Professor of Medical Statistics
Leiden University Medical Center
Homepage
Abstract
The Cox model is the dominant model to quantify the effect of covariates on time-to-event outcomes. It operates under the proportional hazards assumption, which acknowledges the time-dependent behaviour of the hazard, but assumes that the ratio of two hazards for different values of the predictor is constant over time. We will discuss mechanisms that could cause violation of the proportional hazards assumption, leading to time-dependent effects of the predictor, and discuss ways of dealing with possible violations. Seemingly similar, but quite different, is the situation where the value of the predictor can change over time. We will discuss commonly made mistakes in this setting, like immortal time bias, and correct ways of dealing with this situation. The workshop will include illustrations in R, and opportunities for the participants to try out some of the proposed methods in R or another statistical package of their choice.