Study design
Between January and September 2018, we conducted baseline and post-intervention assessments of a stratified cluster-randomised trial designed to test Sugira Muryango’s effects on promoting ECD and preventing violence among families receiving VUP. The trial was conducted within the Nyanza, Ngoma, and Rubavu districts with existing VUP programmes, selected to minimize the overlap with ECD interventions by government or nongovernmental organizations. All families were eligible for one of two versions of the VUP programme: classic public works (cPW), which provides cash for (typically hard) manual labour; or the newer expanded public works (ePW), which provides cash for (typically lighter) labour and also provides access to livestock. All procedures were approved by the Harvard T. H. Chan School of Public Health and Boston College Institutional Review Boards as well as the Rwandan National Ethics Committee, Ministry of Education, and National Committee for Science and Technology.
Participants
Within selected clusters (described in Randomisation, below), families were eligible for inclusion in the study if they 1) belonged to the most extreme level of poverty in the government’s household-ranking system (Ubudehe 1)13 and were eligible for the cPW or ePW programme; 2) had at least one child 6–36 months; and 3) were willing to participate in a home-visiting intervention. The focus of the programme on children aged 6–36 months at enrollment was based on the knowledge that experiences, both social and biological, are particularly important during the first few years of development.3 Furthermore, Sugira Muryango is designed to aid child development to the point where government care becomes available at age 5; at 12-month follow up in the present study the oldest children will be reaching roughly 4.5 years old. Exclusion criteria were a severe, active crisis in the family such as psychosis or suicide attempts by a caregiver or severe mental impairment in the caregiver which may have affected the ability to benefit from the programme. All caregivers gave written informed consent for themselves and their eligible children.
The caregiver who stated that he or she knew the child best—typically the mother—provided reports on child development and health, the home environment, caregiver-child relationships, caregiving practices, feeding practices, child health, as well as information about the household, including family composition and assets. All primary male and female caregivers provided self-reports on mental health and victimization and perpetration of intimate partner violence. Interviews, child assessments, and mother-child observation were conducted in Kinyarwanda in the family’s home. Data were entered on Android tablets by independent local enumerators blinded to intervention status.
The intervention
Sugira Muryango comprises twelve modules (see Table 2) that were delivered by trained, supervised CBCs in the families’ homes, unless contraindicated due to illness or privacy concerns, at a pace of about one module per week (average 90-minute sessions) between May and August 2018. Sugira Muryango offers active coaching of caregivers to promote early stimulation, play, nutrition, hygiene, responsive parenting, nonviolent interactions among household members, and engagement of both female and male caregivers. CBCs also help families navigate formal and nonformal resources (e.g. health and nutrition services and social support). Sugira Muryango was originally developed and tested in Rwanda for families affected by HIV/AIDS.14 During previous pilot studies,15 a version focusing on ECD was developed by integrating United Nations International Children's Emergency Fund (UNICEF) and World Health Organization (WHO) Care for Child Development materials.16 The CBCs were selected from the local community (see Table 1 for selection criteria, training, supervision, and incentives of CBCs). Primary caregivers participated in the modules in interaction with their child(ren); other caregivers and older children were invited to participate. All visits included a 15-minute “active play and communication” session where caregivers received live feedback on parent-child interactions. The usual care (UC) group received VUP services and health services as usual from the Rwandan government and its partners. Intervention and UC families received a stipend (RWF 5,000 equivalent to 3 kilos of rice) after each data collection.
Outcomes
Per our theory of change (Figure 1), immediately following 12 modules of intervention delivered weekly over a 3–4 month period, the primary outcomes were change in parents’ behaviours towards the child including responsive care and play, dietary diversity, care seeking for child health problems, and family violence. Secondary outcomes were caregiver outcomes related to shared decision-making among parents and caregiver mental health, as well as household outcomes related to water, sanitation and hygiene. Questionnaires were developed and tested during pilot intervention research and followed a forward- and back-translation protocol from English to Kinyarwanda.14
With regard to child-level outcomes, responsive caregiving was assessed by trained enumerators using three tools, the Observation of Mother-Child Interaction (OMCI),17 an adapted 43-item version of the infant/toddler Home Observation for Measurement of the Environment (HOME) Inventory, which has previously been used in Uganda,18,19 and the Multiple Indicator Cluster Survey (MICS) Family Care Indicators (FCI).20 The OMCI assesses a five-minute mother-child interaction that is scored according to published guidelines (maximum 57; Cronbach’s α= 0.91). The HOME combines observation of parenting behaviours and household conditions with caregiver report. Items were summed to derive a total score (maximum 43; Cronbach’s α=0.76). The MICS FCI20 is a cumulative score of caregivers’ self-reported engagement in stimulating activities such as singing, reading, and counting with the child during the prior three days (maximum 6; Cronbach’s α=0.74). Children’s nutritional intake was assessed by parent-reported dietary diversity reflecting the number of seven food groups (grains, roots, and tubers; legumes and nuts; dairy products; meat, fish, poultry, and organ meats; eggs; vitamin A rich fruits and vegetables; other fruits and vegetables) the child consumed in the past 24 hours.21 Children’s health status was measured using standard Demographic and Health Surveys (DHS) questions reporting the prevalence of diarrhoea, fever, and cough in the seven days preceding the survey.22 Care seeking at a health facility was defined following DHS guidelines and was assessed only among parents of children who experienced illness.
Caregiver-level outcomes included intimate partner violence which was assessed among parents who were currently married, cohabitating, or in a relationship using items from DHS Domestic Violence Module23 covering emotional, physical, and sexual abuse within the last three months. Among households with a mother-father structure, we also assessed whether caregivers reported equal involvement in decision-making about care for the young child including decisions related to feeding and medical care for illness.24 Finally, caregivers’ mental health was assessed using the Hopkins Symptom Checklist-25 (HSCL-25), a measure of depression and anxiety (internalizing) symptoms validated for use among adults in Rwanda25 (α=0.93). A mean score ≥1.75 was used to define likely clinical mental health problem.
Household-level outcomes were related to hygiene practices were assessed using items from the DHS water, sanitation, and hygiene (WASH) module.22 Indicators included access to clean water, safe treatment of water, and hand-washing facility with soap.
Power calculation
Power calculations drew on previously conducted pilot studies15 and estimated the required sample size for a 0.18 minimum detectable standardized effect size on parenting outcomes and child development outcomes for the 3- and 12-month follow-up period assuming power of 0.8 and a standard alpha level of p<0.05. We used an estimated intraclass correlation of 0.03 for parent-child interactions based on pilot data. The ePW programme was being rolled out during the design phase of the programme, and we assumed based on estimates available to us it would be too scarce to constitute one half of an ideal sample size, so calculations were based on an assumption of 91 ePW clusters and 104 cPW clusters with five households per cluster to be assigned to treatment and control conditions. Because the target number of all ePW clusters did not exist, further adjustments were made, by adding combined clusters, and ultimately adding more cPW clusters to maintain power.
Randomisation
Families were enrolled between February and March 2018. Government staff in the three study districts provided lists of households participating in cPW or ePW. Families’ participation in VUP was determined by governmental policies and was not under the control of the researchers. Non-overlapping, geographically defined clusters were created comprising at least 30 families participating in the cPW programme or 10 families participating in the ePW programme, with some clusters containing both ≥30 cPW and ≥10 ePW households. Clusters were formed by the research team using detailed local maps by combining one or more contiguous villages such that one CBC could provide services to all participating families in the cluster. Villages within the same cluster were selected to be as close to each other and as far apart from other clusters as possible. Due to the relative scarcity of the ePW families, 100% of clusters containing at least 10 ePW families were sampled for participation in the study. Clusters which contained cPW families (including combined clusters containing ePW families) were randomly sampled for inclusion into our study until we reached our target sample size of ≥1,040 households. Randomisation was performed by a data collection contractor and occurred at the cluster level within strata defined by public works type (ePW only, combined ePW/cPW, and cPW only) and geographic sector. Within strata, clusters were assigned random numbers and placed on a ranked list. The first half of clusters on the randomly ranked list were assigned to treatment. In case of an odd number of clusters per strata, a lottery was used to round the number assigned to treatment up or down. After assignment of the cluster, households were contacted by the data collection contractor and invited to participate in the study. Clusters were retained if at least five families in the cPW strata or at least one eligible family in the ePW strata enrolled. We retained 48 ePW-only clusters, 38 ePW/cPW clusters, and 112 cPW-only clusters (Cluster sampling strategy, Figure 2). Neither the caregivers nor enumerators knew the family’s assignment status at the time of the baseline assessments. Enumerators were also not informed about the family’s assignment status during the post-intervention assessment although caregivers’ responses to fidelity questions about the programme following the session may have revealed their treatment status. In total, 1,049 households were enrolled at baseline. After the randomisation n=508 families were allocated to UC and n=541 families were allocated to treatment. Baseline data collection occurred in May 2018 and post-intervention data were collected in August-September 2018.
Statistical analysis
We compared trajectories of outcomes over time among families receiving the Sugira Muryango intervention with UC using linear mixed models for continuous outcomes and generalized linear mixed models with a logit link for binary outcomes. To account for clustering, we included random effects for randomisation cluster and child for outcomes assessed at the child level. For outcomes assessed at the caregiver level we included random effects for cluster and caregiver, and for outcomes assessed at the household level we included random effects for cluster and household. Following intention-to-treat analysis, we used chained equation imputations in STATA to account for missing data.26 We examined whether primary outcomes differed between families who received Sugira Muryango and UC families. We report coefficients for the time-by-treatment interaction term and standardized effect sizes (Cohen’s d for continuous outcomes, odds ratios (OR) for dichotomous outcomes) with 95% confidence intervals. Analyses were conducted using STATA version 15 (StataCorp, College Station, TX). Intraclass correlations can be found in Additional File 2 and auxiliary analyses examining whether a family’s enrolment in either ePW or cPW moderated intervention effects can be found in Additional File 3.
Adverse events in intervention and control households
During the interval between baseline and post-intervention “risk of harm”, cases were reported in 12 families (2.2%) in the intervention group and 12 families (2.4%) in the control group (see Additional File 1). These households were retained in the analyses under intention to treat.
Analytic sample and demographics of the samples
Baseline data were collected on 1,084 children, and 1,498 caregivers and their intimate partners were enrolled in the trial. Instances of loss-to-follow-up from baseline to the postintervention assessment was low (<2.5%). More specifically, three households (0.3%), 36 caregivers (2.4%), and six children (0.5%) did not complete the post-intervention assessment and had post-intervention data imputed. Item-level missing data at both baseline and post-intervention were similarly low (<1%). Descriptive statistics are provided in Table 3. Caregivers ranged in age 18–79 years and were most frequently the biological mother (n=950), the biological father (n=433), or a grandparent (n=96). Sixty-four percent (n=953) of the caregivers were married or cohabitating. At enrolment, 61% of the families reported high levels of food insecurity and 48% of the children were stunted as defined by a standardized height-for-age (HAZ) score below 2 in accordance with WHO growth standards.27