Study design and rationale for the PAASIM project: a matched cohort study on urban water supply improvements and infant enteric pathogen infection, gut microbiome development and health in Mozambique

Introduction Despite clear linkages between provision of clean water and improvements in child health, limited information exists about the health impacts of large water infrastructure improvements in low-income settings. Billions of dollars are spent annually to improve urban water supply, and rigorous evaluation of these improvements, especially targeting informal settlements, is critical to guide policy and investment strategies. Objective measures of infection and exposure to pathogens, and measures of gut function, are needed to understand the effectiveness and impact of water supply improvements. Methods and analysis In the PAASIM study, we examine the impact of water system improvements on acute and chronic health outcomes in children in a low-income urban area of Beira, Mozambique, comprising 62 sub-neighbourhoods and ∼26 300 households. This prospective matched cohort study follows 548 mother-child dyads from late pregnancy through 12 months of age. Primary outcomes include measures of enteric pathogen infections, gut microbiome composition and source drinking water microbiological quality, measured at the child’s 12-month visit. Additional outcomes include diarrhoea prevalence, child growth, previous enteric pathogen exposure, child mortality and various measures of water access and quality. Our analyses will compare (1) subjects living in sub-neighbourhoods with the improved water to those living in sub-neighbourhoods without these improvements; and (2) subjects with household water connections on their premises to those without such a connection. This study will provide critical information to understand how to optimise investments for improving child health, filling the information gap about the impact of piped water provision to low-income urban households, using novel gastrointestinal disease outcomes. Ethics and dissemination This study was approved by the Emory University Institutional Review Board and the National Bio-Ethics Committee for Health in Mozambique. The pre-analysis plan is published on the Open Science Framework platform (https://osf.io/4rkn6/). Results will be shared with relevant stakeholders locally, and through publications. Our study site is the coastal city of Beira, the second largest city in Mozambique (population ~530 000),74 which serves as a gateway for both the central interior portion of the country and a trade corridor to neighbouring land-locked nations. The centre of Beira is bordered by unplanned, informal settlements inhabited by over 300 000 low-income residents.75 A 2018 survey in Beira by Water & Sanitation for the Urban Poor (WSUP) showed that of 5643 respondents, only 28% had a household water connection, and among those households without a connection, 83% used their neighbour’s tap as their main source of water (unpublished data, courtesy of WSUP). Therefore, improvement of water supply and delivery infrastructure is a priority. The World Bank funded the Water Service & Institutional Support (WASIS-II) project in 2016 to address the low access to improved water supply in Mozambique,76 investing US$140 million with the Mozambican public institutions FIPAG (responsible for the public and private investment programme in urban water supply systems that serves as the water utility in Beira) and AURA, IP (the water regulatory authority responsible for the economic regulation and consumer protection of service provision). In addition, improvements in Beira are being augmented by investments from other groups, in particular the Dutch government, through infrastructure upgrades as well as emergency response funds following Cyclone Idai in 2019.77 Improvements in the city of Beira include rehabilitation of water treatment facilities, replacing existing pipe mains that are failing, reticulation of water supply to new areas previously without water service, improving service in areas with poor coverage or low water pressure and subsidising water connection fees for the poor. Several aspects of a city-wide water supply improvement project pose challenges to implementing a rigorous epidemiological study, particularly as people living in neighbourhoods with piped water often differ in myriad ways from people living in neighbourhoods without piped water. Water improvements to communities are often based on the needs or demographics of the community, government or donor priorities and engineering considerations. Provisions of water supply to new areas previously without water service—or dramatic improvements in access and availability—represent a fundamental development that changes the livability and sometimes the makeup of the community.2 These issues lead to difficulty in finding a comparable control group for epidemiological comparison. Furthermore, rollouts of water interventions often happen in continuous phases over time, and these changes might coincide with other events or community improvements that similarly impact health, making it difficult for that community to serve as its own control in a pre–post design. By using a prospective matched cohort in this unique context of an ongoing natural experiment, we are able to overcome many of these difficulties. The prospective nature of our study allows better control of confounding through matching, restriction and rigorous and thoughtful collection of potential control variables. We specifically focus on one region of the city (figure 2), where some neighbourhoods received water system improvements focused on preventing water losses through replacement of the piped water distribution system in dense, low-income settings; other neighbouring areas with similar demographic characteristics did not receive these improvements. Neighbourhood-level matching took place in the context of a natural experiment, where the delayed rollout across the city allows us to find and compare intervention and control neighbourhoods that are similar in many ways, before the rollout eventually reaches all potential control areas. Map of PAASIM study site in Beira, Mozambique. Map of Beira, Mozambique, with enlargement highlighting study site. Red lines indicate the new distribution system water network. Blue lines indicate other parts of the water network. Grey shaded areas indicate neighbourhoods enrolled in the study. The Water Loss Reduction Project represents a subset of the improvements being carried out by FIPAG, with co-funding from the Dutch government and the World Bank WASIS-II project. The improvements are designed to reduce illegal connections, thereby increasing the water pressure and quality and increasing the system’s capacity for household connections. These areas also received some benefits related to improvements to the water intake and distribution systems. FIPAG undertook a campaign to offer new connections to households. These improvements were completed in some informal settlements in these low-income areas of the city in 2019, with other adjacent neighbourhoods with similar density and socioeconomic profile slated for completion in future years but not within the time frame of our study. These specific distribution system upgrades therefore represent a unique opportunity to examine the impacts of community-scale water improvements with neighbouring communities who did not receive the intervention serving as control areas for comparison. The neighbourhoods under study are also in the lowest income—and therefore highest need areas of the city. We will perform analyses that take into account the four factorial possible household types, based on mother–child dyad living in a sub-neighbourhood with or without the improved water network and with or without a household connection (figure 3). Our primary analyses focus on assessing (A) the total network effect, by comparing subjects living in sub-neighbourhoods with the improved water network (Household Types 1 and 2) to those living in sub-neighbourhoods without these improvements (Household Types 3 and 4); and (B) the direct household connection effect, by comparing subjects with household water connections on their premises (Household Types 1 and 3) to those without a connection (Household Types 2 and 4). Depending on the results of these two primary analyses, secondary analyses may evaluate the other comparisons depicted in figure 3. PAASIM study summary diagram. The diagram reflects the summarised (A) study design and data collection, and (B) data analysis approaches to isolate the effects of both overall water supply infrastructural improvements as well as the presence of a household water connection. *The dark blue shading indicates neighbourhoods that benefitted from water distribution system improvements (‘improved water network’) and the light blue shading indicates similar neighbourhoods that have not received improvements (‘unimproved water network’). Within each neighbourhood, some households have a direct household connection to the water network and others do not. The reasons for some neighbourhoods receiving the improvements and others not were a result of resource constraints. According to FIPAG, decisions on the order of improvements in different neighbourhoods were guided by resource constraints as well as engineering logistics. We conducted a population-based survey (described below) that allowed us to both restrict and match study sub-neighbourhoods, thereby creating a statistically appropriate counterfactual for strong internal validity. The evaluation of a real-world intervention delivered in an informal urban setting provides strong external validity for estimating the effects of similar interventions in other LMIC urban sites. Our study design allows us to isolate the effects of both overall water supply infrastructural improvements as well as the presence of a household water connection. The presence of control areas not receiving upgrades adjacent to intervention areas that are matched on socioeconomic and density variables is unique to this study location. We collect data at multiple time points for each study household, allowing us to examine variability in each of the measures taken from each household, rather than at a single point in time, and also allowing for longitudinal analyses of the households and the individual enrolled subjects. We also employ rigorous measures of exposure and novel and objective outcome measures, including gut microbiome composition and molecular detection of enteric pathogens. The executive secretary of AURA, IP, was directly involved in the formation of the research questions, and FIPAG personnel were also engaged from the initiation of the project in helping to develop the study design. Our team also received input from other public agency stakeholders during workshops that were held prior to initiation of the study. Study subjects and members of the general public were not involved in the study design. We provide regular updates with data summaries to public agency stakeholders, and plan to disseminate the main results to all study participants and also through public presentations for stakeholders in both Beira and Maputo. Sub-neighbourhood eligibility, selection and matching of intervention and control sub-neighbourhoods occurred through a two-step process: Our study is a natural experiment, where the investigators had no control over the selection or timing of the intervention implementation. The study flow diagram is shown in figure 4. We worked with FIPAG to determine which neighbourhoods in Beira were to receive water distribution system upgrades prior to initiation of enrolment (2020) and before the end of the study (2023). FIPAG provided maps and timelines for construction works related to the upgrades, and the specific areas participating in the water loss reduction project. We also worked with FIPAG and through satellite imagery to identify similarly dense low-income areas in Beira that were not slated to receive water network upgrades. A total of 17 potential neighbourhoods were considered for inclusion in the study, and neighbourhoods were divided into 80 sub-neighbourhoods, delineated along natural boundaries such as roads or waterways. ‘Intervention’ sub-neighbourhoods include areas with the upgraded water distribution system. ‘Control’ sub-neighbourhoods include areas not receiving these improvements during the time period of the study. Within both intervention and control sub-neighbourhoods, some households have a connection to the water system and others do not. We excluded nine control sub-neighbourhoods that were in close proximity to intervention sub-neighbourhoods or that were scheduled to receive the interventions within the timetable of our project; some control sub-neighbourhoods are slated to receive the intervention after completion of our study. PAASIM study flow diagram. SES, socioeconomic status. We followed the suggestion of Arnold et al to use baseline (preintervention) data at the community level to match intervention to control communities when randomisation is not possible.23 To characterise sub-neighbourhoods for further matching and restriction, we performed a population-based community survey in November–December 2020 of approximately 1700 households; this provided approximately a 5% proportional sample of our potential study sub-neighbourhoods. We used a random grid sampling approach to estimate household density, using Google Earth satellite imagery, where a grid was placed over an area, and a random selection of squares were selected and counted independently in duplicate, and the number of houses per unit was extrapolated across unsampled squares. The survey contained modules regarding household demographics, water access and practices, sanitation access and practices, household assets and wealth indicators, as well as questions related to COVID-19. A socioeconomic status (SES) score was constructed using the ‘simple poverty scorecard’78 developed specifically for Mozambique, and scores were aggregated at the sub-neighbourhood level, and categorised into tertiles. We matched intervention sub-neighbourhoods to control sub-neighbourhoods, using coarsened exact matching,79 80 with intervention sub-neighbourhoods being matched to control sub-neighbourhoods within the same tertile of both SES and population density. Four neighbourhoods (encompassing nine sub-neighbourhoods) were found to be outliers in terms of their sub-neighbourhood-level SES or sanitation, and were excluded from the study sampling frame. Ultimately, we designated 36 intervention sub-neighbourhoods, with an estimated 16 800 households, and 26 control sub-neighbourhoods, with an estimated 9500 households. We recruit pregnant women at the last trimester of pregnancy and follow the infant–mother dyads until the child is 12 months old (figure 5). We selected the first 12 months of life because it is a critical development window,81–83 it is a time when children are most at risk of acute and chronic effects of enteropathogen infection84 and it is a short enough period of time to avoid changes in water access that might occur. We recruit mothers at the end of their pregnancy so we can collect data on household risk factors (including drinking water quality) during the gestational period. Active recruitment occurs through identification of pregnant women in the 2020 population-based survey, lists of pregnant women visiting local health centres for prenatal care and study staff visiting under-enrolled sub-neighbourhoods throughout the recruitment period. Based on Ministry of Health data for Sofala Province (where Beira is located), virtually all mothers attend prenatal clinical visits.85 Passive strategies include referrals of pregnant women by study participants and community leaders. We aim to have complete data on a total of 548 infant–mother dyads, approximately evenly divided between the intervention and control groups. We will continue to enrol dyads into both arms until we reach a minimum of 274 dyads with complete data in each arm, to ensure temporal balance throughout the duration of the study period. Data and sample collection timeline for outcomes in the PAASIM study. Data and sample collection of infant–mother dyads enrolled into the study will be used to address a series of questions about the impact of community-level water system improvements on acute and chronic health outcomes in children in a low-income urban area of Mozambique. During an initial pre-birth visit, pregnant women are assessed for study eligibility: (1) 18 years or older, (2) in third trimester of pregnancy, (3) resides in enrolled study cluster, (4) not planning to move within the next 12 months, (5) carrying a singleton birth and (6) consents to take part in the study. We will reassess study eligibility at each follow-up visit and record if enrolled participants have been lost to follow-up. A local data collection firm (WE Consult) performs the in-country coordination of participant enrolment, data collection and sample collection. Enumerators conduct household visits before birth for consent, eligibility and conditions. At months 3, 6, 9 and 12 we deploy survey instruments to collect data on key indicators through structured observations, reports from respondents and objective measurements (online supplemental table S1). We assess a number of variables related to drinking water, including aspects of water quality, water access, water availability, water security, water consumption and participant satisfaction with water. Enumerators also conduct brief active surveillance calls on a monthly-basis by phone with caregivers to gather supplemental information on prenatal and perinatal environmental exposures and illnesses, on child illness symptoms and intake of medicines, vitamins, breast feeding and introduction of complementary foods (figure 5). To facilitate communication with the study team, participants receive a 150 MZN (Mozambican metical) phone credit at each visit. Aside from these phone credits, there is no financial incentive provided to participants to partake in the study, per Mozambican guidelines for human subjects research. We ask the caregiver to report diarrhoea and blood in the stool (dysentery) of the index child in the previous week at the 3, 6, 9 and 12-month surveys and during active surveillance calls; due to concerns about reporting biases, we also include negative control outcomes.86 At each post-birth visit we measure child: (1) length, weight and head circumference, and (2) calculate length-for-age and weight-for-age Z-scores. Prevalence of stunting and underweight are defined as two SD below median of the reference population.87 All data are collected on electronic tablets using Open Data Kit Collect, an open-source programme which allows offline data collection on a mobile device.88 Additional details are provided in the online supplemental material. bmjopen-2022-067341supp001.pdf We briefly describe sample collection and downstream processing and analysis here, with additional details provided in the online supplemental material. Stool of the index child is collected at months 3, 6, 9 and 12. Three aliquots are placed in temperature stable lysis buffer collection tubes, and two additional aliquots are used to prepare a slide for Kato-Katz analysis of parasite ova.89 Eligible participants are referred for deworming medicine at the 12-month visit, after returning results of the parasitological examination to study subjects in collaboration with Instituto Nacional de Saúde (INS) staff in Beira. Extracted nucleic acids are analysed: (1) using the TaqMan Array Card (TAC, Thermo Fisher Scientific, Waltham, Massachusetts, USA) assay, which allows quantification by real-time PCR via a 384-well microfluidic card for simultaneous detection of multiple viral, bacterial and parasitic enteric pathogen targets as well as antimicrobial resistance genes,90 customised for our targets of interest (online supplemental table S3); and (2) by sequencing of the V4 region of the 16S ribosomal RNA (rRNA) gene amplicon to characterise gut microbiome community structure and composition. Bioinformatic analyses will be completed using the QIIME2 software platform.91 A trained nurse from INS collects up to six dried spots of capillary blood of the index child at 6, 9 and 12-month visits on Tropbio Filter Paper Blood Collection Disks (Cellabs, Sydney, Australia), using a 2 mm lancet. Samples are stored at −20°C at INS facilities in Beira and shipped at ambient temperature.92 We will use the Luminex platform to carry out high throughput, multiplex antibody assays that enable the simultaneous measurement of quantitative antibody responses to dozens of pathogens from a single blood spot.93 Our first measure will occur at 6 months, to avoid detection of maternal antibodies that wane over the first 3–6 months of life.94 We collect 100 mL household drinking water samples from source and stored water at all household visits. To complement the household sampling, we collect samples from a selection of 45 public standpipes located within the study area and 55 additional public standpipes located elsewhere in the city of Beira. At public standpipes we also measure water pressure by measuring time to fill a fixed volume (1 L or 5 L, depending on the pressure). Samples are processed for faecal indicator bacteria within 6 hours of collection using Colilert-18 reagent and the Quanti-Tray/2000 MPN method (IDEXX Laboratories, Westbrook, Maine, USA), as well as for free and total chlorine levels and additional physiochemical parameters (pH, conductivity and turbidity). Large volume samples will be collected from a subset of 50 households (1-2 L, processed by membrane filtration) and 25 public standpipes (50 L, processed by dead end ultrafiltration95) in two different seasons, and tested for enteropathogens using the TAC assay. Our primary outcomes include: any bacteria or protozoa infection at age 12 months after birth; individual pathogens or pathogen groups; child gut microbiome composition; and household source water quality. We include viral, protozoal and bacterial pathogens responsible for the vast majority of enteric pathogen infections and global disease burden.96 97 While we measure viral pathogens using the TAC assay, they will be excluded from the combined enteropathogen prevalence primary outcome measure, because waterborne transmission is unlikely to dominate for these viral pathogens.98–101 In addition to the aforementioned reasons related to child development and infection risk, measuring pathogens at 12 months will give us the greatest power to detect a difference, given higher levels of infection at that age than in younger children. We will measure gut microbiome using 16S rRNA gene amplicon sequencing in the full sample at 12 months and in a random subset of 200 children with complete data at 3, 6 and 9 months, evenly distributed between intervention and control groups; dyads eligible for subset sampling will include those with complete stool sample collection and unchanged intervention exposure conditions. The 12-month samples will allow us to compare all study children at a common time, when all children are consuming drinking water and once the gut microbiome has become relatively established102; the longitudinal samples will allow for comparison of development of the microbiome over time between the two groups. Microbiome outcomes include alpha and beta diversity metrics, and identification of enriched taxonomic groups. We also include household source water quality as a primary exposure outcome, as understanding whether exposure to microbial contaminants is altered is considered a critical aspect of evaluation of WASH projects.71 103 Additional non-primary outcomes include pathogen count, pathogen community similarity (measured using Jaccard Similarity Index), diarrhoea, child growth and prior enteropathogen infection (measured using serology on dried blood spot samples). We will measure additional water quality exposure measures, as well as measures of exposure to the improved water system, such as fidelity of the intervention (eg, improvements to water quantity and coverage of household taps) and receipt of the intervention by community members (eg, reductions in water insecurity, increased water use). These fidelity and uptake measures will be collected at all time points through direct observation and respondent report. Available minimal detectable effect sizes are summarised in table 1 and calculations are further detailed in the online supplemental material. Primary and non-primary health outcomes and exposure outcomes for the PAASIM study Patent enteric pathogen infection in stool is measured via TaqMan Array Card assay; Stool microbiome composition is measured via 16 s ribosomal RNA amplicon sequencing; prior enteric pathogen exposure is measured via serological assays of dried blood spots. Drinking water quality measured by IDEXX as Escherichia coli most probable number/100 mL. See text for further details. Calculations and additional values for the minimum detectable effect are described in the online supplemental material. *Anticipated control group prevalence based on control group prevalence for 10–14 months old in the MapSan trial.32 † Samples also collected at 3, 6 and 9 months of age may also be analysed, depending on results of primary analysis at 12 months. ‡A subset of n=200 samples will be analysed for gut microbiome composition in children at 3, 6 and 9 months of age. All 12-month samples will be analysed. §This is a conservative estimate as it does not account for weekly active surveillance. ¶Defined as two SD below median of the reference population. **Samples will be analysed at 3, 6, 9 and 12 months, but 12-month samples are the primary outcome of interest. The pre-analysis plan for this study is published on the Open Science Framework platform (https://osf.io/4rkn6/). To assess the impact of the intervention on our primary enteric pathogen infection outcomes and water quality exposure outcome (table 1), we will use an intention-to-treat analysis approach to compare children living in intervention versus control sub-neighbourhoods, without regard to uptake/use of the intervention (ie, direct household connection on the premises). This answers the relevant policy question ‘what if an improvement is delivered to an area?’ We will use multivariable log-linear binomial regression models, as pathogen infection is a binary variable, and will use generalised estimating equations to account for clustering at the sub-neighbourhood level. We group matched on sub-neighbourhood-level SES and population density, using weighting to account for unequal numbers between the intervention and control areas within each matching stratum.104 We will additionally control for household-level and individual-level confounders, including household SES, household sanitation, mother’s education-level and child sex. We may adjust for additional variables if there are found to be imbalances in potential confounders in our baseline assessment. We hypothesise that the intervention will lead to reductions in enteric pathogens among children and microbial water contamination of source water. For additional outcomes and exposure variables of interest (table 1), we will use a similar modelling approach, using log-linear binomial regression models for binary outcomes, linear regression models for continuous outcomes and Poisson (or negative binomial) models for count outcomes. For outcomes measured at multiple time points, we will present results separately for each given time point. For these analyses, we will control for sub-neighbourhood-level SES and population density through matching, and will additionally control for household sanitation, mother’s education-level, child sex and any other variables that are imbalanced and are conceivably potential confounders. For previous enteropathogen exposure evaluated using serological measures we hypothesise that those in the intervention group will show delays in pathogen acquisition. To assess the impact of the intervention on microbiome outcomes, we will evaluate alpha diversity (Chao1 species richness estimator, Pielou’s J evenness estimator and the Shannon Diversity Index105) using the same modelling approach as described above for continuous outcomes. Linear discriminate effect size analyses will be used to evaluate specific 16S rRNA gene-based Operational Taxonomic Units (OTUs) that differ between individuals in intervention versus control groups, and will include effect size corrections.106 We will examine the impact of intervention groups, controlling for other covariates, on community similarity using Adonis permutation models,107 based on weighted UniFrac and Bray-Curtis distances, and evaluate and visualise differences using principal components analysis (PCA) and/or non-metric multidimensional scaling (NMDS) plots. We hypothesise that we will be able to observe detectable differences in gut microbiome composition in children living in intervention versus control sub-neighbourhoods and we will report these differences at the individual OTU and bacterial family levels. To assess the effect of having a water connection at the household or compound, we will use models similar to those described above, but accounting for a household network connection. We will also assess the interaction between the household and neighbourhood network variables, which will allow us to contrast and estimate indirect, direct and total effects, as shown in figure 3. We hypothesise that participants with both improved water networks in their sub-neighbourhoods and household water connections will most benefit from the interventions in terms of our primary and non-primary health outcomes and exposure outcomes of interest. For select primary outcomes, we will assess if there is effect modification by a third variable, such as follow-up round/age, participant sex and household sanitation access. We will use interaction terms to identify potential interactions, and will present stratified results (eg, separately by sex) if interactions are detected. The intervention status of sub-neighbourhoods was set at baseline, but if control sub-neighbourhood(s) receive the intervention after the study has started, we will perform sensitivity analyses dropping and/or recategorising sub-neighbourhood(s) that crossed over. There are several analyses where we do not use the matched-design and intervention variable in our analyses. For example, we will assess associations between various water measures on health, without regard to the intervention designation. We will also examine changes in the gut microbiomes of children over time. Additional analyses will be described and documented in OSF. Our minimal sample size of 548 households—half in intervention and half in matched control sub-neighbourhoods—was powered for our primary outcome of prevalence of any non-viral pathogen. Using data from the MapSan trial for children 10–14 months of age (J. Knee, pers comm) we used a control group prevalence of 70% for any non-viral pathogen, and estimated the ability to detect a relative risk of 0.74, alpha=0.05 and power=80% using a two-sided test for significance.32 We estimated a sub-neighbourhood-level interclass correlation coefficient (ICC) of 0.05 (a moderate estimate) among our 62 designated sub-neighbourhoods. We will also report on the final ICC and other assumptions of this power analysis at the end of the study. Estimates of minimum detectable effect sizes based on control prevalence of the outcome of interest (online supplemental table S2) show we may be adequately powered to detect a difference in some individual pathogens if those pathogens have high prevalence and/or if they are strongly associated with the water supply improvement intervention (eg, waterborne pathogens). We target planned recruitment at 900 pregnant women in the third trimester, to account for incomplete data and loss to follow-up. We used sub-neighbourhood enrolment targets proportionate to our density estimates to achieve balance across intervention and control sub-neighbourhoods. All laboratory personnel and field enumerators are blinded to the intervention status of the samples and households. Participants cannot be blinded to their household-level water exposure status or cluster-level exposure status, although participants may or may not know about water improvements in their particular neighbourhood. A primary analyst external to the core data management team is blinded to the group assignments until the data cleaning and primary analysis are completed. Details of these procedures are included in the online supplemental material. Unblinding will occur only after primary outcome models are developed and compared between two independent analysts. Analyses examining the impact of the intervention on non-primary outcomes or exposures of interest will not be unblinded until after analyses that examine the impact of the intervention on our primary outcomes have been completed. Purely observational analyses that do not require information on intervention groups may be completed before unblinding occurs.