In sub-Saharan Africa, unsafe drinking water was estimated to cause more than 200,000 deaths in 2012 [1
]. Information about water quality can be used to manage the safety of water sources. While regulatory structures often exist that specify the responsibilities of health or water surveillance agencies to test water sources, institutions in this region often have limited capacity to carry out these responsibilities [2
]. Water quality testing is expensive and complicated: collecting and testing water samples requires labor, equipment, consumables, transportation, and training [5
Given the large amount of resources required for testing water quality, there is a need to improve its cost-effectiveness. Alternative approaches for identifying high-risk water supplies such as sanitary surveys, which are observational checklists for identifying potential hazards, have been promoted by environmental health specialists and the World Health Organization (WHO) [8
]. Sanitary surveys can be completed quickly and require no special equipment, making them less expensive and easier to implement than microbiological testing [6
]. However, previous research on whether risks assessed through sanitary surveys correspond to measured water quality has produced mixed results and focused primarily on solely groundwater sources. For example, a study in Amuria District, Uganda, found that the water source type was a better predictor of fecal contamination than sanitary scores (the sum of hazards present in the sanitary survey); boreholes had the best microbiological water quality, followed by open dug wells, protected springs, and, finally, surface water, but there was only a weak correlation between thermotolerant coliform (TTC) concentrations and variations in sanitary scores [9
]. A study in County Cork, Ireland, found no correlation between sanitary survey–identified hazards and TTC, although they noted there were too few contaminated samples to make an adequate comparison [10
]. In contrast, a study in Dar es Salaam, Tanzania, found that the sanitary score predicted up to 87% of the measured Escherichia coli
concentrations among wells [6
]. Similarly, a study of shallow protected springs in Kampala, Uganda, identified a positive relationship between the sanitary score and microbial water quality, although no significant correlation was identified when data from low- and high-density population areas were analyzed separately [11
]. It is therefore likely that correlations between microbiological indicator organism measurements and sanitary scores are water source– and context-specific; more research is needed to understand when it is appropriate to use sanitary surveys as a supplement or replacements for water quality testing.
Our objective was to assess whether the results of sanitary survey inspections were associated with measured indicator bacteria levels in rural drinking water sources. We focused on rural areas of Kisii Central, a sub-county in Kenya. In Kenya, while 43% of households used improved water sources in 2015, these improved sources may not provide water that is safe. For example, in Kisii County, Kenya, the population without access to safe drinking water was estimated to be at least nine percentage points higher when improved sources that contained microbial contamination were excluded from the proxy definition of ‘safe’. While water quality testing is performed by public health surveillance agencies in Kisii County, there are more water sources than can be reasonably tested within the three- to five-year recommended time frame [8
]. The results of this study can suggest whether sanitary surveys can be used to complement or replace water quality testing to inform water safety management activities. We also evaluated the variability in fecal contamination between two samples collected from a point water source to assess the information gained or lost from one-time sampling of a source.
2. Materials and Methods
2.1. Study Site
Kisii Central is an administrative division of Kisii County, Kenya, which is located southeast of Lake Victoria with a population of 1.1 million people [17
] (Figure 1
). Average annual rainfall in Kisii is estimated at 1500 mm [18
In 2011, protected springs were the most frequently used water sources in rural areas of Kisii County (used by 55.9% of the population), followed by unprotected springs (used by 27.2% of the rural population) and surface water (used by 7.6% of the rural population) [19
]. Rainwater collection systems were used by 3.6% of the rural population, protected dug wells and unprotected dug wells were used by 2.1% of the rural population each, and less than 2% of the remaining population used piped water or other sources. Aquifers in the Gucha catchment, of which most of Kisii County is included, range from depths of 13–60 m [20
2.2. Study Design
We used stratified sampling to select 61 drinking water sources to sample based on an inventory of drinking water sources obtained from the Kisii County District Public Health Office (KCDPHO) [3
]. In February 2014, KCDPHO Community Health Workers (CHW) had worked with village elders to count all water sources in each village to create the inventory. The total number of selected sources was based on available resources for testing. We selected the number of sources to sample in each rural administrative division (Mosocho, Keumbu, and Kiogoro) roughly in proportion to the total number of water sources in the inventory in each division (Figure 1
). The inventory listed 7400 sources: 3981 in Mosocho (6% springs, 33% dug wells, and 61% rainwater harvesting systems), 2188 at Keumbu (8% springs, 4% wells, and 88% rainwater harvesting systems), and 1231 at Kiogoro (5% springs, 1% wells, and 92% rainwater harvesting systems). Sources accounting for <1% of the inventory included boreholes, rivers, rock catchments, and piped schemes. We sampled 32 water sources from Mosocho, 21 from Keumbu, and eight from Kiogoro.
To select water sources within each division for sampling, we consulted with KCDPHO CHWs who recommended the water sources in each division that were most commonly used by community members. Although rainwater harvesting systems (RWH) accounted for the majority of sources in all divisions, many of these RWH were used only by individual households, while springs were shared by villages and dug wells were frequently shared by multiple households. Therefore, a higher proportion of the population used springs or dug wells, as indicated by the census data. Therefore, we selected 25 springs, 20 wells, and 16 RWH for sampling. Since we prioritized sources used by a large number of people, the sampled springs were shared by communities (the local government was involved in their protection, however, there were no clear structures for managing the springs), the sampled wells were commonly built, shared, and managed by several households, and sampled RWH were constructed and managed by institutions (e.g., schools, hospitals).
We sampled selected water sources twice within one month, one week apart, to capture variability in water quality, in August 2014. The two samples from the same source were collected between seven to 13 days apart.
2.3. Sample Collection and Analysis
Samples from springs were directly collected from the spring outlet. Samples from wells were collected by drawing from a sterilized bucket that had been wiped with ethanol and rinsed with deionized water. Taps at rainwater tanks were sterilized with a flame before sample collection. Water samples were collected in 100 mL Whirl-Pak collection bags (Nasco, Modesto, CA, USA) and put on ice for transport to a laboratory located at the Gusii Water and Sanitation Company in the town of Kisii and tested within 6 h.
We quantified TTC counts using the membrane filtration system included with the Potakit® test kit (Palintest, Ltd., Gateshead, UK). We tested sample blanks daily. We measured temperature, pH, electrical conductivity (EC), and turbidity on site using probes in the Potakit test kit. Calibrations for the equipment were performed before fieldwork began.
2.4. Sanitary Survey
We also conducted a sanitary survey for each selected water source. The sanitary survey questionnaire relied on a list of observations, unique for each water source type (springs, dug wells, and RWH). We used the sanitary surveys in the WHO Guidelines for Drinking Water Quality (GDWQ) and modified these questionnaires during piloting to include additional risk factors recommended by CHWs and exclude questions not applicable to the study area (Table 1
At each water source, questions in the sanitary survey that identified a water quality risk were given a score of one point while those that did not identify a risk were given a score of zero points. We totaled the points for each water source and divided by the total number of questions answered to arrive at a risk score (RS) for each source. We then grouped the RS into low (0%–30%), medium (40%–50%), high (60%–70%), and very high (80%–100%) risk categories [6
]. We also recorded whether it had rained the day before sampling.
2.5. Data Analysis
Data were entered into Excel and analyzed with Excel and R software (R Core Team 2015, Vienna, Austria). TTC data were log transformed and used a value of log10(x+1). We used a non-parametric test, the Wilcoxon rank sum test, to determine the significance of differences in the measured water quality parameters from water sources with different sanitary features. Samples were classified as TNTC if the TTC count exceeded 300 CFU/100 mL. Conductivity was converted to total dissolved solids (TDS) by multiplying the default value in the meter of 0.5 by the EC values. The significance level was set at p < 0.05.
Overall, levels of contamination were high: all samples from dug wells contained fecal indicator bacteria, while 95% of springs and 61% of rainwater harvesting systems were contaminated. We did not find significant correlations between TTC counts and sanitary risk scores. Contamination was instead more closely correlated with the water source type (i.e., dug wells were significantly more contaminated than springs, which were significantly more contaminated than rainwater harvesting tanks). RWH were of better quality than dug wells or springs; in our version of the modified sanitary survey, there were fewer possible risk factors for RWH and, by extension, fewer contamination pathways. We found variability in levels of TTC between the two samples tested from the same water source on two different days or before or after a rainfall event, although results did not vary in aggregate.
One risk factor often used as a proxy for safe water is whether a dug well or spring is protected (the Joint Monitoring Program of the WHO/UNICEF, which monitors global access to water, had previously defined an improved dug well or spring as whether these sources were protected [21
]). In protected springs, the eye of the spring is enclosed in a covered concrete box with an outlet near the bottom to allow the flow of water away from the original site of the spring. In our sample, all but one spring lacked a surface diversion ditch (which would intercept surface water runoff carrying possible contaminants) or lacked a fence (preventing livestock from accessing the uphill area of the spring). Similarly, wells considered by community members as protected often lacked lining or a windlass. Therefore, while we did not find significant differences in TTC concentrations between protected and unprotected springs and wells, inadequacies in protection may have left these water sources vulnerable to contamination. We also noted that many communities identified these springs and dug wells as protected despite these faults. This highlights the need for communication with communities and water source owners on the proper protection of water sources.
Our study focused on water quality at the point of collection. However, during sampling, we observed that some rainwater harvesting tanks may have included a mix of rainwater and groundwater, a practice used to ensure the availability of water and documented in other studies [22
]. This and other water collection and handling-related practices would be expected to further deteriorate water quality [23
]. There is a need to extend the safe management of water past the water source and to collection, handling, and storage of water.
While we found that sanitary surveys were not a substitute for microbial water quality testing results in this context, they did identify many faults in the water sources, particularly in sources considered by communities to be protected. Sanitary surveys could be a useful tool for highlighting hazards to address as part of a comprehensive risk management approach system, such as Water Safety Plans [24
]. For example, sanitary surveys could be conducted as an initial assessment of water sources to identify systemic hazards in an area (e.g., the sanitary surveys in this study revealed that most wells and springs were unfenced, and that most wells were unsealed). Collective data on sanitary risks could help highlight potential key investments. However, for these to be effective tools, their results should be linked to actions to improve the water sources.
In this study, we expanded on the WHO GDWQ sanitary survey. We included water source infrastructure features, which vary little over time (e.g., lack of fencing or diversion ditch), and potential sources of contamination, which could be persistent (e.g., dirty environment, vegetation near the source) or transient (e.g., animals grazing, clothes washing, human activity). Future research could compare the influence on water quality of blocking pathways to contamination through the water source infrastructure to that of controlling sources of contamination. Transient sources of contamination are particularly challenging to study: several of our sanitary survey questions relied on observations at the time of the survey (e.g., observations of children playing near the source). These observations may have differed if the water source had been sampled at a different time of day. Additionally, sanitary risk scores weighed all risks equally; however, it is likely that some factors may be more influential than others (for example, factors related to the structural integrity of the water source could be more influential than transient sources of pollution such as children playing near the source). Future research should focus on weighting the individual elements of a sanitary survey inspection to identify whether some elements have a higher contribution to risk. Such weighting could use an expert opinion or tools such as decision trees or system dynamics approaches to understand the interactions between hazards [25
Finally, the approach used in this study relied on an inventory of water sources to understand where water sources were distributed; we then relied on the knowledge of the Community Health Workers to select from those sources most in use by communities. Water point mapping, a tool frequently used to create inventories of water sources with geo-tagged information, could also serve as useful sampling frames for future studies [28
There are several other limitations to our study. First, the study focused on a small set of sub-counties; therefore, the results obtained are specific to the study area and should not be directly applied elsewhere. Second, the number of sources sampled and the total number of samples were limited. It is possible, due to the small sample size, that a small but significant relationship between sanitary risk scores and TTC existed but that the sample size was not sufficient to detect the association. Finally, there is likely further variation in water quality in other seasons. A review of the seasonal effects of fecal contamination in drinking water sources found that improved sources were more contaminated during wet seasons than dry seasons across all water source types and in various climates [30
]. While our results showed variation in the water quality of individual sources before or after rainfall events, the study took place during only two months. Cross-sectional studies can potentially underestimate contamination due to seasonal variability in fecal contamination [12
]; therefore, it would be important for future studies to sample longitudinally, particularly as effects of the identified hazards may be mediated by rainfall.