https://orcid.org/0000-0002-8956-1458
Figures
Abstract
Microbial contamination of drinking water contributes to disease burdens that disproportionately impact infants and children and are largely preventable through suitable design, operation, monitoring, and management of improved water systems. The World Health Organization (WHO) has published guidance on water safety planning, water quality monitoring, and management approaches, including recommendations on sanitary inspection (SI) of water systems to detect and manage microbial hazards associated with fecal contamination. SI is a low-cost risk assessment tool for water systems based on observable risk factors (RFs) associated with potential water safety hazards. While SI has been previously studied, much of the literature has not quantitatively explored rainfall interactions with SI risk as drivers of fecal contamination. We merged remote-sensing rainfall estimates with SI and water quality data collected from 966 rural boreholes in Ethiopia, Ghana, and Burkina Faso. Logistic regressions (binary and ordinal) were used to characterize associations of total SI score, as well as individual risk factors (RFs), and classes of RFs (i.e., “Source,” “Transport,” and “Barrier” risks) with fecal indicator bacteria (FIB) occurrence, controlling for rainfall (over the past 1–15 days before sampling). We found associations (P < 0.05, OR: 3.5, 95% CI 1.05-11.66) between SI scores and E. coli risk categories controlling for fifteen-day total rainfall. Furthermore, interactions between rainfall and risk factors in the “barrier” category, and the “transport” category were associated with E. coli occurrence. Several individual RFs were also significantly associated with microbial contamination. Incorporating precipitation into models improved model fit characteristics (improved Pseudo R squared and AIC value); specifically, accounting for cumulative rainfall during the fifteen days before sampling improved model fit (increased pseudo-R2 from 0.035 to 0.05) for E. coli contamination. These findings can inform design, construction, maintenance, and monitoring of boreholes and prompt timely remediation of defects in such systems, potentially enhancing water safety.
Citation: Islam SA, Ambelu A, Seidu Z, Cronk RD, Bartram JK, Fisher MB (2025) Sanitary inspection characteristics, precipitation, and microbial water quality - A three-country study of rural boreholes in Sub-Saharan Africa. PLOS Water 4(5): e0000281. https://doi.org/10.1371/journal.pwat.0000281
Editor: Venkatramanan S, National College Autonomous, INDIA
Received: October 3, 2024; Accepted: April 3, 2025; Published: May 14, 2025
Copyright: © 2025 Islam et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: All data are included in Supplemental Information.
Funding: This work was supported in part by the Conrad N. Hilton Foundation, which supported initial field data collection (Project ID: “Monitoring, Evaluation and Learning for WASH”), as well as by the World Health Organization, which supported in part the secondary data analysis included in this study (Project ID: “Review of Sanitary Inspections for Drinking water”). We gratefully acknowledge in-kind support from students and staff at the Water Institute at UNC, which supported in part the secondary data analysis included in this study. The funders played no role in the analysis or writing of this manuscript nor in the decision to publish.
Competing interests: The authors have declared that no competing interests exist.
Introduction
It is estimated that more than two billion people worldwide do not have access to an improved water source (i.e., a source such as piped water or a borehole, which by nature of its construction offers some protection from fecal contamination [1]), while an additional one billion individuals worldwide use water from improved sources that, despite their improved design, are microbially contaminated [1–6]. According to WHO and UNICEF, boreholes are classified as improved water sources [7], and approximately 200 million people in SSA depend on boreholes for drinking water [8]. Rural drinking water systems generally have poorer water quality than urban ones [9], are often smaller in scale than urban systems, and are often community-managed in rural LMIC settings [10] (although multi-village supply schemes managed by rural utilities are increasingly prevalent in some settings such as rural Ethiopia).
Sanitary inspections (SIs) are useful for monitoring and managing water systems. Sanitary inspections are observational checklists that help managers assess the overall likelihood of contamination in a water system by identifying: 1) potential contamination sources, such as inadequate sanitation facilities (i.e., open defecation, absence of toilet facilities, presence of unimproved toilet facilities, presence of animals and their feces, etc), waste disposal practices; 2) deficiencies in the sanitary seals/barriers/protections designed to protect improved sources from contamination; and 3) transport mechanisms/vehicles such as stagnant/ponded water, leaking pipes, etc. [11]. They can be readily implemented using standardized forms developed and recommended by the WHO (World Health Organization) since the 1970s [12] and subsequently adapted by others. These checklists consist of a series of yes/no questions indicating the presence or absence of sanitary hazard indicators, often colloquially referred to as “risk factors” (RFs). These risk factors help identify vulnerabilities in the water source and guide interventions to improve water quality and safety. While SI cannot provide quantitative data on exposures or risks, they can be a valuable and cost-effective tool for prioritizing interventions and reducing the risk of waterborne disease [13].
It is understood that SI scores are preferably interpreted in combination with microbial data. It is reasonable to expect that SI scores and fecal indicator bacteria (FIB) measurements for water systems would be associated in many but not all instances, and published comparisons of sanitary survey data with microbial water quality measures largely bear this out: measures of association are reported in some but not all studies, and vary across contexts and settings [13–20]. A study on shallow tubewells in rural Bangladesh found no association between composite or overall sanitary risk score and water quality [15]. Conversely, a study of protected springs in Uganda and a study of wells in Tanzania found associations between sanitary risk score and microbial water quality [16,19]. If the lack of associations between these variables in some studies and the inconsistency of findings among different studies can be better explained, such findings may tend to strengthen evidence for the continued use of SIs as valuable water safety management tools, even though predicting FIB occurrence is not specifically what SIs are designed to do [14].
One factor with potential to explain the differences among studies is rainfall, as many previous studies did not systematically account for precipitation as a potential covariate or modifier of the relationship between sanitary inspection scores and microbial water quality. Previous studies in Bangladesh and Tanzania on rainfall (in the absence of SI data) found that FIB occurrence in water samples positively correlate with precipitation [21–23], and the effect is stronger when cumulative precipitation measures are used [24–26]. Research in Bangladesh found significant associations between E. coli in wells and heavy rainfall occurring within the 7, 15, and 30 days before sampling [23]. Likewise, a study in Tanzania indicated that heavy rainfall within 14 days of sample collection better predicted E. coli levels in wells [21]. Therefore, a study incorporating rainfall may be able to parse the relationship between SI scores and microbial indicators and better assess the importance of different RFs and RF categories.
This study used data from 966 rural boreholes in three African countries (Ghana, Ethiopia, and Burkina Faso) to explore the relationship between sanitary inspections and microbial water quality, accounting for rainfall. In addition, specific SI risk categories and factors strongly associated with microbial contamination were identified. In our analysis, rainfall events were included as either binary events (present or absent) or as quantitative measures integrated over periods of 1–15 days before sampling. This was done to determine the most effective precipitation lag times associated with microbial contamination. We hypothesized that integrating prior precipitation with SI data and FIB concentrations would improve microbial hazard assessment.
Methods
Study design
A retrospective analysis of secondary data was conducted using SI and microbial water quality data previously collected from 1251 boreholes in three African countries: Ethiopia, Ghana, and Burkina Faso (Fig 1). These data were accessed, anonymized, and merged between June 1 and December 31 of 2021. Water quality data were available for 966 boreholes (water quality samples and data could only be collected from systems that were functional on the day of the visit). Among these systems, most (n = 958) had manual pumps, and the rest (n = 8) were mechanized. International non-governmental organizations (NGOs) World Vision, WaterAid, Living Water, CARE, and Helvetas collaborated in collecting the data between 2014 and 2016. Enumerators conducted sanitary inspections in each country, collected and analyzed water quality samples, and recorded GPS coordinates, system details and images, and other relevant site and water system characteristics. Detailed study design and data collection procedures for each country are provided in Appendices 1 and 2.
Field data collection
Ethiopia.
NGO partners had worked in 222 kebeles (the smallest administrative units) in Ethiopia between 2011 and 2017. Of these, 88 kebeles were randomly selected (44 from the ‘experimental’ arm where water, sanitation, and hygiene [WaSH] program activities had been implemented, and 44 from the ‘comparison’ arm where WaSH programs had not yet been implemented by the NGO partners in question). Data collection occurred from May 18 to July 16, 2015, using Samsung DUOS smartphones equipped with the Akvo FLOW (Akvo Foundation, Amsterdam, The Netherlands) platform. This data collection period spanned the transition from dry to rainy seasons. Additional details on data collection are presented in File S1 Text.
Ghana.
Four districts were identified in the Northeastern Region of Ghana within which NGO partners had implemented WaSH programs in 296 communities. Two hundred and sixteen of these were randomly selected for this study. Data were collected onto Android-operated mobile phones between April and November 2014 using the mobile survey tool Akvo FLOW V 1.6 (Akvo Foundation, Amsterdam, The Netherlands). Details of the data collection have been previously reported (Fisher et al., 2020) and are summarized in Table A in S1 Text.
Burkina Faso.
In Burkina Faso, NGO partners had implemented a water program in 401 villages across six regions from 2003 to 2015. Data were collected from 95 randomly selected villages between September 15, 2015, and January 9, 2016. The data were recorded on the mWater mobile application (New York, USA) utilizing Motorola XT 1021 phones. Additional data collection details are provided in Table B in S1 Text.
Water quality sample collection and analysis
Water quality samples had been previously collected and analyzed according to the protocols detailed in Supplemental Information file S1. Samples were collected from rural boreholes using sterile 100-mL Whirl-Pak Thio bags (Nasco, Ft Atkinson, WI) labeled with unique barcodes. Samples were analyzed for E. coli concentration (most probable number (MPN) per 100 mL) using the Compartment Bag Test (CBT) method (Aquagenx LLC, Chapel Hill, NC, USA) according to the manufacturer’s specifications. Briefly, 100-mL samples were transferred to CBTs within 30 minutes of collection and stored at 4 C until enumerators returned from the field, then incubated at ambient temperature for 24 + /-2 hours for mean daily ambient temperature >30 C and 48 + /- 2 hours for mean daily ambient temp 25–30- C per the manufacturer’s guidelines, as previously described (Brown et al., 2011; E. Kelly et al., 2021; Stauber et al., 2014). Mean daily ambient temperatures were above 25 C in all study settings during data collection periods. E. coli count was recorded as E. coli MPN/100 mL.
For quality control, field blanks and duplicates were collected randomly and analyzed identically to other samples, comprising 5–10% of all samples. Briefly, a random subset of 5–10% of systems was selected to collect field blanks and/or duplicates. For blanks, a bottle of packaged drinking water from a reliable local brand (typically the most widely sold brand in the country) was opened near the water source or system being sampled, and a 100-mL ‘field blank’ sample was collected alongside the experimental field sample being collected at that system. For field duplicates, the field sample to be obtained was simply collected in duplicate. Field blanks and duplicates were labeled using unique barcodes and treated identically to experimental samples concerning collection, transport, and analysis. Because of barcodes, enumerators were blinded to the sample type (experimental, duplicate, or blank) being processed and analyzed after leaving the sample collection location.
Sanitary inspection
Sanitary inspection data were collected using an 18-question SI form (Table C in S1 Text) adapted from published examples (Bartram, 1996; Water & Team, 2006). Sanitary risk factors were divided into three categories, namely “Source,” “Transport,” and “Barrier” risks, as described previously (E. Kelly et al., 2021)(Fig 2). Risk factors in the “Source” category include potential origins or reservoirs of contamination. “Transport” risks denote potential pathways through which pathogens could contaminate water sources, while “Barrier” risk factors define engineered and/or natural barriers to the intrusion of microbial contamination.
(*fencing and presence of animal feces sit in different risk categories but relate to a shared hazard).
The sanitary inspection scores were calculated for each borehole observation and disaggregated (source, transport, and barrier), as described in Table B in S1 Text. Each risk factor was equally weighed and scored items so that ‘0’ denoted that the state of the given item was expected to correspond to lower vulnerability to microbial contamination, while ‘1’ indicated that the state was likely to correspond to greater vulnerability. Table S1 presented disaggregated risk factors and indicated each state’s scoring convention (0 or 1). To calculate the risk score for a given borehole, the states of all risk factors (0 or 1) were summed and divided by the total number of applicable factors (i.e., those factors pertaining to the system type in question) to obtain a composite score between 0 and 1, with 1 representing the greatest vulnerability to contamination and 0 the lowest vulnerability (Appendices 3 and 4).
Imputed precipitation and water table depth estimates
Rainfall.
Satellite rainfall estimates with 0.05-degree resolution were obtained from the University of California, Santa Barbara’s Climate Hazards Center InfraRed Precipitation with Station (CHIRPS) website Funk et al., 2014). For each included borehole, the closest rainfall gauge station was located, and rainfall in mm was summed over one-, two-, three-, seven-, 10-, and 15-day periods prior to the date of the site visit (sanitary inspection and water quality sample collection).
Water table depth.
Annual groundwater depth data for boreholes were estimated from a continuous global map of groundwater depths. This map was generated by Fan et al. based on a groundwater model incorporating more than 100,000 direct well measurements from government archives and published articles [27] and [28]. The groundwater depth corresponding to the closest point from the Fan et al. dataset was used for each borehole site.
Statistical analysis
Summary statistics were calculated for all water systems, by country and overall. Associations between SI scores, FIB occurrence, and rainfall were tested using one-way analysis of variance (ANOVA) and logistic regression. Briefly, ANOVA was used to compare microbial data across the three countries. Logistic regression models explored associations between sanitary inspection scores and microbial data while controlling for precipitation and covariates (country, groundwater depth, etc.).
E. coli MPN and log10 MPN values were tabulated, and binary (presence/absence) and categorical (microbial risk category) variables were constructed from numeric MPN data for statistical analysis. Briefly: a binary variable was created and coded based on “presence” (≥ 1 MPN/100 mL) or “absence (< 1 MPN/100 mL) of E. coli. A categorical variable was created and coded according to WHO microbial risk categories (World Health Organization, 2011): conformity (<1 MPN/ 100 mL), low-risk (1<= MPN/100 mL < 10), intermediate-risk (10<=MPN/100 mL < 100), or high-risk (MPN/100 mL>=100). E. coli presence/absence was then modeled using binary logistic regression, while the WHO health risk category (conformity/low/medium/high risk) was modeled using ordered logistic regression. In both types of regression models, independent variables included sanitary risk scores (overall, disaggregated by risk category, or disaggregated by individual risk factor), precipitation (binary or cumulative [mm] over one, two, three, seven, or 15-day periods), and country (Fig 3). Estimated water table depth (meters) was initially included as a covariate, but dropped after it was found to have negligible effect in all models. Regression models were calculated both with and without the inclusion of plausible interaction terms (i.e., interactions among precipitation and SI risk, as well as among SI risk categories). For models including multiple interactions (e.g., source risk x transport risk x barrier risk x rainfall, etc.), only significant interactions were reported in tabular results to enhance readability of results tables.
The independent variables remained the same for ordered logistic regression, but the dependent variables would be E. coli as WHO risk category (conformity/low/intermediate/high risk).
Various combinations of sanitary risk and precipitation were assessed in our models to determine the most effective model estimates. Pseudo R-squared values were used to evaluate model fit for binary and ordered logistic regressions [29]; the Akaike Information Criterion (AIC) value was also used to assess fit for ordered logistic regression [30].
Software
The Python programming language (version 3.11.4, Python Software Foundation, 2022. https://www.python.org/) was used to run the regression models (binary and ordered logistic), implementing a package called STATSMODELS of version 0.13.1 (statsmodels.org). STATA (version 15.1. StataCorp, College Station, TX) was also used to run selected models. ArcGIS Pro from Esri (Sources: Esri, DeLorme, HERE, TomTom, Intermap, increment P Corp., GEBCO, USGS, FAO, NPS, NRCAN, GeoBase, IGN, Kadaster NL, Ordnance Survey, Esri Japan, METI, Esri China (Hong Kong), Swiss topo, MapmyIndia, and the GIS User Community) was used to extract and merge rainfall data with other variables and to produce maps of study data. Base layers and boundary data for GIS maps were obtained from publicly available data sources: (sub)district boundaries were obtained from GADM (https://gadm.org/license.html); country boundaries and populated places points were obtained from Natural Earth (https://www.naturalearthdata.com/about/terms-of-use/).
Ethics statement
No human subjects’ data or identifiable information were included in this retrospective analysis of environmental monitoring data from communal water systems. Secondary data, including water system details provided by water system committee members in their official capacity, were used. These were determined by the supervising IRB not to comprise human subjects’ data, since they were provided in and pertained to individuals acting in an official capacity. Because no human subjects or identifiable data were included, no such subjects exist to give consent; as a result, informed consent was not obtained for this secondary analysis of environmental monitoring data. Ethical review for this retrospective analysis was undertaken by the University of North Carolina Institutional Review Board (IRB study number 21–0592).
The current study uses environmental monitoring data from prior fieldwork in Burkina Faso, Ghana, and Ethiopia, which were accessed, anonymized, and merged between June 1 and December 31 of 2021 in accordance with IRB study number 21–0592. Ethical approval for the completed prior fieldwork was granted by the University of North Carolina Institutional Review Board (IRB study numbers: 14–0386; 15–1021; 15–1607). Approval for that completed prior fieldwork was also obtained from the applicable Health ministry in each country. The completed prior field work included the collection of both environmental monitoring data (on which the current retrospective study is based) as well as other monitoring and evaluation data including questionnaires with human subjects (which are not included in any aspect of the current retrospective environmental study). For information, we note that informed consent was obtained and documented from human subjects in prior completed fieldwork per the IRB protocols for the studies. However, none of the human subjects, human subjects’ data, or identifiable information are included in the present analysis of environmental monitoring data. For clarity, the present analysis uses environmental data only and therefore does not have any human subjects.
Results
E. coli occurrence
Sanitary inspection scores and water quality data for concurrent grab samples were available for 966 boreholes. Most borehole water samples, totaling 665 (69%), showed no detectable E. coli contamination in a 100 mL sample (Fig 4, Table 1). The highest proportion of borehole samples with detectable contamination was in Ethiopia and the lowest in Burkina Faso (Fig 4), and differences among countries were significant by one-way ANOVA (p < 0.05). The average and geometric mean of E. coli MPN concentrations were 31 MPN/100 mL (standard deviation [SD]= 31) and 15 MPN/100 mL (SD = 31), respectively. Overall, the mean annual water table depth was 20 m, the mean 15-day cumulative precipitation (CP-15) was 53 mm, and an average of 36% of systems had reportedly failed in the past year (Table 1, Table C in S1 Text).
Determinants of microbial contamination
Significant associations (P < 0.05) were observed between rainfall events and E. coli contamination using both binary and ordinal logistic regressions; effect sizes were greater, and model fits were better for longer vs. shorter rainfall integration periods (Table D in S1 Text). Cumulative precipitation using a fifteen-day integration period (CP-15) was more strongly associated with microbial contamination than binary precipitation variables or cumulative precipitation variables using shorter integration periods, and regression results presented in the main findings were therefore based on CP-15 (Tables E-H in S1 Text). No significant association was found between microbial contamination and water table depth. Consequently, this variable was not reported in the main results. Binary regression models achieved slightly better fit (greater R2) values than ordinal logistic regression models, but neither model explained a large proportion of the variability in the dependent variable (Table D in S1 Text).
Aggregated sanitary risk.
For both binary and ordered logistic regression models, a significant association (p < 0.05) between overall sanitary risk score and microbial contamination was found when 15 days’ cumulative precipitation was included in a simple model controlling for the country (Table 2). Odds ratios for overall sanitary risk score were 4.0 and 4.1for binary and ordinal logistic regression, respectively, indicating that systems with the highest vs. lowest possible risk scores were, in theory, 4 times more likely to have any detectable E. coli in 100 mL (binary model) or to have E. coli MPN values corresponding to a higher vs lower risk category (ordinal model) in a one-off grab sample. Effect sizes were even larger when interactions were included (Table 2, Models 3 & 4). Associations between overall SI score and E. coli occurrence measures were positive and significant across the various cumulative precipitation integration periods (one day to fifteen days) studied (Table G in S1 Text), with greatest effect sizes for 15-day cumulative precipitation (CP-15). CP-15 was also significantly associated with E. coli occurrence measures across binary and ordinal logistic regression models (as were CP measures using shorter integration periods). However, the effect size was modest (OR: 1.007 to 1.018 across models 1–4), implying that a system receiving the mean CP-15 of 53 mm over 15 days would have roughly 37%-95% greater odds of E. coli occurrence than a system receiving no precipitation over the same interval.
Disaggregated sanitary risk.
When sanitary risk was broken down into source, transport, and barrier components, the barrier risk component was the only one significantly associated with E. coli occurrence in ordered logistic regression models (OR: 2.55, 95% CI 1.14-5.69, Table 3). This result was consistent across all precipitation integration periods (Table H in S1 Text). For binary logistic regression, barrier failure was associated with decreased E. coli occurrence (OR: 0.42, CI -0.18-0.98). When interaction terms were included, only the interaction of transport risk x barrier risk x rainfall was significant in each model (Table 3, Models 7 & 8).
Individual risk factors.
In binary and ordinal logistic regressions, several individual risk factors were significantly associated with E. coli occurrence. Four were significant in both: drainage channel filled with water; fencing inadequate to keep animals out; visible cracks in concrete pad/floor; and absence of concrete walls extending below the ground surface (Table 4). Detailed results can be found in Table I in S1 Text. The direction of the effects for each of these risk factors was similar for binary and ordinal logistic regression. Odds ratios (ORs) were generally >1 for risk factors structured such that an affirmative response corresponds to greater risk (e.g., “Are there visible cracks on the cement floor around the water point?”), while ORs were <1 for risk factors structured so that a negative response indicates greater risk. At least one risk factor in each category (transport, source, and barrier risk) was significant in each model.
Discussion
This cross-sectional study investigated the relationship between microbial water quality data and sanitary inspections for 966 boreholes across three SSA countries. Results indicated that sanitary inspection risk score and barrier risk score were associated with E. coli occurrence after controlling for rainfall. Occurrence of E. coli was associated with rainfall, and the association was stronger for longer rainfall integration periods. Furthermore, several individual SI risk factors were also associated with E. coli occurrence. When models were updated to include interaction terms, E. coli occurrence was also found to be associated with the interaction of transport risk score, barrier risk score, and integrated rainfall. Ordinal logistic regression models explained slightly more of the variance in measured E. coli occurrence in drinking water than did binary logistic regression models. However, no models in this study explained the majority of the variance.
The observed association between sanitary inspection and microbial contamination was consistent with previous studies from Uganda [16] and Tanzania [19]. However, these previous studies did not account for precipitation. When precipitation was included in our regressions, model fits improved (AIC values decreased, Table D in S1 Text). Ordinal logistic regression models indicated significant associations between overall sanitary inspection score and WHO E. coli risk categories for all precipitation integration periods. Given the importance of 1) controlling for rainfall and 2) controlling for interactions between rainfall, barrier risk, and transport risk, a strong case can be made that sanitary inspection barrier and transport risk scores provide useful and interacting indicators of water system vulnerability to contamination during precipitation events. In other words, these risk categories correspond to measures of vulnerability, not to measures of contamination. We can further speculate that some prior studies reporting no associations between sanitary inspection scores and E. coli occurrence might have obtained a different result if they had included rainfall and/or its interactions with sanitary risk factors in such analyses.
Barrier failure demonstrated consistent and significant associations (OR > 2.4) with microbial contamination. Similar findings were observed by Kelly et al. [17], who found that barrier failure best predicted microbial contamination. Previous studies also observed an association between compromised structural integrity of wellheads (absence or cracked platform, apron failure) and water quality [13,15,20]. To the extent that these observed relationships correspond to broader microbial risks, identifying and proactively addressing engineering, design, and maintenance deficiencies such as absent or failed barriers (e.g., absent or failed sanitary seals, concrete slabs, aprons, and walls extending below the surface of the ground, etc.) and key transport risk factors (such as inadequate drainage around water points) could reduce borehole vulnerability to microbial contamination during wet weather and potentially improve user outcomes.
Univariable regression results suggest that ensuring the presence of fencing could also reduce risk, in agreement with prior studies [16,20]. Other significant factors in our univariable regressions included stagnant water in drainage channels (transport risk), cracks in concrete pads/floors (barrier risk), and lack of adequate walls preventing undercutting or intrusion of water below concrete pads/floors (barrier risk). It is plausible that addressing one or more of these factors as part of monitoring and rehabilitation efforts may be an efficient option for reducing vulnerability to microbial contamination; further intervention studies may be useful to test this hypothesis.
Our study revealed that combining SI score with prior precipitation data better predicts microbial water quality than either metric alone, consistent with the findings of a previous study [17] and consistent with work indicating that heavy rainfall after a prolonged drought can cause “first flush” contaminant spikes in rural water sources if the wellhead is not adequately protected [24]. Future work may explore relevant thresholds for classifying precipitation as sufficiently heavy to cause such events. Further work is also needed to determine whether these findings generalize to mechanized boreholes and other water system types, in addition to the handpumps comprising the majority of systems in the current study.
Limitations
This study was built on cross-sectional data, and therefore did not capture trends or variations in E. coli occurrence over time, which are important in water safety assessment [31,32]. Rainfall data were estimated based on remote sensing data (CHIRPS) with limited resolution. CHIRPS covers the globe and is freely accessible, making it suitable for studies requiring remote rainfall data. However, the average distance from a water sample to the closest rain gauge station in this study was 20 km. This imprecision in the rainfall data limits the ability of the current study to characterize prior rainfall. Another notable limitation was the use of a sanitary risk scoring algorithm that applies equal weights to all risk factors: while this approach replicated standard practice in the field, future studies may explore other weighting approaches to better reflect the relative importance of risk factors.
Another limitation was the limited inclusion of data on local topography and on-site hygiene practices (limited in our study to noting whether nearby latrines are uphill), which were also essential factors when considering the influence of precipitation [26]. Finally, the current study did not control for borehole age, a variable reported to be associated with water quality in prior studies [15]. Water infrastructure deteriorates as it ages and may impact below-ground barrier factors in ways not easily captured by a visual sanitary inspection of the above-ground system. While primary data were not collected with a sole focus on describing and refining sanitary inspection studies, they provide considerable valuable evidence for that purpose. Further work that addresses the above limitations could meaningfully build upon the results of this study.
External validity of this work may be limited by the predominance of rural boreholes over other sources and settings in the study, and by the potential context-specificity of the relative strength of association among FIB occurrence, rainfall, and specific individual SI scores. Results may not generalize well to other geographies, other water supply technologies, and/or to urban/peri-urban systems. Replication of this work in other system types, geographical settings, and contexts may help clarify the generalizability of findings from this approach.
Implications
The finding that sanitary risk score in general, and the Barrier and Transport risk scores in particular, were significantly associated with borehole vulnerability to microbial contamination during/following rainfall events in the study settings has implications for small water system design, construction, monitoring, and management, as well as potential implications for policy and practice. We conclude that sanitary inspections remain highly relevant assessments of the safety/vulnerability of rural boreholes and should continue to be included in monitoring programs and water safety plans covering such systems. Incorporating the results of SI into the risk management process of water safety plans (WSP) would help water utilities and regulatory authorities identify areas for improvement and potential mitigation measures to enhance drinking water supply safety. The finding that barrier risk factors such as the presence and adequacy of concrete pads and the walls around these pads (which should extend below the ground to ensure a sanitary seal), as well as transport factors such as adequate drainage, are strongly associated with safely managed water also suggests that these elements should be included in the specifications, design, and construction of new rural boreholes. Likewise, insistence on adequate fencing is likely still warranted. SI should receive close attention during monitoring, maintaining, and rehabilitating existing systems. Where defects in SI risk factors are detected, prompt repair of these items should help improve microbial water safety, especially during and after rainfall events, when sampling is less common, and the likelihood of microbial contamination is greater. Based on the results of this study, such an approach leverages the fundamental strength of SI: its use of persistently observable risk factors to estimate vulnerability of small and rural water systems to transient contamination during/following rainfall events, when such contamination is least likely to be observed due to difficulties in sampling small/rural systems during rainy seasons.
While the finding that “Barrier” and “Transport” risks (and their interactions with rainfall and each other) were strongly associated with water system vulnerability in the study contexts are worth acting upon, this finding does not necessarily mean that “Source” risks (which were not associated with E. coli occurrence in our models) should be seriously deprioritized per se. It may be the case that the source risk category is less important to the safe management of rural boreholes in the study contexts than implied by previously proposed mechanistic models (E. Kelly et al., 2021). Additionally, it may be the case that the relative importance of various individual SI scores and categories is highly context-specific and can be expected to vary across geographies and settings. However, it is also possible (and perhaps most likely) that the risks associated with the “Source” category are more difficult to adequately capture using current SI observation protocols and risk scores, as compared to those making up the “Barrier” and/or “Transport” categories. This hypothesis seems credible since almost anyone can readily observe the presence/absence of an absent/cracked concrete wall or pad, and these items are, in fact, difficult for a trained observer to miss; by contrast, the presence of human or animal feces, deposited days ago, several meters away under vegetation may easily be missed in a sanitary inspection, particularly if rainfall events may have further reduced their recognizability.
Finally, the findings that including rainfall in regression models improved their performance confirms what WHO and others have long specified: that sanitary inspections were intended to detect and prioritize the VULNERABILITY of water systems to transient fecal contamination during/after weather events and NOT to PREDICT constant year-round fecal contamination or REPLACE direct collection and testing of drinking water for fecal indicator bacteria occurrence. Ongoing monitoring, management, and planning of rural boreholes should leverage sanitary inspections and microbial water quality testing wherever feasible and appropriate, especially during wet weather events. Future work may also evaluate alternative materials and construction methods to reduce the failure of critical contamination barriers identified in this work.
Conclusions
Sanitary inspection, rainfall monitoring/estimation, and water quality analysis are valuable tools for assessing water source vulnerability and safety, and their utility is increased when they are combined. Specifically, this work suggests that while all aspects of safe water management should continue to be prioritized, ensuring the presence, adequacy, and maintenance of essential barrier factors such as adequate and undamaged impervious pads and walls (extending below the surface of the ground to prevent undercutting by surface runoff), as well as factors such as adequately functioning drainage channels and adequate fencing, may be particularly effective in the study setting. Together, these tools can provide better evidence for advancing progress on safely managed water than either tool alone.
While this study focused on boreholes with handpumps, the future importance of boreholes includes both manual and mechanized systems. It is likely that many of the results from this work may prove relevant to both. Simple, low-cost approaches such as sanitary inspection can help identify and prevent vulnerabilities in water systems, as well as help identify and prioritize water system defects for remediation. Many of these defects may be easily identified and corrected during routine on-site monitoring visits, providing low cost options for improving water safety
Supporting information
S1 Text. Study design and data collection in Ethiopia, Ghana, and Burkina Faso; Sanitary inspection questions, risk factors, risk score calculations, and selected descriptive data, regression models, and results.
Table A in S1 Text. Table of adapted sanitary inspection questions with subcategorized risk factors. Table B in S1 Text. Sanitary risk score calculation. Table C in S1 Text. Table of descriptive data of cumulative rainfall for 15 days lag time and annual water table depth. Table D in S1 Text. Table of regression model fit parameters with and without precipitation. Table E in S1 Text. Table of regression results for aggregated sanitary risk with different binary precipitation(presence/absence) lag time. Table F in S1 Text. Table of regression results for dis-aggregated sanitary risk with different binary precipitation (presence/absence) lag time. Table G in S1 Text. Table of regression results for aggregated sanitary risk with different cumulative precipitation lag time. Table H in S1 Text. Table of regression results for dis-aggregated sanitary risk with different cumulative precipitation lag time. Table I in S1 Text. Table of regression results for individual sanitary risk factors with fifteen days cumulative precipitation.
https://doi.org/10.1371/journal.pwat.0000281.s001
(ZIP)
S1 Protocol. Protocol and training materials for field data collection.
https://doi.org/10.1371/journal.pwat.0000281.s002
(ZIP)
Acknowledgments
In addition to the listed authors, many other students, advisors, colleagues, and collaborators supported this work, and it could not have been completed without their generous efforts. Thanks, especially to Bansaga Sage, Hermann Kambou, and many others for logistical support during field data collection, as well as to Matthew Dinwiddie for support with rainfall data acquisition and merging. Thanks to The Odum Institute at UNC for assistance with refining statistical analyses.
References
- 1.
World Health Organization, UNICEF. WHO Global water, sanitation and hygiene: annual report 2022. WHO/UNICEF Joint Monitoring Programme for Water Supply, Sanitation and Hygiene (JMP); 2023
- 2.
World Health Organization , UNICEF . Progress on household drinking water, sanitation and hygiene 2000-2022: special focus on gender. WHO/UNICEF Joint Monitoring Programme for Water Supply, Sanitation and Hygiene (JMP); 2024.
- 3. Onda K, LoBuglio J, Bartram J. Global access to safe water: accounting for water quality and the resulting impact on MDG progress. Int J Environ Res Public Health. 2012;9(3):880–94. pmid:22690170
- 4. Bain R, Cronk R, Hossain R, Bonjour S, Onda K, Wright J, et al. Global assessment of exposure to faecal contamination through drinking water based on a systematic review. Trop Med Int Health. 2014;19(8):917–27. pmid:24811893
- 5. Bain R, Cronk R, Wright J, Yang H, Slaymaker T, Bartram J. Fecal contamination of drinking-water in low- and middle-income countries: a systematic review and meta-analysis. PLoS Med. 2014;11(5):e1001644. pmid:24800926
- 6. Vilane BRT, Dlamini TL. An assessment of groundwater pollution from on-site sanitation in Malkerns, Swaziland. Journal of Agricultural Science .2016.
- 7.
World Health Organization, UNICEF. CORE QUESTIONS ON DRINKING-WATER AND SANITATION FOR HOUSEHOLD SURVEYS. and UNICEF; 2006.
- 8.
Danert K. Stop the Rot Report III: Handpump standards, quality and supply chains with Zambia case study. Action research on handpump component quality and …. Ask for Water GmbH, Skat Foundation and RWSN, St …, 2022; 2022.
- 9. Bain RES, Wright JA, Christenson E, Bartram JK. Rural:urban inequalities in post 2015 targets and indicators for drinking-water. Sci Total Environ. 2014;490:509–13. pmid:24875263
- 10. Harvey PA, Reed RA. Community-managed water supplies in Africa: sustainable or dispensable?. Community Development Journal. 2006;42(3):365–78.
- 11.
World Health Organization. Guidelines for drinking-water quality: incorporating the first and second addenda. Geneva: World Health Organization; 2022.
- 12.
Rajagopalan S, Shiffman MA. Guide to simple sanitary measures for the control of enteric disease. cabidigitallibrary.org; 1974.
- 13. Cronin AA, Breslin N, Gibson J, Pedley S. Monitoring source and domestic water quality in parallel with sanitary risk identification in northern Mozambique to prioritise protection interventions. J Water Health. 2006;4(3):333–45. pmid:17036841
- 14. Kelly ER, Cronk R, Kumpel E, Howard G, Bartram J. How we assess water safety: A critical review of sanitary inspection and water quality analysis. Sci Total Environ. 2020;718:137237. pmid:32109810
- 15. Ercumen A, Naser AM, Arnold BF, Unicomb L, Colford JM, Luby SP. Can sanitary inspection surveys predict risk of microbiological contamination of groundwater sources? evidence from shallow tubewells in rural Bangladesh. Am J Trop Med Hyg. 2017;96(3):561–8. pmid:28115666
- 16. Howard G, Pedley S, Barrett M, Nalubega M, Johal K. Risk factors contributing to microbiological contamination of shallow groundwater in Kampala, Uganda. Water Res. 2003;37(14):3421–9. pmid:12834735
- 17. Kelly E, Cronk R, Fisher M, Bartram J. Sanitary inspection, microbial water quality analysis, and water safety in handpumps in rural sub-Saharan Africa. npj Clean Water. 2021;4(1).
- 18. Misati AG, Ogendi G, Peletz R, Khush R, Kumpel E. Can sanitary surveys replace water quality testing? evidence from Kisii, Kenya. Int J Environ Res Public Health. 2017;14(2):152. pmid:28178226
- 19. Mushi D, Byamukama D, Kirschner AKT, Mach RL, Brunner K, Farnleitner AH. Sanitary inspection of wells using risk-of-contamination scoring indicates a high predictive ability for bacterial faecal pollution in the peri-urban tropical lowlands of Dar es Salaam, Tanzania. J Water Health. 2012;10(2):236–43. pmid:22717748
- 20. Parker AH, Youlten R, Dillon M, Nussbaumer T, Carter RC, Tyrrel SF, et al. An assessment of microbiological water quality of six water source categories in north-east Uganda. J Water Health. 2010;8(3):550–60. pmid:20375484
- 21. Guo D, Thomas J, Lazaro A, Mahundo C, Lwetoijera D, Mrimi E, et al. Understanding the impacts of short-term climate variability on drinking water source quality: observations from three distinct climatic regions in Tanzania. Geohealth. 2019;3(4):84–103. pmid:32159034
- 22. Nijhawan A, Howard G. Associations between climate variables and water quality in low- and middle-income countries: A scoping review. Water Res. 2022;210:117996. pmid:34959067
- 23. Wu J, Yunus M, Islam MS, Emch M. Influence of climate extremes and land use on fecal contamination of shallow tubewells in Bangladesh. Environ Sci Technol. 2016;50(5):2669–76. pmid:26844955
- 24. Powers JE, Mureithi M, Mboya J, Campolo J, Swarthout JM, Pajka J, et al. Effects of high temperature and heavy precipitation on drinking water quality and child hand contamination levels in Rural Kenya. Environ Sci Technol. 2023;57(17):6975–88. pmid:37071701
- 25. Ercumen A, Naser AM, Unicomb L, Arnold BF, Colford JM Jr, Luby SP. Effects of source- versus household contamination of tubewell water on child diarrhea in rural Bangladesh: a randomized controlled trial. PLoS One. 2015;10(3):e0121907. pmid:25816342
- 26. Engström E, Balfors B, Mörtberg U, Thunvik R, Gaily T, Mangold M. Prevalence of microbiological contaminants in groundwater sources and risk factor assessment in Juba, South Sudan. Sci Total Environ. 2015;515–516:181–7. pmid:25723872
- 27. Fan Y, Li H, Miguez-Macho G. Global patterns of groundwater table depth. Science. 2013;339(6122):940–3. pmid:23430651
- 28. Fan Y, Miguez-Macho G, Jobbágy EG, Jackson RB, Otero-Casal C. Hydrologic regulation of plant rooting depth. Proc Natl Acad Sci U S A. 2017;114(40):10572–7. pmid:28923923
- 29. Menard S. Coefficients of Determination for Multiple Logistic Regression Analysis. The American Statistician. 2000;54(1):17–24.
- 30. Chakrabarti A, Ghosh JK. AIC, BIC and Recent Advances in Model Selection. Philosophy of Statistics. 2011:583–605.
- 31. Kostyla C, Bain R, Cronk R, Bartram J. Seasonal variation of fecal contamination in drinking water sources in developing countries: a systematic review. Sci Total Environ. 2015;514:333–43. pmid:25676921
- 32. Harris AR, Daly SW, Pickering AJ, Mrisho M, Harris M, Davis J. Safe Today, Unsafe Tomorrow: Tanzanian Households Experience Variability in Drinking Water Quality. Environ Sci Technol. 2023;57(45):17481–9. pmid:37922469