To assess the impact of inadequate water and sanitation on diarrhoeal disease in low- and middle-income settings.
To assess the impact of inadequate water and sanitation on diarrhoeal disease in low- and middle-income settings.
The search strategy used Cochrane Library, MEDLINE & PubMed, Global Health, Embase and BIOSIS supplemented by screening of reference lists from previously published systematic reviews, to identify studies reporting on interventions examining the effect of drinking water and sanitation improvements in low- and middle-income settings published between 1970 and May 2013. Studies including randomised controlled trials, quasi-randomised trials with control group, observational studies using matching techniques and observational studies with a control group where the intervention was well defined were eligible. Risk of bias was assessed using a modified Ottawa–Newcastle scale. Study results were combined using meta-analysis and meta-regression to derive overall and intervention-specific risk estimates.
Of 6819 records identified for drinking water, 61 studies met the inclusion criteria, and of 12 515 records identified for sanitation, 11 studies were included. Overall, improvements in drinking water and sanitation were associated with decreased risks of diarrhoea. Specific improvements, such as the use of water filters, provision of high-quality piped water and sewer connections, were associated with greater reductions in diarrhoea compared with other interventions.
The results show that inadequate water and sanitation are associated with considerable risks of diarrhoeal disease and that there are notable differences in illness reduction according to the type of improved water and sanitation implemented.
Evaluer l'impact de l'eau et de l'assainissement inadéquats sur les maladies diarrhéiques dans les régions à revenus faibles et intermédiaires.
La stratégie de recherche a utilisé Cochrane Library, MEDLINE, PubMed, Global Health, Embase et BIOSIS, complétés par la recherche dans les listes de références des revues systématiques publiées antérieurement, pour identifier les études rapportant sur des interventions examinant l'effet de l'amélioration de l'eau potable et de l'assainissement dans les régions à revenus faibles et intermédiaires, publiées entre 1970 et mai 2013. Les études comprenant des essais randomisés contrôlés, des essais quasi-randomisés avec un groupe témoins, des études d'observation utilisant des techniques d'appariement et des études observationnelles avec un groupe témoins où l'intervention a été bien définie, étaient éligibles. Le risque de biais a été évalué à l'aide d'une échelle modifiée d'Ottawa-Newcastle. Les résultats de l’étude ont été combinés en utilisant une méta-analyse et une méta-régression pour déduire des estimations du risque globales et spécifiques à l'intervention.
Sur 6819 rapports identifiés sur l'eau potable, 61 études répondaient aux critères d'inclusion et sur 12 515 rapports sur l'assainissement, 11 études ont été incluses. Dans l'ensemble, l'amélioration de l'eau potable et de l'assainissement a été associée à une diminution des risques de diarrhée. Des améliorations spécifiques, telles que l'utilisation de filtres à eau, la fourniture de raccordement à l'eau courante de haute qualité et à des égouts, ont été associées à des réductions plus importantes de la diarrhée par rapport à d'autres interventions.
Les résultats montrent que l'eau et l'assainissement inadéquats sont associés à des risques considérables de maladies diarrhéiques et qu'il existe des différences notables dans la réduction des maladies selon le type d'amélioration de l'eau et de l'assainissement mis en place.
Evaluar el impacto del agua y saneamiento inadecuados sobre la enfermedad diarreica en lugares con ingresos medios y bajos.
La estrategia de búsqueda utilizó la Biblioteca de Cochrane, MEDLINE & PubMed, Global Health, Embase y BIOSIS suplementado con una búsqueda en las listas de las referencias de revisiones sistemáticas previamente publicadas, con el fin de identificar estudios en los que se reportasen intervenciones que examinaran el efecto de mejoras en el agua para consumo y el saneamiento en lugares con ingresos medios y bajos, publicados entre 1970 y Mayo 2013. Los estudios elegibles incluían ensayos aleatorizados controlados, ensayos cuasi-aleatorizados con grupo control, estudios observacionales pareados y estudios observacionales con un grupo control en donde la intervención estaba bien definida. El riesgo de sesgo se evaluó utilizando la escala de Ottawa-Newcastle modificada. Los resultados del estudio se combinaron utilizando el meta-análisis y la meta-regresión para derivar los cálculos específicos de riesgo generales y específicos de la intervención.
De 6,819 registros identificados para agua de consumo humano, 61 estudios cumplieron con los criterios de inclusión y de 12,515 registros identificados para saneamiento, se incluyeron 11 estudios. En general, las mejoras en el agua para consumo y el saneamiento estaban asociadas con una disminución en el riesgo de diarrea. Las mejoras específicas, tales como el uso de filtros de agua, el suministro de agua de calidad mediante conexión al acueducto y al alcantarillado, estaban asociadas con mayores reducciones en la diarrea, comparadas con otras intervenciones.
Los resultados muestran que un suministro de agua y saneamiento inadecuados estaban asociadas con riesgos considerables de enfermedad diarreica, y que existen diferencias notables en la reducción de la enfermedad según el tipo de mejoras en el agua y el saneamiento implementados.
Diarrhoea is among the main contributors to global child mortality, causing one in ten child deaths (WHO 2009; Liu et al. 2012), and inadequate water and sanitation have long been associated with diarrhoea (Esrey & Habicht 1986; Esrey et al. 1991; Clasen et al. 2006, 2010; Waddington et al. 2009; Cairncross et al. 2010). In 2011, 11% of the world population reported using ‘unimproved’ drinking water supplies (defined as unprotected springs and dug wells, surface water and water stored in a tank) and 36% had ‘unimproved’ sanitation (defined as flush toilets not connected to a sewer or septic system, pit latrines without slab, bucket latrines or open defecation). ‘Improved’ and ‘unimproved’ drinking water and sanitation refer to specific sources and facilities as defined by the WHO/UNICEF Joint Monitoring Programme (JMP 2013) and are often taken as proxy indicators for appropriate and inappropriate water and sanitation. ‘Inadequate’ water and sanitation, as we define it for the purpose of this manuscript, means any drinking water or sanitation provision whose use poses a risk to health, which cannot be used safely, which is not available in sufficient quality or quantity or which is too distant for convenient access.
The 2010 Global Burden of Disease Study (GBD), by Lim et al. (2012), concluded that the impact of water and sanitation on diarrhoea was much smaller than previous GBD estimates (Prüss et al. 2002; Clasen et al. 2014). Their conclusion, based on a yet-to-be published systematic review, was that there was an increased risk of diarrhoea associated with unimproved water (RR 1.34, 95% CI 1.02–1.72) and unimproved sanitation (RR 1.33, 95% CI 1.02–1.74). They reported no additional benefit, however, from improved water quality or access over other improved water sources (such as public taps, protected springs or dug wells, boreholes and rainwater) after adjusting for potential bias due to lack of blinding (Lim et al. 2012; Engell & Lim 2013).
The 2010 GBD conclusions, with respect to the health impact associated with water and sanitation, represent a significant departure from previous estimates. This review was undertaken to update previous research and to explore the impact of other methods to adjust for non-blinding. Meta-regression was used to explore the impact of different types of improvement to drinking water or sanitation, as well as other study characteristics. The methods are described in line with the ‘Preferred Reporting Items for Systematic Reviews and Meta-Analysis’ (PRISMA) guideline (Moher et al. 2009) and include a PRISMA checklist (Online-only Appendix 1).
The objective of this study was to estimate the effect of different water and sanitation interventions on diarrhoeal disease morbidity, based on pooled estimates from existing studies. The protocol for this study was agreed, in advance, by an expert group convened by the World Health Organization (WHO) before the searches began.
Studies were sought that reported the effects on diarrhoea at the individual, household or community level of any drinking water or sanitation intervention providing they could be grouped within our conceptual models for drinking water and sanitation (Figures 1 and 2). Eligible study designs included:
Studies were excluded if they mainly targeted institutions such as schools or the work place, or if they used non-representative population groups (e.g. people with HIV). We excluded studies in which the rate of implementation of the intervention was very low and studies that had very low compliance (<20%). A poor implementation rate might be reflected in similar rates of uptake in intervention and control groups: changes in morbidity cannot then confidently be attributed to the water or sanitation source or technology. As an example, Boisson et al. (2009) tested a novel portable water filter technology, but it was reportedly used by only 13% of the participants, and the authors themselves conclude that the health effect was likely not be due to the intervention and we excluded the study. Other studies in which interventions did not lead to differences in drinking water or sanitation access between intervention and control groups included Pradhan and Rawlings (2002) for household sanitation and Walker et al. (1999) for drinking water.
A wide range of single and combined water and sanitation interventions were eligible. Studies were included with participants of all ages from low- and middle-income settings. Due to the limited number of studies reporting mortality, studies had to report our primary outcome of diarrhoeal disease morbidity, regardless of aetiology and case confirmation. The main definition for diarrhoea was the WHO standard of at least three loose stools passed in the previous twenty-four hours (WHO 2005), but alternative case definitions were permitted providing that they could be assessed for validity. Studies were required to be published in a peer-reviewed journal or to have been assessed according to transparent criteria for methodological quality in a previously published systematic review.
Five databases were searched (Cochrane Library, MEDLINE & PubMed, Global Health, Embase and BIOSIS) in May 2013, using keyword and Medical Search Headings. The search terms and strategy are outlined in Online-only Appendix 2. In addition, reference lists of key articles (previously published systematic reviews and an unpublished literature review conducted by WHO) were examined, and subject experts and study authors were contacted to provide additional information and further relevant references where required. The search strategy was prepared and implemented in English, and only reports in English or French were considered. However, if a study published in a language other than English or French had been included in a previously published English or French language systematic review and the relevant data had been extracted and made available, this study was included in our analysis.
Titles and abstracts were screened by a single reviewer, and data extraction and quality assessment was carried out by two independent reviewers, using a structured and piloted form. Differences between reviewers over data extraction and quality assessment were reconciled with the intervention of a third abstractor, where required. The quality assessment criteria were adapted from the Newcastle–Ottawa scale (Wells et al. undated) by Pope et al. (2010) for assessing the quality of studies for the health effects of interventions to reduce indoor air pollution. Specific quality criteria were adapted to each study type (intervention, cohort, case–control, cross-sectional) to assess the risk of sampling bias, bias in exposure and outcome measurements, bias in results analysis and reporting. The criteria are included in the data extraction form (Online-only Appendix 3).
The summary effect estimates were calculated as risk ratios (RR) with 95% confidence intervals (95% CI). For studies with multiple intervention arms (including factorial trials), we derived a single pair-wise comparison of the most comprehensive intervention compared with the least comprehensive intervention (or control) among the categories indicated in Figures 1 and 2, subject to availability of results. Where possible, we combined data across intervention arms falling within the same category (e.g. different methods for filtering at point of use). Whenever possible, effect estimates adjusted for clustering at household or community level were extracted.
Random-effects meta-analyses were conducted to examine, separately, the effect of improvements in drinking water or sanitation on diarrhoeal morbidity. Bayesian meta-regression was used to estimate the impact of different intervention types, baseline water and sanitation conditions and additional study characteristics (Thompson 1994). Other pre-specified covariates were retained in the model if the P-value was smaller than 0.2 or if they changed effect estimates of other variables by at least 15% (Kirkwood & Sterne 2003; McNamee 2003).
Systems for drinking water and sanitation provision lie on a continuum between poor and good supply/quality/facilities. Studies were grouped into categories according to the nature of the improvement, following conceptual models, as shown in Figures 1 and 2 and described in subsequent sections.
As a sensitivity analysis, 20% of studies with the lowest quality rating were excluded. For community- and household-level water interventions, separate sensitivity analyses were conducted as the studies tend to have different characteristics (with household-level interventions, for example, tending to be randomised controlled trials, while community-level interventions are often of a lower quality design – Clasen et al. 2006).
Potential for publication bias was examined with inspection of funnel plots and the use of Begg's and Egger's test. Analyses were performed with Stata 12 (StataCorp. 2011. Stata Statistical Software: Release 12. College Station, TX: StataCorp. LP). Bayesian meta-regression and bias adjustments were performed using WinBUGS (Lunn et al. 2000).
The conceptual model used for the analysis of drinking water interventions is presented in Figure 1. Interventions were grouped as community-level (structural changes in supply) or household-level interventions (point-of-use treatment). Within point-of-use treatments, chlorine, solar disinfection and filter interventions were analysed separately. Within community-level interventions, studies were grouped according to whether the intervention led to an improved water source other than piped water (piped water means piped into premise throughout the article), a basic piped water source or a piped water source with a continuous supply and safe quality (referred to as higher-quality piped water).
We distinguish between ‘basic piped water’ and ‘piped water, continuous and safe quality’. Practically, in all interventions providing piped water to households or premises, piped water was of non-optimal quality and/or supply was non-continuous requiring water storage in the households. The endpoint of these studies was therefore classified as ‘basic piped water’. A ‘piped water source, continuous and safe quality’ is similar to the standard water supply in high-income countries. Studies of interventions that provide a continuous piped water supply of high water quality are currently not available for low- and middle-income settings besides one study (Hunter et al. 2010), which may come closest to the supplies typically encountered in high-income countries. We therefore approximated the transition from ‘basic piped water’ to ‘piped water, continuous and safe quality’ by the effect of safe water storage plus the effect of any quality improvements on a piped water system.
In Figure 1, the transitions a to f represent ‘basic parameters’ in the meta-regression model, each represented by a covariate. All other transitions are coded as combinations of these basic parameters: specifically, r = b – a, s = c – b, t = c – a, u = d – a, v = e – a and w = f – a. The model allows the indirect estimation of transitions that have not been directly observed (including those representing basic parameters), following ideas of network meta-analysis (Salanti et al. 2008).
Safe water storage in the household is an important component to prevent contamination and maintain adequate water quality (WHO 2013a). The effect of safe water storage was estimated by including a binary covariate to indicate either:
The following further study characteristics were explored in meta-regression analyses:
Studies where participants were blinded to point-of-use water quality interventions have consistently failed to show a statistically significant effect on diarrhoeal disease. As there are only three blinded household water interventions in low- and middle-income settings that meet the inclusion criteria (Kirchhoff et al. 1985; Jain et al. 2010; Boisson et al. 2013), it was felt that these were insufficient to define the potential bias associated with non-blinding. As diarrhoea in intervention studies is usually self-reported and non-blinding in subjectively assessed outcomes has been associated with bias (Wood et al. 2008; Savović et al. 2012), an additional analysis was performed, which incorporated bias adjustments based on empirical evidence (as described by Savović et al. (2012) and outlined below).
As community-level interventions are often less apparent to the recipient (study participant) than household-based interventions, it is likely that community-level interventions will be less prone to bias as a result of non-blinding. This idea is supported by the finding of similar results for community water or sanitation interventions when observational studies (examining survey data) and experimental studies were analysed separately. It is assumed that observational studies, using specific matching methods on survey data, are less prone to bias as a result of non-blinding because there is no single study hypothesis; the hypothesis regarding a potential impact of sanitation or water on diarrhoea would be just one of many possible hypotheses investigated in the survey. Such studies therefore offer an opportunity for limiting bias arising from non-blinding.
Meta-regression was repeated after making a bias adjustment in studies of household-level interventions. The result of each non-blinded study was separately adjusted by introducing bias through a prior distribution in a Bayesian framework (Welton et al. 2009). On the basis of the findings of Savović et al. (2012), who examined the distribution of bias due to lack of blinding in a large-scale meta-epidemiological study, three different prior distributions on size and direction of this bias were explored (Welton et al. 2009). These distributions incorporate variability in bias across studies and across meta-analyses. The prior which best represents the findings of the meta-epidemiological study (Savović et al. 2012) is based on the mean bias and the sum of all variance components. This is the preferred approach for the current analysis, as it will adjust the biased studies and should appropriately down-weight them. More information on bias adjustment for non-blinding and results with the other two prior distributions on size and direction of this bias are outlined in Online-only Appendix 4.
Sanitation studies were grouped and analysed according to the conceptual model in Figure 2. We examined, in particular, the possibility of a differential effect of sewer connections over basic household improved sanitation (defined here as all other improved sanitation besides sewer connection). The following study characteristics were explored:
For water, of 6751 records identified through database searches and a further 68 identified through other sources, 3672 records were screened (after de-duplication) and 110 full-text articles were assessed for inclusion of which 61 were included for the meta-regression analysis.
For sanitation, of 12 502 records identified through database searches and a further 13 identified through other sources, 10 057 records were screened (after de-duplication) and 34 full-text articles were assessed for inclusion of which 11 were included for the meta-regression analysis. Figures 3 and 4 provide study flow diagrams of the number of studies screened and assessed for eligibility and included in the review. The Online-only Appendix 5 presents the citation, definitions and characteristics for all studies included in this analysis.
We included 68 comparisons of 61 individual studies from low- and middle-income settings. The number of observations describing each link between study baseline and outcome is listed in Table 1.
|Baseline water||Outcome water||Comparisons||Transition (Figure 1)|
|Unimproved source||Improved community source||8||a|
|Unimproved source||Piped water||4||b|
|Improved community source||Piped water||7||r|
|Piped water||Higher-quality piped water||1||s|
|Unimproved source||POU chlorine treatment||16||d|
|Unimproved source||POU solar treatment||6||e|
|Unimproved source||POU filter treatment||14||f|
|Improved community source||POU chlorine treatment||4||u|
|Improved community source||POU solar treatment||5||v|
|Improved community source||POU filter treatment||3||w|
The summary risk ratio of all observations from the water interventions (all transitions), in a random-effects meta-analysis of the data, is 0.66 (0.60–0.71). Tables 2 and 3 show the results for individual transitions from the meta-regression analysis without and with bias adjustment for non-blinding.
|Baseline water||Outcome water|
|Improved community source||Basic piped water||Piped water, higher qualitya||Chlorine/solar+safe storage||Filter+safe storage|
|Unimproved source||0.89 (0.78, 1.01)||0.77 (0.64, 0.92)||0.19 (0.07, 0.50)||0.82 (0.69, 0.96)||0.53 (0.41, 0.67)|
|0.63 (0.55, 0.72)||0.41 (0.33, 0.50)|
|Improved community source||0.86 (0.72, 1.03)||0.21 (0.08, 0.56)||0.92 (0.76, 1.10)||0.59 (0.45, 0.78)|
|0.71 (0.61, 0.82)||0.46 (0.36, 0.58)|
|Basic piped water||0.25 (0.09, 0.65)||1.07 (0.84, 1.34)||0.69 (0.51, 0.93)|
|0.82 (0.67, 1.01)||0.53 (0.40, 0.69)|
|Baseline water||Outcome water|
|Improved community source||Basic piped water||Piped water, higher qualitya||Chlorine/solar+safe storage||Filter+safe storage|
|Unimproved source||0.89 (0.78, 1.01)||0.77 (0.64, 0.92)||0.21 (0.08, 0.55)||0.99 (0.76, 1.27)||0.66 (0.47, 0.92)|
|0.84 (0.61, 1.16)||0.55 (0.38, 0.81)|
|Improved community source||0.86 (0.72, 1.03)||0.23 (0.09, 0.62)||1.11 (0.85, 1.44)||0.74 (0.52, 1.05)|
|0.94 (0.68, 1.30)||0.62 (0.42, 0.93)|
|Basic piped water||0.27 (0.10, 0.71)||1.29 (0.95, 1.74)||0.85 (0.58, 1.25)|
|1.09 (0.76, 1.56)||0.72 (0.47, 1.11)|
The results from multivariable meta-regression before adjusting for non-blinding were nearly identical between Stata and WinBUGS. The results for chlorine and solar interventions were very similar and so, for convenience, they were combined in all analyses (in the context of Figure 1, this corresponds to setting d = e and hence u = v). Covariates retained in the model were provision of safe water storage and whether the intervention was a combined intervention.
Bias adjustment for non-blinding down weighs mainly estimates for point-of-use water treatment, higher-quality piped water and provision of safe water storage (Table 3).
The multivariable meta-regression model explained 53% of the between-study variance. Improved over unimproved sources are associated with only small reductions in diarrhoea, with a larger effect for piped water compared with other improved sources. The biggest protective effect on diarrhoeal disease was found for higher-quality (i.e. continuous and safe quality) piped water. Among household-level studies, filter interventions that also provided safe storage (for example, ceramic filters) were associated with a large reduction in diarrhoeal disease. After taking account of safe water storage, the effects of ceramic and biosand filters were not significantly different from each other and so are grouped for further analysis. Chlorine and solar interventions did not appear to reduce diarrhoeal disease risk (applied to either unimproved or improved sources) after results were adjusted for non-blinding. There was some evidence of a greater diarrhoea risk reduction from improving household water storage and combining the water intervention with hygiene education and/or improved sanitation than through the water intervention alone (see footnotes of Table 2 and 3).
We included 14 comparisons from low- and middle-income settings. Twelve observations compared improved sanitation facilities (other than sewer connections) with unimproved sanitation, and two observations had sewer connections as their outcome.
The final model explained 97% of the between-study variance. The overall relative risk for improved over unimproved sanitation on diarrhoea, based on meta-analysis, was 0.72 (0.59, 0.88). The results of multivariable meta-regression are shown in Table 4. A larger association between sewer interventions and reduction in diarrhoea was observed compared with other improved sanitation.
|Baseline sanitation||Outcome sanitation|
|Improved sanitation, no sewer||Sewer connectiona|
|Unimproved sanitation||0.84 (0.77, 0.91)||0.31 (0.27, 0.36)|
|Improved sanitation, no sewer connection||0.37 (0.31, 0.44)|
Excluding 20% of studies with the lowest quality rating did not significantly change estimates, either for the water or the sanitation analysis. Funnel plot asymmetry was observed among the studies of household-level water quality interventions, which may be due to publication bias. There was no evidence of funnel plot asymmetry in studies of community-level water or sanitation improvements with or without sewer interventions. Funnel plots and results of statistical tests examining evidence for publication bias are shown in the Online-only Appendix 4.
Water and sanitation intervention studies typically report diarrhoeal levels in children up to 5 years of age, with impacts in other age groups less frequently reported. Data on other age groups were extracted wherever possible, and the results for all ages compared with children under five. The effect estimates were found to be very similar and mostly within the confidence interval of the under-five age group. It has therefore been assumed that the estimates derived here can be used for all ages.
The results show that there are large potential reductions in diarrhoeal disease risk through improvements to both water and sanitation in low- and middle-income settings. For water, the most effective household-level intervention was found to be a point-of-use filter in combination with safe water storage. At the community level, introduction of high-quality piped water (i.e. water supplied continuously to the household of good microbial water quality) was found to be most effective. There were also differences in the impact of sanitation interventions, and there is evidence that sewer interventions are associated with a greater reduction in diarrhoea than basic household sanitation. These results are largely consistent with previously published reviews, where provision of improved community water supply was associated with a limited reduction in diarrhoeal illness (Waddington et al. 2009), and some water quality interventions (especially water filters) had a significant impact on reducing illness (Clasen et al. 2006; Hunter 2009; Waddington et al. 2009; Cairncross et al. 2010). This is also true for sanitation, as sanitation interventions in previous analyses have been shown to reduce diarrhoea by 30–40% (Waddington et al. 2009; Cairncross et al. 2010), with a larger effect observed for sewer connection (Norman et al. 2010).
The effect estimates for higher-quality piped drinking water and sewer connection should, however, be treated with caution. We approximated the transition from ‘basic piped water’ to ‘piped water, continuous and safe quality’ by the effect of safe water storage plus the effect of any quality improvements on a piped water system. We acknowledge that this is likely an underestimate as it accounts for the quality aspect but not any benefits derived through greater water access and its impact on, for example, personal hygiene. Source water quality improvement on piped water was estimated from one single study (Hunter et al. 2010), although the results are consistent with evidence from high-income countries (Payment et al. 1991; Colford et al. 2009). The effect of a sewered system was derived from two observations (Pradhan & Rawlings 2002; Moraes et al. 2003). Given the small number of observations used to derive these results, generalisation should be made only with caution. For example, in the intervention study that provided source water quality improvements on a piped water supply (Hunter et al. 2010), it is possible that the baseline piped water may have been of poorer quality than ‘typical’ piped water in low- and middle-income settings. However, reclassifying the baseline water in this study as unimproved in the analysis barely changed the diarrhoeal effect estimates. Given the limited evidence base, it is likely that these estimates may change considerably as additional evidence becomes available. They do, however, indicate the large potential benefits of improving water and sanitation and call for a disaggregation of the ‘improved’ levels defined by JMP (JMP 2013).
The finding of potentially important disease reduction beyond improved non-piped and also basic piped water sources is eminently plausible. Water from those improved sources is frequently contaminated during collection, transport and household storage (Wright et al. 2004; Rufener et al. 2010). Household piped water in low- and middle-income settings is frequently non-continuous (e.g. Brown et al. 2013) which presents two microbial risks, namely infiltration into non-pressurised distribution systems and recontamination or growth during household storage. In addition, community and non-continuous household water supply may reduce the amount of water available for hygiene purposes. Water availability and distance to the water source are both associated with risk of diarrhoea (Wang & Hunter 2010; RSS 2011; Pickering & Davis 2012). Reliable at-home water supplies were shown to increase water availability and key hygiene practices (Evans et al. 2013). The current analysis further suggests that improved water storage is associated with decreased risk of diarrhoea; a finding which has been previously described (Roberts et al. 2001; Günther & Schipper 2013). The beneficial effect of filters over both unimproved and improved sources remained significant and substantial after bias adjustment for non-blinding. This may reflect the fact that even water from improved sources is frequently of poor quality (Bain et al. 2012, 2014; Wolf et al. 2013). The smaller effect seen from chlorine and solar treatments could be explained if a significant proportion of diarrhoea episodes was caused by agents that are less susceptible to those treatments, non-exclusive use (Mäusezahl et al. 2009), and/or there is low uptake (compliance) of the intervention (as the need for adequate compliance has been shown in previous epidemiological modelling – Hunter et al. 2009; Brown & Clasen 2012; Enger et al. 2013).
Household members with improved sanitation may still be exposed to high levels of pathogens from faecal material if their neighbours have no improved sanitation (Root 2011; Baker & Ensink 2012), or when on-site sanitation is not managed hygienically. In urban areas, especially, latrines have been observed to fill and overflow, which can lead to major contamination of the surrounding area (Carter 2013). Introduction of sewered sanitation at large scale in urban areas in low- and middle-income settings would be expected to have a positive impact on health, although care must be taken that sewage is appropriately treated to avoid the diarrhoeal disease burden being shifted ‘downstream’ to the receiving communities (Baum et al. 2013). As such, it is acknowledged that sewered systems with appropriate sewage treatment are costly, and in some settings, decentralised systems for managing on-site sanitation may be more cost-effective and appropriate (Norman et al. 2010).
Effect estimates are from heterogeneous interventions and therefore only approximate the impact of improving water and sanitation on diarrhoea. Study quality is generally low which confirm previous analyses (Waddington et al. 2009; Cairncross et al. 2010; Clasen et al. 2010). Blinding and randomisation of study participants in water and sanitation interventions is often not possible and sometimes may not be desirable as blinding could negatively influence compliance and community dynamics which are important components for the adoption of interventions (Hartinger et al. 2011). Sanitation studies especially are often quasi-randomised (Capuno et al. 2011; Kumar & Vollmer 2013) which can introduce bias. Additionally, some point-of-use interventions have been shown to have low acceptability to the population (Boisson et al. 2009; Luoto et al. 2012) leading to poor adoption, and even an effective point-of-use treatment will have little impact on health if it is not consistently applied (Enger et al. 2013). In addition, few point-of-use interventions are effective against all typical classes of pathogens, and post-treatment contamination is frequent (Wright et al. 2004; Stauber et al. 2012). Even piped water interventions frequently provide low-quality non-continuous water which therefore requires storage, point-of-use treatment or the use of alternative water sources (Wang et al. 1989; Brown et al. 2013). Better quality water and sanitation interventions showed greater effectiveness in reducing diarrhoeal disease (Clasen et al. 2006). An attempt was made to account for some of these limitations by exploring health impacts beyond basic improved water and sanitation, by the use of specific bias adjustments and different sensitivity analyses.
We applied a bias adjustment to account for non-blinding, based on the findings of Savović et al. (2012). These, however, are based on clinical interventions, and there is little evidence that the pooled estimated bias is representative for the type of interventions covered in this article. The estimate is, however, specific to subjectively assessed outcomes (such as self-reported diarrhoea), and we believe that it represents the best currently available evidence on the effect of bias due to non-blinding.
Currently, only the impact of water and sanitation on diarrhoeal morbidity has been considered. Many other health effects (such as intestinal parasite infections, impaired nutritional status and possibly environmental enteropathy) have been associated with inadequate water and sanitation (Korpe & Petri 2012; Ziegelbauer et al. 2012; Dangour et al. 2013; Lin et al. 2013). Furthermore, inadequate water and sanitation have been associated with reduced school attendance (Freeman et al. 2012) and personal security issues, especially for women (Bapat & Agarwal 2003; Talaat et al. 2011). Unfortunately, quantitative evidence on these effects is currently very limited.
Meta-regression yields observational associations between variables, and is therefore prone to bias (Thompson & Higgins 2002). Use of water sources and sanitation facilities was defined at study level, although it may vary within the community. This can underestimate the true baseline or outcome effect. To include access as a continuous variable is currently not possible as many studies omit this information.
The choice of what level of water and sanitation to consider as representing the highest attainable degree of safety (i.e. the counterfactual) has major implications in terms of the burden of disease that is attributable to inadequate water and sanitation. The analysis demonstrates health benefits beyond those achievable with basic improved water and sanitation, and it seems that health gains can be maximised when high-quality drinking water is available in sufficient quantities in the home and the sanitation system effectively prevents exposure to faecal material (through isolation and/or appropriate treatment). Thus, the results suggest that use of facilities defined as ‘improved’, as used in the 2010 GBD study (Lim et al. 2012), should not be construed as use of fully safe and adequate water and sanitation, devoid of an associated disease burden.
Service levels are frequently lower in low- and middle-income countries than those in high-income countries, but it is suggested that high-level services could represent a reference against which the risk for lower levels of water and sanitation could be estimated. Even defining high-level water services (i.e. high-quality water piped continuously to the home) as the counterfactual may lead to underestimates of the burden of disease. In Iceland, for example, the introduction of water safety plans was associated with a significant reduction of diarrhoea in the population (Gunnarsdottir et al. 2012). Also, tap water in California, USA, meeting all the required quality standards, was still associated with gastrointestinal illness (Colford et al. 2009). However, at present, data limitations preclude the setting of even higher counterfactuals for water and sanitation.
The systematic literature reviews and analyses reported in this paper have led to the identification of areas where evidence is missing on the linkages between water, sanitation and health. It is believed that effect estimates from meta-analyses would greatly benefit from more well-conducted and reported water and sanitation intervention studies complying to, for example, the CONSORT Statement for randomised controlled trials (Schulz et al. 2010) or the STROBE Statement for observational studies (Von Elm et al. 2007). Studies applying a factorial design might be a promising approach to assess different interventions simultaneously and, given a sufficiently large sample size, interactions between different WASH interventions (Montgomery et al. 2003). Studies reporting consistently not only on health outcome but also on implementation and compliance would enable inclusion of this information in future analyses. Additionally, research on underlying factors that strengthen intervention implementation and increase people's acceptance, adoption and sustained use is still rare. Improved methods for using natural experiments or pre-existing development interventions, in which exposure is not artificially manipulated, also seem to be a promising way forward (Arnold et al. 2009; Craig et al. 2012). Furthermore, impacts resulting from inadequate water and sanitation other than diarrhoea morbidity are currently under-researched. More evidence on these topics would enable more meaningful estimates of the potential health benefits of improving water and sanitation to be made.
Inadequate drinking water and sanitation are associated with considerable risks for diarrhoeal disease. The choice of a suitable approach that can differentiate health effects between different improvements in water and sanitation relative to the baseline is crucial for meaningful estimates. However, evidence from well-conducted intervention studies assessing exclusive use of adequate access and supply of safe water or universal use of effective sanitation is still very limited.
The study was partially funded by the United Kingdom Department for International Development (DFID). The funder had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Thomas Clasen has participated in research and consulting services supported by Unilever and Vestergaard-Frandsen, which manufacture and sell household or other point of use water filtration devices. We would like to thank Wolf-Peter Schmidt and Sophie Boisson from the London School of Hygiene and Tropical Medicine for their advice and inputs into the review component. Some authors are staff members of the World Health Organization (WHO) or other institutions. The authors alone are responsible for the views expressed in this publication, which do not necessarily represent the views, decisions or policies of the WHO, DFID or other institutions. This article should not be reproduced for use in association with the promotion of commercial products, services or any legal entity. The WHO does not endorse any specific organisation or products. Any reproduction of this article cannot include the use of the WHO logo.