Types of studies

Randomised (including cluster‐randomised), quasi‐randomised and non‐randomised controlled trials, controlled before and after studies (cohort or cross‐sectional), interrupted time series (ITS) and historically controlled studies.

Types of participants

Children aged under 18 years.

Types of interventions

Types of outcome measures

Electronic searches

We searched the following databases using a keyword search and MeSH terms. We adapted search terms according to the requirements or individual databases.

• Cochrane Public Health Group Special Register

• MEDLINE (general medicine)

• MEDLINE In‐Process

• Web of Science (including Science Citation Index Expanded (SCI‐EXPANDED) 1970 to present; Social Sciences Citation Index (SSCI) 1970 to present; Conference Proceedings Citation Index‐Science (CPCI‐S) 1990 to present; Conference Proceedings Citation Index ‐ Social Science & Humanities (CPCI‐SSH) 1990 to present)

• EMBASE (general medicine)

• Econlit (economics)

• Global Health (public health)

• Greenfile (environment)

• CAB Abstracts (applied life sciences)

• Trial registers (CENTRAL, metaRegister of Controlled Trials (mRCT))

• Grey literature (www.nyam.org/library/online‐resources/grey‐literature‐report/; http://indexmedicus.afro.who.int/; http://www.bireme.br/php/index.php; www.hellis.org; www.emro.who.int/HIS/VHSL/; http://wprim.wpro.who.int/iah/I/index.htm; 3ie Impact; http://scholar.google.co.uk/)

• Chinese‐language databases available under the China National Knowledge Infrastructure (CNKI‐CAJ) (Fung 2008):

  • Chinese Biomedical Literature database (CBM)

  • China National Knowledge Infrastructure (CNKI)

  • VIP information/Chinese Scientific Journals database

We prepared search strategies in English (Appendix 1) and Chinese (Appendix 2). There were no language or date restrictions. We handsearched reference lists of key articles for any additional relevant articles. We contacted subject experts and study authors and asked them to provide additional information and further relevant references.

We performed an initial literature search in July 2011, followed by an update search in June 2012.

Searching other resources

We contacted the following researcher groups and organisations for information on unpublished and ongoing trials: Public‐Private Partnership for Handwashing with Soap (http://www.globalhandwashing.org/); IRC International Water and Sanitation Centre (http://www.irc.nl/); Department of Child and Adolescent Health and Development (WHO) (http://www.who.int/maternal_child_adolescent/en/); World Bank (http://www.worldbank.org/); World Bank Water and Sanitation Programme (http://water.worldbank.org/related‐topics/water‐and‐sanitation‐program); World Health Organization (WHO) (http://www.who.int/en/) and United Nations Children's Fund (UNICEF) (http://www.unicef.org.uk/); International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B) (http://www.icddrb.org/); Water, Sanitation and Health Programme (WHO) (http://www.who.int/water_sanitation_health/en/); Environmental Health Project (USAID) (http://www.ehproject.org/); Foodborne and Diarrheal Diseases Branch, Centers for Disease Control and Prevention (CDC) (http://www.cdc.gov/foodborne/about_fddb.htm); USAID (http://www.usaid.gov/) and UK Department for International Development (DFID) (http://www.dfid.gov.uk/).

We also searched the following for relevant abstracts:

• Waterlines journal;

• International Water Association and the Water, Engineering and Development Centre (Loughborough University, UK);

• public health conferences (e.g. American Public Health Association; European Public Health Association).

Selection of studies

Two authors (LW and SB) reviewed the titles and abstracts retrieved through the English‐language search strategy independently in order to identify and select potentially relevant studies using pre‐defined inclusion criteria, and the full text of all articles selected by either team member were retrieved for a full‐text review. One author (YC) reviewed the results of the Chinese‐language search, undertook the same process and summarised the article in English; LW and SB reviewed the summaries to independently determine study eligibility. Where there was a difference of opinion, disagreement was resolved through discussion with a third review author (ADD). All studies which initially appeared to meet the inclusion criteria but upon inspection of the full text did not meet inclusion criteria are detailed in the Characteristics of excluded studies table with reasons for exclusion. We contacted authors of 12 potentially eligible excluded studies (in English and Chinese) that did not report nutrition outcomes and asked them to provide information on the availability of nutrition outcome data. All authors responded, and one author (two studies) provided additional unpublished data (Luby 2004; Luby 2006).

Data extraction and management

Two authors (LW and SB) independently extracted data from all relevant articles; LW contacted authors to supply missing data where possible. The data extraction forms were based on the data collection form from the Cochrane Effective Practice and Organisation of Care (EPOC) Group and Cochrane Public Health Group, modified for use in this review. Quality criteria questions for the different study designs were built into this form. Any discrepancies between the two review authors were resolved by a third author (ADD). LW entered the extracted data into Review Manager 5 (RevMan 2012). Multiple papers reporting results from one study were considered as one study. We used a standard approach where comparisons of multiple reports and publications of the same study were checked for contradictions and completeness and the data used once.

Assessment of risk of bias in included studies

Two review authors (LW and SB) independently assessed the risk of bias of included studies using the EPOC 'Risk of bias' tool for studies with a separate control group. This tool includes additional items to assess the risk of selection bias and subsequent confounding ("were baseline outcome measurements similar?" and "were baseline characteristics similar?"), as well as an additional item to consider the likelihood of contamination ("was the study adequately protected against contamination?"). We also supplemented the EPOC 'Risk of bias' tool with another additional item to address whether the study authors appropriately adjusted for important confounders in their analysis. We assessed studies for each item with answers of 'Low' indicating low risk of bias, 'High' indicating high risk of bias and 'Unclear' indicating either lack of information or uncertainty over the potential for bias. We contacted study authors for additional information where possible. Any discrepancies were resolved by a third author (ADD).

Measures of treatment effect

We present treatment effect sizes for continuous outcomes and report them using their original scale. All measures of effect are presented with 95% confidence intervals (CI).

Unit of analysis issues

The unit of analysis was the individual. We adjusted data derived from the five cluster‐randomised controlled trials (Du Preez 2010; Du Preez 2011; McGuigan 2011; Luby 2004; Luby 2006) to allow for the clustered design.

Dealing with missing data

We contacted authors of included studies requesting missing information and received responses to all our queries. Through this process we were sent data sets including nutrition outcomes in children from three studies (Du Preez 2010; Du Preez 2011; McGuigan 2011), only one of which had previously been reported (Du Preez 2011). We also contacted authors of 12 potentially eligible excluded studies (in English and Chinese) that did not report nutrition outcomes and were sent two further data sets of previously unpublished data (Luby 2004; Luby 2006). Study authors also provided unpublished trial protocols when available.

Assessment of heterogeneity

Clinical heterogeneity caused by differences in participant characteristics is likely to be moderate as studies only included poorly nourished children from low‐ or middle‐income countries. We assessed statistical heterogeneity using the I² statistic according to guidance of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011 Sections 9.5 to 9.6).

Assessment of reporting biases

Nutritional outcome measures have only rarely been assessed in WASH studies to date and when they are measured they are often classified as secondary outcomes. In our review, we identified four studies in which nutritional status measures had been collected but had not been reported in the published literature. We contacted the principal investigators of these studies and they immediately provided us with full access to these unreported data. Despite a rigorous search, it is possible that other WASH studies have collected nutrition data that have not been reported.

Data synthesis

We conducted data synthesis of study outcomes by group (intervention and control) using Review Manager 5. We created forest plots without meta‐analysis including all possible studies by outcome to enable visual inspection of the available data. Substantial heterogeneity in the designs of studies included in the review limited the amount of meta‐analysis possible. Meta‐analysis combining randomised and non‐randomised studies is not recommended (Higgins 2011) and formal meta‐analysis was therefore restricted to include only the five identified cluster‐randomised controlled trials (Du Preez 2010; Du Preez 2011; Luby 2004; Luby 2006; McGuigan 2011). All trials were primarily designed to reduce diarrhoea incidence, and diarrhoea incidence lies on the direct causal pathway between WASH and nutrition outcomes in children. The pooling of different WASH interventions in meta‐analysis allowed the combined impact of WASH interventions to be assessed. We summarised the remaining nine studies using a narrative synthesis. Our analysis of the effect of the intervention is reported for each study separately (Effects of interventions).

Study authors provided raw anthropometric data for five cluster‐randomised controlled trials (Du Preez 2010; Du Preez 2011; Luby 2004; Luby 2006; McGuigan 2011) from which anthropometric indices (weight‐for‐age, weight‐for‐height and height‐for‐age) were calculated using the WHO Anthro software (available at: http://www.who.int/childgrowth/en/). For these trials, we calculated means and cluster‐adjusted standard deviations (SD) in Stata version 12 (http://www.stata.com/stata12/) prior to analysis in Review Manager 5. Study‐level meta‐analysis included data on 4622 to 4627 children (depending on outcome) who had baseline and final measures in the included cluster‐randomised controlled trials.

Three studies included more than one intervention arm (Bowen 2012; Luby 2004; Luby 2006):

Bowen 2012 had two intervention arms: handwashing with soap and handwashing with soap plus treatment of drinking water with flocculent disinfectant. The overall effect of intervention versus control was examined by pooling the individual effect of each of these intervention arms (mean and SD) and weighting the pooled values for the numbers within each arm.

Luby 2004 had two intervention arms: handwashing with antibacterial soap, and handwashing with plain soap. In study‐level meta‐analysis the overall effect of intervention versus control was examined by pooling the individual effect of each intervention arm (mean and SD) and weighting the pooled values for the numbers within each arm.

Luby 2006 had four intervention arms: handwashing with soap, handwashing with soap plus treatment of drinking water with flocculent disinfectant, treatment of drinking water with bleach and treatment of drinking water with flocculent disinfectant. In study‐level meta‐analysis we combined the effect of water quality interventions (treatment of drinking water with bleach and treatment of drinking water with flocculent disinfectant), and compared this with promotion of handwashing with soap, promotion of handwashing with soap and treatment of drinking water with flocculent disinfectant, and the control arm. Overall values were weighted for the numbers within each arm. In our analysis, Luby 2006 included three intervention groups (soap, soap plus flocculent disinfectant and flocculent disinfectant and bleach) and for this study the shared group (control) was split into three to provide three reasonably independent comparisons (see Cochrane Handbook for Systematic Reviews of Interventions, Section 16.5.4).

We conducted individual participant data (IPD) analysis in Stata version 12 using the raw data provided from the five cluster‐randomised controlled trials only (Du Preez 2010; Du Preez 2011; Luby 2004; Luby 2006; McGuigan 2011). In IPD analysis, we estimated the effect of WASH interventions on weight‐for‐age, weight‐for‐height and height‐for‐age, height and weight using analysis of covariance models (i.e. final follow‐up measure adjusted for baseline measure) additionally adjusting for age at baseline, duration of treatment and sex. The analysis used random‐effects models to allow for both within‐study and within‐cluster variability, where cluster is the unit of randomisation in each study. Heterogeneity of treatment effects across trials was tested by including a random slope for the intervention. Children with missing data were removed from the analysis. In IPD analysis we pooled the data from studies with more than one intervention arm to allow comparison of outcome of children in intervention and control arms. IPD meta‐analysis included data on 5375 to 5386 children (depending on outcome) who had baseline and at least one follow‐up measure in the five included cluster‐randomised controlled trials.

We conducted the analyses on an intention‐to‐treat basis. We collated a 'Summary of findings' table, providing information on primary outcomes, effect sizes and quality of information.

Subgroup analysis and investigation of heterogeneity

We conducted pre‐specified subgroup analysis by sex and age group (two years and under and two to five years) for those studies eligible for inclusion in meta‐analysis. All of the studies in the review were six months or longer in duration, conducted in low‐ or middle‐income countries and conducted in a mixture of rural, peri‐urban and urban locations, and only one study included children aged over five years, precluding analysis by other pre‐specified subgroups (duration of intervention, country setting, community location, age group over five years).

Sensitivity analysis

We planned to carry out a sensitivity analysis on studies judged to be at a low risk of bias. However, no sensitivity analysis was conducted as no included study was judged to be free of bias.