Title: Exposure stratified case-cohort studies
Authors: Martyn Plummer - University of Warwick (United Kingdom) [presenting]
Abstract: The China Kadoorie Biobank (CKB) is a prospective study of over 510,000 adults in 10 regions of China. As part of the CKB we investigated the bacterium H. pylori as a cause of several gastrointestinal diseases: gastric non-cardia cancer, gastric cardia cancer, oesophageal cancer, and gastric and duodenal ulcer. Due to the high cost of the serological tests for H. pylori, a sample of the CKB has been considered. A case-cohort sample was chosen as an appropriate design when there are multiple disease outcomes of interest. We also wanted to optimize the selection of controls to account for the substantial variation in disease risk by age and sex, leading us to the exposure-stratified case-cohort design. The literature on exposure-stratified case-cohort designs contains relatively little advice on optimal design. We, therefore, chose our sub-cohort according to the well-established epidemiological principle of frequency matching. We also found unresolved questions in the analysis of these studies. Previous simulations have shown the superiority of Borgan III, the only unbiased estimating function for exposure stratified case-cohort studies. However, Borgan III is not fully efficient: the price of unbiasedness is that one control is discarded at random. We consider possible advances in the design and analysis of these studies.