10 pathways were not significantly associated. E. coli levels in increase in E. coli on child hands (IRR = 1.29, 95% CI 1.06, food and on child hands were predictors of a child presenting 1.56), while there was no relationship during the rainy season with bloody stool. (IRR = 1.14, 95% CI 0.90, 1.46) (Table S6). Similar results Each log10 increase in E. coli measured on the hands of the were observed for E. coli levels on child hands and caregiver- youngest child in the compound was associated with a 23% defined diarrhea (IRR when dry = 1.43, 95% CI 1.13, 1.81; increase in the incident diarrhea rate among all children <5 IRR when rainy = 1.19, 95% CI 0.94, 1.50; Table S6). During years in the compound. The magnitude of this association the dry season, the incidence rate of blood in stool was 93% appears plausible based on two prospective studies in rural higher with each log10 increase in E. coli on child hands (IRR = Bangladesh that reported each log10 increase in E. coli 1.93, 95% CI 1.26, 2.95) and 48% higher with each log10 measured in stored drinking water increased subsequent increase in E. coli in food (IRR = 1.48, 95% CI 1.16, 1.89); diarrhea prevalence by 14% and 50% (these studies did not these associations with blood in stool were not present or were report incidence estimates).13,14 Molecular analysis of a subset attenuated in the rainy season (Table S6). No other pathways of the child hand rinse samples collected in this study detected were significantly associated with illness outcomes in the rainy rotavirus on 6% of child hands (other pathogens were not or dry season. tested).28 Caregiver hands in Tanzania have also been found to E. coli levels in all transmission pathways were not carry viral and bacterial diarrheal pathogens, including significantly associated with our negative control health pathogenic E. coli, rotavirus, enterovirus, adenovirus, and outcomes (rash and bruising) in the prospective analysis norovirus.12,29 Our finding could reflect the fact that hands are (Table S4). When we restricted the analysis to only include an intermediary pathway between exposure to other pathways data from the control group, E. coli levels on child hands and oral ingestion of fecal contamination; we note that hand E. remained statistically significantly associated with subsequent coli contamination levels were significantly correlated to pond diarrhea by both diarrhea definitions and at similar magnitudes and soil fecal contamination in our study site.30 Although hand as the full sample (WHO-defined IRR = 1.25; caregiver- contamination levels could have varied among children in the defined IRR = 1.26; Table S5). In the control group analysis, E. compound, restricting the model to only include health coli levels in food and on hands were no longer statistically significantly associated with incident bloody stool although the information from children that had provided a hand sample point estimates were in the same direction of effect (Table S5). gave similar results, suggesting hand contamination of one Cross-Sectional Analysis. Simultaneous measurement of child may represent hand contamination among other children child health and environmental fecal indicator bacteria levels within the same compound. This is consistent with a finding did not reveal any statistically significant associations with from rural Tanzania that hand contamination levels among WHO-defined diarrhea, however the magnitude and direction children and caregivers were correlated within the same of the effect estimates were for the most part similar to the household.11 effect estimates for caregiver-defined diarrhea (Figure 1 and The relationship between hand contamination and incident Tables S1 and S2). We estimated statistically significant and diarrhea was strongest among children aged 0−5 months old in positive associations between 7-day prevalence of caregiver- our study. This finding is consistent with the trend of de fi ned diarrhea and concurrent fl y and soil E. coli decreasing hand-mouthing contacts with increasing child age contamination levels (Figure 1, Table S2). In contrast, E. coli documented in our study population.31 In addition, children levels in soil were negatively associated with concurrent bloody aged 0−5 months may be more likely to be exclusively stool (Figure 1 and Table S3). breastfed (limiting exposure to contaminated food or water)32 7933 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936 Environmental Science & Technology Article and are not yet mobile (limiting exposure to soil and other blood.40 The differences in results between prospective and environmental surfaces). cross-sectional analyses suggest studies should not interpret E. coli levels in food and presence of E. coli on child hands cross-sectional associations between fecal indicator bacteria were predictors of incident bloody stool, while only hand E. contamination levels and diarrhea to indicate a causal coli levels were predictive of incident diarrhea. Bloody stool is relationship of environmental contamination leading to child caused by specific pathogens that could have different illness. transmission pathways in the household environment than The two definitions of child diarrhea used in this study gave other diarrheal pathogens.33 For example, Shigella, Campylo- similar results in our analysis (Figure 1). Both definitions are bacter, and Salmonella can cause bloody stool and are widely used in studies of diarrheal disease.41 One advantage of commonly transmitted through food.16 Caregivers may also caregiver-defined diarrhea is that caregivers are familiar with recall bloody stool more accurately than loose stool.34 The their child’s normal bowel movements; for example, a healthy associations we report between E. coli and bloody stool should breastfeeding child with 3 or more loose stools in 24 h would be interpreted with the caveat that bloody stool was a rare be classified as having diarrhea by the WHO definition but may outcome in this study, and underpowered analyses reduce the not by the caregiver. A disadvantage is that caregivers could fail likelihood that detected effects are reflective of true effects. to recognize clinical diarrhea if symptoms are not severe.42 We One potential causal mechanism for child hand contami- cannot rule out that our caregiver-reported diarrhea outcomes nation is contact with soil, including direct contact or contact were subject to bias in this analysis. Caregivers receiving a with soiled surfaces. A typical soil sample in our study had health intervention might be less likely to report that their >100 000 MPN E. coli and >200 000 MPN fecal coliform per child has diarrhea.43 Caregiver-reported diarrhea could there- dry gram.30 A subset of the soil samples in this study was fore provide overestimates of the associations between fecal analyzed for molecular markers of fecal contamination; over indicator bacteria and diarrhea if intervention recipients both two-thirds (331/497 samples) of these had ruminant fecal experience reductions in environmental contamination and contamination detected and one-third (165/497 samples) had underreport diarrhea. However, our research team reports no avian fecal contamination detected (lower prevalences were differences in E. coli levels on child hands, soil, food, or flies detected in stored water and on hands).28 Children 6−23 between the intervention arms and the control group in a months old had the highest levels of E. coli on their hands separate paper (Ercumen 2018, under review). To address (mean log E. coli 0.90, SD 0.84), an age range during which potential reporting bias in intervention arms, we controlled for children spend significant time crawling on the ground.27 intervention status in all models presented in this manuscript. Levels of fecal bacteria on child hands could be representative Additionally, our results were similar when we restricted the of fecal contamination levels in the broader household analysis to only the control group (one-third of the sample environment. size) (Table S5). All identified prospective associations between increased E. This study had several limitations. First, we relied on coli levels in the household environment and subsequent caregiver-reported symptoms of gastrointestinal illness, which gastrointestinal illness were stronger in the dry season and could be subject to bias, as discussed above. Indeed, the were attenuated or not apparent in the rainy season. All measures of diarrhea and negative control outcomes were both pathways had increased levels of E. coli during the rainy season substantially higher in the subset of children included in this compared to the dry season.30 It is possible that E. coli analysis compared with the broader WASH Benefits study contamination on child hands and in food might be dominant population.44 The reasons for these differences are not clear, diarrheal transmission pathways in the dry season, when overall though the children in the present analysis were somewhat fecal bacteria levels in the domestic environment are lower. younger and were assessed by a different separately trained The seasonal differences in our results are consistent with team. However, caregiver-defined and WHO-defined diarrhea previous evidence that weather can affect the transmission of yielded similar associations with measures of environmental diarrheal pathogens. Increased temperatures have been contamination in our prospective analysis. No significant associated with increased diarrhea risk,35−37 while others associations between E. coli and negative control outcomes have documented increased diarrhea after heavy rains and (rash and bruising) suggests that measurement bias was during dry periods.38,39 unlikely to have led to the observed associations between E. Soil E. coli levels and fly E. coli levels were positively coli and diarrhea, assuming the negative control symptoms and associated with caregiver-defined diarrhea in the cross-sectional diarrhea were subject to the same reporting bias. Second, E. analysis but not WHO-defined diarrhea (Figure 1 and Table coli could behave differently in the environment than diarrheal S2). The positive associations could be caused by child pathogens and thus could be a noisy indicator of exposure.45 diarrhea leading to increased contamination in the domestic We measured E. coli as an indicator of bacterial fecal environment (reverse causality). For example, child diarrhea contamination, which could represent bacterial diarrhea could theoretically increase fecal contamination levels in soil if pathogens better than viral or protozoan diarrhea pathogens. the diarrheal feces or anal wash water were not properly Notably, Shigella spp. (which is highly genetically similar to captured and disposed in the latrine, and flies can accumulate enteroinvasive E. coli) was identified as the most attributable fecal indicator bacteria from exposed diarrheal feces. In pathogen to child diarrhea by the GEMS study in Bangladesh, contrast, increasing fecal indicator bacteria contamination while rotavirus, Adenovirus 40/41, and Campylobacter spp. levels on flies and soil were identified as protective for bloody were also identified as important diarrheal pathogens.33 Third, stool in the cross-sectional analysis. One hypothesis to explain our methods to measure E. coli had different detection limits this result is that caregivers may be more likely to engage in and recovery efficiencies across pathways, which could have effective hygiene behaviors (e.g., disposing of feces safely into influenced their relative significance. For example, low recovery the latrine) during a child’s severe illness episode, but not of bacteria from food could have masked the association necessarily during a less severe diarrhea episode without between food bacteria levels and diarrhea. Our results confirm 7934 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936 Environmental Science & Technology Article that E. coli measurements in environmental samples (partic- (2) Walker, C. L. F.; Rudan, I.; Liu, L.; Nair, H.; Theodoratou, E.; ularly solid media such as soil and food) are highly variable, Bhutta, Z. A.; O’Brien, K. L.; Campbell, H.; Black, R. E. Global which necessitates large sample sizes for valid inference. burden of childhood pneumonia and diarrhoea. Lancet 2013, 381 Finally, the results from our study site in rural Bangladesh may (9875), 1405−1416. not be generalizable to other settings with different climates, (3) Kawata, K. Water and other environmental interventions–the minimum investment concept. Am. J. Clin. Nutr. 1978, 31, 2114. diarrheal etiologies, and water and sanitation infrastructure. (4) Eisenberg, J. N. S.; Scott, J. C.; Porco, T. Integrating Disease Prospective analysis may be necessary to accurately Control Strategies: Balancing Water Sanitation and Hygiene characterize the relationship between exposure to fecal Interventions to Reduce Diarrheal Disease Burden. Am. J. Public contamination and diarrhea. Our results suggest that high Health 2007, 97 (5), 846−852. levels of fecal indicator bacteria on child hands is strongly (5) Julian, T. R. Environmental transmission of diarrheal pathogens associated with incident diarrheal illness among young children in low and middle income countries. Environmental Science: Processes in rural Bangladesh. Further research should be done to & Impacts 2016, 18 (8), 944−955. confirm this association in additional settings. (6) Sclar, G. D.; Penakalapati, G.; Amato, H. K.; Garn, J. V.; ■ * ASSOCIATED CONTENT S Supporting Information Alexander, K.; Freeman, M. C.; Boisson, S.; Medlicott, K. O.; Clasen, T. Assessing the impact of sanitation on indicators of fecal exposure along principal transmission pathways: A systematic review. Int. J. The Supporting Information is available free of charge on the Hyg. Environ. Health 2016, 219 (8), 709−723. (7) Wolf, J.; Prüss Ustün, A.; Cumming, O.; Bartram, J.; Bonjour, S.; ACS Publications website at DOI: 10.1021/acs.est.8b00928. Cairncross, S.; Clasen, T.; Colford, J. M., Jr.; Curtis, V.; De France, J.; Prospective and cross-sectional associations between E. et al. Systematic review: Assessing the impact of drinking water and coli contamination along each pathway and World sanitation on diarrhoeal disease in low- and middle-income settings: Health Organization (WHO) defined diarrhea; pro- systematic review and meta-regression. Trop. Med. Int. Health 2014, spective and cross-sectional associations between E. coli 19 (8), 928−942. contamination along each pathway and caregiver-defined (8) Prüss Ustün, A.; Bartram, J.; Clasen, T.; Colford, J. M., Jr.; diarrhea; prospective and cross-sectional associations Cumming, O.; Curtis, V.; Bonjour, S.; Dangour, A. D.; De France, J.; between E. coli contamination along each pathway and Fewtrell, L.; et al. Burden of disease from inadequate water, sanitation bloody stool; prospective associations between E. coli and hygiene in low- and middle-income settings: a retrospective contamination along each pathway and negative control analysis of data from 145 countries. Trop. Med. Int. Health 2014, 19 (8), 894−905. outcomes (rash and bruise); subgroup analysis of (9) Gruber, J. S.; Ercumen, A.; Colford, J. M. Coliform Bacteria as associations between E. coli contamination along each Indicators of Diarrheal Risk in Household Drinking Water: Systematic pathway and illness in the control group only; and Review and Meta-Analysis. PLoS One 2014, 9 (9), e107429. subgroup analyses showing associations between illness (10) Wright, J.; Gundry, S.; Conroy, R. Household drinking water in and E. coli contamination along each pathway by season developing countries: a systematic review of microbiological (rainy and dry) (PDF) contamination between source and point-of-use. Trop. Med. Int. ■ Health 2004, 9 (1), 106−117. AUTHOR INFORMATION (11) Pickering, A. J.; Davis, J.; Walters, S. P.; Horak, H. M.; Keymer, D. P.; Mushi, D.; Strickfaden, R.; Chynoweth, J. S.; Liu, J.; Blum, A.; Corresponding Author et al. Hands, Water, and Health: Fecal Contamination in Tanzanian *E-mail: amyjanel@gmail.com. Phone: 617-627-5163. Communities with Improved, Non-Networked Water Supplies. ORCID Environ. Sci. Technol. 2010, 44 (9), 3267−3272. Amy J. Pickering: 0000-0001-6193-2221 (12) Mattioli, M. C.; Boehm, A. B.; Davis, J.; Harris, A. R.; Mrisho, Ayse Ercumen: 0000-0001-6002-1514 M.; Pickering, A. J. Enteric Pathogens in Stored Drinking Water and on Caregiver’s Hands in Tanzanian Households with and without Author Contributions ▽ Reported Cases of Child Diarrhea. PLoS One 2014, 9 (1), e84939. A.J.P. and A.E are co-primary authors with equal (13) Luby, S. P.; Halder, A. K.; Huda, T. M.; Unicomb, L.; Islam, M. contribution. S.; Arnold, B. F.; Johnston, R. B. Microbiological Contamination of Notes Drinking Water Associated with Subsequent Child Diarrhea. Am. J. The authors declare no competing financial interest. Trop. Med. Hyg. 2015, 93, 904. ■ ACKNOWLEDGMENTS This study was financially supported by the World Bank and by (14) Ercumen, A.; Arnold, B. F.; Naser, A. M. Potential sources of bias in the use of Escherichia coli to measure waterborne diarrhea risk in low-income settings. Trop. Med. Int. Health 2017, 22, 2. (15) Arnold, B. F.; Null, C.; Luby, S. P.; Unicomb, L.; Stewart, C. P.; Grant OPPGD759 from the Bill and Melinda Gates Dewey, K. G.; Ahmed, T.; Ashraf, S.; Christensen, G.; Clasen, T.; Foundation to the University of California at Berkeley. icddr,b et al. Cluster-randomised controlled trials of individual and combined is grateful to the Governments of Bangladesh, Canada, water, sanitation, hygiene and nutritional interventions in rural Sweden, and the UK for providing core or unrestricted Bangladesh and Kenya: the WASH Benefits study design and support. We thank the study participants and excellent field rationale. BMJ. Open 2013, 3 (8), e003476−e003476. staff that made this study possible. ■ (16) CDC. Centers for Disease Control and Prevention (CDC): A- Z Index (website), August 28, 2014. www.cdc.gov/az/e.html REFERENCES (accessed Feb 8, 2017). (1) Naghavi, M.; Abajobir, A. A.; Abbafati, C.; Abbas, K. M.; Abd- (17) Adams, V. D. Water and Wastewater Examination Manual; CRC Allah, F.; Abera, S. F.; Aboyans, V.; Adetokunboh, O.; Afshin, A.; Press, 2017. Agrawal, A.; et al. Global, regional, and national age-sex specific (18) Pickering, A. J.; Boehm, A. B.; Mwanjali, M.; Davis, J. Efficacy mortality for 264 causes of death, 1980−2016: a systematic analysis of Waterless Hand Hygiene Compared with Handwashing with Soap: for the Global Burden of Disease Study 2016. Lancet 2017, 390 A Field Study in Dar es Salaam, Tanzania. Am. J. Trop. Med. Hyg. (10100), 1151−1210. 2010, 82 (2), 270−278. 7935 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936 Environmental Science & Technology Article (19) Julian, T. R.; Islam, M. A.; Pickering, A. J.; Roy, S.; Fuhrmeister, a Systematic Review of Relationships between Diarrheal Diseases and E. R.; Ercumen, A.; Harris, A.; Bishai, J.; Schwab, K. J. Genotypic and Temperature, Rainfall, Flooding, and Drought. Environ. Sci. Technol. phenotypic characterization of Escherichia coli isolates from feces, 2016, 50, 4905. hands, and soils in rural Bangladesh via the Colilert Quanti-Tray (36) Kolstad, E. W.; Johansson, K. A. Uncertainties Associated with System. Appl. Environ. Microbiol. 2015, 81 (5), 1735−1743. Quantifying Climate Change Impacts on Human Health: A Case (20) Arnold, B. F.; Galiani, S.; Ram, P. K.; Hubbard, A. E.; Briceno, Study for Diarrhea. Environ. Health Perspect 2010, 119 (3), 299−305. B.; Gertler, P. J.; Colford, J. M. Optimal Recall Period for Caregiver- (37) Carlton, E. J.; Woster, A. P.; DeWitt, P.; Goldstein, R. S.; Levy, reported Illness in Risk Factor and Intervention Studies: A K. A systematic review and meta-analysis of ambient temperature and Multicountry Study. Am. J. Epidemiol. 2013, 177, 361. diarrhoeal diseases. Int. J. Epidemiol 2016, 45 (1), 117−130. (21) Arnold, B. F.; Ercumen, A.; Benjamin-Chung, J.; Colford, J. M. (38) Carlton, E. J.; Eisenberg, J. N. S.; Goldstick, J.; Cevallos, W.; Brief Report: Negative Controls to Detect Selection Bias and Trostle, J.; Levy, K. Heavy Rainfall Events and Diarrhea Incidence: Measurement Bias in Epidemiologic Studies. Epidemiology 2016, 27 The Role of Social and Environmental Factors. Am. J. Epidemiol. (5), 637−641. 2014, 179, 344. (22) Vandormael, A.; Dobra, A.; Bar ̈ nighausen, T.; de Oliveira, T.; (39) Bhavnani, D.; Goldstick, J. E.; Cevallos, W.; Trueba, G.; Tanser, F. Incidence rate estimation, periodic testing and the Eisenberg, J. N. S. Impact of rainfall on diarrheal disease risk limitations of the mid-point imputation approach. Int. J. Epidemiol associated with unimproved water and sanitation. Am. J. Trop. Med. 2018, 47 (1), 236−245. Hyg. 2014, 90 (4), 705−711. (23) Eisenberg, J. N. S.; Trostle, J.; Sorensen, R. J. D.; Shields, K. F. (40) Khush, R. S.; Arnold, B. F.; Srikanth, P.; Sudharsanam, S.; Toward a systems approach to enteric pathogen transmission: from Ramaswamy, P.; Durairaj, N.; London, A. G.; Ramaprabha, P.; individual independence to community interdependence. Annu. Rev. Rajkumar, P.; Balakrishnan, K. H2S as an Indicator of Water Supply Public Health 2012, 33, 239−257. Vulnerability and Health Risk in Low-Resource Settings: A (24) Zou, G. A Modified Poisson Regression Approach to Prospective Cohort Study. Am. J. Trop. Med. Hyg. 2013, 89, 251. Prospective Studies with Binary Data. Am. J. Epidemiol. 2004, 159 (41) Schmidt, W.-P.; Arnold, B. F.; Boisson, S.; Genser, B.; Luby, S. (7), 702−706. P.; Barreto, M. L.; Clasen, T.; Cairncross, S. Epidemiological methods (25) Yelland, L. N.; Salter, A. B.; Ryan, P. Performance of the in diarrhoea studies–an update. Int. J. Epidemiol 2011, 40 (6), 1678− Modified Poisson Regression Approach for Estimating Relative Risks 1692. From Clustered Prospective Data. Am. J. Epidemiol. 2011, 174 (8), (42) Wilson, S. E.; Oued ́ raogo, C. T.; Prince, L.; Oued ́ raogo, A.; 984−992. Hess, S. Y.; Rouamba, N.; Oued ́ raogo, J.-B.; Vosti, S. A.; Brown, K. H. (26) U.S. EPA. Guidance on Selecting Age Groups for Monitoring and Caregiver Recognition of Childhood Diarrhea, Care Seeking Assessing Childhood Exposures to Environmental Contaminants, EPA/ Behaviors and Home Treatment Practices in Rural Burkina Faso: A 630/P-03/003F; U.S. Environmental Protection Agency, 2006. Cross-Sectional Survey. PLoS One 2012, 7 (3), e33273. (27) Onis, M. WHO Motor Development Study: Windows of (43) Clasen, T.; Boisson, S. Assessing the Health Impact of Water achievement for six gross motor development milestones. Acta Quality Interventions in Low-Income Settings: Concerns Associated Paediatr. 2006, 95, 86−95. with Blinded Trials and the Need for Objective Outcomes. Environ. (28) Boehm, A. B.; Wang, D.; Ercumen, A.; Shea, M.; Harris, A. R.; Health Perspect 2016, 124 (7), 886−889. Shanks, O. C.; Kelty, C.; Ahmed, A.; Mahmud, Z. H.; Arnold, B. F.; (44) Luby, S. P.; Rahman, M.; Arnold, B. F.; et al. Effects of water et al. Occurrence of Host-Associated Fecal Markers on Child Hands, quality, sanitation, handwashing, and nutritional interventions on Household Soil, and Drinking Water in Rural Bangladeshi House- diarrhoea and child growth in rural Bangladesh: a cluster randomised holds. Environ. Sci. Technol. Lett. 2016, 3 (11), 393−398. trial. Lancet Global Health 2018, 6, e302. (29) Mattioli, M. C. M.; Davis, J.; Mrisho, M.; Boehm, A. B. (45) Savichtcheva, O.; Okabe, S. Alternative indicators of fecal Quantification of Human Norovirus GII on Hands of Mothers with pollution: relations with pathogens and conventional indicators, Children Under the Age of Five Years in Bagamoyo, Tanzania. Am. J. current methodologies for direct pathogen monitoring and future Trop. Med. Hyg. 2015, 93 (3), 478−484. application perspectives. Water Res. 2006, 40 (13), 2463−2476. (30) Ercumen, A.; Pickering, A. J.; Kwong, L. H.; Arnold, B. F.; Parvez, S. M.; Alam, M.; Sen, D.; Islam, S.; Kullmann, C.; Chase, C.; et al. Animal Feces Contribute to Domestic Fecal Contamination: Evidence from E. coli Measured in Water, Hands, Food, Flies, and Soil in Bangladesh. Environ. Sci. Technol. 2017, 51 (15), 8725−8734. (31) Kwong, L. H.; Ercumen, A.; Pickering, A. J.; Unicomb, L.; Davis, J.; Luby, S. P. Hand- and Object-Mouthing of Rural Bangladeshi Children 3−18 Months Old. Int. J. Environ. Res. Public Health 2016, 13 (6), 563. (32) Joshi, P. C.; Angdembe, M. R.; Das, S. K.; Ahmed, S.; Faruque, A. S. G.; Ahmed, T. Prevalence of exclusive breastfeeding and associated factors among mothers in rural Bangladesh: a cross- sectional study. International Breastfeeding Journal 2014 9:1 2014, 9 (1), 7. (33) Liu, J.; Platts-Mills, J. A.; Juma, J.; Kabir, F.; Nkeze, J.; Okoi, C.; Operario, D. J.; Uddin, J.; Ahmed, S.; Alonso, P. L.; et al. Use of quantitative molecular diagnostic methods to identify causes of diarrhoea in children: a reanalysis of the GEMS case-control study. Lancet 2016, 388 (10051), 1291−1301. (34) Ercumen, A.; Arnold, B. F.; Kumpel, E.; Burt, Z.; Ray, I.; Nelson, K.; Colford, J. M. Upgrading a Piped Water Supply from Intermittent to Continuous Delivery and Association with Water- borne Illness: A Matched Cohort Study in Urban India. PLoS medicine 2015, 12 (10), e1001892−24. (35) Levy, K.; Woster, A. P.; Goldstein, R. S.; Carlton, E. J. Untangling the Impacts of Climate Change on Waterborne Diseases: 7936 DOI: 10.1021/acs.est.8b00928 Environ. Sci. Technol. 2018, 52, 7928−7936