Plain language summary
Interventions to improve water quality, particularly when implemented at the household level, are effective in preventing diarrhoea in settings where it is endemic
Diarrhoea is a major cause of death and disease, especially among young children in low-income countries. Loss of fluid (dehydration) is the major threat, though diarrhoea also reduces the absorption of the nutrients, causing poor growth in children, reduced resistance to infection, and potentially long-term gut disorders. This review examined trials of interventions to improve the microbiological quality of drinking water. These include conventional improvements at the water source (eg protected wells, bore holes, and stand posts) and point-of-use interventions at the household level (eg chlorination, filtration, solar disinfection, and combined flocculation and disinfection). The review covered 38 independent comparisons from 30 trials that involved more than 53,000 people. In general, such interventions were effective in reducing episodes of diarrhoea. Household interventions were more effective in preventing diarrhoea than those at the source. However, differences in the interventions and the settings in which they were introduced, as well as the methods and measurements of effect, limit the extent to which generalizations can be made. Further research, including blinded trials and longer-term assessments, is necessary to understand the full impact of these interventions.
Diarrhoeal disease, disease agents, and pathways
Diarrhoeal diseases kill an estimated 1.8 million people each year (WHO 2005). Among infectious diseases, diarrhoea ranks as the third leading cause of both mortality and morbidity (after respiratory infections and HIV/AIDS), placing it above tuberculosis and malaria. Young children are especially vulnerable, bearing 68% of the total burden of diarrhoeal disease (Bartram 2003). Among children less than five years, diarrhoea accounts for 17% of all deaths (United Nations 2005). For those infected with the human immunodeficiency virus (HIV) or who have developed acquired immunodeficiency syndrome (AIDS), diarrhoea can be prolonged, severe, and life-threatening (Hayes 2003).
Diarrhoea is a symptom complex characterized by stools of decreased consistency and increased number. The clinical symptoms and course of the disease vary greatly with the age, nutritional status, and immunocompetence of the patient, and the aetiological agent infecting the intestinal system and interfering with normal adsorption. Most cases resolve within a week, though a small percentage continue for two weeks or more and are characterized as 'persistent' diarrhoea. Dysentery is a diarrhoeal disease defined by the presence of blood in the liquid stools (Blaser 1995). About 35% of the deaths from diarrhoea in children less than five years old are believed to be attributable to acute non-dysenteric diarrhoea, with 45% from persistent diarrhoea and 20% from dysentery (Black 1993). Though epidemic diarrhoea such as cholera and shigellosis (bacillary dysentery) are well-known risks, particularly in emergency settings, their global health significance is small compared to endemic diarrhoea (Hunter 1997).
The immediate threat from diarrhoea is dehydration − a loss of fluids and electrolytes. Thus, the widespread promotion of oral rehydration therapy (ORT) has significantly reduced the case-fatality rate associated with the disease. Such improvements in case management, however, have not reduced morbidity, which is estimated at four billion cases annually (Kosek 2003). And since diarrhoeal diseases inhibit normal ingestion of foods and adsorption of nutrients, continued high morbidity is an important cause of malnutrition, leading to impaired physical growth and cognitive function (Guerrant 1999), reduced resistance to infection (Baqui 1993), and potentially long-term gastrointestinal disorders (Schneider 1978).
The infectious agents associated with diarrhoeal disease are transmitted chiefly through the faecal-oral route (Byers 2001). A wide variety of bacterial, viral, and protozoan pathogens excreted in the faeces of humans and animals are known to cause diarrhoea. Among the most important of these are Escherichia coli, Salmonella sp., Shigella sp., Campylobacter jejuni, Vibrio cholerae, rotavirus, norovirus, Giardia lamblia, Cryptosporidium sp., and Entamoeba histolytica ( Leclerc 2002). The importance of individual pathogens varies between settings, seasons, and conditions. While bacterial agents as a group are believed to cause a majority of diarrhoeal disease in developing countries, viral and protozoan agents tend to cause more cases in developed countries (Hunter 1997).
Many of these diarrhoegenic agents are potentially waterborne − transmitted through the ingestion of contaminated water. However, most of the same pathogens are also transmitted by ingestion of contaminated food and other beverages, by person-to-person contact, and by direct or indirect contact with infected faeces. Because of this variety of pathways, interventions for the prevention of diarrhoeal disease not only include enhanced water quality but also steps to improve the proper disposal of human faeces (sanitation), increase the quantity and improve access to water (water supply), and promote hand washing and other hygiene practices within domestic and community settings (hygiene).
While water quality is also adversely impacted by chemical contaminants, the level of disease associated with metals, nitrates, organics, and other chemicals is usually small relative to infectious diarrhoea (WHO 2002). Other important diseases associated with drinking water, such as hepatitis A and E, poliomyelitis, gastroenteritis and typhoid fever, may not cause diarrhoea but are nevertheless associated with potentially waterborne microbes of faecal origin. For this reason, efforts to assess drinking water quality focus primarily on faecal pathogens (WHO 1993).
Because of the difficulty of monitoring water for the presence of all such agents, an indirect approach has been adopted where water is examined for indicator bacteria whose presence implies some degree of contamination. While there is controversy over the preferred indicator (Gleeson 1997), even those that accept the use of the coliform group use different target indicators (total coliforms, thermotolerant coliforms, E. coli) and different methods for assaying the level of indicator present (membrane filtration, multiple tube/most probable number) (Clesceri 1998).
Water quality and diarrhoea
Health authorities generally accept that microbiologically safe water plays an important role in preventing outbreaks of waterborne diseases (Hunter 1997). Accordingly, the most widely accepted guidelines for water quality allow no detectable level of harmful pathogens at the point of distribution (WHO 2004).
However, an estimated 1.1 billion people lack access to improved water supplies (WHO/UNICEF 2000). In settings that are not served by reliable water treatment and distribution systems, diarrhoeal disease is often endemic, that is, present or usually prevalent in the population at all times. In such settings much of the epidemiological evidence for increased health benefits following improvements in the quality of drinking water has been equivocal (Esrey 1986; Lindskog 1987; Cairncross 1989). Because of the multiple pathways of diarrheogenic infection, improvements in water quality alone may not necessarily interrupt transmission (Briscoe 1984). There are also questions about the methods and validity of studies designed to assess the health impact of such interventions (Blum 1983; Briscoe 1986; Imo State Team 1989).
As part of a larger evaluation of interventions for the control of diarrhoeal disease (Feachem 1983), in 1985, Esrey and colleagues reviewed studies to determine the health impact from improvements in water supplies and excreta disposal facilities (Esrey 1985). They updated the review in 1991 and expanded it to include studies addressing a variety of specific pathogens associated with poor water and sanitation (Esrey 1991a). For almost two decades, these reviews have provided guidance on the relative reduction in diarrhoeal disease that was believed to be possible through improvements in water quality, water quantity, sanitation, and hygiene. Important as these reviews have been, there are reasons to consider anew the extent to which interventions to improve water quality impact diarrhoeal disease. This is largely the result of new evidence from interventions at the household level (Clasen 2004a). However, even the league tables in Esrey's reviews comparing the relative impact of various types of environmental interventions, enticingly simple as they are, may obscure the potentially more important finding − the wide range among the studies in the measure of effect. In the case of water quality improvements, for example, Esrey and colleagues cited a median reduction in diarrhoea disease from nine studies of 16%, but a range in effect from 0% to 90%. Because Esrey and colleagues had a relatively small number of studies on water quality interventions, they could not use subgroup analysis to explore some of the potential reasons for this wide range of effect (Mintz 2001; Clasen 2004a).
An update of Esrey's reviews addresses some of these shortcomings (Fewtrell 2005). By using subgroup analysis, for example, Fewtrell and colleagues found important differences in effectiveness of the intervention based on the point of treatment (source versus household). They also observed that interventions were effective even in the absence of improved sanitation (a new finding that challenged the view expressed by Esrey 1986 and VanDerslice 1995), as well as the apparent absence of a cumulative effect from multiple environmental interventions. At the same time, with respect to interventions to improve water quality, the review omitted a number of studies that would seem to have met the inclusion criteria. Moreover, the review presented certain methodological issues, such as the inclusion of observational studies (Ghannoum 1981; Sathe 1996; Iijima 2001), studies where the outcome was other than endemic diarrhoea (Ghannoum 1981; Iijima 2001; Colwell 2003), and the homologous treatment of studies with different measures of effect in their meta-analysis.
A number of interventions have been developed to improve the microbiological quality of water and can be grouped into four main categories.
Physical removal of pathogens (eg filtration, adsorption, or sedimentation).
Chemically treating water to kill or deactivate pathogens, most commonly with chlorine.
Disinfection by heat (eg boiling or pasteurization) and ultraviolet (UV) radiation, either using the sun (solar disinfection) or an artificial UV lamp.
Combination of these approaches (eg filtration or flocculation combined with disinfection).
Water quality can also be enhanced by protecting it from recontamination, for example, by residual disinfection, piped distribution, and safe storage. A combination approach is also common in conventional systems since individual approaches are not effective against the full range of microbial pathogens under all water conditions. Mechanical removal of viruses, for example, presents a challenge to most filters due to their submicron size. Similarly, certain encysted protozoa are resistant to chemical disinfection. The microbiological performance of these approaches may also be impacted by the temperature, pH, turbidity, chemical content, and other characteristics of the water.
In higher income countries, and in many urban settings worldwide, drinking water is treated centrally at the source of supply and is distributed to consumers through a network of pipes and household taps. However, such conventional systems involve significant upfront investment and continued maintenance. In remote and low-income settings, water quality may nevertheless be improved at the source by, for example, providing protected groundwater (springs, wells, and bore holes) or harvested rainwater as an alternative to surface sources (rivers and lakes) that are more susceptible to faecal contamination. Microbial water quality may also be improved at the source or other point in the distribution system by chlorination, filtration, and other means. Improving water at the source is also frequently accompanied by improvements in quantity or access to water by increasing the volume or frequency of water delivery or reducing the time spent in collecting water. This may result in significant benefits not only in health but also in economic and social welfare (Hutton 2004). For purposes of this review, any form of treatment at the water source or otherwise prior to the point of use will be referred to collectively as 'source' water treatment.
For those who have access to sufficient quantities of water but whose water is of poor microbiological quality, an alternative is to treat water at the household or other point of use. Such household treatment may minimize recontamination in the home, a well-known cause of water quality degradation (Wright 2004). At the same time, certain household water filters have been associated with adverse health impacts (Payment 1991a). A review commissioned by the World Health Organization (WHO) identified a wide variety of options for household-based water treatment and assessed the available evidence on their microbiological effectiveness, health impact, acceptability, affordability, sustainability, and scalability (Sobsey 2002). Research on the economics of such interventions also suggests that where adequate quantities of water are already available, household-based water treatment is among the most cost-beneficial and cost-effective approaches in preventing diarrhoeal disease (WHO 2002; Hutton 2004). There is also evidence that the vulnerable population to whom such household-based interventions have been targeted will pay all or a portion of the cost of household water treatment products (Clasen 2004c).
To assess the effectiveness of interventions to improve water quality for preventing diarrhoea.
Criteria for considering studies for this review
Types of studies
Randomized and quasi-randomized controlled trials. The unit of randomization may include individuals, families, households, communities, or other clusters.
Types of participants
Children and adults from settings where diarrhoeal disease is endemic.
Types of interventions
Interventions aimed at improving the microbiological quality of drinking water, including steps to improve water quality by removing or inactivating microbiological pathogens (eg filtration, sedimentation, chemical treatment, heat, or UV radiation) and protecting the microbiological integrity of water prior to consumption (eg residual disinfection, protected distribution, or improved storage). An intervention that has shown elsewhere to reduce the quantity or pathogenicity of waterborne microbes is deemed, for purposes of the review, as an intervention to improve water quality, even if the particular study did not record such an improvement by microbiological examination.
We include interventions that combine improvements in water quality with other components such as improvements in water quantity or access, sanitation or hygiene. We have excluded studies of interventions designed to reduce diarrhoea through improvements in sanitation, hygiene, water quantity or water access, but which do not include a water quality improvement.
People who are following their usual practices with respect to drinking water rather than the prescribed intervention, or who have received a different type of intervention. For example, where a protected well or borehole is introduced, controls may be consuming water that is obtained from the previously available sources, often untreated surface waters. In trials involving household water treatment, controls normally procure their water from the same source as the intervention group but have not received the intervention to treat water in the home. Appendix 3 provides details on the control groups for each study.
Types of outcome measures
The WHO definition of diarrhoea is three or more loose or fluid stools (that take the shape of the container) in a 24-hour period (WHO 1993). We defined diarrhoea and an episode in accordance with the case definitions used in each trial.
Note: We excluded trials that had no clinical outcomes; for example, trials that only report on microbiological pathogens in the stool.
Search methods for identification of studies
We attempted to identify all relevant trials regardless of language or publication status (published, unpublished, in press, and in progress).
We searched the following databases using the search terms and strategy described in Appendix 1: Cochrane Infectious Diseases Group Specialized Register (December 2005); Cochrane Central Register of Controlled Trials (CENTRAL), published in The Cochrane Library (2005, Issue 4); MEDLINE (1966 to December 2005); EMBASE (1974 to December 2005); and LILACS (1982 to December 2005).
We searched the following conference proceedings of the following organizations for relevant abstracts: International Water Association (IWA) (1990 to December 2005); and Water, Engineering and Development Centre, Loughborough University, UK (WEDC) (1973 to December 2005).
Researchers and organizations
We contacted individual researchers working in the field and the following organizations for unpublished and ongoing trials: Water, Sanitation and Health Programme of the World Health Organization; World Bank Water and Sanitation Program; UNICEF Water, Environment and Sanitation (WES); and IRC International Water and Sanitation Centre; Foodborne and Diarrhoeal Diseases Branch, Division of Bacterial and Mycotic Diseases, Centers for Disease Control and Prevention (CDC); US Agency for International Development (USAID), including its Environmental Health Project (EHP); and the UK Department for International Development (DFID).
We checked the reference lists of all studies identified by the above methods.
Data collection and analysis
Selection of studies
Thomas Clasen (TC) and Tamer Rabie (TR) independently reviewed the titles and abstracts located in the searches and selected all potentially relevant studies. After obtaining the full articles, we independently determined whether they met the inclusion criteria. Where we were unable to agree, we consulted Sandy Cairncross (SC) and arrived at a consensus. Those potentially relevant studies that were ultimately excluded are listed together with the reason for exclusion in the 'Characteristics of excluded studies'.
Data extraction and management
TC and TR used a pre-piloted form to extract and record the data described in Appendix 2, and TC attempted to contact authors to supply missing data. TC entered the extracted data into Review Manager 4.2.
TC and TR or Wolf-Peter Schmidt (WS) independently extracted, and where necessary calculated, the measure of effect of the intervention on diarrhoea. We extracted and reported the measure of effect as reported by the authors of each trial, whether it be risk ratios, rate ratios, odds ratios, longitudinal prevalence ratios, or means ratios. In this context, longitudinal prevalence is the number of days with diarrhoea divided by the number of days under observation (Morris 1996). In using these various measures of effect, we note the design effect in treating all such measures of effect as equivalent for common outcomes such as diarrhoea and the debate about methodologies for converting such measures of effect into a single measure (Zhang 1998; McNutt 2003). While it would be possible to calculate a single measure of effect for most trials based on the raw study data, we elected not to do so for the following reasons. Although all trials included in the review assess outcomes on an individual level, the unit of randomization is not the individual but a household, group of households, neighbourhood, or village. As described below, most included trials correct for this design effect by adjusting for the inter-cluster variance. Studies of diarrhoeal disease also frequently adjust for other common covariates, including age and repeated episodes within the same participant. Because these adjustments are generally deemed appropriate, a re-calculation of a measure of effect based on raw data would ignore these important adjustments. In order to avoid the homologous treatment of these different measures of effect, we include the pooled measures of effect in the comparisons only across trials reporting the same measure of effect. In the subgroup analyses, when there were too few trials with the same measure of effect, the comparisons show the forest plots only, with no calculation of pooled measures of effect.
As discussed more fully below, a number of the included trials had multiple intervention arms (eg treating water with bleach or with a flocculant and disinfectant) and compared two or more intervention groups against a single control group. In such cases, a meta-analysis that treats each intervention arm as a separate trial results in counting the control group once for each arm. This violates the important principle in the methodology of meta-analysis that each individual be included only once. However, for the reasons noted in the preceding paragraph, it was not possible to return to the raw data from the trials and thus correct for this by dividing the control group. Because this review is largely descriptive, we elected to include all trial arms but to note this problem. We note, however, that this has occurred and the meta-analysis result will be artificially precise.
Assessment of risk of bias in included studies
TC and TR independently assessed the risk of bias in the trials. We classified the generation of allocation sequence − the process used to generate the randomization list − as 'adequate' if the method used is described and the resulting sequences are unpredictable (eg computer-generated random numbers, table of random numbers, coin toss, drawing lots); 'unclear' if stated that the trial is randomized, but the method is not described; or 'inadequate' if sequences could be related to outcomes (eg according to case record number, date of birth, alternation). We classified allocation concealment − the process used to prevent foreknowledge of group assignment − as 'adequate' if the participants or the investigators enrolling participants cannot foresee assignment; 'unclear' if method is not described; or 'inadequate' if participants and investigators enrolling the participants can foresee their upcoming assignment. We classified blinding − whether the participant or outcome assessor is blind to the intervention group − as 'double blind' if the trial uses a placebo or double-dummy technique such that neither the participants nor the assessor knows whether or not the participants receive the intervention; 'single blind' if the participant or the assessor knows whether or not the participant receives the intervention; or 'open' if both participant and assessor know whether or not the participant receives the intervention. We classified the inclusion of randomized participants in the analysis as 'adequate' if 90% or more of all participants randomized to the trial were included in the analysis; 'unclear' if it is not clear what portion of participants randomized to the trial were included in the analysis; or 'inadequate' if less than 90% of all participants randomized to the trial were included in the analysis.
Additionally, we assessed quasi-randomized controlled trials using the following criteria:
1. Comparability of characteristics between intervention and control groups with respect to relevant baseline characteristics such as water quality, diarrhoeal morbidity, age, socioeconomic status, access to water, hygiene practices, and sanitation facilities. We classified this as 'adequate' if no substantial differences were present, 'unclear' if not reported or not known whether substantial differences exist, or 'inadequate' if one or more substantial difference exists.
2. Data collection for intervention and control groups at the same time. We classified this as 'adequate' if data were collected at similar points in time, 'unclear' if the relative timing was not reported or not clear from trial, or 'inadequate' if data were not collected at similar points in time.
We entered the estimates of effect using the generic inverse variance method on the log scale (Higgins 2005a), and analysed the data using Review Manager 4.2.
We performed tests for heterogeneity by visually examining the forest plots and by using the chi-squared test for heterogeneity with a 10% level of statistical significance (Egger 2001) and the I 2 test for consistency (Higgins 2003). In accordance with our protocol, where there was evidence of heterogeneity we performed the following subgroup analyses: age (all ages versus children less than five years old); intervention point (source versus household); intervention type; water quality only versus compound interventions (ie with hygiene message, vessel, improved sanitation, improved supply); ambient water quality (ie water testing results at pre-intervention or of control group based on log scale levels of thermotolerant coliform per 100 mL); compliance with intervention (< 50% versus ≥ 50%), and effectiveness under various water supply, sanitation, and water access conditions. In the subgroup analyses based on water supply, we followed terminology used by the WHO/UNICEF Global Assessment (WHO/UNICEF 2000), using 'unimproved' to extend to unprotected wells or springs, vendor- or tanker-provided water or bottled water, and 'improved' to extend to household connections, public standpipes, boreholes, protected dug wells or springs, or rainwater collection; we categorized trials as 'unclear' with respect to water supply if they contained insufficient information. We used the same definitions from the WHO/UNICEF Global Assessment to classify sanitation conditions as 'improved' (connection to a public sewer or septic system, pour-flush latrine, simple pit latrine, ventilated improved pit latrine) or 'unimproved' (service or bucket latrines, public latrines, open latrines ); where the necessary information was unclear or unreported, we categorized the sanitation facilities as 'unclear'. To subgroup trials based on access to water source, we used the classifications defined by The Sphere Project 2004, classifying access as 'sufficient' if a consistently available source was located within 500 metres, with queuing no more than 15 minutes and filling time for a 20 litres container no more than three minutes, 'insufficient' if any access failed any such criteria, and 'unclear' if such criteria was unreported or unclear. The quantity of water available to study participants was considered 'sufficient' if consisting of a minimum of 15 litres per person per day. We also used subgroup analyses to compare effectiveness based on methodological quality of the trials.
Where appropriate, we used meta-analyses to derive pooled estimates of effect. Because of the substantial heterogeneity in study results, we used the random-effects model (rather than the fixed-effect model) in such pooling. However, because of important differences in trial methodology, settings, and intervention types, we caution that such pooling of results may be misleading.
Finally, we produced a funnel plot to explore publication bias. We chose not to present results from statistical analysis of publication bias since they are not yet fully accepted as clear evidence of publication bias (Egger 2001).
This review assesses the impact of interventions to improve microbial water quality on diarrhoeal disease. Thirty trials covering 38 interventions and more than 53,000 participants met the review's inclusion criteria. Substantial clinical and methodological heterogeneity among the trials allowed only limited pooling of the results in meta-analyses. Our focus, therefore, has been largely descriptive. Where appropriate, however, we have tried to interpret the evidence, while at the same time noting some of the issues that necessarily limit this.
Interventions to improve the microbiological quality of drinking water are protective against diarrhoea, both for people of all ages and for the vulnerable population of children less than five years old. But the evidence is not absolute and the actual level of effectiveness of such interventions varies considerably. Underlying clinical heterogeneity appears to be responsible for the heterogeneity. The aetiology and epidemiology of diarrhoea is complex and variable, and even the portion of diarrhoea that is waterborne is probably different at different times and places (Luby 2004). Nevertheless, the subgroup analyses provide possible explanations for the differences.
First, household interventions, though also varying considerably in results, are considerably more effective at preventing diarrhoea than interventions at the water source. The reviews from Esrey and colleagues (Esrey 1985; Esrey 1991a) included only studies investigating improvements of water quality at the source, not at the household level. The 15% to 17% median reduction in diarrhoea that they reported is within the range of our findings for source-based interventions. The effectiveness of household-based interventions, on the other hand, is significantly greater than those at the source, and is comparable to certain other environmental interventions to prevent diarrhoea, such as improved sanitation, hygiene (hand washing with soap), and improved water supply (Curtis 2003; Fewtrell 2005). Among household interventions, there is some evidence that filtration offers the most consistent and effective results. It is important to note, however, that none of the source-based interventions included in this review involved piped-in household connections. Thus, the effectiveness of the source systems described herein should not be generalized to reticulated systems.
Second, there is some evidence that effectiveness is enhanced by compliance with the intervention. While this may appear intuitive, it suggests a dose-response association between compliance and results that provides additional evidence of a causal relationship. It also implies the need to address compliance as part of any intervention to improve water quality. This may involve both the inherent acceptability and appeal of the hardware components of the intervention as well as programmatic support to increase utilization. To the extent that interventions are deployed at the household rather than community level, this also implies the need to address compliance during routine activities outside the home such as school and work.
Third, the evidence does not suggest that an 'improved' supply of water (ie household connection, public standpipe, borehole, protected dug well, protected spring, rainwater collection) is essential for water quality interventions to prevent diarrhoea. This finding affirms the WHO's strategy to pursue household water treatment and safe storage as a means of accelerating the health gains of safe drinking water, even though it may not reduce the 1.1 billion currently without access to improved water supplies.
Fourth, water quality interventions appear to be effective in preventing diarrhoea regardless of whether they are deployed in settings where sanitation is 'improved' (ie connection to a public sewer or septic system, pour-flush latrine, simple pit latrine, or ventilated improved pit latrine) or 'unimproved'. This is in contrast to conclusions that interventions to improve water quality are effective only where sanitation has already been addressed (Esrey 1986; VanDerslice 1995).
The subgroup analyses did not demonstrate that the effectiveness of a water quality intervention to prevent diarrhoea is enhanced by adding hygiene instruction, a separate vessel to treat or store water, or by improving sanitation or water supply. This is consistent with our finding that the effectiveness of a water quality intervention does not depend on the baseline conditions in regard to other environmental parameters that are associated with diarrhoea. At the same time, it implies that the cost and effort of combining the water quality intervention with improved hygiene, water storage, water supply, or sanitation may not be justified on the basis of the a synergistic effect on diarrhoeal disease.
Ultimately, the value of water quality interventions in preventing diarrhoeal disease depends not only on their effectiveness but also on their affordability, acceptability, sustainability and scalability within a vulnerable population. Comprehensive cost-effectiveness and cost-benefit analyses will help establish the priority that should be attached to water quality interventions by the public sector and non-governmental organizations. Finally, since household interventions appear especially effective, the private sector, which has particular capacity for addressing the needs of householders, should be explored as a potential source for developing effective, low-cost water treatment interventions on a wide scale.
Trial methodological quality (risk of bias)
The trials with good methodological quality show a greater overall level of effectiveness than those that do not. However, this review included both randomized and quasi-randomized controlled trials, and by necessity, employed different quality criteria for each. Because household-based interventions tended to be randomized controlled trials design while source-based interventions exclusively used quasi-randomization, there is an important bias that may affect the comparison. If, as suggested, the criteria for methodological quality for the two trial designs are not strictly comparable, this bias would affect the comparison of household versus source water interventions. With four criteria for assessing the quality of randomized controlled trials, conclusions about methodological quality could also lead to an unintentional bias. For example, among the three trials that used blinding, only one was deemed adequate on even two other quality criteria. The application of these criteria would thus skew the results against blinded trials.
Only three of 19 randomized controlled trials were blinded, and in each case the intervention had no statistically significant protective effect. This must give pause to any definitive conclusion about the potential value of water quality interventions in the prevention of diarrhoea. The authors of each of these trials suggested possible explanations for their findings. Colford 2002 was the only trial conducted in a developed country setting, and the water there already complied with US standards. Kirchhoff 1985, though a pioneering trial of a potentially important household intervention, had a study population of only 112 persons (smallest of all the included trials) and was rated low on three other criteria of methodological quality. Austin 1993-i also suggested possible methodological issues and used dilute sodium hypochlorite in the control group, an approach that probably improved their water quality thus resulting in an understatement of the intervention's effectiveness. While one trial, Clasen 2004c, cited ethical and other reasons for the decision not to blind the trial, it is not clear why so few of the household-based interventions failed to use placebo controls. Future trials should take steps to address this issue by using blinding design wherever possible.
Trial design and methodology
Subgroup analyses suggest that there are clinical sources of heterogeneity based on the intervention point and also among the different household interventions. However, given the heterogeneity within most of these subgroups, we cannot rule out the possibility that it may be due to differences in the trials' design and methodology.
The design of the trials is not independent of the type of intervention. All six trials involving interventions at the source used quasi-randomization, while 19 of 23 point-of-use interventions were randomized controlled trials. Although this mainly reflects the difficulty in randomizing users of source water interventions, the skewing of design between the two types of interventions could possibly account for differences in the results observed.
The length of the trials was not independent of the point of intervention. The median duration of trials of interventions at the source was more than six times longer than those involving interventions at the household level. Four of six such source-based intervention trials were of three years' duration or longer, while only three of the 32 household-based interventions lasted even one year. Seasonality plays a major role in diarrhoea incidence (Blum 1983), and failure to include at least 12 months' data on diarrhoea may overstate or understate the annual burden of disease in the underlying population and correspondingly influence the measure of effect.
Compliance with the intervention was probably not independent of the point of intervention. Household-based interventions all require some effort on the part of the participants to treat their water, to have treated water consistently available, to avoid recontamination, and to refrain from drinking from untreated sources. Each of these conditions creates an opportunity for non-compliance. Most source-based interventions, on the other hand, extended to the household's entire water supply without any additional compliance steps on the part of the intervention population. Thus, compliance was probably higher among groups using source-based rather than household-based interventions. If compliance is naturally lower among household-based interventions, than this bias may be a natural concomitant. But if compliance can be improved (as it apparently was in some trials), then the higher natural compliance with source interventions may overstate their effectiveness compared to interventions at home.
Participants are more conscious of interventions carried out in their home than those at a distant water source or treatment works. This could lead to bias in trials that are not blinded. Courtesy bias (the tendency of participants who know they are in the intervention group to overstate the effect to please the investigator) and Hawthorne effect (the effect, usually positive, of being under investigation generally) may conspire to overstate or understate the effectiveness of the interventions covered by this review. This is particularly true for the non-blinded trials of household-based interventions that were often research-driven with perhaps more intensive investigator presence.
The availability of water is also an important factor. Interventions at the source are frequently designed primarily to improve the water quantity and availability rather than quality. On the other hand, such improvements in water supply may be a separate and possibly more significant contributor to health than water quality. In the case of the household-based interventions, most appeared to have been conducted in settings with sufficient water, which may mean that these results cannot be generalized to locations where water supplies are inadequate.
The interventions have varying levels of microbiological performance against different types of diarrhoegenic organisms, particularly under different water conditions. In a setting in which diarrhoea was mainly viral, ceramic filters would be only marginally protective. Similarly, where Cryptosporidium or another chlorine-resistant agent is an important cause of diarrhoea, chlorination may provide little if any protection. Even within these categories of interventions, there are important differences in microbiological performance. For example, the filtration subgroup includes ceramic filters that are not generally effective against viruses, and reverse-osmosis filters that are. Similarly, while the sodium hypochlorite used in most chlorination studies has certain antimicrobial capacity, other chlorine studies used mixed oxidants (Quick 1999) that have been shown to have broader biocidal effect (Venczel 1997). Since none of the trials continuously monitored the full range of diarrhoea-causing pathogens present in the drinking water of the study population and few trials attempted to determine clinically the apparent causes of diarrhoea in such population, it is difficult to compare the interventions based on their microbiological performance. This difference in field performance also illustrates another potential flaw in pooling for analysis seemingly similar interventions such filtration and chlorination.
Many of the trials reported results on selected age groups, and not on all ages of persons who would have been affected by the intervention. It is common for research on diarrhoeal disease to specifically target children less than five years of age, a group that is particularly vulnerable to diarrhoea, and many of the trials did provide data for this group. Many others, however, reported results only for a subset of this group, or for some other segment of the population. In most cases, it appears that results were reported for the full population on which data were collected. It is possible, that by omitting a portion of the population affected by the intervention, the design or reporting of results is a source of bias.
Finally, it appears that many if not most of the trials were undertaken primarily for the purpose of investigating the effectiveness of the intervention, and not as an assessment of an ongoing programme. This seems particularly true of the household interventions. While investigators often took special steps to minimize the effect that such a research focus may have had on the study results, the continuous onsite participation of investigators, many of whom were foreign to the study settings, in implementing and assessing the intervention could be a source of possible bias. It may also raise questions about whether the results obtained would be representative of the effectiveness of the interventions outside a research context. Future trials should include assessments of ongoing programmes implemented outside a research context.
We gratefully acknowledge the following people for their research, advice, assistance and other valuable contributions: Greg Allgood, Jamie Bartram, Joseph Brown, Jack Colford, John Crump, Tom Chiller, Val Curtis, Shannon Doocy, Lorna Fewtrell, Carrol Gamble, Bruce Gordon, Stephen Gundry, Bruce Keswick, Steve Luby, Rob Quick, Mark Sobsey, Sara Thomas, and James Wright. We also appreciate the assistance and help provided by other members of the Cochrane Infectious Diseases Group, and by the referees of both this review and the protocol.
The editorial base for the Cochrane Infectious Diseases Group is funded by the UK Department for International Development (DFID) for the benefit of developing countries.
Appendix 1. Search methods: detailed search strategies
|Search set||CIDG SRa||CENTRAL||MEDLINEb||EMBASEb||LILACSb|
|1||water||WATER PURIFICATION||WATER PURIFICATION||WATER PURIFICATION||water|
|2||purification OR treatment OR chlorination OR decontamination OR filtration OR supply OR storage OR consumption||WATER MICROBIOLOGY||WATER MICROBIOLOGY||WATER MICROBIOLOGY||purification OR treatment OR chlorination OR decontamination OR filtration OR supply OR storage OR consumption|
|3||diarrhea||1 OR 2||1 OR 2||1 OR 2||diarrhea|
|4||1 AND 2 AND 3||water||water||water||1 AND 2 AND 3|
|5||—||purification OR treatment OR chlorination OR decontamination OR filtration OR supply OR storage OR consumption OR drink*||purification OR treatment OR chlorination OR decontamination OR filtration OR supply OR storage OR consumption OR drink*||purification OR treatment OR chlorination OR decontamination OR filtration OR supply OR storage OR consumption OR drink$||—|
|6||—||4 AND 5||4 AND 5||4 AND 5||—|
|7||—||3 OR 6||3 OR 6||3 OR 6||—|
|10||—||DIARRHEA/PREVENTION AND CONTROL||DIARRHEA/PREVENTION AND CONTROL||waterborne infection$||—|
|11||—||waterborne infection*||waterborne infection*||cholera OR shigell$ OR dysenter$ OR cryptosporidi$ OR giardia$ OR Escherichia coli OR clostridium||—|
|12||—||INTESTINAL DISEASES||INTESTINAL DISEASES||ENTEROBACTERIACEAE||—|
|13||—||cholera OR shigell* OR dysenter* OR cryptosporidi* OR giardia* OR Escherichia coli OR clostridium||cholera OR shigell* OR dysenter* OR cryptosporidi* OR giardia* OR Escherichia coli OR clostridium||8-12/OR||—|
|14||—||ENTEROBACTERIACEAE||ENTEROBACTERIACEAE||7 AND 13||—|
|15||—||8-14/OR||8-14/OR||LIMIT 14 TO HUMAN||—|
|16||—||7 AND 15||7 AND 15||—||—|
|17||—||—||LIMIT 16 TO HUMAN||—||—|
aCochrane Infectious Diseases Group Specialized Register.
bSearch terms used in combination with the search strategy for retrieving trials developed by The Cochrane Collaboration (Higgins 2005b); upper case: MeSH or EMTREE heading; lower case: free text term.
Appendix 2. Data extracted from included studies
|Trial data||Country and setting (urban, rural)|
|Number of participants/groups|
|Unit of randomization, and whether measurement of effect adjusts for clustering where randomization is other than individual|
|Definition and practices of control group|
|Type and details of water quality intervention (filtration, flocculation, chemical disinfection, heat, or ultraviolet (UV) radiation)|
|Other components of intervention (hygiene message, improved supply, improved sanitation, improved storage)|
|Whether water protected to point of use (ie by pipe, residual disinfection, or safe storage)|
|Case definition of diarrhoea|
|Method for diarrhoea assessment (self-reported, observed, or clinically confirmed)|
|Where self reported, recall period used|
|Prescribed criteria of methodological quality|
|Individual characteristics||Age group|
|Type of water source|
|Level of faecal contamination of control water (low (< 100 thermotolerant coliforms (TTC)/100 mL), medium (100 to 1000 TTC/100 mL), and high (> 1000 TTC/100 mL)|
|Causative agents identified (yes or no)|
|Water collection, storage, and drawing practices|
|Distance to and other constraints regarding water supply|
|Sanitation facilities (improved or unimproved)|
|Outcomes||Pre- and post-intervention faecal contamination of drinking water, and method of assessment (including indicator used)|
|Diarrhoea morbidity and 95% confidence interval for each age group reported|
|Manner of measuring diarrhoea morbidity|
|Mortality attributed to diarrhoea|
|Rate of utilization of intervention and manner of assessing same|
Appendix 3. Primary drinking supply and sanitation facilities (used as control)
|Trial||Description||Sourcea||Access to sourceb||Quantity availablec||Ambient H2O quality||Sanitationd|
|Alam 1989||Shallow, hand-dug wells; some hand pumps||Unimproved||Unclear||Unclear||Not tested||Unclear|
|Austin 1993||Open wells||Unimproved||Sufficient||Unclear||Mean 1871 FC/100 mL in wells; among stored water samples, mean 3358 FC/100 mL in rainy season, 1014 FC/100 mL in dry season||Unclear|
|Aziz 1990||Fewer hand pumps and latrines; no hygiene instruction||Unimproved||Unclear||Unclear||Not tested||Unimproved|
|Chiller 2004||Rivers, springs, taps, and wells||Unclear||Unclear||Sufficient||98% of source samples contained E. coli; precise level not reported||Mostly unimproved|
|Clasen 2004b||80% yard taps supplied by untreated surface source, 20% directly from untreated surface sources||80% improved, 20% unimproved||Sufficient||Sufficient||Baseline mean thermotolerant coliform count of 145/100 mL at taps and 52/100 mL at surface sources||Unimproved|
|Clasen 2004c||Irrigation canals and other surface sources||Unimproved||Sufficient||Sufficient||Baseline mean thermotolerant coliform count of 793/100 mL||Unimproved|
|Colford 2002||Household taps supplied by municipal water treatment||Improved||Sufficient||Sufficient||Data from water treatment plant: met US federal and California drinking water standards||Improved|
|Conroy 1996||Open water holes, tank fed by untreated piped water supply; control households provided same bottles, but were instructed to keep them indoors||Unimproved||Unclear||Unclear||All water sources positive for FC; some counts > 103 colony-forming units (CFU)/mL||Unclear|
|Conroy 1999||Open water holes, tank fed by untreated piped water supply; control households provided same bottles, but were instructed to keep them indoors||Unimproved||Unclear||Unclear||Not tested||Unclear|
|Crump 2004||50% ponds, 49% rivers||Unimproved||Unclear||Insufficient||Baseline mean E. coli level was 98/100 mL||Unclear; 33% defecate on ground|
|Doocy 2004||Surface sources and some tap stands||Unimproved||Unclear||Insufficient||Qualitative measure only||Improved|
|du Preez 2004||Protected wells||Improved||Sufficient||Unclear||Samples with E. coli per 100 mL: 31 < 10; 9 > 10 < 100; 1 > 100 < 1000; 3 > 1000||Improved|
|Garrett 2004||Unclear||Unclear||Unclear||Sufficient||Not tested||Unimproved|
|Gasana 2002||Spring||Unimproved||Unclear||Unclear||Baseline sample range from 4 to 1100 total coliforms/100 mL||Unimproved|
|Handzel 1998||48% tap, 52% tubewell; 61% paid for drinking water||Improved||Sufficient||Sufficient||Baseline geometric mean FC counts/100 mL: tap water -138 at source, 280 stored in home; tubewell water -6.7 at source, 138 stored in home|| |
|Jensen 2003||Some slow sand filters in poor condition; some household taps; majority used ground water||Improved||Unclear||Unclear||Baseline (pre-intervention) geometric mean in intervention village was 13.3 E. coli CFU/100 mL; geometric mean E. coli count of 137/100 mL in control village||Unclear|
|Kirchhoff 1985||Pond water stored in clay pots after filtering with cloth||Unimproved||Unclear||Insufficient||Mean of 970 FC/100 mL from pond sources; 16,000 in control stored water||Unimproved|
|Luby 2004||Tanker trucks, shared municipal taps||Unimproved||Unclear||Unclear||At baseline, 79% of participating stored household samples were free of E. coli||Improved|
|Luby 2004a||Unclear||75% improved||Sufficient||Unclear||Not tested||Improved|
|Lule 2005||16% surface or shallow wells, 50% protected springs, 49% boreholes or taps||Unimproved||Sufficient||Sufficient||Baseline mean E. coli counts: 11 at source, 163 stored water; 54% of source water had some contamination, compared to 89% of stored water||Improved|
|Mahfouz 1995||Shallow wells||Unimproved||Unclear||Unclear||92.3% positive with E. coli; amount not recorded||Improved|
|Messou 1997||Not reported||Unimproved||Unclear||Unclear||Not tested||Unclear|
|Quick 1999||Shallow uncovered wells; 38% treated water||Unimproved||Unclear||Unclear||Baseline median colony count of E. coli: 57050/100 mL for source water and 46950/100 mL for stored water||Unimproved (but 47% used latrine)|
|Quick 2002||Shallow wells; some boiling||Unimproved||Unclear||Unclear||Baseline median colony count of E. coli: 34/100 mL for source water and 44/100 mL for stored water||Unclear|
|Reller 2003||Surface water from shallow wells, rivers and springs||Unimproved||Unclear||Unclear||Baseline median colony count of E. coli: 63/100 mL||Unclear|
|Roberts 2001||Traditional pots or standard ration buckets filled at refugee camp water point||Improved||Unclear||Unclear||At well, 71% of samples were 1 FC/100 mL or less; 100% < 100 FC/100 mL||Unclear|
|Semenza 1998||Households without piped water (procured from street tap, neighbour tap, well, vendor, or river)||Unimproved||Unclear||Unclear||Baseline mean 49 faecal coliform/100 mL||Unclear|
|Torun 1982||Shallow, unprotected, hand-dug wells||Unimproved||Unclear||Unclear||3% of 698 samples from control village had coliform bacteria||Unimproved|
|URL 1995||Household tap (27%), public tap (21%), well (23%)||Improved||Unclear||Unclear||Range 5 to 260 FC/100 mL depending on site||Improved|
|Xiao 1997||Not reported||Unimproved||Unclear||Unclear||Not tested||Unclear|
E. coli: Escherichia coli; FC: faecal coliform.
a'Improved' includes household connection, public standpipe, borehole, protected dug well, protected spring, rainwater collection; 'unimproved' includes unprotected well, unprotected spring, vendor-provided water, bottled water; and 'unclear' means unclear or not reported; definition based on WHO 2000.
b'Sufficient' means located within 500 m, queuing no more than 15 minutes, no more than 3 minutes to fill 20 L container, and maintained so available consistently; 'insufficient' means that it does not meet any of above; and 'unclear' means unclear or not reported; definition based minimum standards established by The Sphere Project 2004.
c'Sufficient' means a minimum of 15 L/day/person; 'insufficient' means less than 15 L/day/person; and 'unclear' means unclear or not reported; definition based on minimum standards established by The Sphere Project 2004.
d'Improved' means connection to a public sewer or septic system, pour flush latrine, simple pit latrine, or ventilated improved pit latrine; 'unimproved' means service or bucket latrine, public latrines, open latrines; and 'unclear' means unclear or not reported; definition based on WHO 2000.
Appendix 4. Interventions
|Trial||Description||Source or household||Compliance measured?||Other components|
|Alam 1989||Improved water supply||Source||Not reported||Hygiene education|
|Austin 1993-i||Household chlorination among children ages 25 to 60 months||Household||60% compliance measured by residual chlorine||None|
|Austin 1993-ii||Household chlorination among children ages 6 to 24 months||Household||68% compliance measured by residual chlorine||None|
|Aziz 1990||Improved water supply||Source||Periodic cross-sectional assessments; rate not reported||Improved sanitation, hygiene education|
|Chiller 2004||Flocculant-disinfectant sachets used at household level||Household||85% compliance measured by residual chlorine||Hygiene education|
|Clasen 2004b||Household ceramic filters||Household||Not reported||Filter included improved storage|
|Clasen 2004c||Household ceramic filters||Household||Not reported||Filter included improved storage|
|Colford 2002||Household reverse osmosis filters||Household||Plumbed-in unit||None|
|Conroy 1996||Solar disinfection in plastic bottles at household level||Household||Random checks by project workers; rate not reported||None|
|Conroy 1999||Solar disinfection in plastic bottles at household level||Household||Not reported||None|
|Crump 2004-i||Sodium hypochlorite used at household level||Household||85% compliance (measured by residual chlorine) at scheduled visits; 61% during unannounced weekly visits||Hygiene education|
|Crump 2004-ii||Flocculent-disinfectant sachets used at household level||Household||86% (measured by residual chlorine) at scheduled visits; 44% during unannounced weekly visits||Hygiene education|
|Doocy 2004||Flocculant-disinfectant sachets used at household level in refugee camp||Household||95% compliance based on residual chlorine sampling||Both controls and intervention group received water storage vessel|
|du Preez 2004||Household ceramic filter||Household||100% based on observation||Filter included improved storage|
|Garrett 2004||Household chlorination using sodium hypochlorite||Household||43% based on residual chlorine||Sanitation, hygiene education, storage, supply|
|Gasana 2002||Source improvements (water pipes, sedimentation tank, ceramic filter, storage tank, communal tap)||Source||Not reported||None|
|Handzel 1998||Household chlorination using sodium hypochlorite solution and special storage vessel||Household||90% compliance based on residual chlorine measurements||None|
|Jensen 2003||Village level chlorination of water supply using calcium hypochlorite||Source||Unclear, though chlorinated water was supplied through distribution system to all intervention households||None|
|Kirchhoff 1985||Household level chlorination with sodium hypochlorite||Household||None reported||None|
|Luby 2004a-i||Bleach + regular vessel||Household||None reported||Vessel provided improved storage|
|Luby 2004a-ii||Bleach + insulated vessel||Household||As above||As above|
|Luby 2004b-i||Dilute bleach + vessel||Household||Yes, though not yet available||Vessel provided improved storage|
|Luby 2004b-ii||Flocculant-disinfectant + soap||Household||As above||Hygiene instruction|
|Luby 2004b-iii||Flocculant-disinfectant + vessel||Household||As above||Vessel provided improved storage|
|Lule 2005||Household level chlorination using sodium hypochlorite + special vessel||Household||Not reported||Vessel provided improved storage; hygiene education was provided to both intervention and comparison groups|
|Mahfouz 1995||Household level chlorination using calcium hypochlorite||Household||Some residual chlorine in all intervention samples||None|
|Messou 1997||Improved water supply||Source||Yes, measured increase in water supplied and change in sanitation and hygiene practices||Sanitation, oral rehydration|
|Quick 1999||Household level chlorination + vessel||Household||70% to 95% compliance based on residual chlorine (increased during course of study)||Improved storage, hygiene education|
|Quick 2002||Household level chlorination + vessel||Household||70% compliance based on residual chlorine||Improved storage, hygiene education|
|Reller 2003-i||Flocculant-disinfectant||Household||Residual chlorine > 0.1mg/L in unannounced visits: 27%||None|
|Reller 2003-ii||Bleach||As above||Residual chlorine > 0.1mg/L in unannounced visits: 36%||None|
|Reller 2003-iii||Bleach + vessel||As above||Residual chlorine > 0.1mg/L in unannounced visits: 44%||Improved storage vessel|
|Reller 2003-iv||Flocculant-disinfectant + vessel||As above||Residual chlorine > 0.1mg/L in unannounced visits: 34%||As above|
|Roberts 2001||Improved storage (bucket with spout and narrow opening to limit hand entry)||Household||Intervention householders received buckets; actual use was not reported||None|
|Semenza 1998||Household level chlorination||Household||73% based on residual chlorine levels at time of visit||Improved storage, hygiene education|
|Torun 1982||Source protection (spring), chlorination facilities, "adequate storage", and water mains with faucets to yards of intervention village||Source||No||Hygiene education|
|URL 1995-i||Household ceramic filters||Household||87% to 93% use of filter by children||None|
|URL 1995-ii||Household ceramic filters||As above||As above||Hygiene education|
|Xiao 1997||Improved water supply||Source||Community intervention; use not otherwise reported||Sanitation, hygiene education|
Protocol first published: Issue 2, 2004
Review first published: Issue 3, 2006
Contributions of authors
Conceived review: SC. Co-ordinated review: TC. Designed review: TC, IR, TR, WS. Drafted protocol: TC. Executed search strategy: TC and Cochrane Infectious Diseases Group (CIDG). Screened search results: TC, TR. Retrieved papers: TC. Applied inclusion criteria: TC, TR, SC. Extracted data: TC, TR, WS. Computed estimates of effect: TC, TR, WS. Applied quality criteria: TC, TR, IR. Contacted authors for additional information: TC. Addressed statistical issues: TC, TR, WS, CIDG. Entered data into Review Manager: TC. Drafted review: TC. Commented on review: IR, TR, WS, SC. Prepared tables: TC. Prepared figures: WS, TC. Guarantor of review: TC.
Declarations of interest
Thomas Clasen, Sandy Cairncross, Tamer Rabie, and Wolf-Peter Schmidt participate in research supported by Unilever, Ltd., which manufactures and sells household-based water treatment devices.
Differences between protocol and review
2006, Issue 3 (first version of review): Wolf-Peter Schmidt became a co-author of the review in June 2005.