Impact on childhood mortality of interventions to improve drinking water, sanitation and hygiene (WASH) to households: systematic review and meta-analysis

Background In low- and middle-income countries (L&MICs), the biggest contributing factors to the global burden of disease in childhood are deaths due to respiratory illness and diarrhoea, both of which are closely related to use of water, sanitation and hygiene (WASH) services. However, current estimates of the health impacts of WASH improvements use self-reported morbidity, which may fail to capture longer-term or more severe impacts. Moreover, reported mortality is thought to be less prone to bias. This study aimed to answer the question: what are the impacts of WASH intervention improvements on reported childhood mortality in L&MICs? Methods and findings We conducted a systematic review and meta-analysis, using a published protocol. Systematic searches of 11 academic databases and trial registries, plus organisational repositories, were undertaken to locate studies of WASH interventions which were published in peer review journals or other sources (e.g., organisational reports and working papers). Intervention trials of WASH improvements implemented under endemic disease conditions in L&MICs were eligible, from studies which reported findings at any time until March 2020. We used the participant flow data supplied in response to journal editors calls for greater transparency. Data were collected by two authors working independently. We included evidence from 24 randomized and 11 non-randomized studies of WASH interventions from all global regions, incorporating 2,600 deaths. Effects of 48 WASH treatment arms were included in analysis. We critically appraised and synthesised evidence using meta-analysis to improve statistical power. We found WASH improvements are associated with a significant reduction of 17 percent in the odds of all-cause mortality in childhood (OR=0.83, 95%CI=0.74, 0.92, evidence from 38 interventions), and a significant reduction in diarrhoea mortality of 45 percent (OR=0.55, 95%CI=0.35, 0.84; 10 interventions). Further analysis by WASH technology suggested interventions providing improved water in quantity to households were most consistently associated with reductions in all-cause mortality. Community-wide sanitation was most consistently associated with reductions in diarrhoea mortality. Around one-half of the included studies were assessed as being at moderate risk of bias in attributing mortality in childhood to the WASH intervention, and no studies were found to be at low risk of bias. The review should be updated to incorporate additional published and unpublished participant flow data. Conclusions The findings are congruent with theories of infectious disease transmission. Washing with water presents a barrier to respiratory illness and diarrhoea, which are the two main components of all-cause mortality in childhood in L&MICs. Community-wide sanitation halts the spread of diarrhoea. We observed that evidence synthesis can provide new findings, going beyond the underlying data from trials to generate crucial insights for policy. Transparent reporting in trials creates opportunities for research synthesis to answer questions about mortality, which individual studies of interventions cannot be reliably designed to address.


Abstract 23
Background 24 In low-and middle-income countries (L&MICs), the biggest contributing factors to the 25 global burden of disease in childhood are deaths due to respiratory illness and 26 diarrhoea, both of which are closely related to use of water, sanitation and hygiene 27 (WASH) services. However, current estimates of the health impacts of WASH 28 improvements use self-reported morbidity, which may fail to capture longer-term or 29 more severe impacts. Moreover, reported mortality is thought to be less prone to bias. 30 This study aimed to answer the question: what are the impacts of WASH intervention 31 improvements on reported childhood mortality in L&MICs? 32 Methods and findings 33 We conducted a systematic review and meta-analysis, using a published protocol.  Briscoe et al. [19] highlighted how diarrhoeal illness becomes normalised among 137 highly exposed groups over time which leads to underreporting, a problem we might 138 expect to become worse when reporting is done by someone other than the patient, 139 in this case the child's carer. Or illness may be acknowledged differently by sex [20], 140 where girls who complain about pain are less likely than boys to be pacified by their 141 carers and therefore may not report it. In other words, we may not see additive 142 effects of multiple WASH technologies provided together if bias in the reporting of 143 disease outcomes, rather than diarrhoea epidemiology, is driving the findings.

145
The key advantage of randomised controlled trials (RCTs) over other methods is the 146 clarity with which randomisation balances unobservable differences across groups in 147 expectation, not in any single trial, but over multiple draws from the population [21]. 148 Thus the "gold standard" for evidence on health impacts from these studies uses 149 meta-analysis of findings from multiple studies [22]. However, meta-analysis can 150 also magnify biases, because is harder to identify errors where they pervade the 151 whole data set. Some approach is clearly needed to address reporting bias. Of great 152 potential concern is publication bias, the phenomenon whereby trials are more likely 153 to be published if they find significant effects, a factor that is made more likely when 154 trials are funded by private manufacturers, as has been common in studies of water 155 treatment (chlorine, water filters) and hygiene (soap) [23]. studies themselves did not aim to do so. We conducted a systematic review of the 161 effects of WASH interventions on child mortality in L&MIC contexts, drawing on a 162 number of sources including losses to follow-up due to mortality as reported in 163 participant flows. It is an established finding that study participants do not misreport 164 death, even in open studies [15,16]. This might be because death of a child is a rare 165 and salient event. The crucial advantage of this approach, therefore, is that reported 166 mortality is less prone to bias.

168
We sought to answer four review questions: is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. 9 183 full description of the procedures followed for searches, study inclusion, outcomes 184 data collection, analysis and reporting is presented in the published protocol [24]. 185 Searches for literature were done as part of an evidence and gap map [10]. Studies  practices (e.g., the washing of food, clothing and fomites). We excluded trial arms . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 207 with a major non-WASH component (e.g., nutrition interventions). We classified 208 WASH interventions according to the "main WASH" technology provided, which was 209 either water supply, water treatment and storage, sanitation or hygiene technologies 210 provided or promoted alone, or multiple combinations of WASH technologies. It was 211 also possible to characterise interventions by whether they provided any 212 improvements in water supply, water treatment, sanitation and/or hygiene alone or in 213 combination with others, which we refer to as "any WASH". This was due to 214 problems in clearly identifying all the components of an intervention. For example, a 215 debate among practitioners suggested that hand hygiene messaging is usually 216 incorporated in CLTS [26].

218
Counterfactual conditions were categorised as "improved" or "unimproved" according 219 to the WHO/UNICEF Joint Monitoring Programme (JMP) classification. Improved 220 water supplies were defined where the majority of households in the sample used 221 drinking water from an improved source (e.g., piped water to the household, a 222 community standpipe or protected spring) within a 30-minute round-trip including 223 waiting time. For sanitation, the counterfactual scenario was defined as "improved" if 224 the majority of controls had a sewer connection to the home or an improved pit  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Template data collection forms are available in the study protocol [24]. Data 242 extracted from included studies is provided in S1 Annex Table A3. The dataset used 243 in analysis is provided in S4 Dataset.

244
Measuring mortality outcomes 245 The primary outcomes for the review were all-cause mortality and mortality due to 246 diarrhoeal disease. Outcomes data were collected independently by two researchers 247 from two sources. The first source was the few studies that reported mortality 248 alongside statistical information [6][7][8]32,33]. Mortality data were also recoverable 249 from studies that reported losses to follow-up (attrition) in sample populations. L&MICs to obtain crude mortality rates for field trials by intervention group. These 252 studies therefore formed the major source of evidence on all-cause mortality. Some . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint 12 253 studies also reported cause-specific mortality rates, including diarrhoea and other 254 infections, defined by carers in verbal autopsy and/or clinicians, or collected from 255 vital registries. All-cause mortality and mortality due to diarrhoea or other infections 256 were defined by carers in self-report or taken from vital registries.

258
Mortality rates were computed over a standard period, as mortality measurements 259 increase over longer exposure periods. Age-specific (e.g., under-2) mortality rates 260 were defined where these data were available [6][7][8]34], or, if they were not, crude 261 mortality rates were taken over the data collection period. Intervention effects were 262 measured as the odds ratio (OR) of the mortality rates, and their 95 percent 263 confidence intervals. Where studies reported multiple intervention arms against a 264 single control arm, we split the control sample assuming an equal mortality rate for 265 each comparison. We applied a continuity correction in study arms where there were 266 no deaths, by adding 0.5 to all frequencies, which can cause bias in meta-analysis of 267 rare events [35]. These studies were assessed as being at 'high risk of bias' in the 268 outcome measurement domain [36][37][38][39][40].

269
Evidence synthesis approach 270 Overall pooled effects were estimated for all-cause mortality (review question 1) and 271 diarrhoea mortality (review question 2) using Stata. We assessed the consistency of 272 the pooled effects using I-squared and tau-squared statistics to measure the relative 273 and absolute heterogeneity between studies. We tested for effect moderators in 274 meta-analysis and meta-regression analysis, including the WASH intervention 275 technology provided to study participants, water supply and sanitation conditions in . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint were pre-specified based on theory and previous reviews, with the exception of the 284 moderator analysis by baseline mortality rate. We used meta-regression plots to 285 assess the predicted effects of the interventions by baseline mortality rate (review 286 question 3).

288
We evaluated the likelihood that potential biases could cast doubt on the findings 289 results through a negative control, formal publication bias assessment, and 290 sensitivity analysis (review question 4). The effects of WASH improvements on 291 mortality are largely expected to occur by blocking transmission of infectious 292 diseases, primarily faeco-oral and respiratory infections, in childhood. People who 293 survive beyond the age of 5 are thought to have developed sufficiently robust 294 immunity to these diseases, hence the effects of WASH improvements on mortality 295 among older groups is expected to be far weaker. Therefore, as a negative control 296 [41,42], meta-analysis was estimated for those studies that reported all-cause 297 mortality among participants aged over 5 years. We also assessed the sensitivity of 298 the pooled effects to exclusion of each single effect, examined whether there was a 299 correlation between risk-of-bias rating and the estimated effect, and tested for small-. CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ; https://doi.org/10.1101/2023.03.13.23287185 doi: medRxiv preprint 300 study effects (publication bias) at the review level using graphical inspection of 301 funnel plots and regression tests.

304
From 13,500 de-duplicated records, 684 full text reports of WASH intervention 305 studies were screened, of which 35 were identified that reported mortality outcomes, 306 30 of which were measured among children aged 5 or under (Fig 1). We were not 307 able to incorporate trials that met the review inclusion criteria but did not report 308 participant flows (e.g., [43]). We found 24 RCTs that measured mortality, all of which 309 were published in peer review journals. RCTs were of water treatment and storage, 310 sanitation and/or hygiene interventions, which mainly used cluster design, with 311 clustering at the community level. We found no RCTs of water supply provision or 312 promotion that reported mortality estimates. Several studies used prospective non-313 randomised trial designs [33,36,44], five analysed cohort data [38,[45][46][47][48] is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint We included in meta-analysis 38 WASH study arms examining all-cause mortality in 355 childhood, of which 26 were from RCTs, and 10 examining diarrhoea mortality, of 356 which 6 were from RCTs. For six studies we could also extract seven estimates of 357 effects on mortality among adults and/or children aged over-5 [32,39,47,52-54] is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint all participants received hygiene education in one study [56]. Most counterfactual 386 populations were assessed as using improved water supplies [7,8,8,34,36-387 38,40,44,45,48-50,53,57,58,63-66]. In a few instances, counterfactuals received 388 piped water inside the compound [36,50], otherwise it was sourced by household 389 members from outside. In one study of continuous water supply ("safely managed 390 drinking water") provision, controls received water for only a few hours a week on 391 average [48]. There were also concerns about reliability of or distance to the water 392 supply in a few studies [8,65], which would have affected ability of study participants 393 to practice improved hygiene. In under half of cases, sanitation was classified as 394 being improved [6,36,38,40,44,[48][49][50]56,[63][64][65]. In all others, the majority of 395 households openly defaecated, or used shared facilities or unimproved facilities like 396 pits without concrete slabs. Imputations were made where it was not clear exactly 397 what types of water and sanitation services were used by households in the 398 counterfactual scenario [33,34,37,38,[44][45][46]49,50,52,55,58,61].

400
In general, just under half of studies (40%) were found to be at 'moderate risk of 401 bias' overall in attributing changes to the intervention, for all-cause mortality (Fig 2a) 402 and mortality due to diarrhoea (Fig 2b). No studies were at 'low risk of bias'. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint One-third of RCTs reported using adequate allocation sequence and concealment, 409 and demonstrated baseline covariate balance, to satisfy a 'low risk' rating on 410 confounding. In some cases, data were collected on water, sanitation and hygiene at 411 pre-test, but balance was not presented for all relevant variables, such as sanitation 412 and hygiene access. Three NRSI were assessed as being at 'moderate risk of bias'  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint participation was largely determined by programme placement, which is thought less 417 problematic to address than self-selection into programmes by participants. In Where participants were recruited before allocation in cluster-RCTs, or where 429 recruiters were blinded to allocation, the studies were judged to be at 'low risk' of 430 selection bias. Where recruitment was done afterwards by those potentially with 431 knowledge of allocation or where individuals needed to be recruited later due to 432 attrition (losses to follow-up during the trial), the study was judged to be at risk of 433 bias. Studies were also assessed as being at 'high risk of bias' when overall attrition 434 rates were greater than 20 percent, or differential attrition greater than 10 percentage 435 points, or where no information was provided about reasons for dropouts by 436 intervention group, tests for covariate balance or robustness of findings. Selection is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint In general, departures from intended interventions due to contamination (controls 441 receive the treatment) or spill-over effects (control outcomes are caused by 442 treatment outcomes) were judged unlikely to be problematic in many studies, which 443 used cluster-randomisation and reported geographical separation of groups. Of 444 specific relevance to mortality estimates, studies providing ORS to severely ill 445 children and/or encouraging mothers to attend health clinic were judged to have high 446 risk of bias in the outcome measure.

448
Regarding outcome measurement, all-cause mortality was usually categorised as 449 being a reliable measure even when self-reported with long recall, owing to the 450 salience and rarity of the event; the longest recall was 6 years [65], the shortest two 451 days [38], and usually it was 12 months or less. However, there is greater suspicion 452 about cause-specific mortality where reporting is through verbal autopsy by the 453 child's carer. If cause-specific mortality was measured, assessment was therefore 454 made as to whether it was verified by a clinician or taken from vital registration, in 455 which case it was assessed as being at 'low risk of bias'. While observational studies 456 of WASH provision have verified cause of death through consultation with a clinician

457
[5], no RCTs and only two NRSI used vital registration data [44,50]. One study [44] 458 was assessed as at 'low risk' of outcome reporting bias for diarrhoea mortality, while 459 another was assessed as at 'high risk of bias' because the study did not attribute . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Nearly all trials were pre-registered, four reported publishing a protocol with pre-464 analysis plan [6][7][8]60], and three blinded data analysts [6][7][8]. In addition, one NRSI 465 was deemed to have 'low risk of bias' on reporting, because it published a baseline 466 report with pre-analysis plan [68].

467
Impacts of WASH on all-cause mortality (review question 1) 468 We conducted meta-analysis across intervention arms reporting all-cause mortality 469 in children aged under 5 years (Fig 3). WASH improvements typically reduced the  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint For the stratified meta-analyses by WASH technology, trial arms incorporating "any 477 WASH" -that is, any single water supply, water treatment, sanitation or hygiene 478 technology, whether provided alone or alongside any other WASH technology -479 were meta-analysed. We found a 34 percent reduction in the odds of mortality for 480 water supply improvements (OR=0.66, 95%CI=0.50, 0.88; I-squared=66%; 7 481 estimates) (Fig 4a)   is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  The overall effect of hygiene promotion was not statistically significant (OR=0.85, 500 95%CI=0.69, 1.04; I-squared=33%; 17 estimates). Five of the studies were assessed 501 as being at 'high risk of bias ' [33,36,45,65,66] and seven were at 'moderate risk of is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint where the water supply was also being improved or had been improved previously 511 (OR=0.71, 95%CI=0.56, 0.90; I-squared=18%; 11 estimates) (Fig 4c).   is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint   is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint We estimated meta-regressions to explore further whether the variation in effects by 540 WASH technology intervention, and the other contextual factors we had identified 541 from theory, might explain differences across studies (Table 1). The regression 542 pooled data from study participants of any age, incorporating the 14 additional 543 estimates measured among all population groups or adults and children aged over 5.

544
The reductions in mortality were significantly larger when interventions were 545 conducted in circumstances where: participants were children aged under 5 years, or 546 data collection was limited to the summer rainy season. Where the study collected 547 data over a shorter follow-up period, the effect on mortality was also significantly 548 larger. Impacts on mortality were significantly greater when water supply 549 improvements were made. The explanatory power of the regression was high (R-  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

562
The meta-analysis of diarrhoea mortality in childhood suggested WASH provision 563 and promotion lead to a reduction in the odds of death due to diarrhoea by 45 564 percent (OR=0.55, 95%CI=0.35, 0.84; 10 estimates) (Fig 5). Six of the studies were 565 assessed as being at 'high risk of bias ' [33,38,44,46,50,65]  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 One of those factors is the degree of movement along the WASH ladders. We tested 574 this hypothesis in moderator analysis according to the type of water supply and 575 sanitation facilities used in the counterfactual group. When the WASH interventions 576 were provided when counterfactuals were using no or unimproved sanitation and 577 water supplies, and therefore exposed to very high risk of environmental is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  and hygiene promotion in circumstances when water supplies were improved, were 611 associated with significantly larger impacts on diarrhoea mortality (S1 Annex Table   612 A4).

614
question 3) 615 We tested for a theoretical relationship between the contextual starting values and 616 programme effectiveness -that is, one might expect higher returns from a lower 617 base -by plotting the relationships between the baseline mortality rate measured in 618 the counterfactual group and the log-odds ratios for all-cause (Fig 6a) and diarrhoea 619 mortality (Fig 6b). The results suggested that, at higher baseline mortality rates, . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023   is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint

634
In this section we present findings from a negative control, analysis of small study 635 effects and the results of sensitivity analyses. Using meta-analysis to power studies 636 adequately with small effect sizes does not necessarily generate effects that are 637 statistically significant if there is no underlying causal relationship [70]. The meta-638 analysis of studies reporting all-cause mortality among participants aged over 5 639 years did not suggest WASH improvements affected mortality when participants 640 were restricted to adults and children aged over 5 (OR=1.05, 95%CI=0.93, 1.19, I-641 squared=0%, 7 estimates) (Fig 7). The study with the largest effect on mortality was 642 of health messaging among 10-year-old school children [39]. Several of the studies 643 were of chlorination [54,56,57]. We might expect to see effects on maternal mortality is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ; https://doi.org/10.1101/2023.03.13.23287185 doi: medRxiv preprint 644 due to sepsis, which improved WASH -particularly in places of birth like health 645 facilities -is thought to alleviate [71]. None of the interventions provided a WASH 646 intervention in a health facility. Since the mortality data were largely collected from participant flow diagrams, the 651 fact that mortality estimates are available at all is indicative of the good standards of 652 reporting in the studies included in this review. This suggested publication bias was 653 likely to be limited, most clearly for prospective trials of WASH interventions, as 654 found in the analysis of small study effects (S3 Annex Fig A5). We also tested the  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023

730
Meta-regression analysis suggested approximately three-quarters of deaths in 731 childhood could be averted when WASH interventions are provided to 732 immunocompromised groups during the peak diarrhoea season, against 733 counterfactuals living in very poor communities with unimproved sanitation services. 734 We found no evidence of publication bias due to small-study effects in trials of 735 WASH interventions, presumably because mortality was not defined as an outcome 736 in these studies. where existing water supply and sanitation services are not available or unimproved, 748 so that community members are not able to practice hand washing and are openly 749 defaecating, or using facilities that are either shared between two or more 750 households or ones that do not adequately separate excreta from the environment. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023  Non-randomised studies at 'high risk of bias' can produce inflated effects, as we 770 found here, because p-hacking would tend to increase effect size magnitudes.

771
However, we estimated the opposite effect for RCTs -that 'high risk of bias' is 772 associated with smaller effects on mortality -a finding which is consistent with site 773 selection bias [80,81]. In other words, trials that are more carefully conducted and 774 reported are of interventions that also tend to be designed and implemented 775 appropriately to the local context, and therefore adhered to, hence being more . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 effective. An example is when interventions promote hand washing (e.g., education, 777 social marketing, soap provision) in contexts where the quantity of water available to 778 households is sufficient to practice domestic cleanliness; or, if it is thought not to be, 779 improvements in water supply access or reliability are made too.

780
Findings in relation to other systematic reviews 781 The evidence presented here, that water supply, hygiene improvements and 782 community-wide sanitation save children's lives in L&MICs, is consistent with 783 findings from an early review [82], but in several respects is quite different from later 784 reviews. These have not tended to find significant effects on diarrhoea morbidity of 785 interventions which aim to improve access to water in quantity for household use.

786
The most recent review by the WHO suggests that clean drinking water provided at 787 the point-of-use, particularly by filtration, reduces reported diarrhoeal illness by 788 around one-half [11]. Reviews have found that HWT appears to be more effective 789 when a protective container is also provided [83], as it may be for example in 790 household filtration devices when drinking water is accessed through a straw or tap.

791
Reviews have also found smaller or null effects for household water treatment is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ; https://doi.org/10.1101/2023.03.13.23287185 doi: medRxiv preprint 798 availability of treatment. However, many of the papers and contexts included in this 799 review are also represented in the reviews of morbidity.  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Hence, regarding the quality of the evidence collected here, reported mortality is not 821 thought to be a biased measure per se. All-cause mortality data can also be 822 triangulated with corresponding data from other sources, such as vital registration, 823 and even the possible effect of other diseases, such as respiratory infections [93].

824
Cause-specific death rates are thought less reliable [16], dependent as they are on a 825 verbal autopsy interview with the bereaved family of the patient, who may be too 826 distraught to give an unbiased, let alone a coherent account of the patient's last 827 days. But, like all-cause mortality, verbal autopsy can be triangulated with, or done 828 by, a physician, which we incorporated in the risk-of-bias assessment. Vital 829 registration and verbal autopsy estimates are also used in GBD calculations.

831
A potentially more serious source of bias is differential attrition. During survey 832 interviews deaths will not be reported for mothers who migrated or died. To the 833 extent that WASH interventions affect migration and adult mortality rates, child 834 mortality rates might be downwards biased in intervention areas. In other words, a 835 potential source of bias affecting the crude death rate calculations used in this study 836 is that they are right-censored: that is, where data are collected contemporaneously 837 among participants regardless of age, children born into the study or who migrate out 838 and younger children will have completed shorter durations than older children; the 839 data on pre-and neo-natal mortality may also be right censored by maternal deaths is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Transparent study reporting is crucial for accountability and learning by enabling 878 effects for relevant outcomes to be measured. A common source of bias in WASH 879 trials is caused by differential losses to follow-up out of the study (attrition). How  Water is an important enabling factor for practising hand and food hygiene and some 890 forms of sanitation (e.g., flush toilets), but articles do not typically report data on . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 distance to the water source, or water consumption (litres per capita per day) and 892 how it is used (e.g., whether consumed or used in bathing). This information is 893 crucial for understanding mechanisms and therefore the generalisability of the   904 We found large and consistent effects of water supply improvements on all-cause 905 mortality in childhood, and of community-wide sanitation improvements on diarrhoea 906 mortality. The contribution of this synthesis -to use participant flow data to provide 907 estimates of changes in child mortality associated with WASH interventions -has 908 been enabled by studies that use agreed standards of reporting such as CONSORT.

909
There is potentially a large number of estimates of mortality in childhood from studies 910 which do not use these methods of reporting, as a recent meta-analysis of 911 household water treatment has indicated [101]. Going forward, the challenge will be 912 for an author collaborative to provide sufficient incentives to obtain unpublished 913 participant flow data, to ensure that future systematic reviews and meta-analyses are . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted March 14, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023