The combined search strategies identified 976 potentially relevant studies of interventions to improve water quality for preventing diarrhoea. The full text of 68 potentially eligible reports was obtained for further assessment after screening of titles and abstracts. Of these 68 reports, 33 with 42 controlled comparisons met the inclusion criteria (fig 1).w1-w33 The meta-analysis includes three new studies not in the original review.w6 w9 w29 Six studies included two or more intervention arms.
Fig 1Flow chart of search
The 33 studies (22 randomised controlled trials, 11 quasirandomised controlled trials) included about 55
650 participants (table 1). Eighteen studies presented results for participants of all ages, 10 included only under 5s or a subgroup thereof, and the remainder reported on both age groups. In most trials the household was the unit of randomisation, although some randomised neighbourhoods, clusters of households, or villages. Trials of interventions at the water source (mainly the quasirandomised controlled trials) were generally longer (median 36 months, range 12 to 60 months) than those of household interventions (median 7 months, range 9.5 weeks to 12 months). All but two trials from the United Statesw8 w9
were carried out in developing countries. Two trials took place in urban settings, two in peri-urban settings, three in urban informal or squatter settlements, two in refugee camps, one in multiple settings, and the others in rural settings.
Table 1 Details of included studies on interventions to improve microbial quality of drinking water for preventing diarrhoea
The interventions to improve drinking water quality were undertaken at the level of either the water source (seven trials) or the household (35 trials). Water source interventions included protected wells, bore holes, or distribution to public tap stands; none included piped in (reticulated) household connections. Household interventions comprised improved water storage (one trial) or one of four approaches for treating water in the home: chlorination (16 trials), solar disinfection (three trials), filtration (eight trials), or combined flocculation and disinfection (seven trials). Apart from solar disinfection and flocculation-disinfection using a water purifying product (PUR sachet; Procter and Gamble), there were potentially important differences in the types of interventions. For example, filtration interventions varied by filter medium and pore size, and chlorination varied by chlorine source, dose, and contact time.
Improvements in water quality were often accompanied by other environmental interventions intended to prevent faecal-oral transmission, including improved sanitation and water supplies, improved water storage in the home, and instruction on basic hygiene regarding contaminated water and diarrhoeal disease (table 1). One study included the introduction of oral rehydration therapy.w24 However, 14 trials consisted solely of water quality interventions, although ceramic filters and solar disinfection interventions may also improve storage.
Compliance (consumption of improved quality water) was not assessed directly. Trials of interventions at the water source tended to assume compliance on the basis that the primary water supply had been improved. Some trials of household water treatment undertook indirect assessments of compliance by measuring residual chlorine levels in stored water,w2w4 w12 w13 w15 w17 w23 w25-w27 w30 comparing microbial water quality of intervention and control households,w4 w5-w6 w12 w19 carrying out periodic or post-study surveys,w4 w27 w29 or counting the amount of intervention product used.w27 Seven trials did not report compliance.w1 w11 w20 w22 w31 w33
Even when efforts were made to document compliance, investigators acknowledged that it was not possible to determine the extent to which participants in the intervention group consumed treated water or avoided consuming untreated water.
Outcome measures and effect estimates
Twenty one trials used the WHO definition of diarrhoea (three or more loose stools in the previous 24 hours); most others used local terms or mothers' definitions. All were based on self report. In most trials participants were visited on a periodic basis, either weekly (14 trials), fortnightly (five trials), or more infrequently (five trials), and were asked to recall episodes of diarrhoea during a previous period, usually seven days (18 trials) to 14 days (six trials). In some trials participants were asked to keep records of days with diarrhoea.
Effect estimates included rate ratios (12 trials), risk ratios (eight trials), longitudinal prevalence ratios (11 trials), and odds ratios (10 trials; table 2). One trial did not provide enough information to calculate the actual measure of effect and was excluded from the meta-analysis.w31 Most studies adjusted for covariates. However, none of the source based interventions and four of the household based trialsw2 w19 w23 w32 did not report adjusting for clustering and may thus receive excess weight in meta-analysis due to artificial precision.
Table 2 Summary of pooled estimates (random effects) for studies reporting rate ratios, risk ratios, longitudinal prevalence ratios, and odds ratios for all studies (source based and household based), by point of intervention (source or household), and (more ...)
Figure 2 present the forest plots for studies reporting effect estimates for all ages and for under 5s. Most trials recorded notable reductions in diarrhoea; none found the interventions to be associated with a statistically significant increase in diarrhoea related morbidity. The evidence from the pooled estimates of effect for all trials by each measure of effect suggests that interventions to improve the microbial quality of drinking water are effective in reducing the occurrence of diarrhoea both for all participants and for under 5s (table 2). Pooled estimates were, however, characterised by considerable heterogeneity (table 2). A subgroup analysis was carried out on the criteria specified in the protocol to attempt to explain such heterogeneity.
Fig 2Forest plot of studies reporting rate ratios, risk ratios, longitudinal prevalence ratios, and odds ratios (random effects model). Squares are point estimates and horizontal lines are 95% confidence intervals. Size of square represents relative (more ...)
Water source versus household interventions
Table 2 shows the pooled estimates of interventions at the water source level and at the household level. Although individual trials of source based interventions reported the intervention to be effective, the pooled estimate for trials using rate ratios fell short of statistical significance, both among four trials reporting data for all ages (0.87, 95% confidence interval 0.74 to 1.02) and three trials reporting on under 5s (0.93, 0.82 to 1.05). The two studies on source based interventions reporting the highest level of effectiveness could not be pooled because they used different measures of effect.w24 w33 Moreover, the small number of clusters and the failure to take clustering into account in the analysis must raise doubts about the validity of such estimates. Household interventions, on the other hand, significantly reduced diarrhoea episodes among people of all ages and among under 5s, as measured with rate ratios, risk ratios, and odds ratios, but these pooled estimates were still heterogeneous. The pooled longitudinal prevalence ratio for household interventions was statistically significant when a possible outlierw13 was excluded from the analysis for all age groups (0.70, 0.56 to 0.88; nine trials) and for under 5s (0.76, 0.66 to 0.88; nine trials).
Type of household intervention
Table 2 also shows the pooled estimates of effect by type of household intervention. Although such subgrouping reduces heterogeneity among certain types of household interventions other pooled estimates were still characterised by considerable heterogeneity. Household chlorination was associated with a statistically significant reduction in diarrhoea among all age groups when measured using rate ratios and risk ratios, and in under 5s when using risk ratios. No statistically significant advantage was found for people of all ages when measured using longitudinal prevalence ratios or for under 5s when measured using rate ratios, longitudinal prevalence ratios, or odds ratios. Household filters were associated with a statistically significant and homogeneous reduction in diarrhoea among all ages and in under 5s for trials measuring risk ratios and odds ratios, but not among trials measuring rate ratios. Excluding the two studies carried out in the United States in settings with high ambient water quality,w8 w9 however, resulted in a single study reporting a statistically significant rate ratio in favour of the intervention (0.21, 95% confidence interval 0.07 to 0.61). Solar disinfection was associated with a reduction in diarrhoea among all ages in both trials measuring odds ratios. A single study that measured the effectiveness of the intervention among under 5s reported a rate ratio of 0.64 (0.41 to 1.00). For household based flocculation-disinfection, pooled estimates from the five trials reporting longitudinal prevalence ratios found no statistically significant difference in the number of diarrhoea episodes compared with the control, either for people of all ages or for under 5s. However, excluding one trialw13 that found a substantial protective effect but which has been identified as a possible outlier rendered the pooled estimate statistically significant in favour of the intervention, both for all ages (0.60, 0.43 to 0.83) and for under 5s (0.66, 0.43 to 0.76). The two trials using odds ratios reported a statistically significant reduction in diarrhoea episodes for all ages from household based flocculation-disinfection but not for under 5s. The one trial that involved improved storage found a protective but, lacking power, not statistically significant difference in diarrhoea episodes, measured with risk ratios, for people of all ages (0.79, 0.61 to 1.03) and for under 5s (0.69, 0.47 to 1.01).w28
Among trials reporting odds ratios, the pooled estimate of effect was substantially higher in settings where compliance with the intervention was higher (≥50% compliance (four trials), odds ratio 0.39, 95% confidence interval 0.39 to 0.51 v <50% compliance (four trials), 0.80, 0.71 to 0.89). These results must be interpreted with caution as the four trials comprising the less than 50% category are from one studyw27 and are compared with only one control group.
Water supply and sanitation
Subgroup analyses for each measure of effect were carried out according to whether the water supply or sanitation was “improved” or “unimproved” on the basis of established criteria. No statistically significant differences were found between pooled estimates on the basis of these criteria. However, pooled estimates show a statistically significant effect in favour of intervention even in settings without improved water supply (seven trials reporting rate ratios, 0.74, 95% confidence interval 0.63 to 0.87; four trials reporting risk ratios, 0.46, 0.36 to 0.58; six trials reporting longitudinal prevalence ratios, 0.83, 0.68 to 1.01; and nine trials reporting odds ratios, 0.66, 0.57 to 0.77). Interventions were also effective in settings without improved sanitation (four trials reporting rate ratios, 0.78, 0.64 to 0.95; two trials reporting risk ratios, 0.55, 0.47 to 0.65).
Water quality only versus compound environmental interventions
Pooled estimates showed that water quality interventions were significantly effective when introduced alone or in combination with other environmental interventions (hygiene instruction, improved water storage vessel, improved sanitation, or improved water supply). Notably, however, no evidence was found for water quality interventions being more effective when combined with other components than when implemented alone. Pooled estimates for water quality interventions alone (seven trials reporting odds ratios, 0.61, 95% confidence interval 0.50 to 0.73; five trials reporting rate ratios, 0.76, 0.52 to 1.02) were not statistically different from pooled estimates for trials combining water quality with instruction on basic hygiene (one trial reporting odds ratio, 0.52, 0.30 to 0.90; three trials reporting rate ratios, 0.85, 0.70 to 1.03), water quality with a storage vessel (three trials reporting odds ratios, 0.77, 0.58 to 0.84; four trials reporting rate ratios, 0.61, 0.46 to 0.81), water quality plus sanitation (three trials reporting odds ratios, 0.60, 0.43 to 0.84; one trial reporting rate ratio, 0.75, 0.70 to 0.80), or water quality with improved water supply (four trials reporting odds ratios, 0.70, 0.59 to 0.84; two trials reporting rate ratios, 0.77, 0.71 to 0.84).
Study design; methodological quality
Subgrouping trials on study design (randomised and quasirandomised controlled trials) did not show a trend in favour of either design approach (table 3). Greater protective effects were generally reported among randomised controlled trials with high quality for sequence generation, allocation concealment, and inclusion or losses to follow-up. Only four studies, however, used double blinding (table 1) and none of these found a statistically significant protective effect from the water quality intervention. Similarly, among quasirandomised controlled trials, effects in studies meeting the specified criteria for methodological quality were larger. Few trials, however, failed to meet these criteria, and subgroup analyses did not explain the heterogeneity.
Table 3 Methodological quality of included studies