Competing risks and multi-state models play an increasingly important role in the analysis of time to event data. Regarding competing risks, there is a lot of confusion regarding the proper analysis. The most important reason for the confusion is conceptual: which quantities can be estimated and what do they represent. Once the concepts are understood and the proper type of analysis has been chosen, most analyses are straightforward and can be performed with standard software for survival analysis. For multi-state models with exactly observed transition times, estimation is reasonably straightforward and the real challenge is in (dynamic) prediction.
The overarching goal of the course is to provide a solid introduction to these topics and thereby increase the analytical validity in this field.
In the first part of the course we cover competing risks analysis: what are competing risks and when do we need to take them into account; the independence assumption; cause-specific cumulative incidence; cause-specific hazard and subdistribution hazard; competing risks as a multi-state model. We will also cover regression models on both cause-specific and subdistribution hazard (Fine-Gray model) and discuss the difference in interpretation. We show how analyses can be performed with standard software. In the second part of the course, the extension to multi-state models is discussed. The course will cover topics including transition intensities and transition probabilities, nonparametric estimation and regression models, as well as methods to obtain predictions of future events, given the event history and clinical characteristics of a patient. With right censored and/or left truncated data, we show that it is possible to perform many types of analyses using standard software, using the same techniques as in multi-state representation of the competing risks model.
