Effects of adding household water filters to Rwanda’s Community-Based Environmental Health Promotion Programme: a cluster-randomized controlled trial in

We conducted a cluster randomized controlled trial in Rwamagana district to determine whether adding a household-based water filter with safe storage to the CBEHPP could be effective in improving drinking water quality.


The intervention under evaluation is the delivery and promotion of the LifeStraw Family 2.0 filters in the CBEHPP program. The filter is a tabletop point-of-use water treatment system that includes an 80 μm pre-filter to remove coarse material, 20 nm hollow-fiber ultrafiltration membrane, backwash lever, and covered storage container with 5.5 L capacity. The system meets the WHO’s “comprehensive protection” guideline for household water treatment technologies36; it can filter up to 18,000 liters of water, which should be able to supply a family of five with clean drinking water for three to five years, without any replacement of parts37.

Delivery and promotion of the filter is through the CBEHPP, which organizes village-level CHCs with a maximum membership of one hundred households. Clubs aim to meet weekly and are led by volunteer CHC facilitators that are trained to deliver a 20-module curriculum designed by the Ministry of Health. The filter-integrated intervention tasks CHC facilitators to additionally serve as the primary service providers of the filter. The CBEHPP filter integration was intended to have “lighter touch” engagement compared to the delivery of filters in the Tubeho Neza campaign. Major differences between the approaches include Tubeho Neza’s additional delivery of improved cookstoves, exclusive targeting of households belonging to the lowest economic quartile, mass media campaigns, and supplementary promotional activities such as regular CHW cooperative and community meetings and frequent household visits (Barstow et al. 2016).

Bradshaw et al.38 publish further details on the intervention and delivery in their process evaluation. CHC facilitators were trained to promote the filter and to repair or replace nonfunctional units. Eligible households were invited to receive the filter at a mass-distribution event held at the main health center serving the geographical sector. Following the distribution, CHC facilitators conducted individual household visits to teach households how to use the filter and provide a promotional poster. Households were instructed not to use the filter until the initial visit was completed. A second promotional household visit by CHC facilitators was completed ~6 months later to monitor upkeep/functionality, use, and satisfaction with the filter. CHC facilitators additionally reinforced messaging in CHC meetings. Households that were eligible to receive the filter included CHC members and had at least one child under the age of 5 or had at least one pregnant woman living in the household. All eligible households were able to receive the filter regardless of being selected to participate in the study.

Catholic Relief Services (CRS) and SNV, two of the government’s primary implementing NGO partners of CBEHPP, delivered the intervention with their local partner African Evangelistic Enterprise (AEE). The NGOs implement CBEHPP and its CHC model through Gikuriro, a USAID WASH and nutrition program. SNV, CRS, and AEE were supported in this initial distribution and promotion by Amazi Yego, the social enterprise that collaborated in the Tubeho Neza filter promotion in Western Province17. Amazi Yego trained CRS, SNV, and AEE and shared experiences in filter delivery. Amazi Yego was also significantly involved in designing the implementation protocol, providing promotional material to be provided to householders and implementing the intervention alongside CRS/SNV/AEE.

Study design

We employed a cluster-randomized controlled trial design to assess the effects of the intervention on point-of-use (POU) drinking water quality as the primary outcome; we also assess intervention coverage and use and effects on reported diarrhea as secondary outcomes. The trial was conducted over 13–16 months in two follow-up visits. Rwamagana is a primarily rural district in Rwanda’s Eastern Province and has a population of 313,461 people39. Rwamagana was selected because it is located in Eastern Province, which has one of the highest rates of fecal contamination of drinking water in the country4 and because it was one of the districts the implementers worked in. SNV and CRS are active in all 474 villages across Rwamagana.

Sixty villages (clusters) were randomly selected, with 30 receiving the intervention (CBEHPP + filter) and 30 serving as controls (CBEHPP alone). Villages were randomly selected from a list of the 474 eligible villages using probability proportional-to-size sampling (PPS) without replacement using samplepps in Stata 16 software40. PPS was done based on the implementer’s reported size of the CHC in each village.

Households in selected villages were eligible to participate in the study if they were verified eligible to receive the intervention (CHC member households who had at least one child under 5 or pregnant person living in the household at time of baseline) and had a household member that was over 18 years of age available to consent to enrollment. A list of eligible households was made for each of the 60 villages by consulting the district registers, CHC registers, and with the CHC facilitators. Eligible households per village ranged from 10–72 households. Twenty-five households were randomly selected to be enrolled in the study from each village list using simple random sampling using the sample function to randomly order households in R statistical software41. Other eligible households were deemed as replacement study households. Enumerators were instructed to attempt each of the randomly selected 25 households twice at least 2 h apart during the day. If households could not be reached or were otherwise found to be ineligible, enumerators enrolled one of the replacement households based on a random order. To complete a village, at least half of the eligible households in the village needed to have been enrolled, but a cap of 25 households per village was enforced due to logistical constraints.

Randomization and blinding

Random allocation of the intervention and control groups was done at the village level. To help ensure geographical balance between arms, random allocation of the intervention was stratified by the 13 sectors within the district. An individual unaffiliated with the project conducted the allocation. The data collection team, village-level implementers/leaders (e.g., CHC facilitators, village leaders, CHWs, AEE staff) and participating households were blinded to the allocation during baseline data collection. Enumerators and households could not be blinded after implementation due to the nature of the intervention. The primary data analyst additionally oversaw and managed the data collection, and therefore, could not be blinded. The principal investigator remained blinded throughout the study duration.

Baseline and follow-ups

A baseline survey was conducted from December 2018 to March 2019 prior to intervention delivery. The intervention was delivered from March to June 2019. A midline survey was conducted 5–7 months (median 6 months) following intervention delivery from October to December 2019. The endline survey was originally planned to be conducted 6 months later. However, due to government lockdowns and restrictions from COVID-19, the endline survey was delayed by approximately 2 months and was completed 13–16 months (median 14 months) after intervention delivery from July-September 2020. We aimed to have equal number of intervention and control villages visited in a day. We collected drinking water samples and information on household and demographic characteristics, reported and observed WASH access based on the WHO/UNICEF Joint Monitoring Programme (JMP) core household survey questions42, reported and observed water treatment and handling practices, and caretaker-reported health of children under 5. Questions were directed to the primary cooks aged 18 and over. If the primary cook was unavailable or under 18, questions were directed to another household member aged 18 and over. Respondents were asked to confirm questions on individual children with their respective primary caregivers if they were available. Survey data were collected and managed using REDCap electronic data capture tools hosted at Emory University43.

Primary and secondary outcomes

The primary outcome is detectable E. coli contamination of drinking water. Following the WHO/UNICEF JMP core household survey questions, each respondent was asked to serve drinking water. A 100 mL sample was collected at each follow-up visit in a sterile Whirl-Pak® bag containing sodium thiosulfate (Nasco, Madison, WI, USA) and kept on ice until tested within 8 h with CompactDry™ (Nissui Pharmaceutical, Tokyo, Japan) media plates using membrane filtration procedures prescribed by UNICEF44. Samples were initially diluted to 50 mL in order to reduce the likelihood of plates that were too numerous to count (TNTC). If drinking water samples were visibly turbid, then they were subsequently diluted to 20 mL, 10 mL, and 5 mL based on the severity of turbidity. Plates were incubated at 30 degrees Celsius for 24 hours using an IncuBox Thermocult (Boehringer, Mannheim, Germany). One technician then counted and recorded individual E. coli CFU on each plate. Random spot checks were performed by managers to validate counts. Water quality results were double entered by two different staff. Plates that were TNTC were assigned a level of 300 CFUs. At least one duplicate and blank of distilled water were tested with samples daily. For duplicate samples, the results of both counts were summed and divided by the total volume processed. In order to obtain standardized totals per 100 mL, we normalized the CFU count by the total volume processed and multiplied the result by 100.

Secondary health outcomes include caregiver-reported diarrhea and healthcare visits for diarrhea within the previous 7-days in children under 5 years of age and under 2 years of age at follow-up visits. For reporting diarrhea in the previous 7-days, we followed the World Health Organization (WHO) standard definition, which defines diarrhea as three or more loose stools in a 24-hour period that can take the shape of a container45. For reporting healthcare visits, we asked caregivers if they sought medical care from a health clinic or CHW for any reported diarrhea cases within the previous 7-days following the WHO definition of diarrhea or caregiver’s interpretation of diarrhea. We also collected data on whether children had a toothache in the previous 7-days to serve as negative control to account for courtesy bias46.

We collected data on filter coverage, use, and acceptability at midline and endline visits. To measure filter coverage, we observed whether the household had the filter and if the filter was in good condition at the time of visit (e.g., assembled properly, working tap, no leaking, undamaged container, adequate flowrate, and ability to backwash). To measure filter use, we collected data on whether the filter was observed to have water in it at the time of visit and whether the household reported using the filter, filling the filter in the previous 7-days, treating drinking water, and if a child under 5 drank filtered water the previous day. To measure filter acceptability, we asked households to rate their acceptability of the appearance of filtered water, smell, taste of filtered water, and time to filter water on a scale from 1 to 4, with 3 and 4 being acceptable and very acceptable, respectively.

Statistical approach

The study was powered to detect a 25% reduction in prevalence of detectable E. coli bacteria in point-of-use water samples, measured at each household visit. The number of households required in each group was derived by first using Diggle, Heagerty, Liang, and Zeger’s47 formula for estimating sample-size requirements for differences in proportions across multiple time points. The result of this equation was then adjusted to account for both village-level clustering and the assumed 15% rate of attrition. We assumed 50% prevalence of E. coli presence in drinking water samples in the control group based on national water quality surveys. We further assumed an intra-village correlation of 0.14 and intra-household ICC of 0.21 based on previous studies18, 2 visits postbaseline, and 25 households per village would meet eligibility requirements. This gave us a sample size requirement of 51 villages to have 80% power for a 25% reduction. To further accommodate the uncertainties of CHC enrollment rates and village size, we aimed to enroll up to 1300 households across 60 villages.

We defined the primary outcome as the presence of E. coli bacteria in 100 mL samples of drinking water. As the samples were diluted for purposes of this analysis, the presence of E. coli CFU follows the limit of detection (LOD) according to the volume processed. The laboratory results showed that the total volume of water processed for household samples that did not display any CFUs (e.g., non-detect plates) ranged from 50 mL to 100 mL. Therefore, results were categorized into a binary variable, where non-detectable E. coli contamination is overall reported as <2 CFU/100 mL water (e.g., LOD for a 50 mL sample). We additionally categorized E. coli presence into two other binary outcomes according to WHO risk category cutoffs for moderate-to-high (≥10 CFU/100 mL) and very high (≥100 CFU/100 mL) contamination20. We examined the latter outcomes based on findings from meta-analysis on water quality and diarrhea, which found a marked increase in disease risk for households when fecal contamination exceeded 10 TTC/100 mL33. We calculated arithmetic and Williams means of CFU counts to account for the skewed distribution. The Williams mean is calculated by adding 1 to all values, taking the geometric mean, and then subtracting the mean by 148. Williams mean were used to account for values less than 1. Non-detect plates were included in the mean calculation as half of their specific LOD.

The effect of the intervention was assessed based on group assignment, regardless of uptake of the intervention (i.e. intention-to-treat). For the household-level primary outcomes on E. coli presence in drinking water and the individual-level secondary outcomes on child health, we used binomial regression with a log link and generalized estimating equations (GEE) with robust standard errors to account for village-level clustering49,50. For the child health models, we adjusted for sex and age in months. We estimated prevalence ratios by calculating the exponential of the model coefficients for the group assignment. Statistically significant effect of the assignment were determined by using a two-sided Type I error rate of 0.05. We provide sample proportions and 95% confidence intervals for outcomes on filter coverage, acceptability, and use in the intervention group.

Covariate adjustment for imbalance

We reviewed the baseline data to see if there were large differences (>10% difference) between arms in socio-economic and household variables that are established determinants of drinking water quality or childhood diarrhoea (Table 1). Covariates that had little variation in the study population (e.g., over 95% prevalence or less than 5% prevalence) were excluded from adjustment. We then examined the relationship between primary and secondary outcomes and imbalanced covariates of concern (e.g., socioeconomic status, access to handwashing location, and access to improved sanitation) in individual bivariate analyses. Socioeconomic status was associated with diarrheal prevalence in children under 5 and 2 (p < 0.05). Access to a handwashing location was associated (p < 0.05) with only very high levels of E. coli bacteria (≥100 CFU/100 mL) and to diarrheal prevalence in children under 5 (p < 0.05). Access to sanitation was not associated with any outcome. We adjusted for socio-economic status and access to handwashing station in separate sensitivity analyses and compared results to unadjusted models to see if there were considerable differences in effects of the intervention. Water quality effects observed in unadjusted models were comparable to models adjusted for access to handwashing. Effects on under-5 child diarrhoea prevalence from the intervention had a 5 percent difference between the unadjusted model and adjusted model with socioeconomic status. Effects on under-5 child diarrhoea prevalence from the intervention had less than one percent difference between the unadjusted model and adjusted model with access to handwashing location. Therefore, we chose to only adjust for socio-economic status in all final models. Unadjusted and adjusted models are presented together in Tables 3 and 4.

Clustering considerations

Current GEE statistical packages are limited in that they only allow for adjusting for one level of clustering. We adjusted at the village-level because it is the highest level of clustering that is of concern and the unit of randomization51, which should intrinsically adjust for lower levels of clustering49. In sensitivity analyses, we adjusted for household-level clustering to account for longitudinal sampling, but did not see major differences in the water quality or diarrhea effects compared to the models adjusted for village-level clustering. The comparison in presented in the water quality results in Supplementary Table 2.

All analyses were done using Stata 16 (Stata Corporation, College station, TX, USA)52.

Ethics and registration

The trial is registered under the Pan African Clinical Trial Registry, Trial ID = PACTR201812547047839. The protocol received ethical approval and was annually renewed by the Emory University Institutional Review Board (CR001-IRB00106424) and Rwanda National Ethics Committee (IRB 0001497). We obtained signed informed consent from the main survey respondent during enrollment.

Next Post

AMA playbook offers a road map to private practice opportunities

Mon Sep 12 , 2022
The resident training landscape is largely populated by big medical institutions and, as a result, new physicians are offered little exposure to private practice settings. An AMA resource helps aspiring independent physicians make up for this knowledge gap by providing tips on how to start, maintain or grow a private […]
AMA playbook offers a road map to private practice opportunities