To assess changes in water quality over time in a real-world setting and to compare four measures of water quality in two types of water storage containers, we conducted a controlled crossover trial of 2 randomly selected groups of households in western Kenya from August 27-October 19, 2012. In one group (Group A), water was chlorinated and stored in clay pots typically used for drinking water storage; in the other group (Group B), water was chlorinated and stored in a plastic safe storage container (Fig. 1). Over the following 72 h, water quality tests were performed for both groups. After a two-week washout period, the container types were switched between the groups, and the process described above was performed (Fig. 2).

We selected a convenience sample of six rural villages in Kisumu County that relied on variety of community drinking water sources and household water storage. Households with the following characteristics were eligible to participate: had ≥ one child <5 years old; collected and transported drinking water in 10 L or 20 L containers (jerry cans or buckets); stored drinking water in a ≥15 L ceramic pots (range 15–30 L) in the home; and were willing to use a plastic safe storage container to store drinking water for half of the study period and their own ceramic pot for the other half of the study. Households that did not store drinking water in ceramic pots with ≥15 L capacity were excluded because of the likelihood that stored water would not last for more than one day.

In each of the 6 study villages, we obtained a list of all households with at least one child <5 years old from the village chief, or conducted a brief census to obtain the list of households. We then used a random numbers table to select a sample of households with children <5 years old in each of the 6 communities. A total of 60 households were initially enrolled in the study. At the time of enrollment, respondents in households were interviewed about demographic characteristics, and water, sanitation, and hygiene practices. Electronic questionnaires were verbally administered in Dholuo, the local language, by trained Kenyan field research assistants.

The 60 households were randomly allocated to two groups — Groups A (30 households) and B (30 households) (Fig. 2). Group A households were asked to use their clay pots during the first half of the study while Group B households were provided a new, 60 L plastic safe storage container with a lid, tap, and stand.

Households were contacted in advance and requested to fill their water collection containers (in most cases, 20 L jerry cans) using water from their usual drinking water source on the morning of the first home visit and to keep it in the transport containers. During the first home visit, investigators collected Time 0 (“pre-dose”) water samples by pouring water directly from the transport containers into test vials and sample bottles.

To assess water quality, three water quality and treatment measures were performed using portable field meters in the home. Water samples collected into 10 mL glass vials were tested for TCR (mg/L) and FCR (mg/L) (Hach^® Pocket Colorimeter^™ II, Loveland, CO, USA); water samples collected into 50-mL polypropylene conical tubes were tested for ORP (mV) (Oakton^® Waterproof ORPTestr^® 10, Vernon Hills, IL, USA). Additionally, a 100 mL sample was collected in a WhirlPak^™ bag containing sodium thiosulfate, stored on ice, and transported to the laboratory within 4–6 h of collection for E. coli quantification (CFU/100 mL) using membrane filtration (0.45 mM, 47 mm filters) with m-ColiBlue24^® media (Hach^®, Loveland, CO, USA). In some cases, because of exceedingly slow filtration rates of water samples due to high turbidity, we limited the volume of filtrate to 20 or 50 mL of sample and multiplied positive results by the appropriate proportion factor; samples with no growth were reported as non-detectable for E. coli.

In addition, because physicochemical parameters can influence chlorine residuals and other water quality measures, we also tested samples collected in 50 mL polypropylene conical tubes for the four following physicochemical parameters: turbidity (NTU) (Hach^® 2100Q Portable Turbidimeter, Loveland, CO, USA), temperature (°C), electrical conductivity (μs/cm), and total dissolved solids (mg/L) (Oakton^® Waterproof Multiparameter PCS Tester 35).

In the presence of the head of household, investigators then treated each water transport vessel with the proper dose of WaterGuard^™, a familiar, locally available water treatment product containing 1.25% sodium hypochlorite solution. The dose was based on turbidity and the volume of water in the jerry can; water with turbidity <10 NTU was dosed with a single 3 mL dose of WaterGuard per 20 L and water with turbidity >10 NTU was dosed with a double dose of 6 mL of WaterGuard per 20 L. Treated water was then poured into either the household’s empty ceramic pot (Group A) or the new plastic safe storage container (Group B).

After 30 min, water samples were collected and tested, as described above. Because the size and weight of clay storage pots precluded pouring water samples, each head of household was asked to wash a cup or ladle that was normally used to obtain water so that water samples could be collected; water samples from the improved plastic storage containers were taken directly from the tap. Heads of households were asked not to retreat the water or refill the container unless it was completely emptied out.

The household was revisited at 24, 48, and 72 h for a short-follow-up interview about water addition or treatment since the previous visit, followed by collection and testing of water samples, as described above. If respondents reported that water or additional disinfectant had been added to the storage container since the previous visit, they were excluded from the remainder of this phase of the study.

The crossover period of the project began after a 2-week washout period. Households were again contacted in advance and requested to fill their transport containers using water from their usual drinking water source on the morning of the first home visit and to keep it in the transport containers. Households in Group A were provided with a plastic safe storage container with a lid, tap, and stand; Group B households were asked to resume using their ceramic pots for water storage. Water treatment and testing proceeded in a fashion identical to phase 1. At the conclusion of phase 2, all households were allowed to keep the plastic safe storage containers, stands, and a bottle of WaterGuard for participation in the study.

Interview data were entered into personal digital assistants (PDAs) and uploaded into an Access (Microsoft, Redmond, WA, USA) database. Chemical and microbial data were collected on hardcopy forms, entered into an Excel database, and analyzed with SAS^® 9.3 software (Cary, NC, USA). TCR, FCR, turbidity, and E. coli concentration had skewed distributions and were categorized according to the following metrics. For descriptive purposes, E. coli was categorized according to WHO risk thresholds as non-detectable or 1–10, 11–100, or >100 CFU/100 mL (WHO, 1997). Since WHO guidelines and public health interventions are aimed at complete removal of E. coli, we further categorized data into a dichotomous presence/absence for modeling. FCR was categorized as <0.2 or ≥0.2 mg/L, as this is the minimum recommended concentration by the WHO Guidelines for Drinking Water Quality for infrastructure treated water (WHO, 2011). TCR was similarly categorized as <0.2 or ≥0.2 mg/L based on previous research that utilized this strategy to assess chlorine treatment efficacy and storage time in ceramic pots (Null and Lantagne, 2012). Water samples were categorized as turbid when turbidity measures were ≥10 NTU, in reference to chlorine dosing recommendation for turbid water. The primary outcomes of interest were TCR, FCR, ORP, and E. coli.

To investigate water quality differences in clay pots and plastic safe storage containers across the five time intervals, two-way within-subjects random effects models were constructed; logistic regression models for the outcomes TCR, FCR, and E. coli and linear regression for ORP. Interaction terms for storage container and time interval were significant for all four primary outcomes (TCR, FCR, ORP, and E. coli). For results stratified by water storage container type, we present estimates from separate repeated measures models for binary outcomes using generalized estimating equations (GEE) and an autoregressive correlation structure. Odds Ratios (OR) and 95% Confidence Intervals (CI) computed from robust standard error estimates are reported. ORP mean differences are computed from random effects linear regression models and Tukey adjusted p-values are reported. All models adjusted for turbidity.

The study protocol was approved by the Ethical Review Committee of the Kenya Medical Research Institute (protocol 2324) and the Institutional Review Board of the Centers for Disease Control and Prevention (protocol 6313). Written informed consent was obtained from all participants. Data were maintained in an encrypted file in a password-protected computer. Personal identifiers were destroyed after all data were collected.

A total of 60 respondents were enrolled in the study. Five households were excluded from the study because respondents weren’t available for one or more of the intervention phases; ultimately 25 respondents remained in Group A and 30 respondents comprised Group B. The median age across participating respondents was 27 (range 17–55) and all were women. Fewer than half (n = 23) had less than a complete primary school education and only one, in Group B, had electricity. The majority (85%) of study households used improved water sources and 60% of respondents reported that they treated water stored in their homes. Of 32 households that reported treating their water, 24 (75%) reported using WaterGuard; 2 (6%) reporting using other chlorine-based products, 5 (16%) reported boiling, and 12 (38%) reported using a cloth to filter water. Fewer than half (47%) of households had an improved sanitation facility. Soap was present in 93% of households and 56% of respondents were able to demonstrate proper handwashing technique.

Water sources used for dosing experiments in clay pots included rain (40%), surface water (24%), springs (18%), piped networks (16%), and ground water (2%). The median turbidity of water samples in clay pots at Time 0 was 37 NTU (range 0–300 NTU) (Table 1). Turbidity measures did not vary widely over the 72 h testing period. Median TCR and FCR values at Time 0 were 0.1 mg/L; over three-fourths of samples were <0.2 mg/L for both TCR and FCR. The median ORP was 393 mV (range 196–597 mV). At Time 0, 83% of water samples were contaminated with E. coli. Water had a median pH 6.8, 25 °C temperature, electrical conductivity of 106 μs/cm, and 78 mg/L total dissolved solids; these median measures did not vary widely over the 72 h testing period.

Thirty minutes after chlorination (Time 0.5 h), median TCR and FCR levels increased to 1.2 and 0.9 mg/L, respectively (Table 1). Median ORP increased to 541 mV (range 392–757 mV), with 93% of samples increasing by >10% of the time 0 value. E. coli were non-detectable in 83% of samples. By 24 h, FCR was <0.2 mg/L in 61% of samples and TCR was <0.2 mg/L in 31% of samples. Approximately 40% of samples had ORP values 10% of the time 0 value. E. coli were non-detectable in 74% of samples. At 48 h, 51% of TCR and 67% of FCR values were <0.2 mg/L and the median ORP measurement decreased to slightly lower than the time 0 value. The percentage of samples with non-detectable E. coli decreased to 48%. By 72 h, median TCR was 0.2 mg/L and FCR was 0.1 mg/L; 35% of samples had no detectable E. coli.

Compared to Time 0 values and adjusted for turbidity, water treated with the recommended amount of chlorine and stored in clay pots was significantly less likely to contain E. coli for up to 48 h (Table 2). Although FCR levels were significantly more likely to be > 0.2 mg/L at 30 min than at Time 0, by 24 h FCR was significantly more likely to have fallen below the threshold of 0.2 mg/L. However, TCR levels were less likely to have fallen below the 0.2 mg/L threshold over the entire 72 h time period than at baseline. At 30 min and 24 h post treatment, ORP was significantly higher than at Time 0. By 48 h, ORP values were not significantly different than at Time 0.

Source water used for dosing experiments in plastic safe storage containers include rain (44%), surface water (22%), springs (18%), piped networks (13%), and ground water (4%). The median turbidity of water samples in plastic containers at time 0 was 28 NTU (range 0–300 NTU) (Table 3). Turbidity measures did not vary widely over the 72 h testing period. At Time 0, median TCR and FCR were 0.1 mg/L and <0.1 mg/L, respectively, with over three-fourths of samples <0.2 mg/L for both TCR and FCR. The median ORP was 387 mV (range 252–556 mV). At Time 0, 87% of water samples were contaminated with E. coli. Water had a median pH 7.2, 24 °C temperature, electrical conductivity of 104 μs/cm, and 69 mg/L total dissolved solids; these median measures did not vary widely over the 72 h testing period.

Thirty minutes after chlorination, median TCR and FCR levels increased to 1.3 and 0.8 mg/L, respectively (Table 3). Median ORP increased to 541 mV (range 392–747 mV), with 89% of samples increasing by >10% of the Time 0 value. E. coli were non-detectable in 91% of samples. By 24 h, median FCR decreased to 0.3 mg/L, median TCR was 0.7 mg/L, and 15% of water samples had fallen to within 10% of the time 0 ORP value. E. coli remained non-detectable in 85% of samples. At 48 h, 17% of TCR and 33% of FCR values were <0.2 mg/L and the median ORP measurement remained higher than the Time 0 value. E. coli were non-detectable in 90% of samples. By 72 h, median TCR was 0.6 mg/L and FCR was 0.3 mg/L, the median ORP value was higher than at Time 0, and 80% of samples had no detectable E. coli.

When compared with Time 0 values and adjusted for turbidity, water treated with the recommended amount of chlorine and stored in a plastic safe storage containers was significantly less likely to contain E. coli across all time points, indicating a protective effect for up to 72 h (Table 4). Both FCR and TCR levels were significantly less likely to be < 0.2 mg/L than the Time 0 values over the entire 72 h time period. Up through 24 h, mean ORP was significantly higher than Time 0 ORP values, however, by 48 h ORP values were not significantly different than Time 0 values.

Using two-way random effects models and adjusting for turbidity, we assessed differences in water storage containers and time points for TCR, FCR, and E. coli. Water container type was a statistically significant effect modifier for time interval, thus we present results stratified by either water container or time interval. There were statistically significant differences between clay pots and plastic safe storage containers for TCR, FCR, and E. coli at 48 and 72 h. Despite no differences in water quality measures between storage containers at pre-treatment, 30 min and 24 h, the odds of having a positive E. coli result were greater in clay pots compared to plastic containers at 48 (p = 0.0002) and 72 h (p = 0.0004). The odds of having TCR <0.2 mg/L were significantly greater in clay pots than plastic containers at 24 (p = 0.0199), 48 (p = 0.0023), and 72 h (p = 0.0061); likewise, the odds of having FCR <0.2 mg/L were significantly greater in clay pots than plastic containers at 24 (p = 0.0370), 48 (p = 0.0014), and 72 h (p = 0.0245) (Table 5).

If TCR or FCR was ≥0.2 mg/L in stored water, regardless of container type or time, there was a decreased likelihood that E. coli was present. This association was stronger for TCR ≥0.2 in plastic containers (OR 0.08, 95% confidence interval [CI] 0.04–0.16) than in clay pots (OR 0.44, 95% CI 0.27–0.75); likewise, this association was stronger for FCR ≥0.2 in plastic containers (OR 0.25, 95% CI 0.14–0.44) than in clay pots (OR 0.43, 95% CI 0.26–0.69).