Diarrhea prevalence in a randomized, controlled prospective trial of point-of-use water filters in homes and schools in the Dominican Republic

23

Lack of sustainable access to clean drinking water continues to be an issue of 24 paramount global importance, leading to millions of preventable deaths annually. Best practices 25 for providing sustainable access to clean drinking water, however, remain unclear. Widespread 26 installation of low-cost, in-home, point of use water filtration systems is a promising strategy. 27

We conducted a prospective, randomized, controlled trial whereby 16 villages were 28 selected and randomly assigned to one of four treatment arms based on the installation location 29 of Sawyer® PointONE™ filters (filter in both home and school; filter in home only; filter in 30 school only; control group). Water samples and self-reported information on diarrhea were 31 collected at multiple times throughout the study. 32

Self-reported household prevalence of diarrhea decreased from 25.6% to 9.76% from 33 installation to follow-up (at least 7 days, and up to 200 days post-filter installation). These 34 declines were also observed in diarrhea with economic or educational consequences (diarrhea 35 which led to medical treatment and/or missing school or work) with baseline prevalence of 36 9.64% declining to 1.57%. Decreases in diarrhea prevalence were observed across age groups. 37 There was no evidence of a loss of efficacy of filters up to 200 days post filter installation. 38 Installation of filters in schools was not associated with decreases in diarrhea prevalence in 39 school-aged children or family members. Unfiltered water samples both at schools and homes 40 contained potential waterborne bacterial pathogens, dissolved heavy metals and metals 41 associated with particulates. All dissolved metals were detected at levels below World Health 42 Organization action guidelines. 43 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 3

This controlled trial provides strong evidence of the effectiveness of point-of-use, 44 hollow fiber membrane filters at reducing diarrhea from bacterial sources up to 200 days post 45 installation when installed in homes. No statistically significant reduction in diarrhea was found 46 when filters were installed in schools. Further research is needed in order to explore filter 47 efficacy and utilization after 200 days post-installation. Trial registration: ClinicalTrials.gov, 48 NCT03972618. Registered 3 June 2019 - Retrospectively registered, 49 https://clinicaltrials.gov/ct2/show/NCT03972618. 50

51 Drinking Water, Point-of-Use Filter, 16S rRNA Community, Diarrhea, Heavy Metals 52

53 Globally, diarrhea caused over 1.65 million deaths in 2016 [1] and half a million deaths in 145 54 low- and middle- income countries in 2012 due specifically to inadequate drinking water [2]. 55 According to the World Health Organization (WHO), there are 1.7 billion cases of childhood 56 diarrheal disease annually and 525,000 children under 5 die from the disease each year, making 57 diarrheal disease the second leading cause of death in children under 5 [3]. Since a major source 58 of diarrhea is fecal pathogens via fecal-oral transmission [4, 5] many of these lives could have 59 been saved through clean drinking water [2] and proper hand hygiene [5, 6]. 60 The Dominican Republic is considered a middle-income country [7], and thus is at potentially 61 high risk for negative impact of diarrheal disease. Results of household surveys confirm high 62 prevalence of diarrhea in children under 5 in the Dominican Republic (32.3% treated for diarrhea 63 with oral rehydration salts in 2002; 46.3% in 2007 and 52.8% in 2013) [8]. Other studies have 64 found a similarly high burden of diarrheal disease. In 2003 the Dominican Republic 65 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 4 Demographic and Health Survey reported that an average of 14% of all children under five years 66 suffered from diarrhea, with rates up to 29% in certain provinces [9]. 67 Other studies have found related concerns about sustainable access to clean drinking water. In a 68 peri-urban district of Santo Domingo, a study of 266 households found that although 57% of 69 interviewees believed their child to be at risk for diarrhea, and 90.6% believed that boiling water 70 could prevent diarrhea, only 42% reported that it was rare for their child to drink untreated water 71 with 34.5% stating ‘insufficient fuel’ as the primary barrier [10]. Another study conducted in the 72 Puerto Plata region to evaluate E. coli levels in household water sources found that in 73 unimproved water sources, 47% were of high to very high risk according to WHO water quality 74 guidelines and in improved sources, 48% were of high to very high risk [11]. 75 While multiple options for reducing diarrheal prevalence by providing clean drinking water and 76 using proper hand hygiene exist, inexpensive but potentially highly effective, point-of-use 77 solutions remain an under-utilized option. One such option is the Sawyer® PointONE™ water 78 filter. Laboratory tests with the Sawyer® PointONE™ water filter suggest it aligns with the United 79 States Environmental Protection Agency standard for bacteria and protozoa removal [12]. Prior 80 studies have identified a significant decrease in diarrhea prevalence [12, 13]. Some studies have 81 argued that filters in the field have been fouled and under-utilized in practice [14, 15], however, 82 others have noted numerous limitations of these studies [16] and reasonably good performance at 83 removing E. coli and coliforms over a one- to three-year period in the field [17]. Thus, there is a 84 continued need for carefully designed field trials to evaluate the efficacy of Sawyer® 85 PointONETM and other hollow fiber membrane filters. 86 In an attempt to better understand filter efficacy, utilization, and the impact of deployment in 87 different community locations, we designed a prospective, randomized, controlled trial whereby 88 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 5 16 villages in the Dominican Republic were selected and randomly assigned to one of four 89 treatment arms based on the location of Sawyer® PointONE™ filter installation (filter in both 90 home and school; filter in home only; filter in school only; control group). Village households 91 were followed over time, monitoring self-reported health characteristics. Drinking water samples 92 were also obtained from a subsample of households to monitor water quality for bacterial and 93 chemical contamination in order to, first, directly assess filter removal of bacteria and 94 particulates and, second, to assess whether the sole use of this type of filter as an intervention 95 was contraindicated due to the presence of toxic levels of dissolved heavy metals. 96 97

98 Sixteen villages in Dominican Republic were selected for inclusion in the study. Each selected 99 village sent school-aged children to a private school within a pre-identified private school 100 network. Each school draws its students from a unique geographic area (i.e., village) surrounding 101 the school. Rural and urban villages across the country were included (Figure 1). The geographic 102 distribution of schools (North, South, East, Capital-1, Capital -2 where “Capital” indicates 103 schools in and near Santo Domingo) is shown in Table 1. Donor funding was available through a 104 non-profit organization to provide point-of-use water filtration systems (Sawyer® PointONE™ , 105 additional details on filter construction described elsewhere [13]) to in-network, sponsored 106 households with a school-aged child that attended the private school. The total number of 107 sponsored households eligible for inclusion at the start of the study in August 2018 was 675, 108 with breakdown by village shown in Table 1. 109 110 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 6 111 112 113 Table 1. Household characteristics by village 114 De-identified village name Total Sponsored Households Eligible for inclusion1 Total households in primary analyses2 Village Response Rate Initial Treatment Group3 Geographic location School received filter?4 A 26 25 96.2% Control North No B 30 21 70.0% Simultaneous Capital – 1 Yes C 35 17 48.6% Control Capital – 2 No D 32 5 15.6% Home Capital – 1 No E 46 38 82.6% Control North No F 60 41 68.3% Home North No G 18 12 66.7% Home North No H 40 36 90.0% School Capital – 2 Yes I 38 22 57.9% Control South Yes J 24 3 12.5% School East Yes K 72 4 5.6% Home East No L 50 16 32.0% Control East No M 47 32 68.1% Home Capital – 2 Yes N 62 8 12.9% Control East Yes O 50 15 30.0% Simultaneous South Yes P 45 27 60.0% Home Capital – 1 No Total 675 322 47.7% -- -- -- 1 Number of sponsored households within the village as of August 2018. 115 2 Use in primary analyses means that at least one household survey pre-household filter installation and at least one 116 household survey post-household filter installation are available, consent of the respondent was received and at least 117 90% of the survey question responses were valid/not missing. 118 3 Since all households and eligible schools ultimately received a filter, treatment groups listed in the table are 119 “Initial” groups with subsequent installation of school/home measurements and commensurately collected survey 120 data. 121 4 Nine schools declined receiving a filter at the school because they determined they were not in need of a filter. 122 Note: No water samples were taken before this determination was made. 123 124 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 7 Randomization procedures 125 In order to evaluate the efficacy of filter installations in both schools and homes, a modified 126 factorial design was used (Figure 2). Each of the sixteen villages was initially assigned to one of 127 the following treatment groups: control (no filter initially), home filter only, school filter only or 128 simultaneous (school and home) filter installation. Assignment of villages to treatment groups 129 (see Table 1) followed a covariate adaptive randomization strategy, whereby predefined 130 covariates are balanced across treatments using the method of minimization [18]. We performed 131 the method of minimization including the following covariates: (a) reported number of sponsored 132 children at each school as of August 2018 (high [51+]) vs. low [50 or fewer]), (b) geographic 133 location/watershed (five groups; [19]), (c) whether the school anticipated receiving a filter or not 134 (some schools declined needing filters due to already having a self-reported ‘safe drinking water 135 solution’) and (d) government public health statistics regarding endemic levels of diarrhea in the 136 region [9] in children under 5 (15% or more vs. 15% or less). Villages initially assigned to the 137 control group subsequently were assigned to the school filter or home filter treatment group. All 138 treatment groups received filters in the home by the end of the study. 139 Survey and data collection 140 We administered a short demographic, health and economic baseline questionnaire to all 141 households upon installation of point-of-use filter in the home [13]. Five local data collectors 142 were trained by the research team. Each data collector was assigned to a village based on pre-143 determined geographic areas (Table 1) and was the same person for pre-filter (control group; 2-8 144 weeks prior to filter installation), baseline (filter installation) and follow-up surveys at respective 145 households and schools. Data collection took place from September 2018 through April 2019. 146 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 8 Filter installation included a brief training on use of the filter by the data collector, mention of 147 importance of handwashing and a demonstration by the respondent of proper filter use and 148 backflushing. After installation of the filter, data collectors attempted to collect two-, eight- and 149 sixteen-week follow-up surveys. This study was approved and monitored by Dordt University 150 IRB and all surveys included a statement of consent for research use. The trial is registered with 151 www.clinicaltrials.gov as NCT03972618. 152 Eligibility for inclusion in analyses in this manuscript required: (a) at least one survey at or 153 before the time of filter installation (pre-survey), (b) at least one survey completed at least two 154 weeks after filter installation (post-survey), (c) obtained consent for data to be used for research 155 purposes and (d) having provided answers to at least 90% of the survey questions. Using these 156 criteria, the overall eligible data response rate was 47.7% (322/675 households; see Table 1) with 157 variation in response rates by village (minimum = 5.6%; maximum = 96.2%) and region 158 (generally lower response rates in the lowlands (East) region). 159 Across the 322 households in the primary analysis, there were 1075 household-level survey 160 administrations. These household surveys consisted of 72 control group baseline survey 161 responses, 322 baseline survey responses upon filter installation, and 681 follow-up survey 162 responses (mean number of follow-up survey responses per household was 2.11). Among the 163 681 follow-up survey responses, 107 of the responses indicated not using the filter and thus were 164 ignored in primary analyses. An intent to treat sensitivity analysis included these households. 165 Survey variables 166 The primary outcome we considered was self-reported, two-week prevalence of diarrhea, 167 including whether diarrhea had economic or educational consequences, causing hospitalization 168 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 9 and/or missed school or work. The household survey administration included questions that 169 determine: (a) the size of the household (number of adults and children living in the household), 170 and then for each household member, whether they have in the past two weeks, (b) had diarrhea, 171 and if so, whether that diarrhea (c) caused missed days of work (adults) or missed days of school 172 (children), and (d) required hospitalization. Additional information used in analysis was (1) 173 season (Fall [Oct, Nov], Winter [Dec, Jan, Feb] or Spring [Mar, Apr]), (2) days since filter 174 installation, (3) region (East, West, South, Central – 1, Central- 2), and (4) water source (city, 175 purchased, well or other (e.g. river, catchment)). 176 Water sampling methods 177 Between two and seven untreated drinking water source samples were collected at each of the 178 sixteen locations (56 total samples). Information on location, date of the sample acquisition and 179 whether the sample came from a home or from the school can be found in Additional File 1. 180 Water samples were tested for the presence of dissolved heavy metals, heavy metals associated 181 with particulate matter, and potential bacterial pathogens using Sawyer® PointONE™ hollow 182 fiber membrane filters to capture particulates and bacteria and using a metal chelating foam to 183 capture dissolved heavy metals. ICP-OES was used to measure dissolved heavy metals that have 184 been identified by the WHO as potential health hazards (Arsenic, Barium, Cadmium, Chromium, 185 Copper, Lead, Nickel and Selenium) [20]. A combination of spectrophotometry, SEM-EDS, and 186 powder X-Ray diffraction (PXRD) were used to determine the concentration of particulates 187 (mg/L) and whether those particulates contained heavy metals of interest. Lastly, the presence of 188 eighteen bacterial genera identified by the WHO as containing potential waterborne pathogens 189 [21] (Acinetobacter, Aeromonas, Burkholderia, Campylobacter, Enterobacter, 190 Escherichia/Shigella, Francisella, Helicobacter, Klebsiella, Legionella, Leptosipra, 191 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 10 Mycobacterium, Pseudomonas, Salmonella, Staphylococcus, Tsukamurella, Vibrio and Yersinia) 192 were assessed through 16S rRNA amplicon sequencing following the Schloss Lab MiSeq SOP 193 [22, 23] for sequencing and initial data processing to produce 97% operational taxonomic units 194 and assigned taxonomies using the Silva Release 132 alignment and database [24]. Downstream 195 analyses of 16S amplicon sequencing data were performed using the R packages Phyloseq [25] 196 and vegan [26]. Details of water sampling, testing and analysis methods are available in 197 Additional File 2. Sequencing data associated with this study have been deposited in the Short 198 Read Archive under PRJNA670359 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA670359). 199 Statistical analysis 200 Generalized linear mixed-effects models with a logistic link function were used to predict 201 diarrhea prevalence by filter status. Random effects were used for repeated measures, with 202 additional fixed effects covariate adjustments for season, water source and household size. 203 Generalized linear mixed effects models with random effects for repeated measures were also 204 used to compare water sampling data between pre- and post-filter installation measurements. All 205 statistical analyses were completed using R version 3.6.2 [27] and used a significance level of 206 0.05. 207

208 Characteristics of the villages 209 Characteristics of the villages are shown in Table 2. Purchased water usage fell from 36% to 210 10% from baseline to follow-up, likely reflecting reliance on the filter to provide safe and clean 211 water. Modest, though statistically significant, changes in region were observed from baseline to 212 follow-up (e.g., Capital – 1 accounted for 21% of follow-up data, but only 12% of baseline data 213 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 11 due to proportionally more follow-up surveys conducted in this region as compared to others). 214 As expected, the percent of households with a filter at their children’s school was somewhat 215 higher at follow-up (31% vs. 18%). 216 Table 2. Village Characteristics 217 Variable Before home filter installation (n=394) Mean (SD) or %(x/n) After home filter installation (n=574) Mean (SD) or %(x/n) P-value for difference1 Household size 4.60 (1.72) 4.41 (1.33) 0.049* Region North 38.8% (153/394) 35.0% (201/574) 0.004** East 8.6% (34/394) 5.9% (34/574) South 15.0% (59/394) 15.5% (89/574) Capital – 1 12.4% (49/394) 21.4% (123/574) Capital – 2 25.1% (99/394) 22.1% (127/574) School filter status Before/Declined 82.0% (323/394) 68.6% (414/594) <0.001*** After 18.0% (71/394) 31.4% (180/594) Season Fall 67.3% (265/394) 19.2% (110/574) <0.001*** Winter 32.7% (129/394) 57.3% (329/574) Spring 0.0% (0/394) 23.5% (135/574) Water source City 49.4% (195/394) 79.6% (457/574) <0.001*** Purchased 36.0% (142/394) 10.3% (59/574) Well 10.4% (41/394) 8.5% (49/574) Other 4.1% (16/394) 1.6% (9/574) *p<0.05; **p<0.01, ***p<0.00 218 1 Logistic regression model with random effects term accounting for repeated measures. 219 220 Two-week diarrhea prevalence by home filter status 221 Before filter installation, 25.6% of households reported at least one member of the household 222 experiencing diarrhea within the previous two weeks (Table 3; Figure 3). Prevalence of diarrhea 223 dropped substantially to 9.8% after filter installation, a difference which remained statistically 224 significant even after adjusting for other variables (adjusted Odds Ratio (aOR) =0.29 (95% CI: 225 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 12 0.16, 0.51), p<0.001). Adjusted odds ratios were statistically significant for adult-only diarrhea 226 and for children under 4 years (0.14 and 0.09, respectively). While the aOR was still small (0.49) 227 it was not statistically significant (p=0.07) for school-aged children after adjusting for other 228 variables (Table 3). We also completed an intent to treat analysis including the 107 households 229 reporting not using the filters (total n=681) which demonstrated higher post filter installation 230 diarrhea prevalence overall (10.3%), as well as in Adults (4.8%) and Children under the age of 5 231 (7.1%, with slightly lower prevalence in school aged children (5.7%; Additional File 3). 232 Prevalence of diarrhea was significantly lower than baseline throughout this study up to 200 days 233 post filter installation (adjusted p<0.001 for all three of: = 7-50 days vs. baseline; 50-100 days 234 vs. baseline and 100-200 days vs. baseline). While diarrhea prevalence continued to decline over 235 time (11.8% 7-50 days; 10.0% 50-100 days; 6.1% 100-200 days), these differences were not 236 statistically significant after accounting for repeated measures, seasonal and regional differences 237 (7-50 days vs. 50-100 days (adjusted p=0.45); 7-50 days vs. 100-200 days (adjusted p=0.38)). 238 Table 3. Prevalence of self-reported diarrhea by home filter status 239 Aggregation Before filter in home1 % (x/n) After filter in home1 % (x/n) OR (95% CI)2 aOR (95% CI)3 Any member of the household 25.6% (101/394) 9.8% (56/574) 0.24 (0.16,0.37)*** 0.29 (0.16,0.51)*** Adults only 17.3% (68/392) 4.4% (25/564) 0.04 (0.01,0.1)*** 0.14 (0.04,0.44)*** School aged children only 12.2% (46/378) 6.0% (33/546) 0.20 (0.10,0.44)*** 0.49 (0.23,1.05) Children less than 5 11.8% (12/102) 4.3% (6/139) 0.15 (0.15,0.15)*** 0.09 (0.011,0.73)* *p<0.05; **p<0.01, ***p<0.001 240 1 Counts are each measurement of each household 241 2 Odds Ratio accounting for repeated measures 242 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 13 3 Adjusted Odds Ratio accounting for repeated measures and adjusted for Household size, Season (Fall, Winter, 243 Spring), Water Source, School Filter Status and Region 244 245 Two-week diarrhea prevalence by school filter status 246 We also examined whether two-week prevalence of diarrhea was impacted by the installation of 247 water filters in schools, irrespective of water filters in the home. Before or after adjusting for 248 covariates, there was no evidence of statistical interaction between school filter status and home 249 filter status for any household member, adults, school-aged children, or children less than 5 250 (Adjusted p-values of 0.63, 0.70, 0.76 and 0.99, respectively). Furthermore, there was no 251 evidence of a statistically significant decline in two-week diarrhea after installation of school 252 filters, for any age group (Table 4) before or after adjusting for covariates, including home filter 253 status. 254 Table 4. Prevalence of self-reported diarrhea in household by school filter status 255 Aggregation Before filter in school1 % (x/n) After filter in school1 % (x/n) OR (95% CI)2 aOR (95% CI)3 Any member of the household 15.9% (114/717) 17.1% (43/251) 1.01 (0.62,1.65) 1.29 (0.69,2.40) Adults only 9.62% (68/707) 10% (25/249) 0.87 (0.40,1.88) 1.58 (0.34, 7.39) School aged children only 7.99% (55/688) 10.2% (24/236) 1.09 (0.51,2.33) 1.17 (0.55,2.52) Children less than 5 7.18% (13/181) 8.33% (5/60) 0.34 (0.018,6.33) 0.47 (0.09,2.54) *p<0.05, **p<0.01, ***p<0.001 256 1 Counts are each measurement of each household 257 2 Odds Ratio accounting for repeated measures 258 3 Adjusted Odds Ratio accounting for repeated measures and adjusted for Household Filter presence (y/n), 259 Household size, Season (Fall, Winter, Spring), Water Source and Region 260 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 14 Economic and educational consequences of diarrhea 261 Prevalence of diarrhea with economic or educational consequences decreased significantly 262 across all age groups, and for each consequence (i.e., missed work, missed school, 263 hospitalizations) across all age groups (all p<0.05 before adjusting for covariates) after filters 264 were installed in homes (Table 5). Before filters were installed in homes, prevalence of diarrhea 265 with economic or educational consequences was 9.64% among any member of the household, 266 compared to 1.57% after filter installation (p<0.001). For children, missed school due to diarrhea 267 decreased from 4.23% to 0.55% (p<0.001). After adjusting for covariates (Table 5), prevalence 268 of diarrhea with any economic or educational consequence for any member of the household still 269 showed a statistically significant decrease after filter installation (aOR 0.07, p<0.05), while other 270 decreases were no longer statistically significant. There was no evidence of an impact of the 271 school filter or interaction between school and home filters on diarrhea with economic or 272 educational consequences (p>0.05 in all cases; detailed results not shown). 273 274 275 276 277 278 279 280 281 282 283 284 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 15 Table 5. Prevalence of diarrhea with economic or educational consequence by home filter 285 status 286 Aggregation Before filter in home1 After filter in home1 Odds ratio2 Adjusted odds ratio3 Any member of the household with any economic or educational consequence 9.64% (38/394) 1.57% (9/574) 0.01 (0.00,0.05)*** 0.07 (0.0014,0.56)* Adult hospitalized due to diarrhea 3.59% (14/390) 0.888% (5/563) 0.02 (0.00,0.18)*** 0.18 (0.0027,11.6) Adult missed work due to diarrhea 5.38% (21/390) 0.888% (5/563) 0.01 (0.00,0.09)*** 0.02 (0.00,1.57) Child missed school due to diarrhea 4.23% (16/378) 0.549% (3/546) 0.01 (0.00,0.09)*** 0.051 (0.0016,1.59) Child less than 5 hospitalized due to diarrhea 7.84% (8/102) 2.88% (4/139) 0.01 (0.00,0.32)* 0.01 (0.00,12.0) *p<0.05, **p<0.01, ***p<0.001 287 1 Counts are each measurement of each household 288 2 Odds Ratio accounting for repeated measures 289 3 Adjusted odds ratio accounting for repeated measures and adjusted for Household size, Season (Fall, Winter, 290 Spring), Water Source, School Filter Status and Region 291 Quality of drinking water sources 292 In order to assess the quality of drinking water sources, 56 unfiltered water samples were 293 collected across the 16 villages (30 home water sources, 25 school water sources, 1 unknown) 294 throughout the course of the study. Thirty-five samples were collected at the time of filter 295 installation and twenty-one after filter installation. Characteristics of each water sample and 296 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 16 outcomes for detection of dissolved heavy metals, particulates and bacteria are provided in 297 Additional File 1. 298 We analyzed bacterial communities of drinking water sources through 16S rRNA amplicon 299 sequencing. Up to 5 gallons (~19 L) of water from each drinking source was filtered through 300 two, 0.1 micron hollow fiber membrane filters linked in a tandem pair. The first filter in the pair 301 (designated “Filter A”) filtered the source water, capturing cells and particulate matter; the 302 second filter in the pair (designated “Filter B”) captured cells and particulate matter (if any) of 303 the filtrate of Filter A. This design allowed the analysis of bacterial communities of both the 304 source water and the filtered water. Two types of controls were also analyzed. First, on each day 305 of processing, backflush controls included 1) water backflushed through an unused filter, 2) a 306 full volume of water used for backflushing all filters was pelleted, and 3) a small volume of 307 water used for backflushing all filters was added to DNA extraction tubes. Second, each 308 sequencing library plate included a positive control of a mock community of 8 bacteria 309 (ZymoBIOMICS Microbial Community Standard) and a negative control consisting of ultrapure 310 water. Evaluation of the pairwise distances between bacterial communities of each filter in the 311 study (Fig 6A) showed that the communities derived from unfiltered source water (Filter A) 312 formed a group that is largely non-overlapping with the communities derived from filtered 313 source water (Filter B). The difference between the two groups was statistically significant 314 (permutational multivariate analysis of variance [PERMANOVA] [28] ; df = 1, F = 9.3361, R2 = 315 0.04755, p = 0.001), however, within group dispersion may also contribute to between group 316 significance (betadisper, F = 4.1054, p = 0.045). Biological replicates of each water source 317 grouped tightly in most cases for the unfiltered water (Filter A), whereas the replicates for 318 filtered water (Filter B) from the same kit did not cluster as tightly (Fig. 6B). Replicate Filter B 319 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 17 samples are generally co-localized with Filter B samples from other kits and with backflush 320 negative controls (Fig. 6B). Consistent with the patterns in Fig. 6B, Filter A replicates from the 321 same kit were less distant from each other than Filter A communities from other kits, explaining 322 a large fraction of the observed variance (PERMANOVA; Filter A: df = 54, F = 7.3144, R2 = 323 0.82808, p = 0.001; betadisper, F = 2.8424, p = 0.002). Similar patterns were observed for Filter 324 B communities (Filter B: df = 32, F = 1.9172, R2 = 0.76354, p = 0.001; betadisper, F = 27.287, p 325 = 0.005). There were distinct bacterial communities associated with samples from different types 326 of water sources. For example, each of the kits in Fig. 6C represents a different water source type 327 and is representative of the positions in the ordination of other kits of the same water source type 328 sampled from different geographic locations. Each of the water type bacterial communities are 329 distinct from the other water type bacterial communities in pairwise comparisons of all samples, 330 with all but one comparison having significant differences both between (PERMANOVA) and 331 within (betadisper) group distances (Table 5). 332 333 334 335 336 337 338 339 340 341 342 343 344 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 18 Table 5 Statistical Tests of Bacterial Communities by Water Source Type 345 Comparison Groups Statistical Test Degrees of Freedom F Statistic R2 p value City Tap Water / City Water Filled Underground Cistern PERMANOVA1 1 4.184 0.07317 0.001 betadisper2 1 19.474 - 0.001 City Tap Water / Purchased Container PERMANOVA 1 4.714 0.05699 0.001 betadisper 1 2.632 - 0.113 City Tap Water / Pumped Well PERMANOVA 1 5.603 0.08673 0.001 betadisper 1 7.365 - 0.010 City Water Filled Underground Cistern / Purchased Container PERMANOVA 1 4.335 0.10004 0.001 betadisper 1 48.137 - 0.001 City Water Filled Underground Cistern / Pumped Well PERMANOVA 1 2.750 0.12089 0.001 betadisper 1 6.022 - 0.026 Purchased Container/ Pumped Well PERMANOVA 1 5.421 0.10752 0.001 betadisper 1 29.216 - 0.001 1 PERMANOVA (adonis) generates a pseudo F statistic that describes the among group differences relative to the 346 within group distances 347 2 betadisper generates an F statistic that describes the difference in within group dispersion of samples relative to the 348 centroid of the group 349 We used the list of bacterial taxa identified in each water source to determine the presence or 350 absence of genera containing known waterborne pathogens as defined by the WHO [21]. A 351 genus was designated as present if it was found in A filter replicates for a kit but not found in B 352 filters or process controls. At least one genus from the list of 18 waterborne bacterial pathogens 353 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 19 was detected in 50 of 51 water sources passing quality control metrics. The single water source 354 without indication of potential pathogens was from a city tap water sample (Kit 233). 355 Additionally, we tested for the presence/absence of heavy metals in water sources. We detected 356 at least one of eight dissolved heavy metals in 38 out of 56 water samples (67.9%). Table 6 357 shows detected bacterial pathogens and dissolved heavy metals stratified by whether the sample 358 was obtained at the time of filter installation or during a follow-up visit. No significant 359 differences were observed in the prevalence of detected bacteria or heavy metals in the source 360 water between installation and follow-up across all measured bacteria and heavy metals 361 measured. Similarly, no seasonal or linear changes in prevalence were observed over time. 362 Furthermore, mean particulate levels were not significantly different at the time of filter 363 installation (0.246 vs. 0.166 mg/L; p=0.57), and also did not differ significantly over time during 364 the study (linear trend p=0.36; seasonal effect p=0.35). Detected particulates were found to 365 contain Ba, Ce, Cl, Cr, Fe, P, Mn, S, Ti, Zr. See Additional File 1 for details about all water 366 quality results. 367 368 369 370 371 372 373 374 375 376 377 378 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 20 Table 6. Summary of test results of source water (unfiltered) 379 Type of water sample testing Bacterial Genus /Heavy Metal Percent of samples present at filter install Percent of samples present at follow-up p-value comparing install to follow-up2 P- value for linear trend3 P-value for seasonal effect4 Bacteria detected Acinetobacter 53.8% (14/26) 35.3% (6/17) 0.25 0.39 0.56 Aeromonas 16.1% (5/31) 20% (4/20) 0.72 0.51 0.64 Burkholderia 6.45% (2/31) 5% (1/20) 0.83 0.31 0.54 Campylobacter 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Enterobacter 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Escherichia/ Shigella 45.5% (10/22) 50% (6/12) 0.80 0.90 0.90 Francisella 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Helicobacter 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Klebsiella 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Legionella 58.1% (18/31) 70% (14/20) 0.18 0.06 0.41 Leptospira 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Mycobacterium 30% (9/30) 57.9% (11/19) 0.00 0.19 0.08 Pseudomonas 85.2% (23/27) 94.4% (17/18) 0.35 0.62 0.40 Salmonella 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Staphylococcus 12.9% (4/31) 15.8% (3/19) 0.78 0.79 0.87 Tsukamurella 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Vibrio 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Yersinia 3.23% (1/31) 0% (0/20) 0.98 0.69 0.47 Dissolved heavy metal ions detected Arsenic 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Barium 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Cadmium 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Chromium 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Copper1 64.7% (22/34) 68.2% (15/22) 0.79 0.38 0.34 Lead1 8.82% (3/34) 4.55% (1/22) 0.55 0.38 0.63 Nickel 0% (0/34) 0% (0/22) 1.00 1.00 1.00 Selenium1 5.88% (2/34) 0% (0/22) 0.96 0.16 0.21 1 Detected levels were below WHO action guidelines in all samples 380 2 From Fisher’s exact test comparing baseline to follow-up prevalence 381 3 From test of linear trend over course of study 382 4 From test of seasonal differences over course of study 383 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 21

384 The purpose of this study was to evaluate the impact of the introduction of point-of-use filters 385 into different community settings on the health of school aged children and their families. 386 Drinking water sources were also evaluated for bacterial, particulate and dissolved heavy metal 387 content to determine the appropriateness of the point-of-use filter technology for long term use 388 by residents. Two-week self-reported diarrhea prevalence in all age groups drops substantially 389 after introduction of filters in homes, with unadjusted declines of between approximately 50 and 390 75%. Similarly, large reductions in economic and educational consequences of diarrhea were 391 also observed after introduction of filters in homes across all age groups. No corresponding 392 change in prevalence was noted after introduction of filters into schools, nor was any moderating 393 effect found for the school filter on the effectiveness of the home filter. Results from water 394 sample testing confirm that bacteria were prevalent in water samples, including genera that are 395 known to contain pathogens associated with waterborne disease. Limited amounts of dissolved 396 metals and metal-containing particulates were observed in the drinking water sources. 397 While previous research [13–15, 17] yielded mixed results from the Sawyer® PointONE™ filter 398 system in laboratory testing and field environments, our randomized, controlled field trial 399 provides additional evidence of their effectiveness in reducing waterborne disease out to at least 400 200 days post-installation in real world settings. The hollow fiber membrane technology is 401 designed to remove bacteria and particulates, but not dissolved ions. Evaluation of unfiltered 402 drinking water sources available to families confirmed the presence of highly diverse bacterial 403 communities that varied with source type. Only a single water source tested in this study was 404 free of bacterial genera that are known to contain waterborne pathogens, indicating that 405 hollowfiber membrane filtration should be an effective method for improving drinking water 406 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 22 quality and health outcomes in these areas. None of the drinking water sources contained 407 dissolved heavy metals above WHO action standards, suggesting that point-of-use filters were an 408 appropriate solution for the villages and schools targeted by this intervention. Furthermore, 409 prevalence of reported diarrhea dropped substantially, though was not eliminated completely, 410 after filter installation in line with expectations that the filter eliminates bacterial and parasitic 411 causes of diarrhea from water while not impacting other diarrheal causes (e.g., viral). However, 412 despite these positives, further field work is needed, especially evaluation of filter effectiveness 413 and utilization over longer time periods. 414 Interestingly, little impact of school filter installation on diarrhea prevalence was observed. 415 There are likely numerous causes for this finding. First, and importantly, only school-aged 416 children in the family receive water at the school and, thus, any potential direct impact of school 417 filters would only be observed on school-aged children. However, in our data, this direct impact 418 was not observed. Secondly, follow-up conversations with school and non-profit workers 419 (personal communication, C. Nelson), suggest that, in part due to a relatively short school day (4 420 hours), children typically bring water from home or buy water (not necessarily clean and safe) 421 from vendors near the school instead of getting water from school. Third, only seven of the 422 sixteen schools in the study were willing to accept a filter for their school, with nine schools 423 having determined that their water was of sufficient quality that no filter was needed, despite no 424 a priori water testing. 425 An important feature of our study evaluating the potential effectiveness of both school and home 426 filters was a sequential study design implementing the method of minimization to balance 427 covariates. As has been noted by others [18] this method is rarely implemented in practice but 428 can be highly effective. In our case, the method of minimization maximized statistical power by 429 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 23 limiting the impact of potential confounding variables over time and ensuring a powerful study 430 design. However, even here, there are additional variables that could have minimized the impact 431 of covariates even further (e.g., proportion of households using unfiltered water in each 432 community; proportion lacking sanitation services per region, etc.); these remain potentially 433 confounding variables in our analysis. This, combined with the use of a self-report diarrhea 434 measurement, represent two areas for future work and limitations of this study. 435 Another limitation of our study worth noting are the widely varying response rates by village. 436 While the impact on our findings is minimal, due in part to the study design, additional statistical 437 power and reduced impacts of potential confounding would be realized through consistently 438 higher response rates across all villages. Some variation and lower response rates are expected 439 when working in remote villages in countries and areas with limited infrastructure, however, 440 having additional community buy-in and support (see previous paragraphs) may serve not only 441 to improve actual intervention efficacy but also to improve statistical aspects of study design 442 leading to more quantifiable and robust findings. 443 This study also combined public health surveys with evaluation of both bacterial and chemical 444 components of drinking water quality using the same hollowfiber membrane filtration 445 technology as was used for the intervention in homes and schools. The tandem filtration design 446 of test kits provided a direct demonstration of removal of potentially harmful bacteria and 447 particulate matter that could be associated with toxic metals. Dissolved heavy metals were 448 assessed from the filtered water sources to determine if potential toxins might persist in the 449 drinking water source. Efforts to evaluate drinking water sources should take into account the 450 biological and chemical components of the water to ensure that proper intervention technologies 451 are used. 452 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 24 It is important to note that this study produced profiles of the bacterial communities in each 453 water sample through sequencing of 16S rRNA amplicons. The use of this sequencing-based 454 evaluation of the presence/absence and relative abundance of bacterial genera is not traditionally 455 applied in the evaluation of drinking water quality, although this is beginning to change [29–31]. 456 More routinely, culture-based methods are used to assess viable loads of fecal indicator bacteria 457 as a proxy for the presence of pathogens, or qPCR-based molecular methods are used to detect 458 the presence of specific pathogens. When interpreting results from the sequencing approach 459 employed in this study, it is important to recognize that the method does not determine whether 460 bacteria detected in the sample are viable or not (this is also true of qPCR approaches). The 461

also does not identify specific pathogens (this is also true of routine culture-based 462 approaches). However, it 1) allows for a single assay that characterizes the entire bacterial 463 community, including all known potential pathogenic groups of bacteria, and 2) it provides an 464 opportunity to study bacterial communities across different drinking water sources from around 465 the world as these approaches become routine. Future studies should include the use of both 466 culture-based and sequence-based approaches to better characterize the relationship between the 467 two data types. 468

469 Our controlled study provides compelling evidence of the efficacy of the PointONE™ filter in 470 homes in a field setting, providing substantial reductions in bacterially-caused diarrhea for the 471 entire length of our study (200 days post filter installation). Water quality sampling provided 472 strong complementary evidence to support self-reported diarrhea information and supported the 473 choice of this filtration technology that removes only microbial and particulate contaminants. 474 While we observed no significant impact of school filters, in other contexts where water is being 475 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 25 consumed more regularly by school children, additional impact may be observed. Further work 476 is needed to evaluate the post-200 day utilization of filters and long-term filter efficacy in a field 477 setting. 478 Abbreviations 479 WHO = World Health Organization 480 EPA = Environmental Protection Agency 481 Additional Files 482 Additional File 1 – AdditionalFile1.xlsx, Spreadsheet (.xlsx), List of water quality kits reported 483 in this study, associated metadata for each kit, and water quality results. 484 Additional File 2 – AdditionalFile2.pdf, Document (.pdf), Supplemental Methods for water 485 quality testing. 486 Additional File 3 – AdditionalFile3.pdf, Table (.pdf), Prevalence of self-reported diarrhea by 487 home filter status: Intent to Treat Analysis. 488 Additional File 4 – AdditionalFile4.xlsx, Spreadsheet (.xlsx), Metadata file for 16S rRNA 489 amplicon data analysis. The tab “DR_MiMarks-Metadata” contains all metadata associated with 490 amplicon sequence data deposited in the SRA. The tab “Key” contains an explanation of all 491 column headers. 492 Declarations 493 Ethics approval and consent to participate 494 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 26 Consent for participation in research was obtained from all participants. This study was 495 conducted under the approval of the Dordt University Institutional Review Board. 496 Consent for publication 497 Not applicable. 498 Availability of data and materials 499 The survey datasets used and/or analyzed during the current study are available from the 500 corresponding author on reasonable request. 501 The 16S amplicon sequencing datasets generated and/or analyzed during the current study are 502 available in the Short Read Archive (SRA), 503 https://www.ncbi.nlm.nih.gov/bioproject/PRJNA670359. 504 Competing interests 505 Portions of the authors’ time (NT, KDVG, RDW, SAB, FSM, JWP, MJP and AAB) were 506 supported by a grant from Sawyer Products, Inc. 507 Funding 508 Portions of the authors’ time and materials were supported by a grant from Sawyer Products, Inc. 509 Representatives of the funding agency were consulted during study design and aided in training 510 for use of point-of-use filters in the field. Neither the funding agency nor its representatives 511 contributed to data collection, data analysis or interpretation of data. 512 Authors’ contributions 513 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 27 NT, KVDG, RU, RDW, SAB, AH and AAB co-led the development of the research questions 514 and study design. DM, EB, OC, CGZ and NT led data monitoring, data cleaning and statistical 515 analyses. AAB, SAB, MJP, JWP, BPK, ADS, RDW, FSM, TMB, CEAC, LE, BF, GKG, MAH, 516 JAL, SML, TRL, JMP, ES, DJS, JES, MJS, MS, DRW designed laboratory protocols and water 517 sampling kits, participated in RNA extraction, metals and particulate testing, final dataset 518 creation and reporting. NT, KVDG, RU, RDW, SAB, BPK, and AAB wrote and revised the 519 manuscript. 520 Acknowledgments 521 The authors thank Darrel Larson, Chad Nelson, Gary Higgins and numerous others in the Child 522 Hope Network for their efforts in administering data collection and their vision for the project. 523

524 1. Troeger C, Blacker BF, Khalil IA, Rao PC, Cao S, Zimsen SR, et al. Estimates of the global, 525 regional, and national morbidity, mortality, and aetiologies of diarrhoea in 195 countries: a 526 systematic analysis for the Global Burden of Disease Study 2016. Lancet Infect Dis. 527 2018;18:1211–28. 528 2. Pruss-Ustun A, Bartram J, Clasen T, Colford JMJ, Cumming O, Curtis V, et al. Burden of 529 disease from inadequate water, sanitation and hygiene in low- and middle-income settings: a 530 retrospective analysis of data from 145 countries. Trop Med Int Health TM IH. 2014;19:894–531 905. 532 3. World Health Organization. Diarrhoeal disease. 2017. https://www.who.int/news-room/fact-533 sheets/detail/diarrhoeal-disease. Accessed 16 Jul 2020. 534 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 28 4. Prüss Annette, Kay David, Fewtrell Lorna, Bartram Jamie. Estimating the burden of disease 535 from water, sanitation, and hygiene at a global level. Environ Health Perspect. 2002;110:537–42. 536 5. Thomas JC, Weber DJ. Epidemiologic Methods for the Study of Infectious Disease. 1st 537 edition. Oxford University Press; 2001. 538 6. Sack RB. Bacterial and Parasitic Agents of Acute Diarrhea. In: Bellanti JA, editor. Acute 539 Diarrhea: Its Nutritional Consequences in Children. New York, NY: Vevey/Raven Press; 1983. 540 p. 53–65. 541 7. The World Bank. World Bank Country and Lending Groups. 542 https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-543 lending-groups. Accessed 16 Jul 2020. 544 8. UNICEF. UNICEF Data. UNICEF DATA - Child Statistics. 2020. https://data.unicef.org/. 545 Accessed 16 Jul 2020. 546 9. Ministerio de Salud Pública. Encuesta Demográfica y de Salud República Dominicana 2013. 547 Santo Domingo, República Dominicana: Centro de Estudios Sociales y Demográficos 548 (CESDEM) y ICF International; 2014. https://dhsprogram.com/pubs/pdf/FR292/FR292.pdf. 549 10. McLennan JD. To boil or not: drinking water for children in a periurban barrio. Soc Sci Med. 550 2000;51:1211–20. 551 11. Baum R, Kayser G, Stauber C, Sobsey M. Assessing the microbial quality of improved 552 drinking water sources: Results from the Dominican Republic. Am J Trop Med Hyg. 553 2014;90:121–3. 554 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 29 12. Hydreion LLC. Microbiological Testing of the Sawyer 7/6B Filter. Report No. S05-03. 2005. 555 13. Tintle N, Heynen A, Van De Griend K, Ulrich R, Ojo M, Boven E, et al. Evaluating the 556 efficacy of point-of-use water filtration units in Fiji. Trop Med Health. 2019;47:48. 557 14. Kohlitz J, Hasan T, Khatri K, Sokota A, Iddings S, Bera U, et al. Assessing reported use and 558 microbiological performance of a point-of-use household water filter in rural Fiji. J Water Sanit 559 Hyg Dev. 2013;3:207–15. 560 15. Murray A, Goeb M, Stewart B, Hopper C, Peck J, Meub C, et al. Fouling in hollow fiber 561 membrane microfilters used for household water treatment. J Water Sanit Hyg Dev. 2015;5:220–562 8. 563 16. Lindquist ED, Norman WR, Soerens T. A review of: Fouling in hollow fiber membrane 564 microfilters used for household water treatment (2015) Murray, A., Goeb, M., Stewart, B., 565 Hopper, C., Peck, J., Meub, C., Asatekin, A. & Lantagne, D. J. WASHDEV 5 (2), 220–228 566 doi:10.2166/washdev.2015.206. J Water Sanit Hyg Dev. 2015;5:229–31. 567 17. Murray AL, Stewart B, Hopper C, Tobin E, Rivera J, Mut-Tracy H, et al. Laboratory efficacy 568 and field effectiveness of hollow fiber membrane microfilters used for household water treatment 569 in Honduras. J Water Sanit Hyg Dev. 2017;7:74–84. 570 18. Scott NW, McPherson GC, Ramsay CR, Campbell MK. The method of minimization for 571 allocation to clinical trials: a review. Control Clin Trials. 2002;23:662–74. 572 19. USGS. USGS EROS Archive - Digital Elevation - Shuttle Radar Topography Mission 573 (SRTM) 1 Arc-Second Global. doi:10.5066/F7PR7TFT. 574 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 30 20. Chapter 12: Chemical fact sheets. In: Guidelines for drinking-water quality. Fourth edition 575 incorporating the first addendum. Geneva: World Health Organization; 2017. p. 307–442. 576 https://www.who.int/water_sanitation_health/publications/gdwq4-with-add1-chap11.pdf?ua=1. 577 Accessed 1 Oct 2020. 578 21. Chapter 11: Microbial fact sheets. In: Guidelines for drinking-water quality. Fourth edition 579 incorporating the first addendum. Geneva: World Health Organization; 2017. p. 231–306. 580 https://www.who.int/water_sanitation_health/publications/gdwq4-with-add1-chap11.pdf?ua=1. 581 Accessed 1 Oct 2020. 582 22. Kozich JJ, Westcott SL, Baxter NT, Highlander SK, Schloss PD. Development of a Dual-583 Index Sequencing Strategy and Curation Pipeline for Analyzing Amplicon Sequence Data on the 584 MiSeq Illumina Sequencing Platform. Appl Environ Microbiol. 2013;79:5112–5120. 585 23. Patrick D. Schloss. MiSeq SOP. MiSeq SOP. https://mothur.org/wiki/miseq_sop/. Accessed 586 13 Oct 2020. 587 24. Quast C, Pruesse E, Yilmaz P, Gerken J, Schweer T, Yarza P, et al. The SILVA ribosomal 588 RNA gene database project: improved data processing and web-based tools. Nucleic Acids Res. 589 2013;41:D590–6. 590 25. McMurdie PJ, Holmes S. Phyloseq: An R Package for Reproducible Interactive Analysis and 591 Graphics of Microbiome Census Data. PLoS ONE. 2013;8. 592 26. Jari Oksanen FGB, Friendly M, Kindt R, Legendre P, McGlinn D, Minchin PR, et al. vegan: 593 Community Ecology Package version 2.5-4. 2019. https://CRAN.R-project.org/package=vegan. 594 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 31 27. R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: 595 R Foundation for Statistical Computing; 2020. https://www.R-project.org. 596 28. Anderson MJ. Permutational Multivariate Analysis of Variance (PERMANOVA). In: Wiley 597 StatsRef: Statistics Reference Online. American Cancer Society; 2017. p. 1–15. 598 doi:10.1002/9781118445112.stat07841. 599 29. Vierheilig J, Savio D, Ley RE, Mach RL, Farnleitner AH, Reischer GH. Potential 600 applications of next generation DNA sequencing of 16S rRNA gene amplicons in microbial 601 water quality monitoring. Water Sci Technol. 2015;72:1962–72. 602 30. Santos QMB los, Schroeder JL, Sevillano-Rivera MC, Sungthong R, Ijaz UZ, Sloan WT, et 603 al. Emerging investigators series: microbial communities in full-scale drinking water distribution 604 systems – a meta-analysis. Environ Sci Water Res Technol. 2016;2:631–44. 605 31. Roguet A, Esen ÖC, Eren AM, Newton RJ, McLellan SL. FORENSIC: an Online Platform 606 for Fecal Source Identification. mSystems. 2020;5. doi:10.1128/mSystems.00869-19. 607 608 Figure Legends 609 Figure 1. Location of 16 selected villages in the Dominican Republic. Stars denote the 610 approximate location of villages included in this study. Villages that were close in proximity 611 were offset on the map to provide distinct representations for each village. 612 613 Figure 2. Depiction of the assignment of schools and households to treatment groups. 614 Depending upon whether the school in the village accepted or declined a filter, villages were 615 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 32 initially assigned to either four or two treatment groups, respectively. In order to ensure that all 616 households and schools that accepted a filter ultimately received one, secondary randomization 617 of control group households was performed as depicted in the figure. 618 619 Figure 3: Prevalence of two-week diarrhea before and after filter installation by age. 620 Unadjusted SE bars are shown. Statistical significance between groups (***p<0.001, **p<0.01, 621 *p0.05) represents statistical significance from repeated measures models adjusted 622 for Household size, Season (Fall, Winter, Spring), Water Source, School Filter Status and Region 623 624 Figure 4. Diarrhea prevalence by days since installation. Unadjusted SE bars are shown. 625 Statistical significance between groups (***p<0.001, **p<0.01, *p0.05) represents 626 statistical significance from repeated measures models adjusted for Household size, Season (Fall, 627 Winter, Spring), Water Source, School Filter Status and Region 628 629 Figure 5. NMDS ordination (k=3, stress = 0.141) of Bray-Curtis pairwise distances of 630 bacterial communities from drinking water sources collected in the study (either home or 631 school locations). Each water source was sampled with a kit consisting of three tandem filters 632 (Filter A and Filter B), yielding three biological replicates of a water source per kit. Each point 633 represents backflushed contents from Filter A, Filter B, laboratory backflush controls, mock 634 community controls, or negative controls. Each kit is represented by up to 3 Filter A points and 3 635 Filter B points. Samples with fewer than 5000 sequencing reads were not included in the 636 analysis. The same ordination is used for all three panels. Panel A: All backflush samples 637 included in the study and associated controls. Shape and color delineate whether a sample is 638 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint 33 from Filter A, Filter B or a control. Panel B: A subset of four kits selected to show the 639 relationship between biological replicates of Filter A and Filter B within and between kits. Shape 640 delineates Filter A, Filter B and control samples. Color delineates which kit (i.e. water source) 641 each filter sample represents along with the type of control. Panel C: The same kits as shown in 642 Panel B with identical coloring showing kit membership. Shape delineates the type of water 643 source. 644 . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint −1.0 −0.5 0.0 0.5 −1 0 1 NMDS1 NMDS2 A Panel A: Sample Type A Filter B Filter Control Pellet Blank Pellet MQ Control Mock Community Panel B and C: Kit ID 224 231 244 248 Blank Pellet Control Pellet Mock Community NA Panel B: Sample Type A Filter B Filter Control Pellet Blank Pellet Mock Community NA Panel C: Water Source City Tap Water Pumped Well City Water Filled Underground Cistern Purchased Container NA −1.0 −0.5 0.0 0.5 −1.0 −0.5 0.0 0.5 1.0 NMDS1 NMDS2 B −1.0 −0.5 0.0 0.5 −1.0 −0.5 0.0 0.5 1.0 NMDS1 NMDS2 C . CC-BY-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted October 23, 2020. ; https://doi.org/10.1101/2020.10.21.20217299doi: medRxiv preprint