|
Abstract:
|
The evaluation of the national ART program in Malawi currently relies on clinic-level aggregated data. To investigate individual-level outcomes, such as treatment adherence, individual-level data is need and the case-control design is appealing. To our knowledge, however, no statistical methods have been developed for case-control data that account for clustering. Furthermore, in the specific context of a collaboration in Malawi, case-control sampling within clinics has been suggested as a more practical strategy. While similar outcome-dependent sampling schemes have been described, the cluster-stratified case-control design is new. Here we describe this design, discuss balanced versus unbalanced sampling techniques, and provide a general approach to analyzing case-control and cluster-stratified case-control studies in cluster-correlated settings based on inverse-probability weighted GEE. Inference is based on a robust sandwich estimator with correlation parameters estimated to ensure appropriate accounting of the outcome-dependent sampling scheme. Comprehensive simulations, based in part on real data from Malawi, are conducted to evaluate small-sample operating characteristics.
|
