Background
Evidence of the impact of community-based nutrition programs is uncommon for two main reasons: the lack of untreated controls, and implementation does not account for the evaluation design. Suchana is a large-scale program to prevent malnutrition in children in Sylhet Division, Bangladesh by improving the livelihoods and nutrition knowledge of very poor households.
Methods
Suchana is being implemented in 157 unions, the smallest administrative unit of government, in two districts of Sylhet. Suchana will deliver a package of interventions to poor people in about 40 randomly selected new unions annually over four years, until all are covered. All beneficiaries will receive the normal government nutrition services. For evaluation purposes the last 40 unions will act as a control for the first 40 intervention unions. The remaining unions will receive the program but will not take part in the evaluation. A baseline survey was conducted in both intervention and control unions; it will be repeated after three years to estimate the impact on the prevalence of stunted children and other indicators.
Results
This stepped wedge design has several advantages for both implementation and evaluation, as well as some disadvantages. The units of delivery are randomized, which controls for other influences on outcomes; the program supports government service delivery systems, so it is replicable and scalable; and the program can be improved over time as lessons are learned. The main disadvantages are the difficulty of estimating the impact of each component of the program, and the geographical distribution of unions, which increases program delivery costs. Stepped implementation allows a cluster randomized trial to be achieved within a large-scale poverty alleviation program.
Conclusions
A stepped wedge design for implementation was used which is particularly suited to evaluations of service delivery. The design allows the intervention to be phased-in and scaled-up over a period of time. The phased or ‘stepped’ process of implementation enables a cluster randomized controlled trial to be achieved without having a permanently untreated control group. However, the design does not allow to assess the impact of each individual component of intervention.