Abstract:
|
Nested case-control studies and other outcome-dependent sampling designs are frequently used when obtaining exposure or other covariate information from the entire cohort is too costly. Risk set sampling is generally based on one particular outcome of interest, and estimation is typically performed for a Cox model. In some settings, this outcome of interest is a non-terminal event. However, such non-terminal outcomes, including disease progression or diagnosis, are often subject to death, and scientific interest may lie in the interplay between the original event and death, as well as the roles of these covariates in both events. This exemplifies semi-competing risks data, which can be analyzed by fitting the illness-death model proposed by Xu et al. (2010). This talk proposes techniques for re-using data from nested case-control studies to estimate components of the illness-death model, leveraging the existing literature for estimating Cox regression coefficients from nested case-control studies. We propose both maximum likelihood and maximum weighted partial likelihood approaches and illustrate them using simulations.
|