Menu
Menu
Menu
Menu
Introduction In Kenya’s Kitui County, 46% of children under 5 years are stunted. Sanitation and nutrition programmes have sought to reduce child undernutrition, though they are typically implemented separately. We evaluate the effectiveness of an integrated sanitation and nutrition (SanNut) intervention in improving caregiver sanitation and nutrition knowledge and behaviours. Methods We conducted a cluster-randomised controlled trial to evaluate the impact of the SanNut intervention on caregiver knowledge, sanitary and hygiene practices, sanitation outcomes and nutrition outcomes. The evaluation included caregivers of children under 5 years across 604 villages in Kitui County. 309 treatment villages were randomly assigned to receive both the SanNut intervention and the standard Community-Led Total Sanitation (CLTS) intervention, while 295 control villages only received the CLTS intervention. 8 households with children under 5 years were randomly selected from each evaluation village to participate in the endline survey, for a total of 4322 households. Results SanNut led to modest improvements in sanitary knowledge and practices emphasised by the programme. Caregivers in treatment villages were 3.3 pp (+32%) more likely to mention lack of handwashing after handling child faeces as a potential cause of diarrhoea, and 4.9 pp (+7.8%) more likely to report safe disposal of child faeces than caregivers in control villages. Treatment households were 1.9 pp (+79%) more likely to have a stocked handwashing station and 2.9 pp ( ‘16%) less likely to report incidences of child diarrhoea. However, SanNut appears to have had no impact on nutritional practices, such as breastfeeding, vitamin A supplementation or deworming. Non-child outcomes traditionally associated with CLTS, including latrine use and homestead sanitary conditions, were similar in treatment and control groups. Conclusion Child-focused messaging can potentially be integrated into CLTS programming, though this integration was more successful for topics closer to CLTS objectives (sanitation practices, including limiting faecal contamination and handwashing) than for more disparate topics (nutritional practices). Trial registration Pan-African Clinical Trials Registry (PACTR201803003159346) and American Economic Association registry for randomised controlled trials (AEARCTR-0002019).
Like other CLTS programmes, the CLTS programme implemented in 2016 by the Kitui County Government, focused on helping communities achieve Open Defecation Free (ODF) status. Within each eligible (non-ODF) village, the programme kicked off with CLTS implementers visiting the village to set a date and location for subsequent activities and to map where open defecation occurs. Following this visit, facilitators led a 2–4 hour community-wide meeting (‘triggering’) to stimulate feelings of shame and disgust around sanitation conditions in the village. This meeting included typical CLTS exercises, such as walking through the village to observe where open defecation occurs and encouraging community members to construct and use latrines. Community health volunteers (CHVs) later visited households to reinforce the messages from the triggering event. The CHVs who implemented both CLTS and SanNut were recruited under the PATUMAIP program, which required CHVs to have primary-level education and be at least 30 years old. One CHV was recruited from each village and was responsible for implementing the community health strategy in their village. CHVs worked exclusively in their assigned village and were supervised by the Public Health Officers in their Ward. They were provided with a monthly stipend of Kshs. 3000 (~US$30) to facilitate their work but did not receive any incentive-based payments. There was no gender requirement for CHV recruitment;the CHV workforce consists of roughly similar numbers of men and women. All CHVs attended ward-level CLTS trainings in July 2016, and CHVs in Treatment villages attended an additional SanNut training in September 2016. Villages that successfully achieved ODF status held a community celebration. The SanNut programme was designed by the Kitui County Government and UNICEF to address recognised gaps in CLTS messaging. SanNut specifically extended the focus of CLTS to children, highlighting the consequences of poor sanitation and nutrition practices on children’s health outcomes and ultimately to the child’s long-term well-being. SanNut’s programming included the topics listed in table 1 that distinguished it from broader CLTS objectives: Community-Led Total Sanitation (CLTS) versus sanitation and nutrition (SanNut) topics Kitui County officials oversaw SanNut programme administration while UNICEF provided technical and financial support. Within randomly selected treatment villages, CHVs invited all caregivers of children under 5 years and pregnant women to participate in SanNut by attending meetings about toddler hygiene and nutrition. These invitations were extended through community leaders, announced in public forums, and in many cases, delivered door-to-door by CHVs. The 1-hour meetings were held at a caregiver’s home or at a public location within the village. The first SanNut caregiver meeting was held 2–3 weeks after CLTS triggering and focused primarily on toddler hygiene and sanitation. The key message from this meeting was ‘Keep faeces away from infants and infants away from faeces.’ The meeting included a discussion linking faecal ingestion to child malnutrition, disease and impaired cognitive development, discussions of how to safely dispose of child faeces and how to wash the child’s hand before feeding, and an interactive exercise where caregivers developed an action plan for keeping children in safe, hygienic environments. Facilitators referred to a session guide with key SanNut messages and used visuals throughout the meeting, including an ‘F-Diagram’ that depicts faecal–oral pathways, brain scan images that contrast a normal health child’s brain with a malnourished child’s brain and Maternal, Infant and Young Child Nutrition (MIYCN) counselling cards that reinforce proper sanitation and nutrition practices. The second SanNut caregiver meeting was held 1–3 weeks after the first caregiver meeting and focused primarily on healthy nutritional practices. Facilitators discussed the importance of breastmilk, how caregivers should exclusively breastfeed children below 6 months and how caregivers should complement breastfeeding with solid food consumption for children 6 months to 2 years. Facilitators also explained how children should be taken to health facilities for deworming treatment, vitamin A supplementation and when they are sick. Facilitators referred to a session guide and supporting materials such as MIYCN counselling cards that reinforced proper sanitation and nutrition practices throughout both meetings. In the months following the caregiver meetings, CHVs visited households in the community up to four times to reinforce the messages from the triggering event and caregiver meetings. These household visits occurred in both treatment and control villages, but the messaging in the visits differed depending on the treatment status. In control villages, CHVs discussed the topics covered in the CLTS triggering event and checked whether households had constructed a latrine and were using it and whether the household had a stocked handwashing station. In treatment villages, CHVs covered the same material as in control villages but added SanNut-specific messaging. This additional messaging included the topics from the caregiver meetings, especially the safe disposal of child faeces, keeping infants away from faecal material, appropriate breastfeeding practices and the importance of regular health facility visits. Figure 1 shows how SanNut activities fit into CLTS implementation in a typical village. The SanNut program was relatively low-cost since it took advantage of the infrastructure and personnel in the CLTS intervention. SanNut required US$34 per village for SanNut facilitators to travel to villages and deliver the two caregiver meetings. However, we do not have access to information on the cost of ward-level trainings for facilitators or any other costs associated with their involvement in SanNut. The CHV workforce was hired and funded under the PATUMAIP CLTS program, and since they performed SanNut activities as part of their routine household visits, they did not require additional funding. SanNut activities within CLTS implementation in treatment villages. CLTS, Community-Led Total Sanitation; ODF, open defecation free; SanNut, sanitation and nutrition. We designed a cluster RCT to measure the effects of the SanNut programme on caregiver practices and sanitation and nutrition outcomes. To construct the sampling frame, we applied several eligibility criteria to the 2100 villages slated to participate in the CLTS programme in Kitui County. First, we excluded three of eight subcounties that are mostly urban or periurban and thus retained the five rural subcounties that were more appropriate for the rural-focused CLTS programme. Second, we excluded one ward (the administrative unit below subcounty and above village) that was more than 90% ODF, since it had few villages participating in the CLTS programme. Third, we excluded six wards where Population Services Kenya, an NGO, was implementing a nutrition intervention with many similarities to the SanNut programme. Fourth, we excluded villages that were far from a health facility (more than 10 km) due to the logistical barriers posed to caregivers in taking children to health facilities; we reasoned that we could only measure the effect of the intervention on a household’s demand for health services if those services were accessible. These criteria left 724 villages in our sampling frame. We conducted power calculations to determine the minimum sample size necessary to detect treatment effects of 0.15 SD or larger, which is near the lower bound of effect sizes of successful sanitation and nutrition programs in our literature review. We used a conservative estimate of 0.2 for the correlation of sanitation and nutrition outcomes between households in the same village, which was the upper bound of intracluster correlations observed in similar studies.16 17 Assuming a sample of five households per village, based on village populations and age distributions from census data, we identified a minimum sample size of 520 villages. To account for potential treatment non-compliance or other factors that could reduce statistical power, we inflated our estimates by ~20%, resulting in a sample size of 627 villages. Since CLTS and SanNut programme implementation would be coordinated at the ward level, we stratified treatment assignment by ward to ensure that each ward had a similar number of treatment and control villages. Within each ward, we randomly assigned half of the villages to receive the SanNut programme (treatment) and half to receive the standard CLTS programme (control). Randomisation was implemented in Stata/IC V.14.0 and documented in .do files. We imported village lists, set the random number seed for reproducibility, generated a random number variable using the runiform() function, sorted the list by ward and the random number variable and assigned the first half of villages within wards to control and the second half to treatment. As such, CHVs, who worked exclusively in their assigned villages, were also effectively randomised to the treatment or control group. If a ward had an odd number of villages, then we assigned the remaining village to the treatment group. Since village lists within wards were randomly sorted, the ‘left-over village’ in wards with an odd number of villages was effectively randomly selected from the pool of evaluation villages so that, within wards, treatment status was orthogonal to village characteristics. We include ward fixed-effects in all analytical models to control for the slightly different probabilities of treatment in odd-number and even-number wards. During data collection, we found that 15 control villages and eight treatment villages did not exist due to errors in the administrative records, resulting in a final sample of 604 villages (309 treatment, 295 control). Villages and households remained balanced on pretreatment covariates: the p value from a joint test of orthogonality on the covariates listed in table 2 is 0.72. Balance check, comparison of means across treatment and control villages for household-level variables from the endline survey Although we expected the SanNut programme to fail to be implemented in a few treatment villages, in fact the opposite occurred: eight villages, or 3% of all villages assigned to control, were incorrectly treated by SanNut staff. (In one ward, Kitui South, the Public Health Officer who oversaw the implementation of SanNut was different from the one trained by the evaluation team on the distinction between treatment and control villages. As such, incorrect information about the list of villages to be targeted for SanNut was relayed and eight CHVs from control villages were trained and subsequently rolled out the SanNut program.) We use original treatment assignment in all analytical models and report intent-to-treat estimates, though the results do not change substantively if we use treatment-on-the-treated estimator. (See our online supplementary table A1 for treatment-on-the-treated estimates for each of the primary outcomes.) bmjgh-2018-000973supp001.pdf To select households within sampled villages we obtained household lists from CHVs. These lists were compiled in all villages in preparation for CLTS and included the number of children below 6 months, the number between 6 months and 2 years and the number between 2 years and 5 years in each household. We obtained these lists in treatment villages immediately prior to SanNut implementation so that CHVs could prepare attendance rolls for caregiver meetings. However, we only obtained household lists in control villages immediately prior to data collection, which was 3 months after obtaining the lists in treatment villages. To ensure that household samples were comparable in treatment and control villages, we ‘trimmed’ the lists to eliminate households that only had a child aged 0–3 months or 57–60 months (and thus would only show up in one group’s list but not the other). The households remaining in the trimmed lists were eligible for inclusion in the study both at the start and end of SanNut implementation (figure 2 shows randomisation and sampling for SanNut study). Randomisation and sampling for SanNut study. The SanNut intervention was targeted at all households with a child under 5 years, but households with children under 2 years were considered high priority by UNICEF since younger children are especially vulnerable to the effects of undernutrition. In order to detect treatment effects within this priority subgroup, we stratified each village list by whether a household had a child under 2 years or a child from 2–5 years (some households had both). To construct our household sample for the endline survey, we randomly selected up to five households in each village from the first sublist of households, ensuring that we had sufficient statistical power to detect effects among households with a child under 2 years. If a village had fewer than five households in this first sublist, all households in that sublist were sampled. After removing households from the second sublist that had already been sampled in the first sublist, we randomly sampled as many households from the second sublist as needed to get eight total households in the village, or all remaining households if the total was fewer than 8. In the final survey sample, the median village contained 7.1 sampled households: 2.1 households with only a child 0–2, 3.3 with only a child 2–5 and 1.8 households with both a child 0–2 and a child 2–5. 8.9% of eligible sampled households in control villages and 9.1% of eligible sampled households in treatment villages were unavailable for the interview and were replaced with other households randomly selected from the eligible pool when possible (Although this left us with only 3.9 households with children 0–2 instead of 5, we still had sufficient statistical power to detect effect sizes of 0.15 SD given the buffer and low rate of non-compliance). Prior to analysis, we calculated the probability that each eligible household would be selected in the final sample. In order to recover estimates of population average treatment effects for caregivers with children under 5, we weight each observation according to the inverse of the probability of being sampled in all regressions. We designed a household questionnaire to measure caregiver knowledge, household sanitary and hygiene practices, sanitation outcomes and nutrition outcomes 3 months after the final SanNut activities. Since SanNut was layered onto existing CLTS programming, we also collected data on standard CLTS indicators to assess whether the additional SanNut activities enhanced or detracted from non-child CLTS objectives. The questionnaire consisted of several modules, including a quiz of caregiver knowledge, a survey of caregiver hygiene and sanitation practices, a survey of diarrhoeal incidence of children in the household, enumerator observation of the household environment and enumerator review of children’s health booklets. In summary, we collected data on 15 sanitation and nutrition outcomes, listed in table 3. These 15 outcomes were selected after extensive consultation with UNICEF to determine what evidence was needed to inform their recommendations about scaling the SanNut program. Summary of results for all primary outcomes All regressions include the control variables listed in table 2, strata fixed effects, sampling weights equal to the inverse probability of selection and standard errors clustered at the village level. *q<0.10, ** q<0.05, *** q<0.01. We recruited, trained and managed local enumeration teams. Our teams conducted surveying from April to July 2017, 3 months after the conclusion of SanNut activities. We did not collect baseline data due to time and budgetary constraints. However, the evaluation was powered to detect sufficiently small effects given our sample size under the assumption of no baseline data. Enumerators were not aware of the treatment status of villages that they visited. When an enumeration team reached a village, they would locate sampled households by asking for the head of the household. Once the correct household was identified, enumerators identified the primary caregiver per the household list; in cases where the individual listed was not the primary caregiver, they substituted for the correct caregiver within the same household (this occurred in only 5% of the households surveyed). All survey modules were completed on tablets using SurveyCTO software. We obtained written or verbal informed consent from all study participants. We registered the SanNut evaluation and preanalysis plan on the American Economic Association’s RCT registry prior to data collection (study ID: AEARCTR-0002019 (https://www.socialscienceregistrysocialscienceregistry.org/trials/2019/history/19731), submitted on 21 February 2017). We also obtained ethical clearance from the Kenya Medical Research Institute (KEMRI) (study ID: Non-KEMRI #547, approved on 31 October 2016) and a research permit from the National Commission for Science, Technology and Innovation (NACOSTI) (study ID: NACOSTI/P/16/57638/12659, approved on 27 July 2016) to conduct the study. For each outcome, we estimate the following weighted least squares regression model: where Per our preanalysis plan, we analyse the impact of the SanNut programme on the 15 primary outcomes listed in table 3. (All regressions are conducted on the full eligible sample. See online supplementary table A2 for results specific to households with a child 0–2 years, a priority subgroup for UNICEF programming prespecified in our analysis plan.) To account for the multiplicity of hypotheses being tested and to reduce the likelihood of incorrectly rejecting null hypotheses, we control for the false discovery rate (FDR) according to the two-stage linear step-up procedure described in Benjamini et al.18 This procedure limits the rate of falsely rejecting null hypotheses to a desired level q. Rather than set an arbitrary level of q for all hypotheses, we follow the algorithm described in Anderson (2008) and perform the procedure for all possible levels of q (from 0 to 1 in increments of 0.0001) and record the smallest level q when each hypothesis is no longer rejected.19 Each estimate’s ‘sharpened q value’ can therefore be interpreted as the expected false discovery rate in the family of outcomes if we reject the null at that level. For convention, in the text, we also report unadjusted p values, but our main effects table and interpretations refer to the FDR-controlled q values. We perform this procedure across all 15 primary outcomes and separately within each of the four indices (sanitation knowledge, caregiver handwashing practices, nutrition knowledge and courtyard cleanliness), which we treat as exploratory analysis to investigate effects on individual components.
N/A
