Environ. Sci. Technol.

2009 43, 89918997

Household Water Treatment in Developing Countries: Comparing Different Intervention Types Using Meta-Regression
PAUL R. HUNTER* School of Medicine, Health Policy and Practice, University of East Anglia, Norwich U.K.

ceramic lters are the most effective form of HWT in the longterm, disinfection-only interventions including SODIS appear to have poor if any longterm public health benet.

Introduction
Inadequate access to safe drinking water is a major cause of morbidity and mortality in developing countries (1). Household water treatment (HWT) or point of use water treatment (syn.) has been extensively promoted as the solution to the problem of poor quality drinking water in poor communities in developing country settings (2). These devices work by increasing the quality but not the availability of drinking water other than by making previously available but unpotable water t to drink. There are a variety of such devices available for use that basically work either through a process of disinfection or ltration. Sobsey classied the most common such devices as follows (2). Disinfection. Chlorination with Safe Storage. Chorine added to drinking water after which it is stored in containers designed to reduce the risk of contamination. Combined Coagulant-Chlorine Disinfection Systems. There are commercial kits that combine dry coagulantocculant and chlorine as tablets or sachets. SODIS. Transparent polyethylene terephthalate (PET or PETE) bottles lled with aerated source water and left in the sun to disinfect the water by solar UV and increased temperature. Filtration. Ceramic Filter. Porous ceramic (red clay) lters remove microbes from drinking water by size exclusion. Biosand Filter. A household version of the slow sand lter where potentially pathogenic microorganisms are removed by a biolm layer (known as the Schmutzdeke) which forms in the top few cm of the lter. There is little doubt that such devices, when operated correctly, can reduce the concentration of indicator bacteria and pathogens in drinking water (2, 3). Systematic reviews of the literature have also suggested that these devices carry signicant public health benets in terms of diarrheal disease reduction (4, 5). However, a recent reassessment of this data concluded that much of the apparent benet of HWT could be explained by recall bias as few of the reported studies were blinded (6). Those few studies that have been blinded found no protective effect on diarrheal disease. This negative analysis generated a swift and clearly impassioned rebuttal from pro HWT researchers (9). Although this author nds the arguments put forward in the reassessment generally compelling, the HWT proponents made an important point when they stated that the substantial heterogeneity in the results of the RCTs on HWT points toward the need to study these inuences and learn how to best target the intervention to maximum public health benet. The fact remains that recall bias in unblinded randomized controlled trials is real and substantial and well-known throughout medical research (10). However, the fact that much of the apparent benet from HWT may be due to recall bias does not necessarily mean that such devices have no or only little public health value. This study aims to quantify the benet of HWT over and above the potential impact of recall bias on published effectiveness. Furthermore, the opportunity has been taken, where possible, to investigate some of the potential causes of heterogeneity in the results to better inform policy regarding HWT interventions.

Received June 30, 2009. Accepted October 15, 2009.

Household water treatment (HWT) is being widely promoted as an appropriate intervention for reducing the burden of waterborne disease in poor communities in developing countries. A recent study has raised concerns about the effectiveness of HWT, in part because of concerns over the lack of blinding and in part because of considerable heterogeneity in the reported effectiveness of randomized controlled trials. This study set out to attempt to investigate the causes of this heterogeneity and so identify factors associated with good health gains. Studies identied in an earlier systematic review and meta-analysis were supplemented with more recently published randomized controlled trials. A total of 28 separate studies of randomized controlled trials of HWT with 39 intervention arms were included in the analysis. Heterogeneity was studied using the metareg command in Stata. Initial analyses with single candidate predictors were undertaken and all variables signicant at the P < 0.2 level were included in a nal regression model. Further analyses were done to estimate the effect of the interventions over time by MonteCarlo modeling using @Risk and the parameter estimates from the nal regression model. The overall effect size of all unblinded studies was relative risk ) 0.56 (95% condence intervals 0.51-0.63), but after adjusting for bias due to lack of blinding the effect size was much lower (RR ) 0.85, 95% CI ) 0.76-0.97). Four main variables were signicant predictors of effectiveness of intervention in a multipredictor meta regression model: Log duration of study follow-up (regression coefcient of log effect size ) 0.186, standard error (SE) ) 0.072), whether or not the study was blinded (coefcient 0.251, SE 0.066) and being conducted in an emergency setting (coefcient -0.351, SE 0.076) were all signicant predictors of effect size in the nal model. Compared to the ceramic lter all other interventions were much less effective (Biosand 0.247, 0.073; chlorine and safe waste storage 0.295, 0.061; combined coagulant-chlorine 0.2349, 0.067; SODIS 0.302, 0.068). A Monte Carlo model predicted that over 12 months ceramic lters were likely to be still effective at reducing disease, whereas SODIS, chlorination, and coagulation-chlorination had little if any benet. Indeed these three interventions are predicted to have the same or less effect than what may be expected due purely to reporting bias in unblinded studies With the currently available evidence

Methods
* Corresponding author e-mail: Paul.Hunter@uea.ac.uk; phone: +44 1603 591004; fax: +44 1603 591750.
10.1021/es9028217 CCC: $40.75 2009 American Chemical Society

This study uses data from RCTs that have been published in the literature. All the studies included in the analysis by
VOL. 43, NO. 23, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9

8991

Published on Web 10/23/2009

Schmidt and Cairncross have been included (6). In addition the recent literature for 2007, 2008, and 2009 has been searched for further studies to add to the database. Search terms included household water treatment or point-of-use. In addition papers referring to one of the prior systematic and other key reviews were screened for new papers using Web of Knowledge (2-7). Effect sizes for all data were taken from the Clasen review (4) or from the original paper where the study had not been included in the Clasen review. However, effect sizes were also checked against those given in the Waddington review (7) and where there were major unexplained differences the original paper was reviewed. There was one paper that fell into this category: that by Doocy and Burnham (4). For this study the Clasen review gave a substantially greater standardized effect size than either the authors or Waddington. For inclusion here the authors effect sizes were used. These were given as both incidence and prevalence ratios and the mean was taken adjusting for variance as described elsewhere (8). Where condence intervals were not given they were back calculated from the incidence rates in the interval and control groups and the presented p value (8). Overall pooled relative risks were calculated using random effects models from published summary estimates using StatsDirect (11). Wood et al. gives two estimates of potential bias in unblinded studies, one due to unclear allocation concealment (ratio of odds ratios 0.69 (95% CI 0.59-0.82)) and one due to lack of blinding (0.75 (0.61-0.93)) (10). To estimate the potential contribution of bias toward the estimated impact of HWT the estimates by the two estimates of Wood et al. were pooled to give (0.71 (0.63-0.81). The real effect size of the pooled HWT was estimated by a MonteCarlo approach using @Risk 5.0 (Palisade Corporation). The log midpoint and standard errors of the pooled effect size of HWT and estimate of bias were each randomly sampled and the difference between the two values was taken to be the real effect size. This calculation was done 10,000 times to give the midpoint and standard error. The investigation of heterogeneity was done using the metareg command in STATA (12). Metareg performs random-effects meta-regression on study-level summary data. Basically this performs a regression analysis where the outcome variable is dened by the log of the effect size for each study with its standard error. One or more other variables can be entered also as predictor variables. This generates mean and standard errors for the constant and regression coefcients in exactly the same way as an ordinary regression analysis. All possible predictor variables were tested in a single predictor model and those variables found to be signicant at the p ) 0.2 level were included in a nal multiple predictor regression model. The log effect sizes and their standard errors for each intervention type for any particular study duration could then be estimated from using the constants and regression coefcients using a MonteCarlo approach as discussed above (13). The estimate of the power of the meta-regression analysis to explain heterogeneity was calculated by determining the percent reduction in Tau2 from the model with no predictor variables and the nal model with all signicant predictors (8). Figures were drawn using Comprehensive Meta-Analysis v2 (http://www.meta-analysis.com/).

Results
Overall 39 data sets of randomized controlled trials of HWT taken from 28 separate studies were included in the analysis. These included all but one of the studies used in the analysis of Schmidt and Cairncross (6). The excluded study was not included because of criticisms raised in the comments against the Schmidt and Cairncross paper that suggested this lack
8992
9

of effect was due to factors other than the intervention (9, 14). Five additional papers covering seven data sets additional to those covered in previous reviews were included in this analysis (15-19). Four of the additional data sets were of lters, one was of SODIS, and two were of chlorine. Of the 39 data sets included in this analysis, 29 were of some form of disinfection and 10 were of some form of ltration. Three data sets were of the Biosand lter, 7 were of the ceramic lter, 7 were of coagulant and disinfectant, 18 were of chlorination (usually with hypochlorite), and 4 were of SODIS. The studies varied in the age range surveyed: 15 were restricted to children and the rest included all ages. The median follow-up time was 26 weeks and only 9 studies followed up the impact of the interventions for the maximum reported follow-up time of 52 weeks or more. Figure 1 shows the individual effect sizes for each study arm and the pooled effect size for each intervention type. For all the studies combined the random effects pooled relative risk was 0.56 (95%CI ) 0.51-0.63), rather lower than the estimates of bias due to lack of blinding 0.71 (0.63-0.81). Accounting for the impact of the lack of blinding would give an effect size of RR 0.85 (95% CI ) 0.76-0.97), still statistically signicant P ) 0.002. As expected there was substantial heterogeneity in the reported effectiveness, Cochran Q ) 222 (df ) 35) P < 0.0001. There was also strong evidence of publication bias (Figure 2), Begg-Mazumdar: Kendalls tau ) -0.368, P ) 0.0012, suggesting that the pooled effect size based on published studies would overestimate the effectiveness of interventions. Table 1 lists the possible explanatory variables along with the z score and p values that were investigated for their potential contribution to heterogeneity. Those variables with p < 0.2 (intervention type, whether or not study was blinded, conducted in an emergency setting, and log duration of follow-up) were combined into a single meta-regression model (Table 2). All four variables remained signicant in the multiple predictor model. In particular it should be noted that the longer the follow-up period, the less effective the intervention becomes. Also, the ceramic lter performs signicantly better than the other intervention types and, as expected, blinded studies indicate lower effectiveness. The one study conducted in an emergency setting was also more effective. This model was able to predict 90% of the heterogeneity of the simple meta analysis. The parameters and standard errors from the regression model, other than the blinding variable, were used to derive estimates of the intervention risk ratios over 52 week followup (Table 3). Table 3 also shows the pooled effect sizes of those studies that were of 52 week follow-up. Although the estimates from the regression model had wider credible intervals than the condence intervals of the 52 week studies they were broadly similar. The impact of the different interventions over different follow-up times is shown graphically in Figure 3 which shows the midpoint effect size by week for each intervention type along with the midpoint for the estimate of bias due to lack of blinding. This illustrates very clearly that the measured health impact of all ve intervention types declines with duration of follow-up. In particular it should be noted that the estimated effect sizes of the three disinfection interventions were less than the mean effect of bias due to lack of blinding derived from the study by Wood et al. (10). This would suggest that the three disinfection interventions did not reduce diarrheal disease over the course of a year after implementation. Figure 4 shows separate meta-regressions for the three interventions technologies with the most studies (chlorine, chlorine-coagulant, and ceramic lter), unblinded studies only. It can be seen that the ceramic lter has the greatest effect size, but also that the slopes for the two disinfection

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 23, 2009

FIGURE 1. Data sets included in the meta-regression analysis ordered by intervention type and ranked by duration of follow-up (* blinded studies; ** conducted in emergency setting).
8993

VOL. 43, NO. 23, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

FIGURE 2. Bias assessment funnel plot of all unblinded studies.

TABLE 1. Single Predictor Variable Analysis of Variables Potentially Contributing to Heterogeneitya

variable age group (all ages: 5 y) combined with hygiene interventions intervention type biosand lter ceramic lterb chlorine and safe storage combined coagulantchlorine SODIS duration of follow-up log duration of follow-up whether or not study blinded conducted in emergency setting coefcient 0.005 -0.019 SE 0.066 0.071 Z 0.07 -0.27 P 0.943 0.791 0.015 0.214 0 0.275 0.252 0.326 0.003 0.242 0.282 -0.324 0.112 0.085 0.095 0.105 0.002 0.123 0.128 0.133 1.91 3.23 2.66 3.11 1.92 1.97 2.20 -2.43 0.056 0.001 0.008 0.002 0.054 0.048 0.028 0.015

TABLE 2. Meta Regression Model of Duration of Follow-up, Intervention Type, and Study Blinding on Effectivenessa
variable log duration of follow-up intervention type biosand lter ceramic lterb chlorine and safe storage combined coagulantchlorine SODIS blinded or not emergency setting constant
a

coefcient 0.186

SE 0.072

2.657 0.010 0.00002

0.247 0 0.295 0.349 0.302 0.251 -0.351 -0.751

0.073 0.061 0.067 0.068 0.066 0.122

3.38 4.80 5.22 4.42 3.82

0.0007 1.6 10-6 1.8 10-7 9.9 10-6 0.0001 3.7 10-6

0.076 -4.63

Outcome variable is log of relative risk so the greater the coefcient the less effective is the intervention. b Dropped from analysis as variable redundant.

a Outcome variable is log of relative risk so the greater the coefcient the less effective is the intervention. b Dropped from analysis as variable redundant.

technologies are steeper suggesting that the effect of these disinfection technologies declines more rapidly than for the lter.

Discussion
Systematic reviews and meta-analyses are really only as good as the quality of the studies on which they are based. One of the problems with the literature on the effectiveness of HWT is that the studies that have been used to support its uptake have often been poorly designed unblinded and have been conducted over short or very short periods of followup. Many of the studies included in theis analysis were of very short durationsthe median duration was only 26 weeks. Both Schmidt and Cairncross and Waddington et al. have commented on the problems that may come from some studies being managed by researchers with potential conicts of interest (7). A substantial proportion of these studies did not have conict of interest statements (7). Waddington et al. also found that there was evidence of marked publication bias and estimated that this led to an overestimate of the effect size by about a relative risk of 0.15 in the pooled effect size.
8994
9

The meta-regression analysis presented here has shown that a large proportion of the heterogeneity in the available RCTs on the public health effectiveness of HWT can be explained by the whether or not the study was blinded, whether or not it was conducted in an emergency setting, the type of intervention, and the duration of follow-up used to monitor the impact on diarrheal disease. In particular the effectiveness of interventions clearly declines with duration of follow-up. This analysis also suggests a difference in apparent effectiveness among the different HWT interventions. In particular, there appears to be a greater protective effect from ceramic lters over all other interventions. In this regard the blinded studies of HWT that found no protective effect were all of disinfection (6). There were no unblinded studies of ltration. Over 52 weeks the median effectiveness of ceramic lters is still much better than the median effect of recall bias because of inadequate blinding. and ceramic lters appear to provide substantial benets over all other interventions. Indeed the analyses presented here suggest that the disinfection interventions have little if any health benet over the course of a 12 month follow-up period. By contrast ceramic lters appear to continue to give signicant benet over the 12 month period. However, care should be taken in the interpretation of the results of the Biosand system as they were based on only three studies and there was no study

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 23, 2009

TABLE 3. Comparison of Risk over 52 Weeks Estimated from Meta-Regression Model (Table 2) with Pooled Mean Effect Size from Studies of 52 Week Follow-Up
risk over 52 weeks in unblinded studies estimated from regression model pooled effect size from studies of 52 weeks duration

lower 90% upper 90% number data sets/ lower 95% upper 95% median credible interval credible interval number of publications risk ratio condence interval condence interval
biosand ceramic combined coagulantchlorine chlorine and safe water storage SODIS 0.65 0.37 0.83 0.72 0.74 0.32 0.19 0.41 0.36 0.36 1.32 0.71 1.66 1.46 1.48 0 3/2 2/1 2/1 1/1 0.44 0.87 0.74 0.69 0.28 0.71 0.63 0.63 0.70 1.06 0.87 0.75

with a duration of >26 weeks. Nevertheless, the pooled effects for this intervention irrespective of duration of follow-up are still less effective than the estimated 52 week effectiveness of the ceramic lters. It is not clear why intervention effectiveness declines with duration of follow-up. Nor is it clear whether effectiveness continues to decline after 52 weeks. Possible reasons for the decline include participants stopping using the intervention, the intervention starting to fail, the impact of reporting bias declining with time, or illness rates nding a new equilibrium with the changed exposure rate and host immunity levels. Certainly the evidence would support the observation that in a number of studies continued use of the intervention within the target population declines with time, either due to choice, failure of the device, or inability to purchase replacements (2). Stopping using the intervention for whatever reason would certainly reduce effectiveness of the intervention within the study population. There is also some evidence that reduced effectiveness of an intervention may occur because of reduced population immunity through lower exposure increasing susceptibility (20). This author is not aware of any direct evidence that the magnitude of recall bias in unblinded randomized controlled trials varies with duration of follow-up. However, this is a plausible explanation, especially considering the problem of reporting fatigue with duration of follow-up in studies of subjective outcomes (21). It is for this reason that the estimates of 52 week effectiveness (Table 3) were calculated assuming no blinding.

This observation that disinfection alone may not provide adequate public health protection should not be surprising. Several waterborne pathogens, especially Cryptosporidium and enteric viruses, are more resistant to chlorine than are bacteria (22). Furthermore, it was shown some time ago, albeit from a European study, that fecally polluted water carries with it increased risk of illness even after adequate chlorination has killed all indicator organisms (23, 24). There have also been studies suggesting that eld effectiveness of HWT disinfection is poor compared to laboratory-based studies of its effectiveness (25). However, this may not be only a problem for disinfection as the microbiological effectiveness of Biosand lters in the eld may also be variable (26, 27). Finally, there is also evidence that continued use of disinfection based systems may be poorer than for ltration (2). It has been argued that it is more important to maintain water at a constant quality rather than improve it but then allow it to deteriorate again from time to time (28). Returning to the comment mentioned in the Introduction that what is needed is to best target the intervention to maximum public health benet (9), this author would agree wholeheartedly with this comment. Probably one of the most useful approaches to help us understand this has been the work of Eisenberg and colleagues who used epidemiological modeling to set waterborne transmission in the context of other transmission pathways (29). They basically concluded that the effectiveness of single pathway interventions such as HWT depend on the how much of disease risk is associated

FIGURE 3. Impact of duration of follow-up on effectiveness of the different intervention types unblinded and the midpoint estimate of the likely impact of reporting bias due to lack of blinding.
VOL. 43, NO. 23, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9

8995

FIGURE 4. Meta-regression of the three intervention types with the most studies indicating the impact of log follow-up duration of effectiveness. Log risk ratio is natural log, and follow-up is log10. with that transmission pathway relative to all other transmission pathways. For some situations, they argued, single pathway interventions have real public health benets and in others they do not. This conclusion was criticized on the basis of the evidence from the randomized trials that indicated success in almost all situations (30). However, the very different impression of the effectiveness of HWT in this study and in the recent systematic review by Waddington et
8996
9

al. and the Schmidt and Cairncross paper (6, 7) would suggest that Eisenbergs approach be reconsidered. Based on this analysis alone, this author would not recommend that any of the three disinfection interventions continue to be implemented in developing country settings. Before chlorination, coagulation-chlorination, or SODIS be part of future interventions there would need to be large double-blinded placebo-controlled studies undertaken to

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 43, NO. 23, 2009

prove these interventions effectiveness. For Biosand lters there needs to be one or more large, preferably blinded, randomized trials of adequate follow-up duration, at least 52 weeks. By contrast the clear effectiveness of the ceramic lter in this analysis would make further controlled trials unethical. Research should focus primarily on how to increase uptake and sustainability of the intervention.

Literature Cited
(1) Pruss, A.; Kay, D.; Fewtrell, L.; Bartram, J. Estimating the burden of disease from water, sanitation, and hygiene at a global level. Environ. Health Perspect. 2002, 110, 537542. (2) Sobsey, M. D.; Stauber, C. E.; Casanova, L. M.; Brown, J. M.; Elliott, M. A. Point of Use Household Drinking Water Filtration: A Practical, Effective Solution for Providing Sustained Access to Safe Drinking Water in the Developing World. Environ. Sci. Technol. 2008, 42, 42614267. (3) Arnold, B. F.; Colford, J. M., Jr. Treating water with chlorine at point-of-use to improve water quality and reduce child diarrhea in developing countries: A systematic review and meta-analysis. Am. J. Trop. Med. Hyg. 2007, 76, 354364. (4) Clasen, T.; Schmidt, W. P.; Rabie, T.; Roberts, I.; Cairncross, S. Interventions to improve water quality for preventing diarrhoea: Systematic review and meta-analysis. Brit. Med. J. 2007, 334, 782785. (5) Fewtrell, L.; Kaufmann, R. B.; Kay, D.; Enanoria, W.; Haller, L.; Colford, J. M. Water, sanitation, and hygiene interventions to reduce diarrhoea in less developed countries: A systematic review and meta-analysis. Lancet Infect. Dis. 2005, 5, 4252. (6) Schmidt, W.-P.; Cairncross, S. Household water treatment in poor populations: Is there enough evidence for scaling up now? Environ. Sci. Technol. 2009, 43, 986992. (7) Waddington, H.; Snilstveit, B.; White, H.; Fewtrell, L. Water, Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries; International Initiative for Impact Evaluation (3ie), 2009; http://www.3ieimpact.org/ admin/pdfs_synthetic2/1.pdf. (8) Borenstein, M.; Hedges, L. V.; Higgins, J. P. T.; Rothstein, H. R. Introduction to Meta-Analysis; Wiley and Sons: Chichester, 2009. (9) Clasen, T.; Bartram, J.; Colford, J.; Luby, S.; Quick, R.; Sobsey, M. Comment on Household Water Treatment in Poor Populations: Is There Enough Evidence for Scaling up Now? Environ. Sci. Technol. 2009, 43 (14), 55425544. (10) Wood, L.; Egger, M.; Gluud, L. L.; Schulz, K. F.; Juni, P.; Altman, D. G.; Gluud, C.; Martin, R. M.; Wood, A. J. G.; Sterne, J. A. C. Empirical evidence of bias in treatment effect estimates in controlled trials with different interventions and outcomes: Meta-epidemiological study. Brit. Med. J. 2008, 336, 601605. (11) Stats Direct; http://www.statsdirect.com/. (12) Sterne, J. A. C.; Bradburn, M. J.; Egger, M. Meta-analysis in STATA. In Systematic Reviews in Health Care; Egger, M., Davey Smith, G., Altman, D. G., Eds.; BMJ Publishing: London, 2001; pp 347369. (13) Hunter, P. R.; Wilkinson, D. C.; Catling, L. A.; Barker, G. C. Metaanalysis of experimental data concerning antimicrobial resistance gene transfer rates during conjugation. Appl. Environ. Microbiol. 2008, 74, 60856090. (14) Jain, S.; Sahanoon, O.; Blanton, E.; Schmitz, A.; Imoro, T.; Hoekstra, M.; Quick, R. The impact of sodium dichloroisocyanurate treatment on household drinking water quality and health in peri-urban Ghana: a randomized placebo controlled, doubleBlinded trial; International Symposium and 4th Annual Network Meeting, Accra, Ghana, 2-5 June, 2008.

(15) Stauber, C. E.; Ortiz, G. M.; Loomis, D. P.; Sobsey, M. D. A randomized controlled trial of the concrete biosand lter and its impact on diarrheal disease in Bonao, Dominican Republic. Am. J. Trop. Med. Hyg. 2009, 80, 286293. (16) Du Preez, M.; Conroy, R. M.; Wright, J. A.; Moyo, S.; Potgieter, N.; Gundry, S. W. Use of ceramic water ltration in the prevention of diarrheal disease: A randomized controlled trial in rural South Africa and Zimbabwe. Am. J. Trop. Med. Hyg. 2008, 79, 696701. (17) Brown, J.; Sobsey, M. D.; Loomis, D. Local drinking water lters reduce diarrheal disease in Cambodia: A randomized, controlled trial of the ceramic water purier. Am. J. Trop. Med. Hyg. 2008, 79, 394400. (18) Mausezahl, D.; Andri Christen, A.; Pacheco, G. D.; Tellez, F. A.; Iriarte, M.; Zapata, M. E.; Cevallos, M.; Hattendorf, J.; Cattaneo, M. D.; Arnold, B.; Smith, T. A.; Colford, J. M., Jr. Solar Drinking Water Disinfection (SODIS) to Reduce Childhood Diarrhoea in Rural Bolivia: A Cluster-Randomized, Controlled Trial. PLoS Med. 2009, 6 (8), e1000125. (19) Sobsey, M. D.; Handzel, T.; Venczel, L. Chlorination and safe storage of household drinking water in developing countries to reduce waterborne disease. Water Sci. Technol. 2003, 47, 221 228. (20) Swift, L.; Hunter, P. R. What do negative associations between potential risk factors and illness in analytical epidemiological studies really mean? Eur. J. Epidemiol. 2004, 19, 219223. (21) Strickland, M. J.; Mac Crawford, J.; Shen, L.; Wilkins, J. R., III. Time-dependent recordkeeping fatigue among youth completing health diaries of unintentional injuries. J. Safety Res. 2006, 37, 487492. (22) Payment, P. Poor efcacy of residual chlorine disinfectant in drinking water to inactivate waterborne pathogens in distribution systems. Can. J. Microbiol. 1999, 45, 709715. (23) Ferley, J. P.; Zmirou, D.; Collin, J. F.; Charrel, M. Etude longitudinale des risques lies a la consommation deaux non ` conformes aux normes bacteriologiques. Rev. Epidemiol. Sante Publique 1986, 34, 8999. (24) Zmirou, D.; Rey, S.; Courtois, J.-P.; Ferley, J.-P.; Blatier, J.-F.; Chevallier, P.; Boudot, J.; Potelon, J.-L.; Mounir, R. Residual health risk after simple chlorine treatment of drinking water in small community systems. Eur. J. Publ. Health 1995, 5, 7581. (25) Meschke, J. S.; McLaughlin, L. A.; Levy, K.; Beck, N. K.; Shin, G.-A.; Eisenberg, J. N. An observational study on the effectiveness of point-of-use chlorination. J. Environ. Health 2009, 7, 4853. (26) Stauber, C. E.; Elliot, M. A.; Koksal, F.; Ortiz, G. M.; DiGiano, F. A.; Sobsey, M. D. Characterisation of the biosand lter for E. coli reductions from household drinking water under controlled laboratory and eld use conditions. Water Sci. Technol. 2006, 54, 17. (27) Murphy, H. M.; McBean, E.; Farahbakhsh, K.; Sampsen, M. Integrated microbial and chemical assessment of ceramic and Biosand water lters in rural Cambodia; 15th Health Related Water Microbiology Symposium, Naxos, 2009; pp 152-154. (28) Hunter, P. R.; Zmirou-Navier, D.; Hartemann, P. Estimating the impact on health of poor reliability of drinking water interventions in developing countries. Sci. Total Environ. 2009, 407, 26212624. (29) Eisenberg, J. N. S.; Scott, J. C.; Porco, T. Integrating disease control strategies: Balancing water sanitation and hygiene interventions to reduce diarrheal disease burden. Am. J. Publ. Health 2007, 97, 846852. (30) Clasen, T. F.; Schmidt, W.-P.; Cairncross, S. Evidence from intervention studies questions need for integrated diarrhoea control strategies; http://www.ajph.org/cgi/eletters/97/5/846.

ES9028217

VOL. 43, NO. 23, 2009 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

8997