Differential Effectiveness of Water, Sanitation, and Handwashing Interventions to Reduce Child Diarrhea in Dry and Rainy Seasons: A Systematic Review and Meta-Analysis of Intervention Trials

Abstract

Background:

Studies evaluating how water, sanitation, and/or handwashing (WASH) interventions in low- and middle-income countries impact diarrheal diseases have shown inconsistent results. The prevalence of enteric pathogen infections and diarrhea are highly seasonal and climate-sensitive, which could explain heterogeneous findings. Understanding how season influences the effectiveness of WASH interventions is critical for informing intervention approaches that will be resistant under the varying weather conditions that climate change will bring.

Methods:

We conducted a systematic review of the literature and meta-analysis to test whether and to what extent the impact of WASH interventions on diarrhea differs by season. We searched the literature for randomized and nonrandomized controlled WASH intervention trials and identified the season in which data were collected—rainy, dry, or both—for each study using proximate land station weather datasets. We compared the relative risk (RR) estimates for the impact of interventions on diarrhea for each study, stratified by season, and analyzed estimates using meta-analysis and meta-regression. This study is registered with PROSPERO, CRD42021231137.

Results:

A total of 50 studies met the inclusion criteria, resulting in 34 drinking water intervention estimates, 8 sanitation intervention estimates, and 14 handwashing intervention estimates. Of the total studies, 60% ($n=30$) spanned more than one season, with most single-season studies (75%, $n=15$) occurring exclusively in the dry season. The effect of WASH interventions was stronger in dry seasons than in rainy seasons, with a 33% [95% confidence interval (CI): 24%, 41%] and 18% reduction (95% CI: 5%, 29%) in diarrhea risk, respectively. When stratified by type of intervention, the stronger effect size in dry seasons was consistent for water and handwashing interventions but not for sanitation interventions.

Conclusions:

Estimates of the seasonal impact of WASH interventions revealed larger effects in the dry season than in the rainy season overall and for water and handwashing interventions in particular. These patterns likely affected previous estimates of intervention effectiveness, which included more dry season estimates. These findings suggest the need to collect data across seasons and report seasonally stratified results to allow for more accurate estimates of the burden of disease impacted by WASH investments and to improve projections of potential impacts of these interventions under future climate conditions. These findings also underscore the need for robust WASH interventions designed to be resistant to seasonal variations in temperature and rainfall now and under future climate change scenarios. https://doi.org/10.1289/EHP14502

Introduction

Climate change and its inextricable links to infectious disease transmission and food, water, and financial insecurity disproportionally impact the most vulnerable populations and threaten to reverse decades of public health progress.^1,2 Diarrheal diseases, the ninth leading cause of disability-adjusted life-years (DALYs) and the fourth leading cause of death among children under 5 y of age,^3,4 are particularly sensitive to both seasonal patterns and climate change.^5–7 Increases in precipitation, storm surges, and sea-level rise can contribute to flooding and runoff that can bring sewage, chemicals, and other enteric disease-causing agents into contact with humans.⁸ In addition, drought can concentrate enteropathogens in scarce water supplies and increased temperatures can lead to greater consumption of potentially contaminated water sources.^6,8 The Intergovernmental Panel on Climate Change (IPCC) states with “high confidence” that water-related risks are projected to increase with every degree of global warming, and more vulnerable populations and regions are expected to face greater risks.⁹ The WHO estimates that between the years 2030 and 2050, there will be an additional 48,000 diarrhea-related deaths per year in children under 15 y of age due to climate change.¹⁰ The ability of water and sanitation (WASH) infrastructure to prevent enteropathogen transmission across seasonal variability is paramount to mitigating health impacts under both current conditions and future climate scenarios.¹¹

Interventions that target improvements in WASH infrastructure, behaviors, and/or conditions have been shown to lower the risk of childhood diarrhea morbidity and mortality.^12–16 A recent systematic review and meta-analysis of the impact of WASH interventions on diarrhea found that drinking water interventions, specifically point-of-use (POU) water treatment and improved drinking water supply on-premises, and improved sanitation, specifically sewer connections, each reduced childhood diarrhea risk by $\sim 50\%$.¹² Handwashing interventions reduced diarrhea risk by 30%.¹² The analysis revealed substantial heterogeneity in point estimates, and several recent large-scale trials have found null effects on childhood diarrhea.¹⁷

Seasonal and climatic drivers of variability in diarrheal diseases are well-recognized, with heavy rainfall, flooding, and increases in ambient temperature—meteorological conditions closely associated with climate change—all being positively associated with diarrhea incidence.⁸ Diarrheal diseases also peak seasonally, typically during rainy seasons in tropical and subtropical climates,^18–22 and enteropathogens are highly affected by ambient temperature conditions.^5,8 In high-income countries with well-developed water and sanitation infrastructure, the effects of seasonal meteorological variability manifest primarily under extreme conditions when infrastructure is overwhelmed.^23–25

The highly seasonal incidence of enteric pathogen infections and diarrhea could help explain heterogenous WASH trial results, yet seasonal variability in the effectiveness of interventions remains unexplored. Here, we examine if and how the effect of WASH interventions on diarrheal diseases differs under rainy vs. dry conditions.

Methods

Our systematic review and meta-analysis of the impact of WASH interventions on diarrheal diseases by season followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines (see PRISMA Checklist, Appendix).²⁶

Search Strategy and Selection Criteria

We searched Ovid MEDLINE, Embase, Scopus, Cochrane Library, and BIOSIS Citation Index using both keywords and Medical Subject Headings terms. The search strategy and a full list of search terms are provided Excel Table S1. This search included literature published between 1 January 1970 and 25 May 2021 and was used to inform Wolf et al.¹² We subsequently updated the search through 9 October 2023. Including the results from a previous systematic review, Wolf et al.,¹² this systematic review covered studies published from 1 January 1970. Studies were included if they were published in peer-reviewed scientific journals in English or French. Included studies examined the impact of one or more WASH interventions on diarrheal morbidity in children under 5 y of age (children under 5 y were the primary focus; however, if studies did not provide data for this age group, we then extracted data for broader age groups, including older children and adults). Eligible comparators were study participants without involvement in the eligible WASH interventions or those who were given control or placebo interventions. Interventions included those that improved water supply, water quality, sanitation, and/or handwashing. For the water and sanitation intervention analyses, we included studies of interventions in household or community settings in low- to middle-income countries (LMICs).²⁷ In addition to these interventions, for the handwashing intervention analysis only, we included studies conducted in high-income countries that related to children in schools or daycare centers.²⁷

We included individual and cluster randomized controlled trials (RCTs), and the following nonrandomized controlled studies: quasi-randomized and nonrandomized controlled trials; studies using time-series and interrupted time-series design; cross-sectional studies that include a comparison group (controlled before and after studies, historically controlled studies); and case–control and cohort studies when they were related to a specified intervention. Studies that included population groups not representative of the general community were excluded. These included specific subgroups, such as individuals with human immunodeficiency virus (HIV) or children of HIV-positive mothers, people residing in refugee camps, and hospital patients, whose health statuses or living conditions may not reflect those of the broader population.

For the seasonal component of this review and meta-analysis, we excluded studies if they occurred across multiple seasons but did not present seasonally or longitudinally disaggregated diarrheal results. Studies were also excluded if they did not state the dates of data collection, did not indicate the study location, and/or did not have a proximate (within 100km) land station weather dataset available to classify rainy and dry seasons.

Data Extraction and Analysis

Data extraction.

From the included studies, we gathered data on setting, age group used for measurement of outcome, and type of WASH intervention, and we conducted quality and risk of bias assessments. We extracted the dates of diarrhea morbidity data collection post-intervention implementation, sample size, and diarrhea effect estimates for each study. We examined each study for diarrhea morbidity data that could be disaggregated by time (e.g., weekly, biweekly, monthly point estimates), extracted these data when available, and contacted authors when they were not. For studies that included disaggregated diarrhea morbidity data in figures only, we used PlotDigitizer (https://plotdigitizer.com) to extract point estimates of effect. For studies that occurred entirely in one season, we extracted diarrhea effect estimates without modification. All data were extracted by two people independently and cross-checked.

Calculating seasonal parameters.

To improve the precision of our key stratifying variable of interest (i.e., season), we categorized data from each study as either rainy or dry season using date and location-specific parameters for data collection.¹⁸ We adapted methods from Chao et al., who identified data-driven seasons for a seasonal diarrheal pathogens analysis by applying principal component analysis (PCA) and k-means clustering to monthly land station weather data, empirically categorizing distinct seasons based on environmental conditions rather than literature-based fixed calendar dates.¹⁸ This approach allowed for the creation of seasons based on a composite of weather variables (precipitation, temperature, humidity) that are likely to influence environmental pathogen prevalence and transmission. In addition, this method provided a quantitative and objective way to define seasons based on actual weather data, acknowledging the substantial year-to-year variability that can influence the onset and duration of these seasons.

Following this approach, we used land station weather data to create weather datasets for each study. To create these weather datasets, we used the location (city/region and country) and dates of data collection (months and years of diarrhea morbidity data collection post intervention implementation) for each study to select corresponding data from the National Oceanic and Atmospheric Administration’s (NOAA) Global Surface Summary of the Day (GSOD) catalog of land station weather datasets.²⁸ We gathered these GSOD weather datasets using the R package gsodr. The GSOD weather datasets queried were for the complete year or years surrounding the dates of data collection. For example, if a study collected data from March until August of 2019, we gathered weather data for the entirety of 2019 and then determined seasons for that entire year. This approach meant that we defined our seasons relative to each year not relative to the study timeframe, which could artificially impose a multiseason component to that particular study. Each GSOD dataset included daily estimates of key weather variables that are primary components of season: mean temperature (°C), total precipitation (mm), elevation (km), and dewpoint. We used the GPS coordinates of the center point of the city/region for each study site to select the most proximate (no farther than $100\text{\hspace{0.17em}km}$ away) and the most complete (no more than one-third missing in weather variables) GSOD weather dataset. Daily relative and specific humidity values are not measured in these datasets; therefore, we calculated these daily values manually using the equations below.

Relative humidity²⁹:

$$RH=100\times \left[\left(112-\left(0.1\times T\right)+DP\right)/\left(112+\left(0.9\times T\right)\right)\right],$$ (1)

where RH is relative humidity in percentage, T is temperature in Celsius, and DP is the dewpoint in Celsius.

Specific humidity¹⁸:

$$P=101.3\times e^{-EL/\mathrm{8,200}},$$ (2)

$$E_s=0.622\times e^{(17.502\times T)/(T+240.97)},$$ (3)

$$E=\left(0.01\times RH\right)\times E_s,$$ (4)

$$W=\left(0.622\times E\right)/\left(P-E\right),$$ (5)

$$SH=1000\times W\left(1+W\right),$$ (6)

where P is air pressure in kPa, EL is elevation in meters, $E_s$ is saturation vapor pressure in kPa, E is the vapor pressure in kPa, W is the mixing ratio, and SH is the specific humidity in g/kg.

Once we created these weather datasets relative to each study, we reduced the dimensionality of each dataset using PCA. PCA served to reduce the dimensionality of the data by identifying principal components that captured the majority of variance in the weather parameters. We then performed k-means clustering to analyze the principal components and group similar months together. This process generated clusters of months that exhibited similar weather patterns, thereby creating distinct “seasons” within the data. Each cluster was labeled as either a “rainy” or “dry” season based on the predominant weather patterns it represented—specifically, clusters with months showing higher averages of precipitation, temperature, and humidity were classified as “rainy seasons,” whereas those with lower averages were designated as “dry seasons” within each study context.

Analysis.

Risk estimates for individual studies.

We calculated relative risks (RR) for each study and the standard errors of the log RR using standard formulae.³⁰ For studies occurring across seasons, we averaged longitudinal point estimates (e.g., weekly, biweekly, monthly estimates) to the seasonal level and created stratified seasonal RR estimates.³¹ This approach was necessary to standardize the data across studies that varied in the frequency of outcome measurement, ensuring that each provided a single summary measure per season. This intrastudy averaging was conducted without weighting the individual point estimates, because they were derived from the same study cohort and thus inherently carried equivalent statistical weight.

Odds ratios were converted to RR when the control group risk was reported.³⁰ For those studies that contained multiple intervention comparisons, we combined effect estimates for similar interventions that shared a control group, or if the interventions were not similar enough for combination, we included only the most comprehensive intervention in the analysis.³¹ For studies that reported on multiple intervention comparisons, if each intervention had its own control group, both RR estimates were included in the analysis.³¹

Multilevel meta-analyses stratified by season.

Using the seasonally stratified RR estimates, we conducted multiple random effects meta-analyses of:

1. the effect of all types of WASH interventions on diarrhea stratified by dry versus rainy season (Figure 1, Boxes 2 and 3);

2. the effect of each type of WASH intervention (i.e., water, sanitation, or handwashing) on diarrhea stratified by dry versus rainy season (Figure 1, Boxes 2 and 3); and

3. the effect of WASH interventions that occurred in a single season on diarrhea for the dry and rainy seasons (Figure 1, Boxes 4 and 5).

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram of study selection and seasonal stratifications for systematic review of water, sanitation, and handwashing (WASH) interventions by season conducted from 1 January 1970 to 9 October 2023. We ultimately included 50 studies in our analyses that contributed 56 effect estimates in total (4 studies contained effect estimates for multiple intervention types therefore the total number of estimates is higher than the total number of studies). When we applied seasonal stratifications to these studies, 30 studies (contributing 36 effect estimates) occurred across seasons and 20 studies occurred in a single season (contributing 20 effect estimates, Boxes 4 and 5). We calculated seasonally stratified effect estimates for all included studies, resulting in 92 seasonally stratified effect estimates for our main analyses (Box 1) with 51 stratified estimates in the dry season (Box 2) and 41 stratified estimates in the rainy season (Box 3).

We performed a weighted meta-analysis for each seasonal stratification, calculating the pooled RR for each scenario. The weights for this meta-analysis were derived from the inverse variance of each study’s seasonal RR estimate. This approach ensured that the influence of each study on the pooled results was proportional to its precision, with larger, more statistically precise studies exerting greater influence on the pooled result.

We used random effects meta-analysis to address potential dependency or correlation between effect estimates that came from the same study.^32,33 There were two scenarios where the assumption of independence among our study effect estimates was violated: a) Some studies contributed more than one effect estimate to our pooled analyses. For example, Luby et al.³⁴ presented an effect estimate for a POU water intervention and an effect estimate for a separate handwashing intervention; and b) multiseason studies contributed two seasonally stratified effect estimates (each from the same study) to the pooled analyses. We used random effects models that added an additional heterogeneity term (in addition to the sampling error of individual studies and the between-study heterogeneity estimator in a standard random effects meta-analysis model) to capture within-study heterogeneity or variance between effect sizes extracted from the same study.^32,33 We used the restricted maximum-likelihood (REML) estimator to estimate the variance of the distribution of true effect sizes (${\mathrm{\tau }}^2$) and Knapp-Hartung adjustments to control for the uncertainty in our estimates of between-study heterogeneity.^35,36

To compare the pooled effect estimates in dry vs. rainy seasons, we used Q tests to determine whether differences across seasons were large enough that they were not explainable by sampling error ($\mathrm{\alpha }=0.10$, common standard for this test).^37,38 To explore between-study heterogeneity among our pooled dry and rainy season studies, we used the $I^2$ statistic and prediction intervals.³⁷ We employed forest plots to provide a visual representation of individual study result as well as compare dry and rainy season results.

Multilevel meta-regression.

We fit a mixed-effects meta-regression model to further explore differences in effect by season in terms of other study covariates using seasonally stratified effect estimates from all of our included studies (Figure 1, Box 1). We set season as our main predictor of diarrhea morbidity along with study-level covariates for type of WASH intervention and setting (urban vs. rural) based on a priori hypotheses of association ($\mathrm{\alpha }=.05$). The basic form of the meta-regression model is given in the following equation:

$$y_{ij}={\mathrm{\beta }}_{1}Seaso{n}_i+{\mathrm{\beta }}_{2}Wate{r}_i+{\mathrm{\beta }}_{3}Sanitatio{n}_i+{\mathrm{\beta }}_{4}Hygien{e}_i+{\mathrm{\beta }}_{5}Settin{g}_i+{ϵ}_{ij}+{\mathrm{\zeta }}_{\left(2\right)ij}+{\mathrm{\zeta }}_{\left(3\right)j},$$ (7)

where $y_{ij}$ is the stratified log relative risk of diarrhea $i$ from study $j$, ${\mathrm{\beta }}_1-{\mathrm{\beta }}_5$ represent the change in the stratified log relative risk of diarrhea for a unit increase in each listed dichotomous variable, ${ϵ}_{ij}$ is the sampling error through which the stratified log relative risk estimate $i$ from study j deviates from the true effect, ${\mathrm{\zeta }}_{\left(2\right)ij}$ is an error term capturing within-study heterogeneity, and ${\mathrm{\zeta }}_{\left(3\right)j}$ is an error term capturing between-study heterogeneity. Between-study variance ${\mathrm{\sigma }}_n^2$ was estimated using common estimation procedures for random effects models (REML estimators). Because the outcome variable represents the log relative risk of diarrhea, we exponentiated the beta coefficients for each variable and the 95% confidence intervals (CIs) for ease of interpretation. Each exponentiated coefficient represents the estimated ratio of relative risks [or relative risk ratio (RRR)] of diarrhea for that particular variable, comparing two levels of the variable while holding other variables constant.

We performed three sensitivity analyses. The first excluded single-season studies to examine only those studies in this review that collected data across multiple seasons. The second excluded studies not located in tropical climate zones (i.e., excluding subtropical and temperature climate zones), to compare the seasonally stratified effect estimate trends derived for all studies to those derived only from studies in countries with more pronounced dry and rainy seasons. The third excluded comparisons from high-income settings within the handwashing interventions.

We assessed the risk of bias in each study using a modified Newcastle-Ottawa scale (NOS).³⁹ The NOS is a quality assessment scale originally designed for nonrandomized studies that was later modified by Pope et al. to include randomized studies in their assessment of the health impacts of air quality interventions.⁴⁰ The modified NOS considers the following areas of bias: selection bias, response bias, follow-up bias, misclassification bias, bias in outcome assessment, bias in outcome measurement, and bias in analysis, with each area of bias was adapted to evaluate mixed study designs. We assigned each included study a score of up to nine stars, with a higher number of stars indicating a smaller risk of bias. Further details about scoring criteria for each category of the modified NOS are included in Excel Tables S2, S3, and S4. We visually inspected a funnel plot to assess the risk of publication bias.

We conducted all analyses in R (version 4.0.5; R Development Core Team).

Results

Our final review included 50 studies: 34 drinking water intervention estimates (from 30 separate studies), 8 sanitation intervention estimates (from 6 separate studies), and 14 handwashing intervention estimates (from 10 separate studies) (Figure 1). Four additional studies provided multiple estimates, or data on more than one type of intervention, two of which provided data on all three (i.e., water, sanitation, and handwashing interventions). The included studies spanned 25 countries, with the majority taking place in tropical climate zones ($n=38$; 76%) where dry and rainy seasons are more pronounced (Figure 2).

Figure 2.

Geographical distribution of studies included in the systematic review. This map visualizes the locations of 50 studies addressing the impact of water, sanitation, and handwashing (WASH) interventions on diarrheal morbidity by season. Details on the specific countries and characteristics of each study can be found in Excel Tables S5, S6, and S7. World shapefile is from ArcGIS Hub.

Thirty studies occurred across seasons (contributing 36 effect estimates), and 20 studies were contained entirely in one season, 75% exclusively in dry seasons ($n=15$) and 25% exclusively in rainy seasons ($n=5$) (Figure 1). After we stratified effect estimates by season, there were 51 stratified effect estimates in the dry season (Figure 1, Box 2) and 41 stratified effect estimates in the rainy season (Figure 1, Box 3), for a total of 92 seasonally stratified effect estimates (Figure 1, Box 1).

Meta-Analysis

The effect of all types of WASH interventions on diarrhea morbidity stratified by dry vs. rainy season.

In dry seasons ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=51$; Figure 1, Box 2), meta-analysis showed that WASH interventions reduced risk of diarrhea by 33% ($\text{RR}=0.67$; 95% CI: 0.59, 0.76) (Table 1; Figure 3), whereas in rainy seasons ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=41$; Figure 1, Box 3), WASH interventions reduced risk of diarrhea by 18% ($\text{RR}=0.82$; 95% CI: 0.71, 0.95) (Table 1; Figure 4). The Q test for subgroup differences showed that this $15\text{\hspace{0.17em}percentage}$ point difference in the effectiveness of WASH interventions by season was statistically significant ($p=0.020$).

Table 1.

Results of meta-analyses assessing the impact of WASH on diarrheal morbidity by WASH category and by seasonal stratification.

Intervention type	Number of seasonally-stratified estimates	Effect size (95% CI)	Number of dry season stratified estimates	Dry season effect size (95% CI)	Number of rainy season stratified estimates	Rainy season effect size (95% CI)
All	92	0.73 (0.66, 0.80)	51	0.67 (0.59, 0.76)	41	0.82 (0.71, 0.95)
Water	54	0.69 (0.60, 0.80)	31	0.63 (0.53, 0.74)	23	0.82 (0.68, 1.00)
Sanitation	13	0.97 (0.82, 1.14)	7	0.97 (0.91, 1.03)	6	0.97 (0.61, 1.52)
Handwashing	25	0.71 (0.62, 0.82)	13	0.66 (0.54, 0.80)	12	0.77 (0.62, 0.95)

Note: The results are presented overall and stratified by intervention type (water, sanitation, handwashing) and season (dry, rainy). Each entry includes the number of seasonally stratified estimates, the effect size (RR), and the 95% CIs. Note: CI, confidence interval; RR, risk ratio; WASH, water, sanitation, and handwashing.

Figure 3.

Forest plot illustrating the effect sizes of WASH interventions on diarrheal morbidity during dry seasons. The analysis included 51 stratified effect estimates derived from studies conducted between 1 January 1970, and 9 October 2023. The plot shows RRs with 95% CIs, indicating a 33% reduction in risk ($\text{RR}=0.67$; 95% CI: 0.59, 0.76). The meta-analysis model used random effects to account for heterogeneity among studies. Note: CI, confidence interval; RR, risk ratio; SE, standard error; WASH, water, sanitation, and handwashing.

Figure 4.

Forest plot illustrating the effect sizes of WASH interventions on diarrheal morbidity during rainy seasons. This analysis included 41 stratified effect estimates derived from studies conducted between 1 January 1970, and 9 October 2023. The plot presents RRs with 95% CIs, indicating an 18% reduction in risk ($\text{RR}=0.82$; 95% CI: 0.71, 0.95). The meta-analysis model used random effects to account for heterogeneity among studies. Note: CI, confidence interval; RR, risk ratio; SE, standard error; WASH, water, sanitation, and handwashing.

Overall, WASH interventions included in this analysis ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=92$; Figure 1, Box 1) reduced diarrhea risk by 27% (pooled $\text{RR}=0.73$; 95% CI: 0.66, 0.81) (Table 1). This should not be considered a new estimate of the overall impact relative to Wolf et al., because this only represented a subset of eligible studies where seasonally stratified data were available.

The effect of each type of WASH intervention (i.e., water, sanitation, or handwashing) on diarrhea morbidity stratified by dry vs. rainy season.

In our subgroup meta-analyses stratified by season, water interventions reduced the risk of diarrhea by 37% ($\text{RR}=0.63$; 95% CI: 0.53, 0.74) in dry seasons ($n\text{\hspace{0.17em}stratified\hspace{0.17em}}\text{estimates}=31$; Figure 1, Box 2) and by 18% ($\text{RR}=0.82$; 95% CI: 0.68, 1.00) in rainy seasons ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=23$; Figure 1, Box 3) (Table 1). Similarly, handwashing interventions reduced risk by 34% ($\text{RR}=0.66$; 95% CI: 0.54, 0.80) in dry seasons ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=13$; Figure 1, Box 2) and by 23% (pooled $\text{RR}=0.77$; 95% CI: 0.62, 0.95) in rainy seasons ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=12$; Figure 1, Box 3) (Table 1). Sanitation interventions showed no difference by season, with a 3% decreased risk ($\text{RR}=0.97$; 95% CI: 0.91, 1.03) in dry seasons ($n\text{\hspace{0.17em}stratified}\text{\hspace{0.17em}estimates}=7$; Figure 1, Box 2) and a 3% decreased risk ($\text{RR}=0.97$; 95% CI: 0.61, 1.52) in rainy seasons ($n\text{\hspace{0.17em}stratified}\text{\hspace{0.17em}estimates}=12$; Figure 1, Box 3) (Table 1).

Without seasonal stratification, water interventions ($n\text{\hspace{0.17em}stratified}\text{\hspace{0.17em}estimates}=54$; Figure 1, Box 1) on average reduced the risk of diarrhea by 31% ($\text{RR}=0.69$; 95% CI: 0.60, 0.80) and handwashing interventions ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=25$; Figure 1, Box 1) reduced the risk of diarrhea by 29% ($\text{RR}=0.71$; 95% CI: 0.62, 0.82) (Table 1). Sanitation interventions ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=13$; Figure 1, Box 1) decreased the risk of diarrhea by 3% ($\text{RR}=0.97$; 95% CI: 0.82, 1.14) (Table 1).

The effect of WASH interventions that occurred in a single season on diarrheal morbidity for the dry and rainy season.

Among studies that only reported data from a single season, average results were similar; those that occurred in the dry season ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=15$; Figure 1, Box 4) on average found that WASH interventions reduced diarrhea by 36% ($\text{RR}=0.64$; 95% CI: 0.51, 0.80), and those that occurred in the rainy season ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=5$; Figure 1, Box 5) on average found that WASH interventions reduced diarrhea by 39% (pooled $\text{RR}=0.61$; 95% CI: 0.35, 1.08).

Meta-Regression

Table 2 shows the results of the meta-regression model of the 90 seasonally stratified effect estimates (Figure 1, Box 1), with season as the main predictor (rainy vs. dry season as reference) and covariates for setting (urban vs. rural as reference) and type of WASH intervention (water, sanitation, or handwashing). Season was significant in the regression model, with an estimated RRR of 1.22 (95% CI: 1.05, 1.43), indicating that WASH interventions were 22% less effective in protecting against diarrhea in rainy seasons in comparison with dry seasons while holding other variables in the model constant. Study setting showed an estimated RRR of 0.82 (95% CI: 0.65, 1.03), indicating that WASH interventions were 18% more protective against diarrhea in urban settings ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=29$) in comparison with rural settings ($n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=63$). Water and handwashing interventions both showed significant protective effects against diarrhea in the model, with estimated RRRs of 0.67 (95% CI: 0.58, 0.78) and 0.73 (95% CI: 0.58, 0.91), respectively.

Table 2.

Multiple meta-regression assessing the impact of WASH interventions on diarrheal morbidity.

Covariate	RRR	95% CI	$p$-Value
Season (rainy vs. $\text{Ref}=\text{dry}$)	1.22	1.05, 1.43	0.01
Setting (urban vs. $\text{Ref}=\text{rural}$)	0.82	0.65, 1.03	0.09
Water intervention	0.67	0.58, 0.78	$<.0001$
Sanitation intervention	0.86	0.67, 1.10	0.23
Handwashing intervention	0.73	0.58, 0.91	$<.01$

Note: This table presents the RRRs derived from a multiple meta-regression model with season as the main predictor, while controlling for setting (urban vs. rural) and type of WASH intervention. CI, confidence interval; Ref, reference; RRR, relative risk ratio; WASH, water, sanitation, handwashing.

We evaluated the quality of included studies using a robust quality assessment framework. Each study was assessed for various risk of bias domains using a modified Newcastle-Ottawa Scale, tailored to our specific review criteria.³⁹ Of the 56 effect estimates in our assessment, 46 (82%) were within a range of 3–5, revealing some heterogeneity across studies included in our review. Sixty-one percent of studies achieved a score of 4 or higher, and only 36% of studies scored 5 or above, suggesting a moderate level of confidence in their results. We found that randomized trials exhibited strong treatment of selection bias, with some possibility among quasi-experimental trials. Most included studies showed strong treatment of misclassification bias and nearly all appropriately treated clustering (bias in analysis), all of which bodes well for internal validity. Alternatively, there was some risk of bias associated with lack of blinding among respondents (response bias) and blinding associated with outcome measurement. All studies were at risk of bias related to the outcome assessment, given that diarrhea (the main outcome) is measured by caregiver-reported symptoms. There was some heterogeneity associated with follow-up bias, with about half of studies reporting $>10\%$ loss to follow-up. A detailed summary of the quality assessment score for each study can be found in Excel Tables S2, S3, and S4.

Sensitivity Analyses

Sensitivity analyses showed similar results as the primary analysis: higher effectiveness of WASH interventions in dry seasons in comparison with rainy seasons. We narrowed our analyses to a) only those studies that occurred across multiple seasons, b) only those studies that occurred in tropical climate zones, and c) only those studies occurring in low- to middle-income countries. When examining only the multiseason studies ($n=30$; $n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=72$), we found a pooled dry season RR of 0.68 (95% CI: 0.58, 0.81) and a pooled rainy season RR of 0.88 (95% CI: 0.75, 1.01), a $20\text{\hspace{0.17em}percentage}$ point difference, in comparison with a $15\text{\hspace{0.17em}percentage}$ point difference when including both multiseason and single-season studies. When we excluded comparisons from studies not located in tropical climate zones (i.e., excluding subtropical and temperature climate zones), we found a pooled dry season RR of 0.65 (95% CI: 0.55, 0.78, $n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=33$) in comparison with 0.67 (95% CI: 0.59, 0.76; $n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=51$) in the primary analysis. The pooled rainy season RR with these studies removed was 0.81 (95% CI: 0.66, 0.99; $n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=28$) in comparison with 0.82 (95% CI: 0.71, 0.95; $n\text{\hspace{0.17em}stratified\hspace{0.17em}estimates}=41$) in the primary analysis. When we excluded comparisons from high-income settings within the handwashing interventions ($n=4$), we found a pooled dry season RR of 0.63 (95% CI: 0.50, 0.78) in comparison with 0.66 (95% CI: 0.54, 0.80) in the primary analysis. The pooled rainy season RR with these studies removed was 0.72 (95% CI: 0.56, 0.92) in comparison with 0.77 (95% CI: 0.62, 0.96) in the primary analysis.

A funnel plot including the 50 studies showed no evidence of publication bias (Figure S1).

Discussion

This study examined the extent to which the effects of WASH interventions on diarrheal diseases differed by season, and the results demonstrate that WASH interventions on average were $15\text{\hspace{0.17em}percentage}$ points more protective against risk of diarrhea in dry seasons than in rainy seasons ($20\text{\hspace{0.17em}percentage}$ points when considering only multiseason studies). Larger impacts on diarrhea risk in dry seasons were found among subanalyses of water and handwashing interventions but not sanitation interventions. The greater impact of WASH interventions on diarrhea risk in the dry season has wide-ranging implications, including how we interpret the overall effects of WASH interventions and how interventions should be designed to ensure they are effective across all meteorological conditions and support collecting and reporting seasonally stratified data.

Potential Explanations for Stronger Protective Effects in Dry Seasons

Evidence suggests that diarrheal diseases peak in tropical climate rainy seasons.^18–22 We, therefore, would have expected a higher protective effect from WASH interventions in rainy seasons if they interrupt seasonally higher transmission of diarrhea-causing pathogens. However, we found a stronger protective effect in dry seasons. Two explanations could account for this main finding.

First, lower effectiveness of WASH interventions in the rainy season could result from WASH interventions and infrastructure that are not adequately designed to withstand the increased precipitation, temperature, and other associated challenges during these periods. Differential seasonal or meteorological conditions can overwhelm WASH systems and hardware, attenuating their effectiveness in interrupting enteric pathogen transmission. Heavy rainfall and flooding, typically associated with rainy seasons, can damage or destroy water infrastructure, leading to environmental intrusion and contamination of drinking water.⁸ These events can also overwhelm both water and sanitation systems, causing enteric pathogens to move into drinking water sources and degrade water treatment systems, run directly into groundwater, or directly into human contact.⁸ Even high-income settings with advanced water and sanitation systems are at increased risk of infrastructure disruption or failure due to the varying climatic conditions associated with climate change.^41,42 Thinking more broadly, there are also distal, community-level factors that could influence the effectiveness of WASH interventions. For example, community management of water systems often faces increased challenges in maintaining infrastructure integrity and operational efficiency during periods of heavy rainfall.⁴³ This seasonal dynamic can lead communities to revert to using less-secure water sources, thereby undermining the effectiveness of interventions and highlighting the importance of integrating management strategies into WASH programs to ensure effectiveness across different meteorological stressors.⁴³

Second, seasonally differential prevalence and/or differences in dominant diarrhea-causing pathogens and seasonally differential infection pathways could explain the observed results. Dominant diarrhea-causing pathogens differ by season or time of year across a variety of settings.^{5,8,18,20,44,45} Diarrhea peaks in dry, cooler seasons are largely caused by viral pathogens, specifically rotavirus, one of the largest drivers of viral diarrhea.^20,46–48 The pathways to infection for these viral peaks are frequently driven by human behavior, with individuals moving indoors and in closer proximity, facilitating the person-to-person/airborne transmission that these viruses need to survive.^49,50 As a result, viral diarrhea most commonly occurs in outbreaks within close communities like daycare centers, homes, and hospitals.⁴⁹ These outbreaks can normally be controlled by proper handwashing, cleaning surfaces, and, in some cases, vaccines. This observation may underlie the relatively greater effectiveness of handwashing interventions during the rainy season observed in this study, in comparison with water and sanitation interventions. The global introduction of the rotavirus vaccine specifically has significantly decreased worldwide rotavirus-associated diarrhea morbidity and mortality over the last decade.^51,52 Alternatively, diarrheal peaks in rainy, hotter seasons are commonly caused by bacterial pathogens like Shigella, Salmonella, and Campylobacter.^{5,18,20,44,45} Under favorable conditions (bacteria can replicate at higher temperatures with proper nutrient levels and can even express more virulence genes at elevated temperatures), bacteria can persist in environmental media (e.g., soil and water) for long periods, especially in warm weather.⁸ This environmental persistence lends itself to numerous, primarily environmentally mediated, pathways of infection for these pathogens (Figure S2). WASH interventions typically target one or more of these pathways but may not target all relevant pathways or have sufficient coverage to control community disease transmission.^17,53,54 Most of these pathways are additionally catalyzed by the warmer and wetter weather conditions of rainy seasons. For example, heavy rains mobilize pathogens into groundwater or drinking water or bring pathogens directly into human contact via flooding.⁸ This combination of higher infection pressure and more pathways of infection during the rainy seasons could overwhelm WASH interventions and attenuate their effectiveness in preventing transmission of diarrhea-causing pathogens.

In subgroup analyses, sanitation interventions did not demonstrate an effect on diarrhea, displaying null effects in both dry and rainy seasons. This observation contrasts with the protective effects identified in water and handwashing interventions, which were more pronounced in dry seasons in comparison with rainy seasons (in line with the trend observed with all WASH interventions combined). A critical factor in interpreting these findings is the relatively low number of sanitation studies available for these subgroup analyses. For instance, although Wolf et al. reported a significant reduction in diarrhea from improved sanitation, drawing on 20 effect estimates for their analyses, our review included only 8 effect estimates providing seasonally stratified data for sanitation. This limited sample size considerably reduces the statistical power to detect potential effects and may not adequately capture the true efficacy of sanitation interventions, particularly in the context of seasonal variations. Furthermore, practical aspects of sanitation interventions contribute to their distinct outcomes. Sanitation interventions often involve substantial infrastructural and behavioral changes that are less susceptible to short-term seasonal shifts compared with interventions that directly modify water quality or hygiene behaviors.^55–57 A recent systematic review evaluating RCTs of sanitation interventions alone (not in combination with other WASH interventions) also found little evidence of a significant impact on diarrhea, supporting the notion that the effects of sanitation might not be as immediately observable as those of other WASH interventions.⁵⁸

Implications of the Differential Impacts of WASH Interventions in Dry vs. Rainy Seasons

Wolf et al. found that, in comparison with untreated water from an unimproved source, the risk of diarrhea was reduced by up to 50% with water treated at POU, and the promotion of handwashing with soap reduced diarrhea risk by 30%.¹² The findings from the present study suggest that these estimates of intervention effectiveness could be influenced by seasonality. A large percentage of studies included in our review (40%) collected data entirely in one season, with the majority of those being in the dry season. This selection may be due to convenience and field logistics, because WASH intervention studies are conducted in areas that are often impassable in the rainy season. On the one hand, this imbalance in timing of data collection, leading to a predominance of data from the dry season, may have resulted in an overestimate of intervention effectiveness by Wolf et al., given the findings that show a higher protective effect of WASH interventions in the dry season. Specifically, the effect sizes for water interventions were 6% higher in the dry season in comparison with overall across both seasons and 5% higher in the dry season in comparison with overall across both seasons for handwashing interventions. The majority of effect estimates (∼78%) from both single-season and multi-season studies demonstrated protective effects during the dry season. On the other hand, the imbalance may have inadequately characterized the potential impact of interventions under dryer, cooler conditions when systems are not overwhelmed by rainfall and high rates of enteropathogens in circulation. Thus, the predominance of data collected in the dry season may overestimate overall effects and underestimate the potential effects of WASH interventions in controlling transmission under dryer conditions.

The findings from this study also suggest that current WASH interventions are not resistant to meteorological stressors, and this vulnerability may increase in the future under increased stressors of climate change. For example, sanitation systems that are able to better sequester human waste during flooding events or reticulated water networks that better prevent groundwater intrusion with standing water and high water tables may be needed to handle increasingly strong storms and more frequent flooding. In addition, our findings point to the need to measure diarrhea across seasons, report seasonally stratified data, and extend data collection over longer time periods to encompass different seasons. For instance, a recent reanalysis of the WASH Benefits Bangladesh trial by Nguyen et al. demonstrated the influence of seasonality on WASH intervention effectiveness, highlighting how interventions were particularly effective during the monsoon season and following heavy rainfall events.⁵⁹ These findings underscore the necessity of integrating seasonal variability into the planning and evaluation of WASH programs to ensure their effectiveness under varying climatic conditions. The results also underscore the importance of considering multiple different study contexts, because the conclusions drawn from Nguyen et al.’s results from one study differ qualitatively from those of this meta-analysis.

In considering the variations in the effectiveness of WASH interventions by season, it is important to also consider the potential impact of compliance and behavioral changes. Compliance with WASH practices, such as the consistent use of safe water sources and adherence to handwashing protocols, may vary across different seasons due to a variety of factors, including availability of resources, cultural practices, and personal convenience. Although this study did not have the data necessary to directly analyze the impact of compliance on WASH effectiveness by season, we propose that future research should include detailed tracking of compliance metrics across seasons. This future work could greatly enhance our understanding of how behavioral factors contribute to the seasonal variations observed in the effectiveness of WASH interventions.

Meta-regression allowed for additional exploration of setting (urban vs. rural) on WASH intervention effectiveness. WASH interventions were 18% more protective against diarrhea in urban settings in comparison with interventions in rural settings. However, only approximately one-third of the estimates included in the review occurred in urban settings (as detailed in the meta-regression model results in the “Results” section). Settings were categorized as urban or rural based on original study classifications, including peri-urban and informal settlements as urban. This simplification may affect the applicability of results across varied socioeconomic contexts, highlighting the need for more nuanced geographical classifications in future research. With rapid urbanization occurring across the globe, and the majority of this urban growth occurring in LMICs,⁶⁰ more information and research is needed in urban settings to better understand links among WASH, diarrhea, and climate in these settings.

Limitations

This systematic review and meta-analysis had several limitations. First, nearly half of the studies included in the Wolf et al.¹² review did not contain health outcome data presented longitudinally or by time of year. Thus, we had to exclude these studies from our seasonal analysis, creating a potential bias based on the availability of studies that could be seasonally disaggregated. Our estimates of the overall effect should therefore be interpreted with this caveat in mind, and these results should not be considered as an overall estimate of the impact of WASH interventions on diarrhea, given that we excluded many studies from the more comprehensive Wolf et al. review. However, we believe that these studies are representative of the broader set of studies evaluating the impact of WASH on diarrhea. Although some bias may have been introduced in comparison with the full set included in the Wolf et al. review, if anything the subset included here likely comprises higher quality studies, given their more thorough reporting of data collection dates, which is aligned with trial reporting standards. The exclusion of such a significant number of studies due to the lack of detailed seasonal, location, or date information underscores the need for future research to incorporate these elements. This finding highlights a major gap in the current WASH literature and reinforces the importance of considering environmental factors such as weather and seasonality in the study design, data collection, and analysis of WASH interventions.

Second, we created seasonal classifications using land station weather data instead of satellite data because weather stations tend to give more accurate measures of ground conditions with less interpolation, especially with regard to precipitation measurements.^61,62 This approach led to the exclusion of eight studies from our analysis for which proximate and/or complete land station weather data were unavailable. However, precipitation ground measurements have been shown to outperform satellite surface wetness estimates, and precipitation is one of the main meteorological drivers of the outcome of interest in this review.⁶²

A third limitation was the categorization of estimates into a binary dry/rainy season for studies conducted in regions where distinct dry and rainy seasons may not exist. However, our sensitivity analyses on studies that occurred in tropical climates with more distinct dry and rainy seasons yielded the same trend of WASH interventions conferring a more protective effect against diarrhea in dry seasons, suggesting that our estimates were not biased by studies with less intense climatic/seasonal pressures. This binary classification might obscure finer meteorological nuances such as the bimodal rainy seasons in East Africa in comparison with the extended monsoon periods in South Asia. We acknowledge that although our approach maintains the feasibility of a global systematic review, it overlooks regional variations and a broader range of meteorological conditions like extreme weather events that could affect WASH interventions and outcomes. Future research should explore these unique climate profiles and extreme weather to provide a more nuanced understanding of how seasonal variations in climate affect WASH effectiveness.

Fourth, another limitation of our review was the variable quality of the included studies, which added complexity to interpreting their outcomes. Our assessment of bias—specifically associated with blinding of the intervention to both the recipient and the evaluator—is typical of field evaluations of environmental health interventions. Similarly, biases associated with caregiver-reported diarrhea are well documented, leading to many in the sector to call for more objective measures of effect for WASH interventions. All of these biases could lead to courtesy bias, biasing estimates away from the null. Biases associated with short follow-up were also noted, potentially biasing our results to the null, because WASH interventions may take longer to yield anticipated health gains. However, although each of these biases may impact internal validity of individual studies, this bias is not likely related to season. As such, our overall interpretation of the differences by season are unlikely to be impacted. This variation in study quality particularly affects the robustness of our meta-analytical conclusions. Therefore, although our findings indicate a significant impact of WASH interventions on health outcomes, these results should be interpreted with caution. The mixed quality of the underlying evidence necessitates a cautious approach, and future research should prioritize enhancing methodological rigor and transparency.

Conclusions

The major findings from this study provide a clearer understanding of the effectiveness of WASH interventions by season and highlight potential shortcomings that these interventions might have in the face of a changing climate. These results show the importance of better quantification of seasonal impacts of WASH interventions, to both inform policymakers on the true benefits of these interventions and to improve effectiveness across a variety of meteorological conditions.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.