Abstract:
|
Analysis of time-to-event data typically considers full covariate information on all available subjects. When measurements are difficult or expensive to collect and the outcome is rare, the nested case-control design proves to be a useful alternative approach. This method only requires full covariate information for all cases and selected controls, thus reducing collection time, cost, and burden on patients. Under proportional hazards, this approach leads to consistent estimation of regression parameters with little impact on efficiency in many cases. This presentation focuses on the nested case-control design under time-varying effects. We present simulation results revealing that under a correctly specified model, the standard nested case-control design yields biased estimates of model parameters in the presence of time-varying effects and high censoring. The censoring distribution is implicitly modified when using the nested case-control design thereby contributing to the observed bias. As such, we propose a new estimator that re-weights the contribution of each subject to the partial likelihood based on a covariate-dependent inverse probability of censoring estimate.
|