Title: Efficient and reliable inference in nested case-control studies
Authors: Dennis Dobler - Vrije Universiteit Amsterdam (Netherlands) [presenting]
Jan Feifel - Ulm University (Germany)
Abstract: Nested case-control designs are of great importance in time-to-event studies with rare outcomes or expensive covariate evaluations. Such designs allow for a (random) selection of only very few controls for each case while, at the same time, not much power is lost compared to the full evaluation. We propose a resampling algorithm for the approximation of the distribution of estimators - like the cumulative hazard - whose complexity grows only linearly in sample size and independently of the number of controls per case. The algorithm works for various sampling designs (counter-matching and more customized designs) and minimal assumptions on the dependence between censoring and event times are required. Theoretical validity with growing sample size is proved rigorously, and simulation results confirm the practical usefulness of our approach.