Selection and randomisation of study sites
The evaluation is being conducted at 24 PHCs in South Africa. All study sites follow the current guidelines for HIV care and treatment, dated December 2014. Six clinics were chosen from one district each in Gauteng, KwaZulu-Natal, Limpopo and North West Provinces.
These provinces were chosen in consultation with NDOH to represent high HIV burden regions with high-burden districts and high-volume clinics. The study team developed a list of all sites in each participating province that met these criteria and selected three matched pairs of clinics per province. Pairs were matched on ART patient volume (1000-1999, 2000-4999, or =5000 current ART
patients), setting (urban, informal settlement, or rural), location (pairs should be located relatively nearby one another) and HIV viral suppression rate (see table 2 of the report).
In each pair, one clinic was randomly assigned (using a computer-generated randomisation) to receive early implementation of the minimum package of interventions, while the other continued to provide standard of care. No blinding was used.
Table 4 of the report provided under Related Materials shows the sample size that is required to detect meaningful differences for objectives 1-5. Sample sizes were determined using PASS software for cluster-randomised designs. Each sample size was determined to measure our short-term outcome for the objective. All calculations assume a site-clustered design with the clinic as the cluster and 24 clusters evenly split between intervention and comparison groups.
We assumed power of 80% and an alpha of 0.05. Sample sizes accounted for the cluster-randomised design by assuming a coefficient of variation of 0.1. Each sample size was calculated assuming a baseline proportion of patients achieving the outcome in the absence of the intervention as determined from the literature or experience. Sample sizes were calculated based on being able to detect an absolute increase on outcomes deemed to be clinically meaningful, ranging from 15% to 20% as determined by consensus of the investigators. The total sample size was calculated to be 3456 including all of the five HIV cohorts.
