Two methods for dynamic prediction

An important extension of standard survival analysis takes into account intermediate events, both as an outcome of interest and as a predictor for terminating events. We will present two methods for predicting the impact of an intermediate event on subsequent outcomes which can both be considered as extensions of the Cox model. In the first, a Markov multi-state model is used to estimate the transition probabilities between all events of interest and the effect of covariates on them. In the second, a series of landmark models is built which are connected by a supermodel. Both models will be illustrated on a dataset describing outcomes after allogeneic hematopoietic stem cell transplantation for patients with CLL.

This presentation is partly based on joint work with Hein Putter and partly on work by Hein Putter and Hans van Houwelingen.