Stepped-wedge study design
Given the series of design revisions made, a review of the study’s sampling framework was re-visited. In general, the design revisions have very little impact on the study’s sampling framework. However, refinement and further clarity on the steps involved in performing the two stages of sampling process was considered vital at this stage in preparation for the start-up of the study.
In particular, the sampling framework review focused on the following key points:
- How the stage 1 sampling fits WFP’s request for sampling only in areas ‘covered’ by their distribution sites?
- How many quadrats / hexagonal grids will be used per cluster?
- How many villages / settlements per hexagonal grid will be sampled per cluster?
For the first point, we intend to bound the spatial sampling for stage 1 based on the extents of the area distribution of the programme sites. For some areas such as Kassala Urban, the programme sites cover almost the whole area hence the extent is almost similar to the extent of the borders of the locality itself. For other clusters, the programme sites are evenly distributed spatially hence the stage 1 sampling will have a lesser extent than the administrative boundaries of the locality. This addresses the concern of WFP because doing this type of sampling will focus the selection of sampling villages in areas covered by a programme site while still allowing for the selection of villages that may not be covered by the programme. This is important because we want to have a sample that can say something about the way programme sites are selected geographically.
For the second and third point, after deliberating on the expected number of under 5 children and PLW in a village, it has been agreed that we will select 20 quadrats per study cluster and then select 2 villages per quadrat. This will be a total of 40 sampling villages per study cluster. Assuming that we can get at best about 4 PLW per village (up to 8 in theory)1, this sampling frame will give us on average a total of 160 PLW per study cluster. This is a worst case scenario. This number will be enough to classify PLW indicators. At best case scenario of up to 8 PLW per village, we will get a sample of about 320 PLW per study cluster which would be enough to give an estimate of the indicators for PLW at useful precision. Overall, the 40 sampling villages will be more then enough to provide at least 192 under 5 children consistently.
We will test these assumptions on baseline or during the first data collection round and will calibrate the stage 1 sample accordingly based on the sample sizes we achieve.
For stage 2 sampling, we will do an exhaustive sample of all children under 5 and all PLW in each selected sampling village. This is to ensure that we get enough sample size for PLW and for the coverage metrics.
We have originally planned to recruit study participants into the incidence sub-study through the various health clinics and programme sites during routine growth monitoring sessions or during well-baby clinics. However, feedback gained from Kassala state consultations indicate that the health facility-based recruitment and follow-up for the incidence study will most likely provide a sample that is skewed towards the younger children (under 2) and very few age 4 to 5 years and some 3 years. This will be an uneven sample to begin with. Second issue is that there will most likely not be enough sample to complete the sample size requirement if only facility-based recruitment was utilised. This feedback echoes comments provided by the FMoH.
This is a well-founded feedback and something that requires a re-thinking of the approach to the recruitment and follow-up of participants to the incidence study. In light of this, the recruitment and follow-up process is to be adjusted as follows:
- For the study clusters for both intervention and control arms, villages within the catchment area of the programme sites will be selected. Number of villages to select will depend on the estimate of how many potential participants can be recruited in each vis-à-vis the total sample size needed each arm.
- Selected villages will then be visited and then all children age from 6 months to 4 years old who fit the study criteria will be recruited.
- A month after, the same villages will be visited and then the study participants will be followed-up. There will be 5 follow-up visits.
This obviously has cost implications particularly on transport when visiting villages. This is something we have to consider.