Approaches to promote handwashing and sanitation behaviour change in low- and middle-income countries: a mixed method systematic review

1.2.1 Approaches to promote behaviour change

To improve effectiveness of WASH interventions, increasing attention is currently being focused on the design of programmes and the selection of approaches to promote WASH behaviour change. Several approaches have been developed over the last two decades, and are currently being applied in practice to promote uptake of WASH interventions and to achieve WASH behaviour change (Peal et al., 2010). The approaches can be grouped in the following categories: •

Community-based participatory approaches (as in the case of programmes such as Community Led Total Sanitation (CLTS), Participatory Rural Appraisal (PRA), Participatory Hygiene and Sanitation Transformation (PHAST), Self-esteem, Associative Strengths, Resourcefulness, Action-Planning, and Responsibility (SARAR), community reunion, community hygiene club/mother club, community health clubs (CHC), child-to-child approach (CtC), Urban Led Total Sanitation (ULTS), Community Approaches to Total Sanitation (CATS), Methodology for Participatory Assessments (MPA), Community Action Planning (CAP), Child Hygiene and Sanitation Training/Transformation (CHAST), and the model home approach). A promotional approach is considered a “community-based approach” when one of the above-mentioned programmes is reported, or where it is clearly indicated that community members are invited and there is shared decision-making. A community-based approach works with the whole community, and typically community meetings which trigger behaviour change are conducted.

A promotional approach can contain different promotional elements, depending on the context for which the programme was developed. Based on the main focus or major element of the promotional approach, we classified the promotional approaches/promotional elements for the purpose of this review in 4 groups: community-based approaches, social marketing approaches, sanitation and hygiene messaging, and elements of psychosocial theory (detailed explanation below).

Any of the approaches above can be delivered using one or more different communication strategies: •

Interpersonal communication: peer to peer, home visits, focus group; either of these approaches could work with change/transformation agents such as hygiene promotors, WASH Committees, champions/natural leaders who are not part of community leadership system, community leaders (chefs, elected village/ appointed village leaders, councillors, etc.), religious leaders, teachers, Village Health Workers, Local Government Staff (dealing with WASH, Social Services, Health, etc.), volunteers (e.g. Red Cross volunteers), lecture, workshops, games, material provision with demonstration, quiz.

It is not always clear which of these approaches is the most effective in relation to sanitation and hygiene behaviour change, and other outcomes leading to behaviour change (e.g. learning outcomes) or longer term outcomes that follow from behaviour change (e.g. mortality, morbidity). In the WASH sector, the evaluation of programmes tends to focus on intended outcomes and impacts (whether the intervention worked and what effect it had on outcomes) but not on appraising the process of implementation and establishing how the use of a specific approach leads to changes in outcomes. However, decision makers need to know the critical factors in the process of implementation that ensure that impacts are achieved and sustained, and how scaling up is best achieved.

For the purpose of this review we focused on approaches to promote handwashing and sanitation interventions, with behaviour change as the main outcome. To be able to make this choice we developed a review of existing systematic reviews (see below, 1.4). Since adherence to water, sanitation and hygiene programmes is known to be highly associated with factors such as gender, socioeconomic status, education and occupation, equity factors are also considered in this systematic review (DFID, 2013). Since the effect of WASH interventions on health outcomes (such as diarrhoea, cholera, trachoma, helminth infections) has been shown in many existing individual studies and systematic reviews (Cairncross et al., 2010; Dangour et al., 2013; Fewtrell et al., 2005; Peletz et al., 2013; Stocks et al., 2014; Strunz et al., 2014; Taylor et al., 2015; Waddington, 2009), and practicing/showing the right behaviour is a pre-requisite for health impacts, health outcomes are also looked at in those studies that measured behaviour change. Although it would be relevant to include studies that measured cost-effectiveness, this is outside the review scope.

1.2.2 Definitions

In the context of this review, we used the following definitions:

Behaviour change: Influencing the intention, use and habit in the performance of a certain behaviour (Mosler, 2012).

Intention: Intention represents a person's readiness to practice a behaviour: how willing the person is to implement a behaviour (Mosler, 2012). Intention can include for example “partial construction” or “savings for latrine construction”.

Use: Refers to the execution of actions. Both the desired behaviour and competing behaviours must be considered (Mosler, 2012). “Use” consists of uptake, adherence and longer-term use: •

Uptake: Uptake is defined as the actual use or non-use (Lillevol et al., 2014). For the purpose of this project we define this outcome as use during the implementation of the programme.

Habit: Habits are routinized behaviours that are executed in specific, repeating situations nearly automatically and without any cognitive effort (Mosler, 2012; Neal et al., 2015).

Promotional approach: a planned and systematic method which encourages people to adopt a specific behaviour (Peal et al., 2010; Aunger & Curtis, 2015; Mosler, 2012; Dreibelbis et al., 2013). Detailed promotional approaches are described below in the selection criteria.

1.4.1 Key debates in current policy

As part of its 2030 Agenda, the United Nations (UN) set as Goal 6 of the 17 Sustainable Development Goals (SDGs) the ambition to “Ensure access to water and sanitation for all”, including the target to “achieve access to adequate and equitable sanitation and hygiene for all and end open defecation, paying special attention to the needs of women and girls and those in vulnerable situations.” The importance of influencing behavior in order to achieve these goals is widely recognized.

In the eighties and nineties health promotion was based mainly on cognitive psychology (Aunger and Curtis, 2015). Behavior change policies in the WASH sector were predominantly influenced by different theory models such as the ‘Health Belief Model’ or ‘Theory of planned behavior’ among others (Rosenstock, 1974). When translated into policies, these theories shared a major commonality in assuming that people make rational decisions about protecting their health based on knowledge, skills and facilities. This is the era of participatory methodologies like PHAST ¹ (Participatory Hygiene and Sanitation Transformation) which aimed at increasing collective understanding about health risks and promoting preventive actions. This is also the time of extensive health and/or hygiene campaigns which would aim at educating the public by raising awareness and public understanding about risk behavior.

With the spread of social marketing theories in the early 2000's, the ‘education campaign’ approach in WASH policies have shifted into new emerging approaches such as Communication for Behavioral Impact (COMBI) ² or Change for Development (C4D) ³ . The incorporation of social marketing principles in behavior change approaches has led to the massive production of Information, Education and Communication (IEC) materials, often without considering the relevance of these materials to the desired behavioral outcome. Little attention was given on how to sustain these campaign approaches within targeted populations.

The last 10 years new developments on behavior change models were introduced, with emphasis on non-cognitive models and psychosocial theory, shaping again policies and resulting in approaches such as the current widely spread ‘Community Led Total Sanitation’ (CLTS) ⁴ or ‘Behaviour Centered Design’ ⁵ . This new vision emphasized the importance of attitudes and beliefs that influence certain behavior and social choices that shape what people think. Many variations of these approaches currently exist and it is still questionable if there is any added value of subsidies or incentives to this type of behaviour change approaches.

In summary, different behavioral theories and models have informed (and still inform) policy makers, donors and implementers about the issues to consider and the likely success of initiatives and interventions. Despite the efforts by the WASH sector in developing approaches to influence WASH behaviors, there still is no guidance on which are the most succesful techniques.

1.4.2 Overview of existing systematic reviews

In a first scoping phase (September 2015 – January 2016) an extensive overview of existing systematic reviews was performed, to answer the following research questions:

Research question 1: What is the effectiveness of approaches aiming to promote WASH behaviour change in low- and middle-income countries?

Research question 2: How do the perceptions and experiences of participants in terms of the programme's feasibility, appropriateness and meaningfulness influence WASH behaviour change?

We identified systematic reviews on the following WASH interventions : water quality (Fiebelkorn et al., 2012), hygiene hand sanitizers (Mah et al., 2008; Ejemot-Nwadiaro et al., 2015) and multiple WASH interventions (water, sanitation, hygiene) (Evans et al., 2014; Hulland et al., 2015;Joshi & Amadi, 2013). No systematic review focused on water supply or sanitation promotion programmes only.

The (multiple) WASH interventions were promoted using different approaches as follows: via social marketing principles (Mah et al., 2008; Evans et al., 2014), via community-led total sanitation (Hulland et al., 2015), via educational and/or communication channels (Ejemot-Nwadiaro et al., 2015; Hulland et al., 2015; Joshi & Amadi, 2013) or via multiple promotional approaches (community mobilization, health education, motivational interviewing, role modeling, and social marketing: Fiebelkorn et al., 2012). No systematic reviews on the use of financial incentives or other approaches to promote WASH interventions were found.

There was a paucity of information on promotional approaches of interventions in the systematic reviews, which prevented us from making any further conclusions. Population heterogeneity, type of intervention and outcome measurement were some of the reasons why meta-analyses were not performed in systematic reviews.

Only one systematic review reported data on implementation factors that could influence WASH behaviour (sustained adoption) (Hulland et al., 2015). Systematic reviews concerning other factors influencing implementation were not identified. Evidence from the systematic review by Hulland et al. (2015) suggests that the most influential programme factors associated with sustained adoption include frequent, personal contact with a health promoter over a period. While the Hulland review investigated factors that affect sustained adoption of WASH technologies (e.g. promotion via frequent, personal contact), this review focuses on factors that influence the implementation of approaches to promote WASH behaviour (e.g. culture as a barrier to use a financial incentive).

More details on the methodology used in this scoping phase can be found in Appendix 2, and detailed information about the methodology, results, and conclusions will be published in a separate peer-reviewed publication.

Based on our scoping review, we concluded that in the context of our two research questions, there is still an evidence gap. For example, no systematic collection of evidence is available regarding specific promotional approaches (e.g.community-based approaches) or specific WASH components (e.g. sanitation), in relation to behaviour change as an outcome. In addition, systematic reviews lack qualitative information about factors that can influence implementation of WASH promotional approaches. Therefore, we concluded that the systematic collection, extraction and analysis of qualitative/quantitative data on the effectiveness of promotional approaches aiming to promote handwashing and sanitation behaviour change outcomes was relevant and timely.

The objective of this systematic review is to identify promotional elements and those factors in the implementation process that influence behaviour change. This study objective is answered by a mixed-methods systematic review: findings from quantitative studies that identify effective promotional approaches (quantitative arm) were enriched with insights from qualitative studies that explore factors that hinder or facilitate the implementation of these promotional approaches (qualitative arm), focusing on people's lived experiences and perceptions. The findings of this review will provide guidance to governments and international bodies in selecting promotion strategies that positively influence behaviour change.

3.2.1 Types of studies

The type of study design is different for the quantitative and qualitative component of the review.

To answer Question 1 (effectiveness of promotional approaches), the following study types were selected: •

Impact evaluations using an experimental design (Randomised Controlled Trials (RCTs) with assignment at individual or household/community (cluster) level; Quasi-randomised controlled trials, using a quasi-random method of allocation (e.g. alternation))

Quasi-experimental and observational analytic studies were included since these were prevalent in the WASH literature, because randomised assignment is not always feasible or ethical.

Uncontrolled studies, case series, research methodology reports/manuscripts, editorials and economic analyses were excluded.

To answer Question 2 (implementation aspects), all qualitative study designs addressing factors influencing implementation of the promotional approaches were considered for inclusion. This includes for example grounded theory, case studies, phenomenological studies, ethnographic research, action research and thematic approaches to qualitative data analysis. The following types of studies were excluded: studies that did not use formal qualitative research study designs (e.g. surveys) or data collection techniques (e.g. interviews, focus group discussions, observations), and purely descriptive studies such as editorials and opinion pieces.

3.2.2 Types of participants

Participants included both children and adults from low- and middle-income countries (LMIC), as defined by the World Bank, at the time the intervention was conducted. Studies performed at an individual, household, school or community level were included, whereas studies conducted in institutional settings (e.g. hospitals) were excluded.

3.2.3 Types of interventions

Programmes conducted to promote uptake and use of handwashing, and the following sanitation interventions were included: latrine/toilet use, safe faeces disposal practices, and discouraging the practice of open defecation. Any combination of the interventions listed above were included. The following programmes were excluded: programmes conducted to promote water treatment, water supply for drinking only, menstrual hygiene, food hygiene, animal waste disposal, facial cleansing. Any combination of the interventions listed above with water treatment, drinking water supply or other hygiene interventions were included if individual outcomes, as listed below, were present.

The programme contained a direct promotional approach related to one of the following categories: community-based approaches, social marketing approaches, sanitation and hygiene messaging, elements of psychosocial theory, incentives, advocacy, or any combination of the promotional approaches or promotional elements mentioned above (multichannel approach) (details on these approaches can be found in paragraph 1.2.1).

Programmes using no promotional approaches were excluded.

3.2.4 Comparison

For Question 1 (effectiveness of promotional approaches), the comparison is the use of a programme with other forms of behaviour change promotional approach, or no promotional programme.

3.2.5 Types of outcome/evaluation measures

To answer Question 1 (effectiveness of promotional approaches), studies reporting the following outcomes were selected:

Primary outcomes

The primary outcome is behaviour change, operationalized in the following way: (a) use of handwashing and sanitation interventions (handwashing:handwashing with or without soap (or alternatives such as ash) and/or hand disinfection with alcohol based gels, handwashing at key times (before eating, before food preparation, after visiting the toilet, after children's faeces disposal or cleaning the baby's bottom, or other key times used in the studies); sanitation: latrine/toilet use, safe faeces disposal, number of people practicing open defecation): uptake of the interventions, adherence to the interventions, longer-term use of the interventions, (b) intention to practice handwashing and sanitation interventions (readiness, willingness), (c) habit to practice handwashing and sanitation interventions (routinized behaviour, adherence, longer-term use). Other indirect outcomes, such as “presence of soap” were not considered. Outcomes concerning animal faeces were not included if it was explicitly mentioned that faeces were from animals. Outcomes that could not be categorised under one of the outcome measures listed above were not included (e.g. cleaning of child after defecation).

Secondary outcomes

The secondary outcomes are: behavioural factors (knowledge, skills, attitude, norms, and self-regulation concerning the practice of handwashing and sanitation interventions); morbidity and mortality due to agents associated with faecal-oral transmission. Indirect outcomes, such as “pupil absence”, were not considered. Symptom-based health outcomes, such as cough, general illness, fever and congestions were not included. Studies reporting data on morbidity and mortality were only included if data on primary outcomes (behaviour change) were also available. Studies reporting only behavioural factors, and no primary outcomes, were included. We included outcomes that were measured via direct observation/demonstration (where a participant is asked to show how a behaviour is practiced), as well as self-reported, parent-reported or teacher-reported outcomes. To answer the Question 2 (implementation aspects), perceptions, experiences, opinions, or viewpoints of implementers or recipients of the programme concerning factors influencing implementation were extracted. These factors included for example public commitment, motivation, culture, gender, social capital, etc. From an analytical point of view, we focused on aspects of feasibility, appropriateness and meaningfulness of the promotional approach as experienced by the people involved in the implementation of the promotional programmes.

3.2.6 Duration of follow-up

No restrictions in timing of outcome measurement were used. Outcomes measured during the implementation of the programme were categorised as “uptake”, outcomes measured within 12 months after the programme implementation were categorised as “adherence”, and outcomes measured >12 months after the end of the programme implementation were categorised as “longer-term” outcomes.

3.2.7 Language

No language restrictions were used.

3.2.8 Publication date

Studies from 1980 to March 2016 were included. This date is based on the introduction of the Millennium Development Goals in 1990 (MDG7: “To ensure access to drinking water and sanitation for all”), which was followed by the development of evidence-based interventions for hygiene promotion (DFID, 2013). We also checked the publication dates of the included studies in the identified systematic reviews (scoping phase), but since one study was published in 1985, we chose 1980 as cut-off date (Stanton & Clemens, 1985).

3.3.1 Electronic databases

We searched the following databases from 1980 to March 2016: •

3ie Impact Evaluation Database

3.3.2 Searching other resources (grey literature)

To find unpublished material and relevant programme documents, we contacted the following research groups and organizations and/or checked the following websites (March 2016): •

CLTS Foundation (www.cltsfoundation.org)

3.4.1 Selection of studies

Study selection was performed independently and in parallel by two evidence reviewers, using EPPI-Reviewer software. In the first phase, titles and abstracts of the references identified during the search were scanned. Full text versions of relevant articles were retrieved, and references that met the selection criteria were included for further analysis. The references resulting from grey literature sources were screened, based on title and abstract, by only one reviewer. Full text assessment of the grey literature was done by 2 reviewers. Any discrepancies between the two reviewers were resolved by consensus, and in case of disagreement, a third reviewer was involved. A PRISMA study selection flowchart was developed (Moher et al., 2009), and a list of excluded studies with the reasons for exclusion was provided. References were labelled as “unavailable”, when it was not obtainable through the libraries of the institutions involved (Stellenbosch University (South Africa), KU Leuven (Belgium)).

3.4.2 Data extraction and management

Data extraction (including quality assessment) was performed by two reviewers independently.Question 1 (effectiveness of promotional approaches): Data concerning publication date, study design, study population, details of the intervention, outcome type, and study quality were independently extracted by the two reviewers. For the intervention, information on the targeted activity (handwashing, sanitation) as well as information on the promotional approach, was extracted. For the promotional approach we extracted the following data: (1) who is providing the approach, (2) who is receiving the approach, (3) the exact content of the promotional approach (presence of promotional elements such as sanitation and hygiene messaging, psychosocial theories, community-based participatory approach, social marketing, incentives, advocacy, and other elements such as pride/disgust/behaviour change techniques), and (4) process evaluation factors (recruitment, attrition, reach, dose, fidelity, adaptation, engagement, satisfaction, acceptability). All these different elements were extracted separately. Study authors of all included papers were contacted by email (in July 2016) to ask for any relevant information, related to the population, intervention or outcomes, that was missing or not reported in the paper. A reminder to authors was sent in August 2016. All relevant information received by the latest, on 19^th of September, was screened and included in the code book. Outcomes measured at different time points following the intervention were extracted separately. For each dichotomous outcome, we either extracted the number of participants experiencing the event, and the number of participants in each treatment group, or the information necessary to estimate odds and risk ratios, including group means and sample sizes. For each continuous outcome that can be assumed to be normally distributed, we extracted means, standard deviations (or information to estimate standard deviations), and number of participants in each group. For skewed continuous data, medians, ranges, and p-values for non-parametric tests were extracted. Any discrepancies between the two data extractors were resolved through discussion, or by consulting other review co-authors. If studies used different conventions/scales, the direction of interpretation is explained and it is clearly indicated when directions were reversed. Data were entered into meta-analysis software, and checked for accuracy. A table was developed with the characteristics of the included studies, containing a summary of the characteristics of the participants, interventions, outcomes and other relevant information. In addition, a visual overview of the findings was created, in addition to the forest plots with pooled and unpooled findings.Question 2 (implementation aspects): For Question 2, data concerning publication date, study design, study population, details of the intervention, and evaluation measures were extracted by one reviewer, and double checked by the second reviewer. A third reviewer resolved any disagreements. Similar information on the intervention was extracted as described for Question 1. Implementation factors (such as programme environment factors, recipient-related factors, and socio-cultural, physical and personal contextual factors) of our ToC were used as a-priori themes. Subsequently, inductive coding on both the original statements of the interviewees (defined as PE (“primary evidence”)) and the author statements (defined as AS (“author statements”)) was performed. Both data extraction and inductive coding was double checked by the second reviewer.Use of codebook for data extraction: Quantitative as well as qualitative data were extracted using a codebook developed for this purpose (see Appendices 5 and 6). The codebook is based on the elements of the ToC. All items of the codebook were incorporated in EPPI-Reviewer software, so that data extraction could be performed easily in parallel by two reviewers. In the codebook, variables were theoretically and operationally defined if this was necessary to guarantee intercoder and intracoder agreement during the data extraction process.

3.4.3 Quality assessment of included studies and determination of certainty of evidence

Question 1 (effectiveness of promotional approaches): Risk of bias in the individual studies (experimental studies) was analysed at the study level by using the Cochrane Risk of Bias tool (Higgins and Green, 2011). For quasi-experimental studies, a combination of the risk of bias tool provided by 3ie and the Cochrane tool for non-randomised studies (ACROBAT-NRSI), was used (see Appendix 7). The different choices made during the risk of bias assessment were justified by providing information directly from the study. A specific question was added to the risk of bias assessment concerning the rigour of the outcome measurement, especially for handwashing, since it is known that over-reporting often takes place when using questionnaires (Manun’Ebo et al., 1997; Contzen et al., 2015). The GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach was used to assess the overall quality/certainty of the evidence included in this review. This approach is based on the limitations in study design, imprecision, inconsistency, indirectness, and publication bias (Atkins et al., 2004). As part of the GRADE process (Atkins et al., 2004), for each type of promotional approach, the certainty of evidence for the “body of evidence” was assigned per outcome category. The final certainty of evidence ranged from high (i.e. further research is very unlikely to change our confidence in the estimate of effect), moderate (i.e. further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate), low (i.e. further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate) to very low (i.e. we are very uncertain about the estimate). Because of a very large number of data and analyses, it was decided not to determine the certainty of evidence in the following cases: (1) when statistical heterogeneity > 50%, (2) individual outcomes, and (3) secondary outcomes. The online tool of the GRADE Working Group (“GDT” or “Guideline Development Tool”) was used for the GRADE assessment process. Standardised qualitative statements were used to link the findings to their corresponding level/certainty of evidence in the description of the meta-analyses (Section 4.3.1.1) and the “Summary of main results” (Section 6.1): use of the wording “probably” with moderate certainty evidence, use of wording “may” with low certainty evidence, and a statement about being uncertain about the effect of the intervention on the outcome for very low certainty evidence (EPOC 2015).Question 2 (implementation aspects): A quality appraisal was done at the study level by using the CASP Qualitative Checklist to reveal limitations in study design (Critical Appraisal Skills Program 2014), as a baseline measure of quality of the included studies (see Appendix 8). We did not exclude any studies from our review. Instead, we conducted a sensitivity analysis exploring the impact of including low quality studies in the review on the overall findings.

3.4.4 Measures of treatment effect

Binary outcomes were used to calculate risk ratios (RR) (+ 95% confidence intervals (CI)). For continuous data, (weighted) mean differences (MD) (+ 95% CI) were calculated. We only used the (unadjusted/adjusted) effect measures calculated by the study authors in case the binary/continuous data were not available. If outcome measures were opposite to the intervention categories we defined (e.g. “no latrine use” instead of “latrine use”), binary data were reversed. This was indicated on the forest plots with an asterisk. Unit of analysis issues were carefully considered in order to adjust for the clustering effect (in case of cluster RCTs) and/or for multiple testing (in case of multi-arm trials). For cluster RCTs a cluster adjustment on the raw data (binary/continuous outcomes) was made. For the binary outcomes, the raw data (e.g. number of handwashing at key times events) were divided by the calculated design effect. For the continuous outcomes, the raw data (e.g. mean number of people washing their hands at key times) was multiplied by square root of the calculated design effect. The design effect was calculated by the formula: design effect = 1 + ((average cluster size -1) x ICC (intra-cluster correlation coefficient)), as detailed in the Cochrane Handbook of Systematic Reviews Chapter 16.3 (Higgins and Green, 2011). We used the ICC as reported by the original study. In cases where the ICC was not reported, we estimated the ICCs using the following strategy: within each category of promotional approaches we used the mean of the ICCs of studies for which an ICC was reported; in two categories of promotional approaches (i.e. sanitation and hygiene messaging and social marketing approach) none of the studies had reported ICCs, in which case the most conservative ICC value of the other categories was used. We calculated synthetic effects for any instances of dependent effects (e.g. shared control groups in multi-arm trials), according to the method described in the Cochrane handbook chapter 16.5.4 (Higgins and Green, 2011): for dichotomous outcomes both the sample sizes and the numbers of people with events were summed across groups.

3.4.5 Data synthesis

Evidence relating to Question 1 (effectiveness of promotional approaches) was synthesized in a quantitative way (meta-analysis), where possible. Meta-analyses were performed for 13 different outcomes across promotional approaches and timing of measurement of outcomes, to be able to make conclusions about the effect of “any promotional approach versus no promotional approach”. As soon as an outcome was present more than once, but within the same study type, it was included in a meta-analysis. Meta-analyses were conducted using Stata version 14 software. Meta-analysis results are displayed using forest plots. We used random-effects meta-analysis to produce an overall summary, if an average treatment effect across trials was considered meaningful. Fixed effect meta-analysis was not applied because its homogeneity assumption was not applicable in this systematic review. Included experimental studies were categorised and analysed according to the different promotional approaches. Experimental and quasi-experimental/observational studies were analysed separately. Mantel-Haenszel (M-H) methods were used for binary outcomes in the random-effects meta-analysis, and for calculating the effect measures, and the Inverse-Variance (I-V) method was used for continuous outcomes. Effect measures of binary outcomes were expressed as RRs (as described in 3.4.4), however a sensitivity analysis using risk differences (RD) was also made and tabulated. Forest plots reporting RDs are available upon request. Where meta-analysis was not possible, we reported results from individual studies separately. The data were grouped in separate forest plots according to the promotional approach and outcome. Data were included in forest plots if possible, or reported narratively otherwise. Evidence conclusions were formulated in a narrative way, but mentioning where possible the effect sizes (and CI), and considering risk of bias. Where possible, differences in results are explained by describing likely explanatory factors. A statistically non-significant p-value was interpreted as a finding of uncertainty (“no evidence of effect”) unless confidence intervals were sufficiently narrow (no imprecision according to the GRADE approach) to rule out an important magnitude of effect (“evidence of no effect”). Accuracy of numeric data in the review were checked against the data as available from the original study.

3.4.6 Assessment of statistical heterogeneity

The measures I²and τ² were used as a measure of presence of heterogeneity, which was then further explored. An I² value of greater than 50% was considered as a substantial measure of heterogeneity.

3.4.7 Subgroup analysis

Subgroup analyses were performed according to the type of promotional approach (community-based approaches, social marketing approaches, sanitation and hygiene messaging, psychosocial theory). Because of an insufficient number of studies per meta-analysis, no other subgroup analyses were made. The following factors were used in a descriptive way as likely explanatory factors for differences in results: (1) different types of promotional approaches,(2) the targets of the study (individual, household, community), (3) the setting where the approach has been applied (rural, urban, informal-urban; see Peal et al., 2010)(Fiebelkorn et al.(2012) reported differential behaviour change near the city and among the rural population; see also DFID, 2013), (4) the scale at which the approach has been applied (small scale (one village, several villages) vs larger scale (sub-district, district, province or region, national); see Hulland et al., 2015), and (5) other equity factors such as socioeconomic status, occupation and education (O'Neill et al., 2014) (adherence to water, sanitation and hygiene programmes is known to be highly associated with these confounding factors; see DFID, 2013).

3.4.8 Sensitivity analysis

No sensitivity analyses were performed due to insufficient number of studies per meta-analysis, however the risk of bias of the individual studies was considered when interpreting results.

3.4.9 Synthesis of qualitative research

For the qualitative evidence synthesis, we used the “Best fit framework synthesis” approach (Booth, 2015; Carroll, 2013). The first step of this approach was to identify an existing model for a particular health behaviour, in this case “WASH behaviour”. In the scoping phase of this project existing models for WASH behaviour change were identified, including the RANAS model and IBM-WASH model (Mosler, 2012; Dreibelbis et al., 2013). These models, that were included in our ToC, were used as an “a priori framework”. In addition to the information from the WASH behaviour change models, elements from the “Checklist for implementation” (Cargo et al., 2015), the SURE framework for implementation of a policy option (The SURE Collaboration 2011), and the PROGRESS framework to consider equity issues (O'Neill et al., 2014), were used to inform the a priori framework. In the second step of this approach, we coded data from individual qualitative studies against the a priori themes of our ToC model, representing factors that can influence the implementation of the promotional approaches ToC model (i.e. programme environment factors and recipient-related moderators, process evaluation factors and recipient-related contextual factors). Inductive, thematic analysis techniques were used if data could not be accommodated within these themes. Information from the critical appraisal items (CASP tool) was not used a-priori to exclude low-quality or high-quality studies. A sensitivity analysis was conducted by excluding low-quality studies and to test the impact of these exclusions on the overall synthesis of findings (Carroll et al., 2012). The conclusions of both strands of evidence were integrated at the end of the review process in the conclusion and discussion section. In addition, the conclusions were coupled back to the ToC. Conclusions were based only on findings from the synthesis (quantitative or narrative) of studies included in the review. In the discussion section of the review, policy implications of the findings are discussed, taking into account local considerations. In addition to the policy messages, implications for research are formulated.

4.1.1 Results of the search

We identified 23,435 records through database searching. In addition, 2,132 references were identified through grey literature searching. Following title and abstract screening, 522 records were selected for full text screening, including 401 references from database searching and 121 records via screening of the grey literature. The full texts of these references were read in detail, and after applying the pre-specified selection criteria, 342 database papers and 111 grey literature reports were excluded. This finally resulted in 35 quantitative, 19 qualitative and 5 mixed-methods studies from databases, and 6 quantitative and 4 qualitative studies from grey literature. A mixed-methods study was defined as a study fulfilling the criteria of our first and second research question. Taken together, we identified 46 references to quantitative studies (individual quantitative and mixed-methods studies), and 28 references to qualitative studies (individual qualitative and mixed-methods studies). For the quantitative papers published by Contzen et al. (2015a and 2015b), Galiani et al. (2012 and 2015), Hoque et al. (1994 and 1996) and Patil et al. (2013 and 2015), two separate references (with complementary information) for each study were included resulting in a total number of 41 quantitative studies (from 45 references). The study selection flowchart can be found in Figure 3.

OPEN IN VIEWER

Figure 3

Study selection flowchart

The promotional approaches differed considerably across the studies. For each study, we indicated if elements of sanitation and hygiene messaging, psychosocial theory, community-based working, social marketing, incentives or advocacy were used, leading to 27 different combinations of elements and thus 27 different promotional approaches (see Figure 5). Based on the main promotional element in each approach we classified the promotional approaches/promotional elements in 4 groups. This was done independently by 4 team members (methodological and content experts), followed by discussion to resolve disagreements. In addition, we also discussed this with a large group of stakeholders who agreed with the classification approach

Based on the major component of the promotional approach used in each study, we distinguished these 4 major approaches: 1.

Community-based approaches: in this category we included the studies that used a formal community-based approach or those approaches that contained elements of community-based working as the major strategy. Other elements that could be part of these approaches were: education, incentives, and/or theory-based elements.

According to these criteria we classified the promotional approach as a community-based approach in 13 studies, a social marketing approach in 7 studies, and sanitation and hygiene messaging in 15 studies. Elements of psychosocial theory were investigated in 6 studies. Table 2 gives an overview of which studies were grouped under each category.

Figure 6 also lists the specific approach in each study and the WASH component for each study. Community-based approaches all contained at least a sanitation component (except for one study with a handwashing-only intervention), social marketing approaches and sanitation and hygiene messaging interventions focused in the majority of the cases at least on handwashing, and the approaches based on elements of psychosocial theory almost in all cases only had a handwashing component.

Seven studies only looked at the relative effectiveness of a promotional approach versus another promotional approach and 1 study compared programmes with a similar promotional approach (i.e. sanitation and hygiene messaging) but with different communication channels (interpersonal+mass media communication versus mass media only).

Since (non-) financial incentives were always part of a broader promotional approach listed above, we did not create a separate category for this type of promotional elements. However, in Table 3 an overview of the types of incentives is provided, and in the results section below, incentives are dealt with as a possible moderating factor. Financial incentives included a modest salary and subsidies, and non-financial incentives included a motorcycle, lunch, food, gifts and soap. We make the distinction between incentives given to the secondary implementer (community-member involved in the implementation) and the recipients (villagers/household members, receiving the promotional approach).

•

Communication strategies

All intervention programmes (n=55) used (at least) interpersonal communication channels: 22 interventions (40%) used interpersonal communication only, 16 interventions (29%) used interpersonal+mass media communication, 7 interventions (13%) used interpersonal+traditional communication and 10 interventions (18%) used interpersonal+mass media+tradional communication.

The programmes with a promotional approach in the control group (n=10) were promoted via interpersonal communication only (n=5), via mass media communication only (n=1), via traditional communication only (n=1), via interpersonal+mass media communication (n=1) or via interpersonal+mass media+traditional communication channels (n=2). •

Implementers (see Figure 7)

Almost all studies (n=40, 95%) reported who the implementers of the programme were. Information about training/qualification of the implementers (n=24, 57%), the role of the evaluator (n=18, 43%) and gender of the implementers (n=11, 26%) was less frequently reported. Information about ethnicity (n=4, 9%), age (n=4, 9%) and socio-economic status (n=4, 9%) of the implementers was rarely reported. •

Implementing organization (see Figure 8)

In general, information about the implementing organization was not frequently reported: about 30% of the studies provided information about leadership (n=15), the quality of the training materials (n=14), technical support or supervisory guidance (n=14). Funding information (about the programme (not the study)) was provided in 10 studies (24%) and only 2 studies (5%) provided information on partnership/coordination between providers. •

Process evaluation factors (see Figure 9)

OPEN IN VIEWER

Figure 5

Promotional elements present in the interventions of the 41 included quantitative studies

OPEN IN VIEWER

Figure 6

Main categories of promotional approaches with detailed indication of WASH component and specific promotional approach for each included quantitative study

OPEN IN VIEWER

Figure 7

Reported information about the implementers

OPEN IN VIEWER

Figure 8

Reported information about the implementing organization

OPEN IN VIEWER

Figure 9

Reported information about the process evaluation factors

Recruitment (n=34, 81%) and dose (n=33, 78%) were frequently reported. Forty-three percent of the studies provided information on reach (n=18) or adaptation (n=21, 50%) whereas information on fidelity (n=5), implementer engagement (n=5), participation engagement (n=7) or co-intervention (n=4) was only reported in 10-20% of the studies. No studies had information on composite implementation measures. •

Outcomes

In total, 559 different outcomes (i.e. different outcome descriptions, timing of measurement, method of assessment, and reported statistics) were measured across all studies.

Raw data were available in most of the studies (n=39, 93%): binary data (n=18), continuous data (n=12), binary+continuous data (n=7), continuous+correlation data (n=1) and binary data+calculated effect sizes (n=1). Three studies (7%) only reported calculated effect size measures.

Primary (behaviour change) outcomes were reported in 39 studies: intention in 2 studies, handwashing (with or without soap) in 12 studies, handwashing at key times in 21 studies, latrine use in 9 studies, faeces disposal practices in 9 studies and open defecation in 9 studies. The following behavioural factors (secondary outcomes) were assessed: knowledge in 12 studies, skills in 6 studies, attitude in 5 studies, and self-regulation in 4 studies. Morbidity and mortality (secondary outcomes) were measured in 11 studies and 1 study, respectively.

Outcomes were assessed via self-reported measures in 27 studies (64%), via direct observation in 10 studies (24%), or via self-reported measures plus direct observation in 5 studies (13%).

The timing of outcome assessment was different across studies: 18 studies assessed the outcomes during the programme implementation (i.e. uptake), 16 studies assessed the outcomes within 12 months after the end of the implementation (i.e. adherence) and only 5 studies measured the outcomes more than 12 months after the end of the implementation (i.e. longer-term use). Three studies assessed outcomes at two different time points: 1 study at uptake+adherence, 1 study at uptake+longer-term use and 1 study at adherence+longer-term use.

4.1.3 Excluded studies

After title and abstract screening, 522 full texts (401 from databases and 121 from grey literature) were screened for eligibility. The majority of these full-texts were excluded (n=461, 88%) for different reasons: study design (n=242, 52%), intervention (n=95, 21%), outcome (n=77, 16%), population (n=12, 3%), duplicates (n=24, 5%), not available (n=11, 2%). Detailed information can be found in Appendix 9 (List of excluded database studies) and 6 (List of excluded grey literature studies), and the reference list of excluded studies.

4.2.1 Experimental studies (n=32)

A visual overview of the risk of bias of the experimental studies can be found in Figure 10. •

Random sequence generation

OPEN IN VIEWER

Figure 10

Risk of bias in the experimental studies

Many studies did not provide clear information on the way the randomization sequence was generated. In 14 of the 32 studies (44%) the randomization sequence was clearly described, and assigned as being at low risk of selection bias. In 18 of the 32 studies (56%), not enough information was provided to determine if the method of random sequence generation was adequate. •

Allocation concealment

In two studies (6%), Guiteras et al. (2015b) and Pattanayak et al.(2009), allocation concealment was described, and was assessed to be a low risk of bias. In two studies (6%), Pickering et al.(2015) and Huda et al.(2012), allocation concealment was not conducted and thus assessed as high risk of bias. The majority of studies (n=28, 88%) did not provide any information to assess risk of bias and were thus assigned as unclear. •

Blinding of participants

Blinding of participants to a treatment group was not easy for this type of intervention, and only one study (2%), Biran et al.(2014), reported on blinding of participants. In 18 studies (56%), there was a lack of information about blinding, and these studies were rated as unclear. Thirteen studies (42%) reported no blinding of participants. •

Blinding of outcome assessors

No information on blinding of outcome assessors was given in 12 of the studies (37%), with 11 studies (34%) reporting no blinding and 9 studies clearly indicating that outcome assessors were blinded (28%). Self-reported outcomes were assessed in 18 studies (56%) whereas 14 studies (44%) measured outcomes via direct observation techniques. •

Incomplete outcome data

Incomplete outcome data was clearly dealt with in 5 studies (16%), with the many studies (n=13, 40%) having not dealt with this issue. In the remaining 14 studies (44%), there was no information on how incomplete outcome data was dealt with. •

Selective reporting

Selective reporting bias was found to be present in many studies (20/32, 62%), with only 5 studies (16%) reporting having dealt adequately with this bias. No information was present in 7studies, and these were rated as unclear. •

Other risks of bias

There were no other risks of bias in the majority of the studies (20/32, 62%). There were other risks of bias in 10 studies (high risk, 31%) and two studies (6%) did not provide any information regarding other risks of bias. No intra-cluster correlations (ICC) were reported in 15 of the 22 cluster RCTs.

4.2.2 Quasi-experimental (n=8) and observational studies (n=2)

A visual overview of the risk of bias of the quasi-experimental and observational studies can be found in Figure 11. The observational studies both were cohort studies (Arnold et al., 2009, Seimetz et al., 2016). •

Bias in selection of participants

OPEN IN VIEWER

Figure 11

Risk of bias in the quasi-experimental and observational studies

Three studies (30%) were assessed to be at a critical level for this category. Three studies (30%) were judged to have serious bias and three were moderate. Only the Arnold et al. (2009) study was judged to be of low bias, as the selection into the study (or into the analysis) was unrelated to intervention or unrelated to outcome. The start of follow-up and start of intervention coincided for most participants, and there were adjustment techniques used that were likely to correct for the presence of selection biases. The allocation mechanism was also appropriate to generate equivalent groups. •

Bias due to confounding

There were 4 studies (40%) judged to have critical level of bias due to confounding. An equal number had a low risk of bias, as the authors used an appropriate analysis method that controlled for all the important confounding areas (baseline confounding). The authors also used an appropriate analysis method that controlled for time-varying confounding, if present, and confounding areas that were controlled for measured validly and reliably by the variables available in this study. The remaining studies were judged to be of moderate (1) and serious (3) bias. •

Bias in measurement of interventions

Three studies (30%) had a low bias in measurement of interventions, with 3 studies (30%) being judged as moderate and 4 studies (40%) being judged as serious. One study, Kochurani et al. (2009), was evaluated to have critical bias as the intervention was not well defined, the information used to define intervention groups was not recorded at the start of the intervention, and information on intervention status was affected by knowledge of the outcome or risk of the outcome. •

Bias in measurement of outcomes

Four studies (40%) showed moderate bias in this category and 5 studies (50%) were judged as serious. One study, Kochurani et al. (2009), was deemed to show critical bias as this study did not have an objective outcome measure. The methods of outcome assessment were not comparable across intervention groups, and outcome assessors were aware of the interventions that the groups received. •

Bias due to departures from intended intervention

The Contzen et al. (2015a/2015b) study had a low risk of bias and three other studies were of moderate bias. Five studies (50%) were shown to have serious bias, and the Kochurani et al. (2009) study was assessed to have critical levels of bias as the important co-interventions were not balanced across intervention groups, the study participants did not adhere to the assigned intervention regimen, and the intervention was not implemented successfully for most participants. •

Reporting bias (missing data + selective outcome reporting)

The reporting biases as discussed here incorporate biases because of missing data and selective outcome reporting. The Arnold et al. (2009) study showed low bias for both aspects of reporting bias. Contzen et al. (2015a/2015b) showed low bias in the selective outcome reporting category, but moderate for the missing data category. Andrade (2013), Dickey et al. (2015), Kochurani et al. (2009), Seimetz et al. (2016), Waterkeyn & Cairncross (2005) and Whaley & Webster (2011) provided no information on reporting bias and were assessed as unclear. Both Pinfold (1999), and Seimetz et al. (2016) were assessed as moderate for the selective outcome reporting category.

4.3.1 Promotional approach versus no promotional approach

In 34 studies the effect of using a promotional approach was compared with not using a promotional approach. Of these studies, 12 studies described a community-based approach, 6 studies described a social marketing approach, 12 studies described sanitation and hygiene messaging, and 4 studies described a small-scale intervention based on elements of psychosocial theory. An overview of the studies included in each category of promotional approaches (compared to not using a promotional approach) can be found in Table 4.

4.3.1.1 Any promotional approach

For the list of predefined outcomes (see above) meta-analyses were performed across the different promotional approaches and different times of measurement. For each meta-analysis, subgroup analyses according to the promotional approach were performed, and where possible according to timing of measurement. However, for 11 of the 13 outcomes there was too much heterogeneity to be able to make conclusions across the different types of promotional approaches. The pooled value per promotional approach is reported below in case no statistical heterogeneity was present. Below we describe the results for the 1 different outcomes: •

Behaviour change: handwashing after toilet use (Analysis 1). Since there was too much heterogeneity it was not possible to pool the outcomes across promotional approaches. Only for the community-based approaches, a level of heterogeneity < 50% was found. A community-based approach may make little or now difference in handwashing after toilet use (RR 1.06, 95 %CI [0.99, 1.14]; level of certainty: low, Table 5) (Huda et al., 2012; Phuanokoonnon et al., 2013).

•

Behaviour change: handwashing before cooking (Analysis 2). There was no significant increase in handwashing for the community-based approach (RR 0.94, 95% CI [0.31, 2.91]) (Huda et al., 2012). Sanitation and hygiene messaging may improve handwashing before cooking (RR 1.23, 95% CI [1.09, 1.39]; level of certainty: low (Table 6)) (Bowen et al., 2013; Stanton & Clemens, 1987). The effect of elements of psychosocial theory on handwashing before cooking is uncertain(RR 33.06, 95% CI [6.72, 162.69]; level of certainty: very low (Table 7)) (Langford & Panter-Brick, 2013; Luby et al., 2010).

•

Behaviour change: handwashing after cleaning a child's anus (Analysis 3). There was no A significant increase in handwashing for the community-based approach (RR 1.34, 95% CI [0.85, 2.12]) (Huda et al., 2012). For the other approaches and “overall promotional approach” there was too much heterogeneity to be able to make overarching conclusions.

•

Behaviour change: handwashing before eating (Analysis 4). A community-based approach may lead to slightly improved handwashing before eating (RR 1.12, 95% CI [1.02, 1.22]; level of certainty: low (Table 8)) (Huda et al., 2012; Phuanukoonnon et al., 2013), while elements of psychosocial theory may improve it (RR 34.73, 95% CI [4.90, 246.39]; level of certainty: low (Table 9)) (Langford & Panter-Brick, 2013; Luby et al., 2010). In case of sanitation and hygiene messaging, there was too much heterogeneity to be able to make overall conclusions.

•

Behaviour change: handwashing before feeding a child (Analysis 5). The effect of a community-based approach is uncertain (RR 1.04, 95% CI [0.94, 1.15]; level of certainty: very low (Table 10)) (Huda et al., 2012, Phuanukoonnon et al., 2013). A theory-based approach may improve handwashing before feeding a child (RR 3.63, 95% CI [1.91, 6.88]; level of certainty: low (Table 11)) (Langford & Panter-Brick, 2013; Luby et al., 2010).

•

Behaviour change: latrine use (Analysis 6). High heterogeneity across the studies (all using a community-based approach) did not make it possible to pool the outcomes. Therefore, we were not able to make any overall conclusions for this outcome. However, when a subgroup analysis was performed according to timing of measurement (adherence and longer-term use), a community-based approach may improve latrine use less than 12 months after the end of programme implementation (adherence) (RR 2.63, 95% CI [1.62, 4.29]; level of certainty: low (Table 12)) (Jinadu et al., 2007; Pattanayak et al., 2009).

•

Behaviour change: safe faeces disposal practices and safe child faeces disposal practices (Analysis 7 and 8). Since there was too much heterogeneity it was not possible to pool the outcomes across and within the promotional approaches. For sanitation and hygiene messaging, only one study was included, showing statistically significant increased safe faeces disposal practices(RR 1.68, 95% CI [1.21, 2.32]), however a significant effect on safe child faeces disposal practices could not be demonstrated (RR 1.07, 95% CI [0.70, 1.65]) (Yeager et al., 2002).

•

Behaviour change: open defecation (Analysis 9). A community-based approach resulted in a statistically significantly decrease in open defecation (RR 0.40, 95% CI [0.37, 0.44]) (Pickering et al., 2015). Sanitation and hygiene messaging may make little or no difference in open defecation (RR 0.99, 95% CI [0.72, 1.37]; level of certainty: low (Table 13)) (Lansdown et al., 2002; Stanton & Clemens, 1987; Wang et al., 2013).

•

Behavioural factors: skills: using soap for handwashing (Analysis 10). Sanitation and hygiene messaging probably slightly improves using soap for handwashing (handwashing technique) (RR 1.05, 95% CI [1.02, 1.08]; level of certainty: moderate (Table 14)) (Bowen et al., 2013; Luby et al., 2009). No studies on other approaches measured this outcome.

•

Behavioural factors: skills: rubbing hands together at least 3 times (Analysis 11). Only studies using sanitation and hygiene messaging measured if there was an improvement in rubbing the hands together at least 3 times (Bowen et al., 2013; Luby et al., 2009). Since there was too much heterogeneity it was not possible to pool the data, and it was not possible to make any overall conclusions for this outcome.

•

Behavioural factors: skills: lathering hands > 10 seconds (Analysis 12). Only studies using sanitation and hygiene messaging measured if lathering hands for more than 10 seconds (handwashing technique) had increased (Bowen et al., 2013; Luby et al., 2009). Since there was too much heterogeneity it was not possible to pool the data, and it was not possible to make any overall conclusions for this outcome.

•

Behavioural factors: skills: drying hands with a clean towel (Analysis 13). Only studies using sanitation and hygiene messaging measured if drying hands with a clean towel (handwashing technique) had resulted in an increase (Bowen et al., 2013; Luby et al., 2009). Since there was too much heterogeneity it was not possible to pool the data, and it is not possible to make any overarching conclusions for this outcome.

OPEN IN VIEWER

Figure 1

Analysis

OPEN IN VIEWER

Figure 2

Analysis

OPEN IN VIEWER

Figure 3

Analysis

OPEN IN VIEWER

Figure 4

Analysis

OPEN IN VIEWER

Figure 5

Analysis

OPEN IN VIEWER

Figure 6

Analysis

OPEN IN VIEWER

Figure 7

Analysis

OPEN IN VIEWER

Figure 8

Analysis

OPEN IN VIEWER

Figure 9

Analysis

OPEN IN VIEWER

Figure 10

Analysis

OPEN IN VIEWER

Figure 11

Analysis

OPEN IN VIEWER

Figure 12

Analysis

OPEN IN VIEWER

Figure 13

Analysis

OPEN IN VIEWER

Table 5

Assessment of the certainty of evidence for handwashing after toilet use (pooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing after toilet use (Huda 2012 and Phuanukoonnon 2013)
2	2 randomised trials a	very serious b	not serious	not serious	not serious	none	324/382 (84.8%)	90/150 (60.0%)	RR 1.06 (0.99 to 1.14)	36 more per 1.000 (from 6 fewer to 84 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT (Huda 2012) and 1 quasi-RCT (Phuanukoonnon 2013)

Selection bias (Huda 2012), attrition, detection and reporting bias (Phuanukoonnon 2013)

OPEN IN VIEWER

Table 6

Assessment of the certainty of evidence for handwashing before cooking (pooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing before cooking (Bowen 2013 and Stanton 1987)
2	2 randomised trials a	serious b	not serious	not serious	serious c	none	256/333 (76.9%)	118/201 (58.7%)	RR 1.23 (1.09 to 1.39)	135 more per 1.000 (from 53 more to 229 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 cluster RCTs

detection bias (Bowen 2013) and attrition bias (Stanton 1987)

low number of events

OPEN IN VIEWER

Table 7

Assessment of the certainty of evidence for handwashing before cooking (pooled data), elements of psychosocial theory vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Theory-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing before cooking (Langford 2013 and Luby 2010)
2	2 randomised trials a	very serious b	not serious	not serious	serious c	none	85/356 (23.9%)	1/155 (0.6%)	RR 33.06 (6.72 to 162.69)	207 more per 1.000 (from 37 more to 1.000 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT (Luby 2010) and 1 quasi-RCT (Langford 2013)

Detection bias (Langford 2013), attrition bias (Langford 2013 and Luby 2010), reporting bias (Langford 2013 and Luby 2010) and other bias (Langford 2013)

Low number of events

OPEN IN VIEWER

Table 8

Assessment of the certainty of evidence for handwashing before eating (pooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing before eating (Huda 2012 and Phuanukoonnon 2013)
2	2 randomised trials a	very serious b	not serious	not serious	not serious	none	326/2209 (14.8%)	91/2045 (4.4%)	RR 1.12 (1.02 to 1.22)	5 more per 1.000 (from 1 more to 10 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT (Huda 2012) and 1 quasi-RCT (Phuanukoonnon 2013)

Selection bias (Huda 2012), attrition, detection and reporting bias (Phuanukoonnon 2013)

OPEN IN VIEWER

Table 9

Assessment of the certainty of evidence for handwashing before eating (pooled data), elements of psychosocial theory vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Elements of psychosocial theory	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing before eating (Langford 2013 and Luby 2010)
2	2 randomised trials a	serious b	not serious	not serious	serious c	none	92/472 (19.5%)	0/131 (0.0%)	RR 34.73 (4.90 to 246.39)	0 fewer per 1.000 (from 0 fewer to 0 fewer)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT (Luby 2010) and 1 quasi-RCT (Langford 2013)

Attrition and reporting bias (Langford 2013 and Luby 2010) and detection and other bias (Langford 2013)

Low number of events

OPEN IN VIEWER

Table 10

Assessment of the certainty of evidence for handwashing before feeding a child (pooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing before feeding a child (Huda 2012 and Phuanukoonnon 2013)
2	2 randomised trials a	very serious b	not serious	not serious	serious c	none	292/890 (32.8%)	80/653 (12.3%)	RR 1.04 (0.94 to 1.15)	5 more per 1.000 (from 7 fewer to 18 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT (Huda 2012) and 1 quasi-RCT (Phuanukoonnon 2013)

Selection bias (Huda 2012), attrition, detection and reporting bias (Phuanukoonnon 2013)

Low number of events

OPEN IN VIEWER

Table 11

Assessment of the certainty of evidence for handwashing before feeding a child (pooled data), elements of psychosocial theory vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Elements of psychosocial theory	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing before feeding a child (Langford 2013 and Luby 2010)
2	2 randomised trials a	serious b	not serious	not serious	serious c	none	34/64 (53.1%)	8/52 (15.4%)	RR 3.63 (1.91 to 6.88)	405 more per 1.000 (from 140 more to 905 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT (Luby 2010) and 1 quasi-RCT (Langford 2013)

Attrition and reporting bias (Langford 2013 and Luby 2010) and detection and other bias (Langford 2013)

Low number of events

OPEN IN VIEWER

Table 12

Assessment of the certainty of evidence for latrine use (adherence) (pooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Latrine use: adherence (Jinadu 2007 and Pattanayak 2009)
2	2 randomised trials a	serious b	not serious	not serious	serious c	none	47/174 (27.0%)	18/177 (10.2%)	RR 2.63 (1.62 to 4.29)	166 more per 1.000 (from 63 more to 335 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 RCT (Jinadu 2007) and 1 cluster RCT (Pattanayak 2009)

Reporting bias (Jinadu 2007) and attrition bias (Pattanayak 2009)

Low number of events

OPEN IN VIEWER

Table 13

Assessment of the certainty of evidence for open defecation (pooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Open defecation (Lansdown 2002, Stanton 1987 and Wang 2013)
3	3 randomised trials a	serious b	not serious	not serious	serious c	none	172/197 (87.3%)	168/191 (88.0%)	RR 0.99 (0.72 to 1.37)	9 fewer per 1.000 (from 246 fewer to 325 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 RCT (Lansdown 2002) and 2 cluster RCT's (Stanton 1987 and Wang 2013)

Detection bias (Lansdown 2002), attrition bias (Stanton 1987) and reporting bias (Lansdown 2002 and Wang 2013)

Low number of events

OPEN IN VIEWER

Table 14

Assessment of the certainty of evidence for skills, using soap for handwashing (pooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Skills: using soap for handwashing (Bowen 2013 and Luby 2009)
2	2 randomised trials a	serious b	not serious	not serious	not serious	none	592/626 (94.6%)	291/326 (89.3%)	RR 1.05 (1.02 to 1.08)	45 more per 1.000 (from 18 more to 71 more)		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 cluster RCT's

Detection bias (Bowen 2013) and attrition bias (Luby 2009)

We also expressed the effect measures as Risk Differences (RD), showing the absolute effect, instead of Risk Ratios (RR) (Table 15).

OPEN IN VIEWER

Table 15

Risk ratio and risk difference

Outcome		RR, [95% CI]		RD, [95% CI]
	Number of studies	Results	I2 (%)	Results	I2 (%)
Handwashing after toilet use
Total	8	1.24, [1.00, 1.54]	96.5	0.12, [0.02, 0.22]*	94.0
Community-based approach	2	1.06, [0.99, 1.14]	0.0	0.06, [-0.00, 0.11]	0.0
Sanitation and hygiene messaging	4	1.12, [0.80, 1.57]	97.8	0.07, [-0.06, 0.20]	95.4
Elements of psychosocial theory	2	1.99, [0.15, 25.93]	99.0	0.31, [-0.20, 0.83]	97.7
Handwashing before cooking
Total	5	2.42, [0.97, 6.04]	88.3	0.23, [0.01, 0.44]*	98.7
Community-based approach	1	0.94, [0.31, 2.91]	-	-0.00, [-0.01, 0.01]	-
Sanitation and hygiene messaging	2	1.23, [1.09, 1.39]*	0.0	0.15, [0.07, 0.23]*	0.0
Elements of psychosocial theory	2	33.06, [6.72, 162.69]*	0.0	0.43, [-0.13, 0.98]	98.2
Handwashing after cleaning a child's anus
Total	5	1.24, [0.97, 1.59]	60.9	0.13, [0.01, 0.26]*	82.7
Community-based approach	1	1.34, [0.85, 2.12]	-	0.09, [-0.05, 0.23]	-
Sanitation and hygiene messaging	2	1.10, [0.64, 1.90]	80.7	0.03, [-0.11, 0.17]	82.9
Elements of psychosocial theory	2	2.23, [0.27, 18.63]	90.5	0.33, [-0.05, 0.71]	87.7
Handwashing before eating
Total	6	1.34, [0.83, 2.18]	97.8	0.13, [0.04, 0.22]*	96.7
Community-based approach	2	1.12, [1.02, 1.22]*	0.0	0.05, [-0.07, 0.16]	88.7
Sanitation and hygiene messaging	2	1.06, [0.81, 1.39]	54.9	0.05, [-0.14, 0.23]	52.7
Elements of psychosocial theory	2	34.73, [4.90, 246.39]*	0.0	0.32, [-0.08, 0.71]	96.9
Handwashing before feeding a child
Total	4	1.82, [0.71, 4.66]	87.3	0.16, [-0.01, 0.34]	92.6
Community-based approach	2	1.04, [0.94, 1.15]	0.0	0.01, [-0.01, 0.02]	0.0
Elements of psychosocial theory	2	3.63, [1.91, 6.88]*	0.0	0.35, [0.07, 0.63]*	73.2
Latrine use
Total	4	3.63, [0.79, 16.78]	99.1	0.31, [-0.04, 0.67]	99.4
Community-based approach: Adherence	2	2.63, [1.62, 4.29]*	0.0	0.13, [-0.05, 0.30]	86.3
Community-based approach: Longer-term use	2	4.02, [0.44, 37.13]	99.7	0.50, [-0.04, 1.03]	99.7
Safe faeces disposal
Total	3	1.63, [1.29, 2.08]*	57.2	0.17, [0.01, 0.32]*	92.8
Community-based approach	2	1.67, [1.10, 2.53]*	76.5	0.17, [-0.06, 0.40]	95.9
Sanitation and hygiene messaging	1	1.68, [1.21, 2.32]*	-	0.17, [0.07, 0.27]*	-
Elements of psychosocial theory	-	-	-	-	-
Safe child faeces disposal
Total	3	1.65, [0.62, 4.39]	92.8	0.14, [-0.15, 0.43]	96.8
Community-based approach	2	2.07, [0.59, 7.22]	88.0	0.20, [-0.18, 0.59]	96.7
Sanitation and hygiene messaging	1	1.07, [0.70, 1.65]	-	0.01, [-0.07, 0.10]	-
Open defecation
Total	4	0.61, [0.21, 1.81]	99.6	-0.18, [-0.46, 0.10]	98.1
Community-based approach	1	0.40, [0.37, 0.44]*	-	-0.33, [-0.36, -0.31]*	-
Sanitation and hygiene messaging	3	0.99, [0.72, 1.37]	36.0	-0.11, [-0.38, 0.16]	73.2
Skills: using soap for handwashing
Total	2	1.05, [1.02, 1.08]*	1.4	0.05, [0.02, 0.08]*	0.0
Sanitation and hygiene messaging	2	1.05, [1.02, 1.08]*	1.4	0.05, [0.02, 0.08]*	0.0
Skills: rubbing hands together at least 3 times
Total	2	5.78, [0.84, 39.71]	97.0	0.61, [-0.09, 1.31]	99.6
Sanitation and hygiene messaging	2	5.78, [0.84, 39.71]	97.0	0.61, [-0.09, 1.31]	99.6
Elements of psychosocial theory	-	-	-	-	-
Skills: lathering hands > 10 sec
Total	2	6.25, [1.03, 38.11]*	95.9	0.56, [-0.07, 1.19]	99.5
Sanitation and hygiene messaging	2	6.25, [1.03, 38.11]*	95.9	0.56, [-0.07, 1.19]	99.5
Skills: drying hands with a clean towel
Total	2	1.68, [0.62, 4.55]	95.2	0.14, [0.02, 0.26]*	78.0
Sanitation and hygiene messaging	2	1.68, [0.62, 4.55]	95.2	0.14, [0.02, 0.26]*	78.0

All risk ratios and risk differences are presented as Cochran-Mantel-Haenszel Estimate, [95% CI]. RR: Risk Ratio; CI: Confidence Interval; RD: Risk Difference; I²: heterogeneity; *p≤0.05

We performed a sensitivity analysis for the use of incentives as part of the promotional approach (see Table 16, forest plots available upon request). Three studies made use of financial or non-financial incentives, including providing a modest salary to the secondary implementer as part of a community-based approach (Huda et al., 2012), providing small subsidies to the households as part of a community-based approach (Pattanayak et al., 2009), and providing a bar of soap as part of a theory-based approach (Langford & Panter-Brick, 2013).

OPEN IN VIEWER

Table 16

Risk ratios in studies describing programmes including incentives versus programmes without use of incentives

Outcome	RR, [95% CI] (incentives)			RR, [95% CI] (no incentives)
	Number of studies	Results	I2 (%)	Number of studies	Results	I2 (%)
Handwashing after toilet use
Community-based approach	1	1.27, [0.72, 2.23]	-	1	1.06, [0.99, 1.14]	-
Elements of psychosocial theory	1	1.10, [0.99, 1.22]	-	1	3.62, [2.20, 5.93]*	-
Handwashing before cooking
Community-based approach	1	0.94, [0.31, 2.91]	-	-	-	-
Elements of psychosocial theory	1	30.58, [4.37, 214.06]*	-	1	38.75, [2.41, 622.42]*	-
Handwashing after cleaning a child's anus
Community-based approach	1	1.34, [0.85, 2.12]	-	-	-	-
Elements of psychosocial theory	1	1.19, [1.04, 1.37]*	-	1	4.74, [1.29, 17.44]*	-
Handwashing before eating
Community-based approach	1	1.14, [0.63, 2.04]	-	1	1.12, [1.02, 1.22]*	-
Elements of psychosocial theory	1	43.21, [2.71, 688.87]*	-	1	27.89, [1.74, 446.44]*	-
Handwashing before feeding a child
Community-based approach	1	1.35, [0.63, 2.92]	-	1	1.04, [0.94, 1.14]	-
Elements of psychosocial theory	1	3.58, [1.85, 6.92]*	-	1	4.50, [0.27, 75.60]	-
Latrine use
Community-based approach: adherence	1	2.59, [1.58, 4.25]*	-	1	4.74, [0.24, 95.33]	-
Safe child faeces disposal
Community-based approach	1	1.11, [0.50, 2.49]	-	1	1.44, [1.27, 1.65]*	-

All risk ratios are presented as Cochran-Mantel-Haenszel Estimate, [95% CI]. RR: Risk Ratio; CI: Confidence Interval; I²: heterogeneity; *p≤0.05

In Table 16 we present the findings of the studies describing programmes with incentives versus studies where no incentives were used. Focussing on findings from studies with low heterogeneity (< 50%), we found: (1) statistically significant improvement in handwashing after toilet use (RR 3.62, 95% CI [2.20, 5.93], elements of psychosocial theory), handwashing before eating (RR 1.12, 95% CI [1.02, 1.22], community-based approach) and safe child faeces disposal (RR 1.44, 95% CI [1.27, 1.65], community-based approach) when using programmes without incentives, while this was not the case for similar programmes using incentives; (2) for handwashing before cooking, handwashing after cleaning a child's anus and handwashing before eating, both programmes (based on elements of psychosocial theory) with and without incentives had statistically significant positive effects, but the RR was larger for the programmes without incentives; (3) programmes that used elements of psychosocial theory: statistically significant improvement in handwashing before feeding a child (RR 3.58, 95% CI [1.85, 6.92]), and in latrine use (RR 2.59, 95% CI [1.58, 4.25]) was found when using programmes making use of incentives, while this was not the case for programmes not using incentives; (4) no positive effects on handwashing after toilet use or before feeding a child were present in community-based interventions with or without incentives. Overall, the number of studies is too limited, and the type of incentives is too variable, to be able to make any firm conclusions based on these data.

In summary, because of a high degree of heterogeneity it was very difficult to make overall conclusions about the effectiveness of using any promotional approach versus no promotional approach, and about the effectiveness of a specific promotional approach. Since many other specific outcomes were measured that were not included in the meta-analyses because these were unique outcomes, we provide a more complete overview below, however without statistically pooling these.

4.3.1.2 Community-based approaches

From the 12 studies that we categorised as describing a community-based approach, 8 clearly described the approach as a formal community-based approach, and the following formal approaches were identified: community-led total sanitation (Guiteras et al., 2015b; Patil et al., 2013/2015; Pattanayak et al., 2009; Pickering et al., 2015), community-based interventions (Andrade, 2013; Jinadu et al., 2007) and community health clubs or women's groups (Waterkeyn & Cairncross, 2005; Younes et al., 2015). The other studies did not formally describe their approach as community-based approach, but clear elements of community involvement and engagement were described (Hoque et al., 1994/1996; Huda et al., 2012; Kochurani et al., 2009; Phuanokoonnon et al., 2013). One study was a school-based study (Kochurani et al., 2009), and Andrade (2013) worked at household, community and school level at the same time. All but one study had a sanitation component in the intervention: four studies only focused on sanitation, 7 studies looked at a mixed intervention (all WASH components in 6 cases, water supply/water quality and sanitation in one case) component, and only one study contained a handwashing only programme (see Figure 6).

Below we narratively describe the findings for the different outcome types. We specifically mention when the programme only consisted of a sanitation intervention, or handwashing intervention. In all other cases the programme contained all WASH elements. •

Behaviour change: handwashing (Analysis 14). One study, implementing a handwashing only intervention, measured handwashing at key times during the intervention period (“uptake”) (Younes 2015). A significant increase in handwashing with soap before food preparations (RR 4.31, 95% CI [3.40, 5.45]), or before feeding a child was measured (RR 2.83, 95% CI [2.50, 3.20]) (certainty of evidence: low (Table 17)) (Younes et al., 2015). In two studies adherence outcomes were measured. In a sanitation only study with a moderate risk of bias a statistically significant increase in handwashing after cleaning children's faeces, and after defecation was found (RR 2.23, 95% CI [1.21, 4.10]) (Jinadu et al., 2007). A significant increase in “handwashing before eating” was shown (RR 1.12, 95% CI [1.02, 1.22]) in a smaller experimental study with serious risk of bias, however a significant change could not be shown for 5 other key times (Phuanokoonnon et al., 2013). The certainty of evidence for the adherence outcomes was found to be low (Table 18). In addition, three studies measured longer-term use outcomes (Huda et al., 2012; Pickering et al., 2015; Kochurani et al., 2009). The community-based intervention, only containing a sanitation component, significantly improved handwashing with soap (MD 0.50, 95% CI [0.33, 0.67]) (Pickering et al., 2015). Kochurani et al. (2009), a school level study, found that the community-based intervention significantly increased the frequency of handwashing before eating (96% versus 61%, n=7,835; p<0.0001). However, a significant effect in handwashing at 7 different key times (including handwashing before eating) could not be demonstrated in an experimental study with serious risk of bias (Huda et al., 2012). The level of evidence for handwashing at longer term was found to be very low (Table 19).

•

Behaviour change: latrine use (Analysis 15). A statistically significant increase in latrine use during the intervention period (“uptake”) was measured (RR 1.88, 95% CI [1.39, 2.55]) (Waterkeyn & Cairncross, 2005). In Hoque et al. (1994/1996) it was shown that latrine use after the intervention increased by 89%, however no standard deviations were provided, so it was not possible to calculate confidence intervals. Adherence outcomes were measured in two different experimental studies, describing a sanitation only intervention, and a significant increase in overall latrine use (RR 2.59, 95% CI [1.58, 4.25]), and latrine use in children up to 24 months (RR 7.95, 95% CI [4.72, 13.40]) was shown (Pattanayak et al., 2009; Jinadu et al., 2007), however no difference in latrine use in children between 25 and 60 months could be shown (RR 4.74, 95% CI [0.24, 95.33]) (Jinadu et al., 2007). The adherence outcomes had a low certainty of evidence (Table 20). In the longer term statistically significantly increased overall latrine use (RR 1.48, 95% CI [1.37, 1.59]), latrine use by males (RR 10.40, 95% CI [7.59, 14.26]), latrine use by females (RR 11.70, 95% CI [8.36, 16.37]), and potty use by children (RR 3.28, 95% CI [2.90, 3.71]) was shown (Hoque et al., 1994/1996; Pickering et al., 2015). The certainty of evidence for the longer-term outcomes was found to be low (Table 21). The study by Pickering et al. (2015) was a sanitation-only intervention, while Hoque et al. (1994/1996) combined sanitation with a water supply/water quality intervention.

•

Behaviour change: safe faeces disposal (Analysis 16). Two studies measured outcomes during the study period (“uptake”) (Waterkeyn & Cairncross, 2005; Patil et al., 2013/2015). A statistically significant increase of “not disposing faeces in the open” (RR 2.41, 95% CI [1.99, 2.90]) was demonstrated in a quasi-experimental study (Waterkeyn & Cairncross, 2005). No difference in the presence of child faeces in the yard was shown (Waterkeyn & Cairncross, 2005). Patil et al. (2013/2015), describing a sanitation-only intervention, reported this outcome result as means, but no standard deviations were given. From the paper, the ITT adjusted difference between intervention and control was 0.075, 95% CI [0.036, 0.113] for child faeces disposal (in favour of the community-based intervention) and 0.019, 95% CI [-0.026, 0.065] for “no faeces observed in living area”, the latter being non-significant. The certainty of evidence for the uptake outcomes was assessed as very low (Table 22). Significant outcomes were also shown in the period less than 12 months after the programme period (“adherence”): child faeces disposal (RR 2.16, 95% CI [1.60, 2.91]) and no faeces lying around (RR 1.44, 95% CI [1.27, 1.65]), in a study implementing a sanitation-only intervention (Jinadu et al., 2007). The certainty of evidence for the adherence outcomes was assessed as moderate (Table 23). In the longer term a significant increase in not leaving human faeces in the compound was shown in an experimental study (sanitation-only) with moderate risk of bias (RR 2.07, 95% CI [1.40, 3.05]) (Pickering et al., 2015), but a significant effect on child faeces disposal could not be demonstrated in an experimental study with serious risk of bias (RR 1.02, 95% CI [0.45, 2.35]) (Huda et al., 2012). The certainty of evidence for longer-term outcomes was found to be low (Table 24).

•

Behaviour change: open defecation (Analysis 17). One experimental study, describing a sanitation-only programme, measured outcomes during the study period (“uptake”) (Patil et al., 2013/2015). The study reported this outcome result as means, but no standard deviations were given. The ITT adjusted difference between intervention and control was -0.087, 95% CI [-0.135, -0.038] for men, -0.091, 95% CI [-0.141, -0.041] for women and -0.054, 95% CI [-0.088, -0.020] for children, thus the community-based intervention significantly reduced open defecation in men, women and children. The certainty of evidence for the uptake outcomes was moderate (Table 25). One study, implementing a sanitation-only intervention, measured adherence outcomes, and found a statistically significant decrease of open defecation in case of a latrine promotion program combined with use of subsidies (MD -9.00, 95% CI [-13.70, -4.30]) or a combination of subsidies and a supply intervention (MD -9.00, 95% CI [-14.10, -3.90]). No significant effect was shown in case of the supply intervention alone (MD -2.50, 95% CI [-10.73, 5.73]) (Guiteras et al., 2015b). The certainty of evidence for the adherence outcomes was found to be moderate (Table 26). Three studies measured open defecation in the longer term (Guiteras et al., 2015b; Pickering et al., 2015; Kochurani et al., 2009). A statistically significant decrease in open defecation on the longer term was shown in adult women, adult men, and children younger and older than 5 years in one study with a sanitation-only intervention (Pickering et al., 2015), however this could not be shown in case of a latrine promotion program in the study by Guiteras et al. (2015b) (MD -2.10, 95% CI [-7.20, 3.00]). Kochurani et al. (2009) found that the community-based intervention in schools significantly reduced the number of girls practicing open defecation (1% versus 9%, n=7,835; p=0.004), however for boys no significant difference was found (30% versus 23%; p=0.12). Open defecation at the longer term had a certainty of evidence of very low (Table 27).

•

Behavioural factors (Analysis 18). Three studies measured knowledge (Andrade, 2013; Kochurani et al., 2009; Phuanukoonnon et al., 2013). Andrade (2013) showed statistically significantly increased disease transmission knowledge and knowledge of key handwashing times at 1 and 2 years following the implementation of the intervention (see forest plot). For Kochurani et al. (2009), a quasi-experimental study with critical risk of bias, there was no difference in knowledge of handwashing before eating, in a group of school boys and girls. However, the community-based intervention significantly increased knowledge of handwashing after using the toilet (girls: 100% vs 93%, p=0.001; boys: 100% vs 85%, p<0.001) and knowledge on the health advantages of handwashing (girls: 98% vs 88%, p=0.002; boys: 100% vs 77%, p<0.001). For Phuanukoonnon et al. (2013) significantly higher mean knowledge scores were observed in the community-based intervention compared to the control group, concerning the fact that diarrhoea can cause weight loss among children (3.66 versus 3.47 (out of 4), n=395, p<0.05). No effect was shown for 6 other outcomes concerning knowledge about causes and consequences of diarrhoea (Phuanukoonnon et al. 2013).

•

Health outcomes (Analysis 19-20). A significant decrease in diarrhoea in children over 5 years old (RR 0.45, 95% CI [0.31, 0.64]) (Hoque et al., 1994/1996), and in acute respiratory tract illness (RR 0.58, 95% CI [0.45, 0.75]) (Younes et al., 2015) was shown. However, a significant effect on overall diarrhoea, and diarrhoea in children under 5 years old, could not be demonstrated in three studies (Hoque et al., 1994/1996; Pickering et al., 2015; Huda et al., 2012). In addition, using the ITT adjusted mean difference for the mean number of cases reported in the previous 7 days, Patil et al. (2013/2015) found no difference in cases of diarrhoea (-0.002, 95% CI [20.019, 0.015]) and high credible gastrointestinal illness (-0.002, 95% CI [20.024, 0.020]), but found that there were more cases of acute lower respiratory tract illness in the control group than in the intervention group (0.049, 95% CI [0.009, 0.089]) (Patil et al., 2013/2015). One experimental study with moderate risk of bias measured mortality outcomes (Pickering et al., 2015). A significant decrease of all-cause mortality and diarrhoea-related mortality was not found (Pickering et al., 2015).

•

Influence of incentives in programs. From the 12 studies describing community-based approaches, 5 studies described the use of incentives (see Table 3), including a modest salary to the hygiene promotors (Huda et al., 2012), a motorcycle and lunch to the health technicians (Waterkeyn & Cairncross, 2005), and subsidies to households (Guiteras et al., 2015b; Patil et al., 2013, 2015; Pattanayak et al., 2009). For these studies: (1) when providing additional incentives to the secondary implementers, there was a significant improvement of latrine use and safe faeces disposal on the short term (uptake) (Waterkeyn & Cairncross, 2005), but no significant effects on handwashing and safe faeces disposal on the longer term (Huda et al., 2012, serious risk of bias); (2) when providing incentives to the recipients of the programme, a significant improvement of safe faeces disposal and open defecation (uptake, adherence), and latrine use (adherence) was found. When comparing absolute effect measures of the individual outcomes between the studies with or without use of incentives, no major differences were found. However, Guiteras (2015b) compared a community-based intervention with and without use of subsidies (i.e. latrine vouchers), and found significant better results for open defecation when subsidies were given as an additional incentive.

OPEN IN VIEWER

Figure 14

Analysis

OPEN IN VIEWER

Figure 15

Analysis

OPEN IN VIEWER

Figure 16

Analysis

OPEN IN VIEWER

Figure 17

Analysis

OPEN IN VIEWER

Figure 18

Analysis

OPEN IN VIEWER

Figure 19

Analysis

OPEN IN VIEWER

Figure 20

Analysis

OPEN IN VIEWER

Table 17

Assessment of the certainty of evidence for handwashing at key times (uptake) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (uptake) (Younes 2015)
1	1 Quasi-experimental study a	not serious	not serious	not serious	not serious	none	930/2164 (43.0%)	321/2376 (13.5%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 non-randomised controlled trial

OPEN IN VIEWER

Table 18

Assessment of the certainty of evidence for handwashing at key times (adherence) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (adherence) (Jinadu 2007 and Phuanukoonnon 2013)
2	2 randomised trials a	serious b	serious	not serious	not serious	none	1220/1982 (61.6%)	326/617 (52.8%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 RCT (Jinadu 2007) and 1 quasi-RCT (Phuanukoonnon 2013)

Attrition bias (Phuanukoonnon 2013) and detection bias (Phuanukoonnon 2013)

OPEN IN VIEWER

Table 19

Assessment of the certainty of evidence for handwashing at key times (longer-term use) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (longer-term use) (Huda 2012, Kochurani 2009 and Pickering 2015)
3	2 randomised trials a and 1 quasi-experimental study b	very serious c	not serious	not serious	serious d	none	Three studies measured longer-term use outcomes (Huda 2012, Pickering 2015, Kochurani 2009). The community-based intervention significantly improved handwashing with soap (MD 0.50, 95% CI [0.33, 0.67]) (Pickering 2015). Kochurani (2009), a school level study, found that the community-based intervention significantly increased the frequency of handwashing before eating (96% versus 61%, n=7,835; p<0.0001). However, a significant effect in handwashing at 7 different key times (including handwashing before eating) could not be demonstrated in an experimental study with serious risk of bias (Huda 2012).					CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 cluster RCT's (Huda 2012 and Pickering 2015)

1 non-randomised controlled trial (Kochurani 2009)

Selection bias (Huda 2012, Kochurani 2009 and Pickering 2015), attrition/reporting bias (Pickering 2015), bias due to confounding/bias in measurement of outcomes/interventions/bias due to departures from intended interventions (Kochurani 2009)

Lack of data

OPEN IN VIEWER

Table 20

Assessment of the certainty of evidence for latrine use (adherence) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Latrine use: adherence (Jinadu 2007 and Pattanayak 2009)
2	2 randomised trials a	serious b	not serious	not serious	serious c	none	163/397 (41.1%)	32/391 (8.2%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 RCT (Jinadu 2007) and 1 cluster RCT (Pattanayak 2009)

Reporting bias (Jinadu 2007) and attrition bias (Pattanayak 2009)

Low number of events

OPEN IN VIEWER

Table 21

Assessment of the certainty of evidence for latrine use (longer-term use) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Latrine use: longer-term use (Hoque 1994/1996 and Pickering 2015)
2	2 randomised trials a	serious b	serious	not serious	not serious	none	1860/2367 (78.6%)	526/1817 (28.9%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 RCT (Hoque 1994/1996) and 1 cluster-RCT (Pickering 2015)

Selection bias (Huda 2012/Pickering 2015) and attrition/reporting bias (Pickering 2015)

OPEN IN VIEWER

Table 22

Assessment of the certainty of evidence for safe faeces disposal (uptake) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Safe faeces disposal practices (uptake) (Patil 2013/2015 and Waterkeyn 2005)
2	1 randomised trial a and 1 quasi-experimental study b	very serious c	not serious	not serious	serious d	none	Two studies measured outcomes during the study period (“uptake”) (Waterkeyn & Cairncross, 2005; Patil et al., 2013/2015). A statistically significant increase of “not disposing faeces in the open” (RR 2.41, 95% CI [1.99, 2.90]) was demonstrated in a quasi-experimental study (Waterkeyn & Cairncross, 2005). No difference in the presence of child faeces in the yard was shown (Waterkeyn & Cairncross, 2005). Patil et al. (2013/2015) reported this outcome result as means, but no standard deviations were given. From the paper, the ITT adjusted difference between intervention and control was 0.075, 95% CI [0.036, 0.113] for child faeces disposal (in favour of the community-based intervention) and 0.019, 95% CI [-0.026, 0.065] for “no faeces observed in living area”, the latter being non-significant.					CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT (Patil 2013/2015)

1 non-randomised controlled trial (Waterkeyn 2005)

Detection bias (Patil 2013/2015), selection bias/bias due to confounding/bias in measurement of interventions/outcomes/bias due to departures from intended interventions (Waterkeyn 2005)

Lack of data

OPEN IN VIEWER

Table 23

Assessment of the certainty of evidence for safe faeces disposal (adherence) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Safe faeces disposal practices (adherence) (Jinadu 2007)
1	1 randomised trial a	serious b	not serious	not serious	not serious	none	226/300 (75.3%)	132/290 (45.5%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 RCT

Reporting bias

OPEN IN VIEWER

Table 24

Assessment of the certainty of evidence for safe faeces disposal (longer-term use) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Safe faeces disposal practices (longer-term use) (Huda 2012 and Pickering 2015)
2	2 randomised trials a	serious b	not serious	not serious	serious c	none	76/652 (11.7%)	46/726 (6.3%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 cluster RCTs

Selection bias (Huda 2012 and Pickering 2015) and attrition/reporting bias (Pickering 2015)

Low number of events

OPEN IN VIEWER

Table 25

Assessment of the certainty of evidence for open defecation (uptake) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Open defecation (uptake) (Patil 2013/2015)
1	1 randomised trial a	serious b	not serious	not serious	not serious	none	The study reported this outcome result as means, but no standard deviations were given. The ITT adjusted difference between intervention and control was -0.087, 95% CI [-0.135, -0.038] for men, -0.091, 95% CI [-0.141, -0.041] for women and -0.054, 95% CI [-0.088, -0.020] for children, thus the community-based intervention significantly reduced open defecation in men, women and children.					CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT

Detection bias

OPEN IN VIEWER

Table 26

Assessment of the certainty of evidence for open defecation (adherence) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Open defecation (adherence) (Guiteras 2015b)
1	1 randomised trial a	serious b	not serious	not serious	not serious	none	The study reported this outcome as a % mean difference. It was shown that a latrine promotion program (LPP) in combination with subsidies (and a supply intervention) resulted in a statistically significant decreased open defecation (MD -9%, 95% CI [-13.70, -4.30] for LPP+subsidy and MD -9%, 95%CI [-14.10, -3.90] for LPP+subsidy+supply. A statistically difference in open defecation after receiving the supply intervention only could not be demonstrated (MD -2.50%, 95%CI [-10.73, 5.73]					CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT

Other bias

OPEN IN VIEWER

Table 27

Assessment of the certainty of evidence for open defecation (longer-term use) (unpooled data), community-based approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Community-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Open defecation (longer-term use) (Guiteras 2015b, Kochurani 2009 and Pickering 2015)
3	2 randomised trials a and 1 quasi-experimental study b	very serious c	not serious	not serious	serious d	none	A statistically significant decrease in open defecation on the longer term was shown in adult women, adult men, and children younger and older than 5 years in one study (Pickering et al., 2015). Kochurani et al. (2009) found that the community-based intervention in schools significantly reduced the number of girls practicing open defecation (1% versus 9%, n=7,835; p=0.004), however for boys no significant difference was found (30% versus 23%; p=0.12). Finally, in 1 study (Guiteras 2015b), a statistically difference in open defecation after receiving a latrine promotion program could not be demonstrated (MD -2.10%, 95%CI [-7.20, 3.00]					CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 cluster RCT's (Guiteras 2015b and Pickering 2015)

1 non-randomised controlled trial (Kochurani 2009)

Selection bias (Kochurani 2009 and Pickering 2015), attrition/reporting bias (Pickering 2015), bias due to confounding/bias in measurement of outcomes/interventions/bias due to departures from intended interventions (Kochurani 2009) and other bias (Guiteras 2015b)

Lack of data

4.3.1.3 Social marketing approaches

From the 6 studies that we grouped in the category “social marketing approaches”, 5 studies formally described that they used a marketing campaign or social marketing techniques or interventions (Biran et al., 2009; Briceno et al., 2015; Cameron et al., 2013; Galiani et al., 2012/2015; Pinfold, 1999). Two of these studies implemented their intervention at school level (Galiani et al., 2012/2015; Pinfold, 1999). One study did not describe their approach as a formal social marketing approach, but used several elements that are generally part of a social marketing approach (infrastructure promotion, use of incentives) (Arnold et al., 2009). Since for the study of Galiani et al. (2012/2015) no raw data were available, the data represented are adjusted for confounding factors (gender and education of household head, children's age and gender, mother living in the home, rainfall and geographical region). All but one study had a handwashing component in the intervention (in contrast to the community-based approaches, where the focus was a sanitation intervention): four studies described a handwashing-only intervention, with one of these also studying an intervention arm with sanitation-only and a combined intervention, one study combined the handwashing intervention with a water supply/water quality component, and two studies included a sanitation-only intervention (see Figure 6).

Below we narratively describe the findings for the different outcome types. Since the majority of the studies had a handwashing-only intervention, we only mention the intervention specifically in case of a sanitation or combined programme. •

Behaviour change: handwashing (Analysis 21). Ony study, implementing a sanitation-only intervention, measured outcomes during the study period (“uptake”) (Cameron et al., 2013). Handwashing after toilet use was measured, but no significant increase in handwashing could be demonstrated (Cameron et al., 20132013). Two experimental studies (Galiani et al., 2012/2015; Briceno et al., 2015), and one observational study, with a combined handwashing and water supply/quality intervention (Arnold et al., 2009), measured outcomes less than 12 months after the programme period (“adherence”), and some differences across these studies were found. In a study with moderate risk of bias (Briceno et al., 2015), for the outcome “handwashing before food handling” a significant effect was shown when implementing a handwashing intervention (MD 7.70, 95% CI [3.78, 11.62]), or a combined handwashing and sanitation intervention (MD 1.60, 95% CI [0.03, 3.17]), however results were not consistent when measured by observation or in a self-reported way (Briceno et al., 2015). In addition, this effect could not be shown in a second study with moderate risk of bias, implementing a community level or school level intervention (Galiani et al., 2012/2015). For “handwashing with water and soap prior to eating” a significant effect was shown in the case of a school level intervention (self-reported: MD 0.09, 95% CI [0.01, 0.18]; observation: MD 0.12, 95% CI [0.02, 0.21]) (Galiani et al., 2012/2015), but not for the community level intervention (Galiani et al., 2012/2015) or in the observational study (Arnold et al., 2009). Finally, no significant effect could be demonstrated for handwashing with soap during the period “the last 24 hours” (Briceno et al., 2015), or handwashing at other key times (before feeding a child, after faecal contact, before cooking, before eating, after changing baby) (Arnold et al., 2009; Briceno et al., 2015; Galiani et al., 2012/2015). The certainty of evidence was very low for the adherence outcomes (Table 28). No longer term outcomes were found in studies using social marketing approaches.

•

Behaviour change: latrine use (Analysis 22). In one experimental study latrine use adherence was measured (Briceno et al., 2015). A significant effect on shared latrine use could not be demonstrated in the case of a handwashing intervention only (MD -3.1, 95% CI [-8.98, 2.78]), however in the case of a sanitation intervention, or a combined handwashing and sanitation intervention, a significant decrease of shared latrine use (indicating more private latrine use) was shown (MD -9.2, 95% CI [-14.49, -3.91] and MD -7.6, 95% CI [-70.90, -81.10] respectively) (Briceno et al., 2015). The certainty of evidence for this outcome was moderate (Table 29).

•

Behaviour change: safe faeces disposal (Analysis 23). Only outcomes for the period “less than 12 months after the end of the implementation period” were measured (“adherence”). In an experimental study with moderate risk of bias, a positive effect was seen for the observation of faeces outside the latrine in the case of a combined handwashing and sanitation intervention (MD -4.3, 95% CI [-8.42, -0.18]), but not for the handwashing or sanitation intervention alone. A significant increase of safe child faeces disposal was seen in the case of a sanitation or combined intervention (MD 11.7, 95% CI [5.04, 18.36] and MD 8.4, 95% CI [1.93, 14.87] respectively)), but not for the handwashing intervention alone (MD 4.3, 95% CI [-2.76, 11,36]) (Briceno et al., 2015). No significant increase in safe faeces disposal could be demonstrated in an observational study with serious risk of bias where a handwashing and water supply/quality programme was implemented (RR 0.91, 95% CI [0.83, 1.01]) (Arnold et al., 2009). The certainty of evidence for these outcomes was very low (Table 30).

•

Behaviour change: open defecation (Analysis 24). No statistically significant decrease of open defecation could be shown during the progam period (“uptake”) in an experimental study with low risk of bias, describing a sanitation-only intervention (RR 0.92, 95% CI [0.80, 1.05]) (Cameron et al., 2013). In case of a sanitation, or combined sanitation and handwashing intervention, a statistically significant decrease of people that always or regularly practice open defecation, and that usually defecate in fields, bushes or rivers, could be shown for the period less than 12 months after the end of the implementation (“adherence”), but not for the handwashing intervention alone (Briceno et al., 2015). The certainty of evidence for this outcome was found to be moderate (Table 31).

•

Behavioural factors (Analysis 25). Three experimental (Cameron et al., 2013; Briceno et al., 2015; Galiani et al., 2012/2015), and one quasi-experimental study performed in schools (Pinfold, 1999), measured the effect of social marketing approaches on knowledge. In a study with low risk of bias (Cameron et al., 2013), no effect could be demonstrated concerning knowledge about causes of diarrhoea, and building of a latrine. In a study with a moderate risk of bias, a significant increase in the knowledge that “not washing hands with water and soap is the main cause of diarrhoea”, was seen for the community level intervention (Galiani et al., 2012/2015). A statistically significant increase in handwashing knowledge was reported in 3 studies (Pinfold, 1999; Briceno et al., 2015; Galiani et al., 2012/2015). In Briceno et al. (2015), only the combined handwashing and sanitation intervention led to improved knowledge concerning the best method to wash hands and when to wash hands. In Galiani et al. (2012/2015), this result was only seen in the school level intervention. In one study, the knowledge of the key events when handwashing was required, was tested, but no effect on this knowledge could be demonstrated as a result of the intervention (Galiani et al., 2012/2015). One study looked at skills, and more specifically at the practice of handwashing with one or both hands (Biran et al., 2009). An effect on washing one hand or both hands could not be demonstrated (RR 1.01, 95% CI [0.62, 1.64] and RR 0.70, 95% CI [0.48, 1.02] respectively) (Biran et al., 2009). A third behavioural factor, attitudes, was investigated in one experimental study with a low risk of bias (Cameron et al., 2013), but no effect on the attitude to open defecation could be demonstrated. The outcome “norms” was measured in one experimental study (Briceno et al., 2015); the combined sanitation and handwashing intervention resulted in a significant decrease in the number of households that were aware of community members practicing open defecation (MD -6.6, 95% CI [-12.87, -0.033]), but this was not the case for the sanitation (MD -5.50, 95% CI [-11.18, 0.18]) or handwashing (MD -5.20, 95% CI [-10.88, 0.48]) intervention alone (Briceno et al., 2015).

•

Health outcomes (Analysis 26). Morbidity outcomes were studied in three experimental (Cameron et al., 2013; Briceno et al., 2015; Galiani et al., 2012/2015) and one observational study (Arnold et al., 2009). An effect of the social marketing approach could not be shown for any diarrhoeal, and acute respiratory tract infection outcomes (Cameron et al., 2013; Briceno et al., 2015; Arnold et al., 2009; Galiani et al., 2012/2015).

•

Influence of incentives in programs. From the 6 studies describing social marketing approaches, 2 studies described the use of incentives (see Table 3), including food (Arnold et al., 2009), and gifts (Biran et al., 2009) to the progam recipients. A third study compared two different promotional approaches, both with use of incentives, and is described below (Dickey et al., 2015). When focusing on these studies we found no significant effects on handwashing (adherence) and safe faeces disposal. When comparing the studies with or without use of incentives, there were no major differences.

OPEN IN VIEWER

Figure 21

Analysis

OPEN IN VIEWER

Figure 22

Analysis

OPEN IN VIEWER

Figure 23

Analysis

OPEN IN VIEWER

Figure 24

Analysis

OPEN IN VIEWER

Figure 25

Analysis

OPEN IN VIEWER

Figure 26

Analysis

OPEN IN VIEWER

Table 28

Assessment of the certainty of evidence for handwashing at key times (adherence) (unpooled data), social marketing approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Marketing approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (adherence) (Arnold 2009, Briceno 2015 and Galiani 2012/2015)
3	2 randomised trials a and 1 observational study b	very serious c	serious	not serious	not serious	none	Two experimental studies (Galiani et al., 2012/2015; Briceno et al., 2015), and one observational study (Arnold et al., 2009) measured outcomes less than 12 months after the programme period (“adherence”), and some differences across these studies were found. In a study with moderate risk of bias (Briceno et al., 2015), for the outcome “handwashing before food handling” a significant effect was shown when implementing a handwashing intervention (MD 7.70, 95% CI [3.78, 11.62]), or a combined handwashing and sanitation intervention (MD 1.60, 95% CI [0.03, 3.17)], however results were not consistent when measured by observation or in a self-reported way (Briceno et al., 2015). In addition, this effect could not be shown in a second study with moderate risk of bias, implementing a community level, or school level intervention (Galiani et al., 2012/2015). For “handwashing with water and soap prior to eating” a significant effect was shown in the case of a school level intervention (self-reported: MD 0.095, 95% CI [0.01, 0.18]; observation: MD 0.12, 95% CI [0.02, 0.21]) (Galiani et al., 2012/2015), but not for the community level intervention (Galiani et al., 2012/2015) or in the observational study (Arnold et al., 2009). Finally, no significant effect could be demonstrated for handwashing with soap during the period “the last 24 hours” (Briceno et al., 2015), or handwashing at other key times (before feeding a child, after faecal contact, before cooking, before eating, after changing baby) (Arnold et al., 2009; Briceno et al., 2015; Galiani et al., 2012/2015).					CRITICAL

CI: Confidence interval; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 cluster RCTs (Briceno 2015 and Galiani 2012/2015)

1 cohort study (Arnold 2009)

Attrition/other bias (Briceno 2015), bias in measurement of outcomes/bias due to departures from intended interventions (Arnold 2009)

OPEN IN VIEWER

Table 29

Assessment of the certainty of evidence for latrine use (adherence) (unpooled data), social marketing approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Marketing approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Latrine use (adherence) (Briceno 2015)
1	1 randomised trial a	serious b	not serious	not serious	not serious	none	A significant effect on shared latrine use could not be demonstrated in the case of a handwashing intervention only (MD -3.1, 95% CI [-8.98, 2.78]), however in the case of a sanitation intervention, or a combined handwashing and sanitation intervention, a significant decrease of shared latrine use (indicating more private latrine use) was shown (MD -9.2, 95% CI [-14.49, -3.91] and MD -7.6, 95% CI [-70.90, -81.10] respectively) (Briceno et al., 2015).					CRITICAL

CI: Confidence interval; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT

Attrition/other bias

OPEN IN VIEWER

Table 30

Assessment of the certainty of evidence for safe faeces disposal (adherence) (unpooled data), social marketing approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Marketing approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Safe faeces disposal practices (adherence) (Arnold 2009 and Briceno 2015)
2	1 randomised trial a and 1 observational study b	very serious c	serious	not serious	not serious	none	In an experimental study with moderate risk of bias, a positive effect was seen for the observation of faeces outside the latrine in the case of a combined handwashing and sanitation intervention (MD -4.3, 95% CI [-8.42, -0.18]), and on safe child faeces disposal in the case of a sanitation or combined intervention (MD 11.7, 95% CI [5.04, 18.36] and MD 8.4, 95% CI [1.93, 14.87] respectively)), but not for the handwashing intervention alone (MD 4.3, 95% CI [-2.76, 11,36]) (Briceno 2015). No significant increase in safe faeces disposal could be demonstrated in an observational study with serious risk of bias (RR 0.91, 95% CI [0.83, 1.01]) (Arnold 2009).					CRITICAL

CI: Confidence interval; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCTs (Briceno 2015)

1 cohort study (Arnold 2009)

Attrition/other bias (Briceno 2015), bias in measurement of outcomes/bias due to departures from intended interventions (Arnold 2009)

OPEN IN VIEWER

Table 31

Assessment of the certainty of evidence for open defecation (adherence) (unpooled data), social marketing approach vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Marketing approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Open defecation (adherence) (Briceno 2015)
1	1 randomised trial a	serious b	not serious	not serious	not serious	none	In case of a sanitation, or combined sanitation and handwashing intervention, a statistically significant decrease of people that always or regularly practice open defecation, and that usually defecate in fields, bushes or rivers, could be shown for the period less than 12 months after the end of the implementation (“adherence”), but not for the handwashing intervention alone (Briceno et al., 2015).					CRITICAL

CI: Confidence interval; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT

Attrition/other bias

4.3.1.4 Sanitation and hygiene messaging

Sanitation and hygiene messaging is a predominantly directive educational approach, consisting mainly of one-way communication, designed to help individuals and communities improve their health, by increasing their knowledge and/or skills. We identified an approach using sanitation and hygiene messaging as the major element of the promotional approach in 12 studies, of which 4 studies described school-based interventions (Abiola et al., 2012; Caruso et al., 2014; Lansdown et al., 2002; Pickering et al., 2013). All but one study had a handwashing component in the intervention (comparable to the social marketing approaches): eight studies described a handwashing-only intervention, with three of these also studying an intervention arm where handwashing was combined with either a water supply/quality or sanitation component. Six studies described a combined intervention (either handwashing with water supply/quality, handwashing with sanitation, or all three WASH components). One study included a sanitation-only intervention (see Figure 6).

Below we narratively describe the findings for the different outcome types. We specified the intervention if it was not focused on handwashing alone. •

Behaviour change: handwashing (Analysis 27-28). A significantly improved frequency of handwashing (MD 18.00, 95% CI [17.31, 18.69]) during the programme period (“uptake”) was shown in an experimental study (Kaewchana et al., 2012). In another experimental study, with a combined handwashing and sanitation intervention, a significant decrease was seen in washing hands only with water (MD -11.6%, p<0.001) (Mascie-Taylor et al., 2003). In addition, a statistically significant increase in handwashing with product after toilet use and before lunch was shown in the case of an educational intervention with hand sanitizer provision in schools (Pickering et al., 2013). In the case of an educational intervention with soap in schools, a significant increase in “handwashing with soap” after toilet use (RR 18.66, 95% CI [11.58, 30.08]) was shown, but not in “any type of handwashing” (Pickering et al., 2013), meaning that handwashing already regularly occurred before the handwashing with soap intervention was implemented. A significant increase in “handwashing with soap” before lunch was also shown in the case of the soap intervention, but again not in “any type of handwashing” (RR 19.00, 95% CI [1.22, 295.91]) (Pickering et al., 2013). For “handwashing after toilet use” at less than 12 months after the programme period (“adherence”) results were inconsistent (RR 1.15, 95% CI [1.05, 1.26]) (Abiola et al., 2012); RR 0.72, 95% CI [0.40, 1.31] (Yeager et al., 2002 (sanitation-only))), and for none of the other adherence outcomes a significant effect was demonstrated (Stanton & Clemens, 1987; Yeager et al., 2002; Abiola et al., 2012). For the uptake outcomes the certainty of evidence was found to be moderate and for the adherence outcomes it was low (Tables 32 and 33). Finally, two experimental studies, both with moderate risk of bias, measured longer-term outcomes (Bowen et al., 2013; Luby et al., 2009); in one study (Bowen et al., 2013) the handwashing intervention was combined with a water supply/quality component. No significant difference in handwashing with or with soap was shown in the first study (RR 1.00, 95% CI [0.97, 1.04]; RR 1.02, 95% CI [0.99, 1.06]) (Luby et al., 2009). However, in the second study the promotional approach had a positive effect on 9 out of 14 “handwashing at key times” outcomes (Bowen et al., 2013). The certainty of evidence for the longer-term outcomes was low (Table 34).

•

Behaviour change: latrine use (Analysis 29). Latrine use was measured in one experimental study, less than 12 months following the end of the study period (“adherence”) (Caruso et al., 2014). No statistically significant difference in latrine use was shown in this study (handwashing intervention: MD 1.80, 95% CI [-0.17, 3.77], latrine cleaning + handwashing intervention: MD -1.00, 95% CI [-2.91, 0.91]) (Caruso et al., 2014). The certainty of evidence for this outcome was found to be very low (Table 35).

•

Behaviour change: safe faeces disposal (Analysis 30). In one experimental study, with a moderate risk of bias, describing a sanitation-only intervention, a statistically significant increase in “no child faeces on the ground” was shown (RR 1.68, 95% CI [1.21, 2.32]), but an effect on “safe child faeces disposal” could not be demonstrated, in the period less than 12 months after the end of the study period (“adherence”) (RR 1.07, 95% CI [0.70, 1.65]) (Yeager et al., 2002). The certainty of evidence for this outcome was assessed to be low (Table 36).

•

Behaviour change: open defecation (Analysis 31). A significant effect of an education approach on open defecation in a short term (“uptake”) and less than 12 months after project implementation (“adherence”) could not be demonstrated in 3 experimental studies, all with moderate risk of bias (Lansdown et al., 2002; Wang et al., 2013; Stanton & Clemens, 1987). All studies had an intervention with a handwashing and sanitation component, and in 2 of the 3 also a water supply/quality component was included. The certainty of evidence for both the uptake and adherence outcomes was assessed as low (Tables 37 and 38).

•

Behavioural factors (Analysis 32). Knowledge was measured in 3 experimental (Lansdown et al., 2002; Mascie-Taylor et al., 2003; Abiola et al., 2012) and one observational study (Seimetz et al., 2016). In one study an effect of the school-based educational intervention on knowledge could not be demonstrated 9 months after the start of the intervention, however a statistically significant increase in knowledge (health causation and prevention) was measured 15 months after the end of the implementation (MD 2.71, 95% CI [0.36, 5.06]) (Lansdown et al., 2002). In a second study, no effect on perceived vulnerability, severity, or health knowledge was shown (Seimetz 2016). In Mascie-Taylor et al. (2003), the percent difference in knowledge from baseline to 18 months between intervention and control was calculated. The promotional intervention improved the level of health knowledge regarding whether worms are good for health (MD 31.1%, p<0.001), whether defecation in the courtyard is associated with worms (MD 68.2%, p<0.001), whether defecation in the bushes is associated with worms (MD 58.1%, p<0.001), and whether removal of all worms is good for a person (MD 54.7%, p<0.001). In Abiola et al. (2012) a significant increase in knowledge about the meaning of personal hygiene (RR 1.16, 95% CI [1.06, 1.27]), and eating with unclean hands as the cause of diarrhoea (RR 1.65, 95% CI [1.31, 2.08]) was shown after implementing a school-based intervention, but not for 2 other outcomes on personal hygiene knowledge. Next, three studies also measured skills (Bowen et al., 2013; Luby et al., 2009; Seimetz et al., 2016). In two of the studies (Bowen et al., 2013; Luby 2009) a statistically significant increase in using soap for handwashing (handwashing skills) was shown (RR 1.05, 95% CI [1.02, 1.08], see pooled value in Analysis 10). Also a significant increase in “rubbing hands together at least 3 times” (skills) and “lathering hands for at least 10 seconds” was shown. For “drying hands with a clean towel” a significant effect could not be shown in 2 of the 4 intervention arms (Bowen et al., 2013; Luby et al., 2009). In Seimetz et al. (2016), no difference in maintenance self-efficacy (confidence in abilities to maintain the behaviour) and recovery self-efficacy (confidence in abilities to successfully return to the behaviour) could be demonstrated, and, surprisingly, a decrease in action self-efficacy, which is the confidence in the abilities to successfully perform the behaviour, was shown (MD -0.20, 95% CI [-0.31, -0.09]). Finally, attitude outcomes were measured in two studies (Seimetz et al., 2016; Abiola et al., 2012), however the effect of sanitation and hygiene messaging on the majority of the outcomes could not be demonstrated (beliefs about costs, belief that the behaviour will lead to the outcome (response), feelings of liking washing hands, feelings of dirtiness when not washing hands, necessity to wash hands after going to the toilet, willingness to recommend practice of personal hygiene to friends), except for feelings of attractiveness when using soap to wash hands, which was significantly decreased (MD -0.27, 95% CI [-0.48, -0.06]). Seimetz et al. (2016) also measured “norms” and “self-regulation”, but no significant effects were demonstrated except a significant decrease in action control (“self-regulation”), the determination to execute and control the behaviour, was shown.

•

Health outcomes. Health outcomes were not measured in studies using sanitation and hygiene messaging approaches.

•

Influence of incentives in programs. From the 12 studies describing sanitation and hygiene messaging, only one study described the use of incentives (see Table 3), which was the provision of soap bars to the programme recipients (Seimetz et al., 2016). This study only reported outcomes such as skills, attitude and self-regulation and could not show any improvement of these outcomes. No difference were shown in these outcomes when in- or excluding this study making use of soap bars as incentives.

OPEN IN VIEWER

Figure 27

Analysis

OPEN IN VIEWER

Figure 28

Analysis

OPEN IN VIEWER

Figure 29

Analysis

OPEN IN VIEWER

Figure 30

Analysis

OPEN IN VIEWER

Figure 31

Analysis

OPEN IN VIEWER

Figure 32

Analysis

OPEN IN VIEWER

Table 32

Assessment of the certainty of evidence for handwashing at key times (uptake) (unpooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (uptake) (Pickering 2013)
1	1 randomised trial a	serious b	not serious	not serious	not serious	none	2089/3692 (56.6%)	686/3482 (19.7%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT

Reporting, detection and other bias

OPEN IN VIEWER

Table 33

Assessment of the certainty of evidence for handwashing at key times (adherence) (unpooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (adherence) (Abiola 2012, Stanton 1987, Yeager 2002)
3	3 randomised trials a	very serious b	not serious	not serious	not serious	none	313/631 (49.6%)	290/598 (48.5%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 quasi-RCTs (Abiola 2012 and Yeager 2002) and 1 cluster RCT (Stanton 1987)

Reporting bias (Abiola 2012 and Yeager 2002), attrition bias (Stanton 1987 and Yeager 2002) and detection bias (Abiola 2012)

OPEN IN VIEWER

Table 34

Assessment of the certainty of evidence for handwashing at key times (longer-term use) (unpooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (longer-term use) (Bowen 2013 and Luby 2009)
2	2 randomised trials a	serious b	serious	not serious	not serious	none	No significant difference in handwashing with or without soap was shown in the first study (RR 1.02, 95% CI [0.99, 1.05) (Luby et al., 2009). However, in the second study the education approach had a positive effect on 8 out of 14 “handwashing at key times” outcomes (Bowen et al., 2013).					CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 cluster RCTs

Detection bias (Bowen 2013 and Luby 2009) and attrition bias (Luby 2009)

OPEN IN VIEWER

Table 35

Assessment of the certainty of evidence for latrine use (adherence) (unpooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Latrine use (adherence) (Caruso 2014)
1	1 randomised trial a	serious b	serious	not serious	serious c	none	The school-based handwashing educational approach in this study resulted in statistically significantly increased latrine use (MD 1.80, 95% CI [0.81, 2.79]), however, surprisingly, when the same intervention was combined with a latrine cleaning element, a significant decrease in latrine use was measured (MD -1.00, 95% CI [-1.97, -0.03]) (Caruso et al., 2014).					CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 quasi-RCT

Attrition/reporting bias

Large variability in results

OPEN IN VIEWER

Table 36

Assessment of the certainty of evidence for safe faeces disposal (adherence) (unpooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Safe faeces disposal practices (adherence) (Yeager 2002)
1	1 randomised trial a	serious b	not serious	not serious	serious c	none	103/323 (31.9%)	72/323 (22.3%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 quasi-RCT

Reporting/attrition bias

Low number of events

OPEN IN VIEWER

Table 37

Assessment of the certainty of evidence for open defecation (uptake) (unpooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Open defecation practices (uptake) (Lansdown 2002 and Wang 2013)
2	2 randomised trials a	serious b	not serious	not serious	serious c	none	56/71 (78.9%)	101/115 (87.8%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 RCT (Lansdown 2002) and 1 cluster RCT (Wang 2013)

Reporting and detection bias (Lansdown 2002 and Wang 2013)

Low number of events

OPEN IN VIEWER

Table 38

Assessment of the certainty of evidence for open defecation (adherence) (unpooled data), sanitation and hygiene messaging vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Sanitation and hygiene messaging	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Open defecation practices (adherence) (Stanton 1987 and Wang 2013)
2	2 randomised trials a	serious b	not serious	not serious	serious c	none	116/125 (92.8%)	67/76 (88.2%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

2 cluster RCTs

Attrition bias (Stanton 1987) and reporting/detection bias (Wang 2013)

Low number of events

4.3.1.5 Elements of psychosocial theory

The 4 studies that we included in this category all described theoretical elements or a formal psychosocial theory as the basis of the intervention. One study used the Theory of Planned Behaviour (Langford et al., 2013) and one study the RANAS model (Tumwebaze & Mosler, 2015). Biran et al. (2014) describes the SuperAmma approach, based on emotional drivers of behaviour, and Luby et al. (2010) describes an approach based on the stages of change theory. It should be noted that all these studies were conducted at small scale, and that elements of psychosocial theory should be incorporated in a larger promotional approach for a programme at scale. All studies implemented a handwashing-only intervention (see Figure 6).

Below we narratively describe the findings for the different outcome types. •

Behaviour change: handwashing (Analysis 33). Two different experimental studies describing interventions based on elements of psychosocial theory, measured handwashing at key times during the study period (“uptake”) (Langford & Panter-Brick, 2013; Luby et al., 2010). The study by Luby et al. (2010) had two different intervention arms, one with a theory-based intervention with soap, and one with a theory-based intervention with hand sanitizer. A significant effect on handwashing at different key times could be shown for 7 of the 9 outcomes (excluding the programme with hand sanitizer) (Langford & Panter-Brick, 2013; Luby et al., 2010). For the hand sanitizer intervention, a significant effect for handwashing in 3 out of 10 key times was shown (Luby et al., 2010). The certainty of evidence for the uptake outcomes was found to be low (Table 39). In one experimental study, with a low risk of bias, adherence outcomes were measured (Biran et al., 2014). Handwashing at key times was significantly improved, both at 6 weeks (MD 15.00, 95% CI [10.71, 19.29]) and 6 months (MD 31.00, 95% CI [29.45, 32.55]). For the adherence outcomes, the certainty of evidence was moderate (Table 40).

•

Behavioural factors (Analysis 34). One experimental study with moderate risk of bias measured knowledge, skills and attitudes (Tumwebaze & Mosler, 2015). An effect on knowledge about disease severity (MD 0.09, 95% CI [-0.06, 0.24]) and knowledge about disease vulnerability (MD 0.02, 95% CI [-0.05, 0.09]) could not be demonstrated. An additional public commitment element in the promotional approach also did not result in any significantly improved outcomes. An intervention based on elements of psychosocial theory improved skills in cooperation confidence in both treatment arms (MD 0.44, 95% CI [0.06, 0.82]; MD 0.42, 95% CI [0.06, 0.78]), but improved skills in cleaning ease (confidence in the ability to participate in cleaning a shared sanitation facility) and using a cleaning roster (planning showing who is responsible for cleaning at a certain time point) could not be demonstrated. Finally, no differences in attitudes regarding time cost, cleaning affect and cleaning effort could be shown in any of the treatment arms (Tumwebaze & Mosler, 2015).

•

Health outcomes. Langford et al. (2013) measured morbidity outcomes. The intervention based on elements of psychosocial theory significantly reduced the “median days of diarrhoea” from 16.3 to 9.7 (intervention vs controls, n=88, p=0.023).

•

Influence of incentives in programs. From the4 studies describing elements of psychosocial theory, only one study described the use of incentives (see Table 3), which was the provision of soap bars to the programme recipients (Langford & Panter-Brick, 2013). This study found a significant increase in handwashing at the short term, however absolute effects were similar as with the studies not using incentives.

OPEN IN VIEWER

Figure 33

Analysis

OPEN IN VIEWER

Figure 34

Analysis

OPEN IN VIEWER

Table 39

Assessment of the certainty of evidence for handwashing at key times (uptake) (unpooled data), elements of psychosocial theory vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Theory-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (uptake) (Langford 2013 and Luby 2010)
2	2 randomised trials a	very serious b	not serious	not serious	not serious	none	743/3422 (21.7%)	144/2884 (5.0%)	not pooled	not pooled		CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 quasi-RCT (Langford 2013) and 1 cluster RCT (Luby 2010)

Attrition/reporting bias (Langford 2013 and Luby 2010) and detection and other bias (Langford 2013)

OPEN IN VIEWER

Table 40

Assessment of the certainty of evidence for handwashing at key times (adherence) (unpooled data), elements of psychosocial theory vs no promotional approach

Quality assessment							N№ of patients		Effect		Quality	Importance
N№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	Theory-based approach	no promotional approach	Relative (95% CI)	Absolute (95% CI)
Handwashing at key times (adherence) (Biran 2009)
1	1 randomised trials a	serious b	not serious	not serious	not serious	none	Handwashing at key times was significantly improved, both at 6 weeks (MD 15.00, 95% CI [10.71, 19.29]) and 6 months (MD 31.00, 95% CI [29.45, 32.55]).					CRITICAL

CI: Confidence interval; RR: Risk ratio; quality of the evidence ranges from very low quality (⨁◯◯◯), low quality (⨁⨁◯◯), moderate quality (⨁⨁⨁◯) to high quality (⨁⨁⨁⨁).

1 cluster RCT

Reporting bias

4.3.2 Comparison of different promotional approaches

In 7 studies, certain promotional approaches were compared with one another. In this way, the effect of specific additional elements to a promotional approach could be studied. We discuss the different comparisons below (Contzen et al., 2015a/2015b; Dickey et al., 2015;Graves et al., 2011; Guiteras et al., 2015a; Lhakhang et al., 2015;Whaley & Webster, 2011; Zhang et al., 2013).

An overview of the findings on studies comparing different communication strategies is given in Table 41 and described in detail below.

4.3.2.1 Hygiene messaging and elements of psychosocial theory versus hygiene messaging alone

In Contzen et al.(2015a/2015b) three intervention arms were compared (Analysis 35). A health education approach (hygiene messaging) based on psychosocial theories (elements of infrastructure promotion, public commitment, reminders) was compared with health education (hygiene messaging) alone, and only handwashing was included in the intervention. In one intervention arm, education was combined with infrastructure promotion and reminder, in another intervention arm, education was combined with a focus on public commitment and reminder, and in a third arm, both elements were included. These 3 intervention arms were compared with a control arm, consisting of health education alone. A statistically significant increase of stool-related and food-related handwashing were shown in case of using the infrastructure promotion (stool-related: MD 0.20, 95% CI [0.04, 0.36]; food-related: MD 0.21, 95%CI [0.06, 0.36]) or the combined infrastructure promotion and public commitment (stool-related: MD 0.27, 95% CI [0.11, 0.43]; food-related: MD 0.32, 95% CI [0.17, 0.47]) interventions, however in case of a programme only using public commitment this could not be demonstrated (stool-related: MD 0.09, 95% CI -0.07, 0.25]; food-related: MD 0.08, 95% CI [-0.07, 0.23]).

OPEN IN VIEWER

Figure 35

Analysis

In addition, several behavioural factors were also measured in this study. A statistically significant correlation was shown between the educational approach together with infrastructure promotion, public commitment and reminder, and the following behavioural factors, regarding changes in food- and stool-related handwashing: descriptive norm (correlation coefficient food-related handwashing: 0.87; stool-related handwashing: 1.05), injunctive norm (correlation coefficient food-related handwashing: 0.65; stool-related handwashing: 0.60), commitment strength (correlation coefficient food-related handwashing: 0.53), forgetting (correlation coefficient food-related handwashing: -0.66; stool-related handwashing: -0.66), motivational self-efficacy (belief in ability to initiate and execute the behaviour) (correlation coefficient food-related handwashing: 0.47; stool-related handwashing: 0.54), volitional self-efficacy (belief in ability to maintain the behaviour) (correlation coefficient food-related handwashing: 0.44; stool-related handwashing: 0.44) and impediments (anticipated barriers and distractions to a behaviour) (correlation coefficient food-related handwashing: -0.49; stool-related handwashing: -0.49). For the educational intervention with infrastructure promotion, a significant correlation was found for most of the behavioural factors, while for the educational intervention with public commitment, significant correlations could only be found for less than half of the factors studied.

4.3.3 Effect of different communication strategies

An overview of the findings on studies comparing different communication strategies is given in Table 42 and described in detail below.

4.3.3.1 Mass media and interpersonal communication versus mass media alone

Only in one experimental study, with a moderate risk of bias, two types of communication strategies were compared (Chase & Do, 2012). The programme in the study focused on handwashing and was based on psychosocial theory (based on the FOAM framework), and a combination of mass media and interpersonal communication activities was compared with mass media alone.

The additional component of interpersonal communication resulted in a statistically significant increase in handwashing, less than 12 months after the programme period (“adherence”) (MD 0.01, 95% CI [0.01, 0.01]) (Analysis 38). In addition, an increase in handwashing at different key times (after faecal contact, before food preparation, before (breast) feeding a child, when hands look or feel dirty) was measured. An increase in “handwashing while doing laundry” could not be demonstrated, and, surprisingly, a significant decrease in “handwashing before eating” was measured (Chase & Do, 2012) (Analysis 39). Finally, a significant decrease in diarrhoea (MD -0.02, 95% CI [-0.02, -0.02]), and acute respiratory tract infection (MD -0.04, 95% CI [-0.05, -0.04]) was shown when using additional interpersonal communication activities (Chase & Do, 2012) (Analysis 40).

OPEN IN VIEWER

Figure 36

Analysis

OPEN IN VIEWER

Figure 37

Analysis

OPEN IN VIEWER

Figure 38

Analysis

OPEN IN VIEWER

Figure 39

Analysis

OPEN IN VIEWER

Figure 40

Analysis

OPEN IN VIEWER

Figure 41

Analysis

OPEN IN VIEWER

Figure 42

Analysis

OPEN IN VIEWER

Figure 43

Analysis

5.1.1 Results of the search

The identification of qualitative studies was performed in parallel with the identification of quantitative studies, since the same search strategy was used. Therefore, full text screening of 400 records, as described in 4.1.1, also resulted in a number of qualitative studies. We finally identified 28 qualitative studies, of which 24 were found through database searching (19 qualitative studies and 5 mixed-methods studies) and 4 from the grey literature. In addition, 5 mixed-methods studies were identified, as described above. The study selection flowchart is depicted in Figure 3 (see 4.1.1).

5.1.2 Included studies (n=28)

An overview of the characteristics of the included qualitative studies can be found in Table 43. The majority of the studies (n=19, 68%) was published in the last 5 years, with only 9 studies published between 2002 and 2011. •

Countries (see Figure 12)

Most of the studies (n=15, 53%) were performed in Sub-Saharan Africa (Kenya (n=3), Tanzania (n=3), Zimbabwe (n=2), Nigeria (n=1), Ethiopia (n=1), Malawi (n=1), Uganda (n=1), Zambia (n=1), Somalia (n=1) and South Africa (n=1)). Ten studies (36%) were performed in Asia:7 studies in South Asia (Bangladesh (n=3), India (n=3) and Nepal (n=1) and 3 studies in South-East Asia (Vietnam (n=2) and Cambodia (n=1)). Only 4 studies (11%) were conducted in Latin America and the Caribbean (El Salvador (n=1), Haiti (n=1) and Peru (n=1)).

Considering country income at the time the studies were performed, 19 studies (68%) were conducted in low-income countries (Bangladesh, Cambodia, Ethiopia, Haïti, Kenya, Malawi, Nepal, Somalia, Tanzania, Uganda, Vietnam (until 2008) and Zimbabwe) and 9 studies (34%) in lower middle-income countries (El Salvador, India, Nigeria, Peru, South Africa, Vietnam (from 2009) and Zambia). •

Setting and target level

OPEN IN VIEWER

Figure 12

World map indicating in which countries the included qualitative studies were performed

Most (68%) of the studies were executed in a rural setting (n=19), 3 studies (11%) were performed in an urban setting, 2 studies (7%) were executed in both a rural and urban setting and 3 studies (11%) were performed in an informal-rural setting (i.e. slums, settlements). One study (3%) did not provide any information about the setting in which the study was conducted. The intervention was targeted at a community level in 22 studies (12 on a community level, 4 on a (sub-) district level, 2 on a household level, 2 on a village level, 1 on a compound level and 1 on a county level) and at a school level in 3 studies. Two studies investigated interventions on both a school level and a community level (n=1) or village level (n=1). One study did not provide any information about the target level in which the study was conducted. •

WASH components

The following (combination of) WASH components were present in the interventions: WASH (general) in 11 studies, sanitation only in 9 studies, handwashing only in 4 studies, handwashing/sanitation in 1 study, handwashing/sanitation/water supply in 1 study, and handwashing/sanitation/hygiene/water quality in 1 study. •

Promotional approach

We classified the promotional approaches in 4 main groups according to the same criteria used for the quantitative studies (see 4.1.2: promotional approach). The approach in 18 studies (64%) was considered as a community-based approach, a social marketing approach in 2 studies, sanitation and hygiene messaging in 5 studies, and the intervention was based on elements of psychosocial theory in 3 studies.Table 44 shows which studies were grouped under each category, and Figure 13 in addition also provides the WASH component of each study.

OPEN IN VIEWER

Figure 13

Main categories of promotional approaches with detailed indication of WASH component and specific promotional approach for each included qualitative study

5.1.3 Excluded studies

Since study selection was performed in parallel for both the quantitative and qualitative studies, the main reason for exclusion of papers is described for both study types in paragraph 4.1.3. Detailed information can be found in Appendix 9 (List of excluded database studies) and 10 (List of excluded grey literature studies), and the reference list of excluded studies.

5.3.1 Process evaluation factors

Barriers/facilitators related to almost all (7/9) pre-identified process evaluation themes (acceptability, dose, engagement, fidelity, reach and satisfaction) were extracted from the qualitative studies. No information was available for 2 factors: recruitment and attrition. An overview of all barriers and facilitators identified can be found in Table 45 and Appendix 11. •

Acceptability

Acceptability refers to the quality or state of meeting one's needs adequately. Evidence from 3 studies identified recipients not willing to change their habits (Andrade, 2013), the mind-set of communities to demand free or subsidized materials (Malebo et al., 2012), and the possible safety risk of activities for children on the street (Rajaraman et al., 2014) as potential barriers. Household interviewees from 1 Indian study about a rural handwashing with soap programme (Rajaraman et al., 2014) indicated an intervention team being polite and entertaining and cooperation of the intervention team with the villagers as positive factors (facilitators) for making the handwashing programme more acceptable. •

Dose

Dose refers to the content, frequency, duration and coverage of the programme. Several of the included studies identified the following barriers related to dose of the programme: the messages are too long (O'Donnell, 2015 and Rajaraman et al., 2014), short programme duration (Bruck & Dinku, 2008), a lack of follow-up by the implementers (Malebo et al., 2012 and Whaley & Webster, 2011) or giving recipients only verbal information. Interventions of longer duration (Xuan et al., 2013), relevant messages (Andrade, 2013), frequent and external visits by the implementers or health promoters (Langford & Panter-Brick, 2013;Andrade, 2013; Whaley & Webster, 2011) and a broad/detailed (Whaley & Webster, 2011), step-wise approach (Andrade, 2013) were considered as potential facilitators. •

Engagement

Engagement refers to the subjective attributes that define the recipient's participation in interaction with or receptivity to an intervention. It also refers to the subjective attributes of programme staff that can influence their capacity to deliver intervention strategies (Cargo et al., 2015). The following barriers at the level of the implementer and related to recipient engagement were found: lack of follow-up by the implementers (Whaley & Webster, 2011), lack of communication (Emerging Markets Consulting, 2014), overlap with other programmes (Lawrence et al., 2016), the personal career of the implementer (Hueso & Bell, 2013), and lack of enthusiasm from outside experts (Lansdown et al., 2002). We also identified barriers at the level of the recipient: lack of interest from the recipients (Xuan et al., 2013), and not willing to give up unhealthy habits (Akter & Ali, 2014).

People from the interviews or focus group discussions also indicated several positive factors (facilitators) at the level of the implementer, including enthusiasm of the members of the Village Development Committees (Sarker & Panday, 2007 and Smith et al., 2004) and leadership of the implementer (Pardeshi, 2009). In addition, the following facilitators at the level of the recipient were found: income generating activities for participants of the health club (Whaley & Webster, 2011), and the praise and recognition of having a pretty home (Andrade, 2013). •

Fidelity

Fidelity reflects the extent to which an intervention is implemented as originally intended by programme developers (Cargo et al., 2015). One school-based study conducted in India suggested that school closures can act as a barrier to the fidelity of the programme (Rajaraman et al., 2014). •

Reach

Reach refers to the degree to which the intended audience participates in an intervention by ‘their presence’ (Cargo et al., 2015). In at least one study with a primarily social marketing approach, the small scale of the intervention was linked to not reaching the population of interest by stating that “the organization is not interested in offering individual sanitation loans because they are too small and will not reach very poor populations…” (Emerging Markets Consulting, 2014). On the other hand, the intention (e.g. intention to read a leaflet at home, Yeager et al., 2002) and motivation of people targeted by the promotional approach (e.g. motivation to adopt sanitation technology, Malebo et al. 2012) may act as facilitators in reaching a substantial amount of people when implementing sanitation and handwashing promotion programmes. •

Satisfaction

Satisfaction refers to the fulfilment of a need or want. Several of the included studies contained potential barriers related to the satisfaction of the recipients/implementers. The following barriers at the level of the implementer and related to the satisfaction of the recipient were found: a lack of interaction between recipient and trainer when using passive teaching methods (Xuan et al., 2013), a lack of collaboration with experts (Rheinländer et al., 2012 and Whaley & Webster, 2011), lack of training of the implementer (Hueso & Bell, 2013 and Rheinländer et al., 2012), lack of communication by the implementer (Whaley & Webster, 2011) and inappropriate attitude of the implementer (e.g. the manner and language towards villagers was not appropriate) (Emerging Markets Consulting, 2014). Other barriers related to recipient satisfaction were a lack of privacy (e.g. during open defecation) (Akter & Ali, 2014), cost of the hardware (e.g. water) (Kiwanuka et al., 2015), and political strategies (e.g. priorities for borehole locations during political campaigns because politicians want votes) (Kiwanuka et al., 2015). In case of social marketing approaches and use of a loan system, the loan repayment method (e.g. high interest rates) and slow loan processing times were found to be barriers (Emerging Markets Consulting, 2014). Barriers related to the satisfaction of the implementer were: criticism by authorities (e.g. for not achieving improved sanitation despite the effort) (Rheinländer et al., 2012), and frustration about not achieving enough results (i.e. no effective programme) (Rheinländer et al., 2012).

Other evidence identified 9 potential facilitators to keep recipients/implementers satisfied: interactive teaching methods and dialogue between villagers and trainers (Xuan et al., 2013 and Yeager et al., 2002), confidence in the health promoter's competence, training and ability to make change. (Andrade, 2013 and Malebo et al., 2012), innovative training materials (i.e. soap opera style of the video), full participation to the programme (Emerging Markets Consulting, 2014), collateral benefit of a WASH loan/fund (i.e. a contribution toward loan repayment and funeral expenses on the death of any member of the client's household) (Emerging Markets Consulting, 2014), respect toward and the proudness of the recipient (Andrade, 2013).

5.3.2 Programme environment factors

Barriers/facilitators related to all programme environment themes (training materials, funding/resources, intent of a programme to change a specific outcome, providing leadership to the implementing organization and partnerships) were extracted from the qualitative evidence. One additional theme was developed after coding the primary evidence/author statements: community capacity. An overview of all barriers and facilitators identified can be found in Table 46 and Appendix 12. •

Training materials

Evidence from five studies identified the following potential barriers related to training materials: safety risk (e.g. risk of stealing education materials) (Lansdown et al., 2002), limited availability of marketing materials (Emerging Markets Consulting, 2014), lack of detailed instruction guides (Brooks et al., 2015), or cost price (of a latrine). Another barrier was cultural insensitivity, e.g. the use of bodnas, which are traditionally used for anal cleansing after defecation, as handwashing station in both urban and rural (Hulland et al., 2013).

Two studies with a major community-based component and 1 study promoting water and sanitation via educational messaging identified sufficient availability (Graves et al., 2013 and Lawrence et al., 2016) and distribution of the training materials (Jimenez et al., 2014) as potential facilitators. •

Community capacity

Several of the included studies identified the following barriers: knowledge dissemination by children to their parents, which was perceived as improper (Lansdown et al., 2002), the lack of accountability of WASH Committees (Bruck & Dinku, 2008), the lack of support in constructing latrines (Bruck & Dinku, 2008), the lack of involvement of the Education Office (Bruck & Dinku, 2008) or village and ward leaders (Jimenez et al., 2014), insufficient capacity building (e.g. village leaders receiving little training on sanitation software) (Hueso& Bell, 2013; Silali & Njambi, 2014), the lack of sense of ownership (e.g. community owners are only called to implement projects, and are not involved in the development of the project) (Silali & Njambi, 2014; Schouten & Methenge, 2010) and the involvement of government-dominated stakeholders (Rheinländer et al., 2012).

In two school-based programmes focusing on sanitation (Lansdown et al., 2002) or sanitation, handwashing and water supply (Graves et al., 2013), teachers and mothers indicated that knowledge dissemination by children toward the parents could also be considered as proper. In line with this evidence, one study revealed that there was a multiplier effect from parents to children and that this led an improved connection (Langford & Panter-Brick, 2013). During a community-based handwashing programme conducted in El Salvador, individuals identified instrumental support of health promoters, the promoter's dedication to the hygiene and well-being of the community, and guiding/educating people of the community, as potential facilitators (Andrade, 2013). During the Total Sanitation Campaign in India, sanitation key informants indicated that capacity building and village leadership had a positive influence on community connectivity (Hueso & Bell, 2013). Indeed, community leadership and the use of programme leaders were also considered as potential facilitators in 2 other community-based WASH programmes conducted in Sub-Saharan Africa (Katsi, 2008; Smith et al., 2004). Evidence from four different community-based studies found that sense of ownership by the community members may serve as a positive driver to improve community capacity (Kiwanuka et al., 2015; Sarker & Panday, 2007; Schouten & Mathenge, 2010; Jimenez et al., 2014). A final beneficial factor to increase community capacity was creating financial self-management capacity, which is the practice of sharing resources among community members to enhance the integration and solidarity in the village (Sarker& Panday, 2007). •

Funding/resources

The most frequent reported barrier, identified in different community-based approaches (such as the MTUMBA approach in Tanzania, RUWASA in Uganda, and CLTS in Zambia) and social marketing programmes, was the limited financial, technological or facilitation capacity. An example of this is the lack of construction materials, expensive loans, insufficient programme funding, increased governmental charge, or inadequate budget allocation (Bruck & Dinku, 2008; Jimenez et al., 2014; Katsi, 2008; Lawrence et al., 2016; Malebo et al., 2012;Emerging Markets Consulting, 2014; Whaley & Webster, 2011, Kiwanuka et al., 2015; Schouten & Mathenge, 2010). During 2 community-based sanitation programmes performed in Tanzania (Jimenez et al., 2014) and India (Hueso & Bell, 2013) specific payment modalities (e.g. upfront payments from clients) also served as potential barriers to the recipient's resources. Finally, during a social marketing-based WASH programme implemented by WaterSHED in Cambodia (Emerging Markets Consulting, 2014), late payments by the implementer to the sanitation teachers was indicated as a barrier.

From interviews and focus group discussions conducted during the CLTS approach in Tanzania, it was noted that affordable technology was raised as a potential facilitating factor. Evidence from other community-based programmes conducted in Bangladesh, Kenya and Zambia suggested other facilitators such as the financial assistance of the Bangladesh Rural Advancement Committee (BRAC) (Akter & Ali, 2014), fundraising/income-generating activities by the community members (e.g. membership fee, collection of seasonal crops and indirect support of partner NGOs) (Sarker & Panday, 2007), reasonable payment modalities (e.g. monthly charges) and the use of local/traditional building materials (Lawrence et al., 2016). •

Intent of a programme to change a specific outcome

Community Health Club facilitators indicated that changing their mentality may serve as a positive driver to behaviour change of the community (Brooks et al., 2015) •

Providing leadership to the implementing organization

During the Total Sanitation Campaign in India, the decision-making process of government officers and engineers was seen as a barrier because they neglected sanitation in favour of more stimulating and costly water projects (Hueso and Bell, 2013). Stakeholders that were interviewed during the SANIVAT project (“Water supply, sanitation, hygiene promotion and health in Vietnam) also indicated that a lack of collegial support or supervision by experts may play a negative role (Rheinländer et al., 2012). During another community-based programme in South Africa, household heads said that open discussion promoted the credibility of each leader (Smith et al., 2004). •

Partnership, coordination between providers of the same intervention or other health interventions

Several community-based studies implemented in different continents (Sub-Saharan Africa, The Caribbean and South-East Asia) criticized the lack of partnerships ranging from the lack of partnerships between members of Community Health Clubs (Brooks et al., 2015), the lack of partnerships with the government/NGO (Brooks et al., 2015), the lack of partnership with the private sector (Bruck & Dinku, 2008) to the lack of inter-sectoral collaboration (Rheinländer et al., 2012). Evidence from 2 community-based and 1 social marketing study suggested that a lack of coordination (Bruck & Dinku, 2008; Malebo et al., 2012), information (Malebo et al., 2012), communication (Malebo et al., 2012; Emerging Markets Consulting, 2014), or involvement (of the loan officers) (Emerging Markets Consulting, 2014) may hinder well-constructed partnerships. Households during the MTUMBA approach raised the lack of quality and skills of the partners as a major limitation to get a successful programme (Malebo et al., 2012). During the SANIVAT project in Vietnam, different stakeholders complained about the lack of responsibility by both the implementers and the recipients (Rheinländer et al., 2012). Finally, evidence from 3 community-based WASH programmes and 2 social marketing-based WASH programmes indicated that coordination (with health offices) (Bruck & Dinku, 2008), decentralized systems (Hueso & Bell, 2013) and partnerships with government and/or NGOs (Kiwanuka et al., 2015; Emerging Markets Consulting, 2014; Whaley & Webster, 2011) would be beneficial factors for durable partnerships. •

Training/qualification of the implementers

Evidence from 1 educational promotional programme and 1 community-based sanitation/water supply intervention, both conducted in Sub-Saharan Africa, suggested a lack of financial resources as a barrier to train implementers appropriately.

5.3.3 Implementer-related factors

In our initial ToC, we only defined recipient-related factors in addition to the programme environment factors and process evaluation factors. However, in community-based approaches the recipients are typically involved as (secondary) implementer, called for example a health promoter or community leader. However, at the same time they are also recipient of the approach. We therefore created a separate category “implementer-related factors”, containing the same factors as were predefined for the recipients. Barriers/facilitators related to most (4/6) pre-identified factors were extracted from the qualitative studies. No information was available for 2 factors: self-efficacy and awareness about personal risk. An overview of all barriers and facilitators identified can be found in Table 47 and Appendix 13. •

Awareness about costs and benefits

For this factor, we only identified evidence from a study describing a social marketing approach and making use of a loan system (Emerging Markets Consulting, 2014). The availability and sustainability of sanitation loans was found to be a facilitator for programme implementation (Emerging Markets Consulting, 2014). Prices of the latrine business (delivering latrines) that not seemed to be competitive with prices of latrines supplied in the market, was found as a barrier for the awareness about cost and benefits, and consequently programme implementation (Emerging Markets Consulting, 2014). •

Motivation

Motivation was a newly identified theme, compared to our initial ToC. A factor negatively influencing the motivation of sanitation teachers was late payment of their salary, since they earn an income from selling latrines on commission (Emerging Markets Consulting, 2014). A facilitator for motivation was the feeling of responsibility of community health educators (Smith et al., 2004). •

Planning skills

Time constraints were found to be a barrier for the planning skills of the implementer, and thus for programme implementation. This was found in 3 studies with a community-based, education and social marketing approach, respectively. Time constraints were present at different levels, from teachers not making time to visit parents (Lansdown et al., 2002) to pressure to present positive results (Hueso & Bell., 2013), and workload and time in promoting sanitation loans (Emerging Markets Consulting, 2014). In addition, having other priorities (Yeager et al., 2002) and the bureaucratic loan application process (Emerging Markets Consulting, 2014) were barriers for timely planning by the implementer. •