Recent Developments in Censored, Non-Markov Multi-State Models

摘要

Nonparametric estimation of transition probabilities for a censored multi-state model is traditionally performed under a Markov assumption. However, this assumption may (and will) fail in some applications, leading to the inconsistency of the time-honoured Aalen-Johansen estimator. In such a case, alternative (non-Markov) estimators are needed. In this work we review some recent developments in this area. We also review the key problem of testing if a given (censored) multi-state model is Markov, giving modern ideas for the construction of an omnibus test statistic.