Friday, February 11, 2000, 12:00-1:00 p.m.
In this talk we shall illustrate the use of these models in the study of bone marrow transplantation for leukemia. In the first example we use a multistate model to study the effects of graft-versus-host disease on the probability of relapse and death in remission. By varying various rates we can examine the effects of treatment or alternative donor cell sources on the probability of relapse and death in remission. The projections are useful in designing clinical trials for GVHD prophylactics. In the second example we show how a multistate model can be used to model current leukemia free survival in a study where patients who relapse post transplant are given a donor leukocyte infusion to induce a second remission. We show that the method suggested in the medical literature estimator of CLFS based on the differences of several Kaplan-Meier estimators.
Multistate models for complex survival experiments are a useful tool for providing predictions for patient outcomes. In these models transition intensities are estimated for the rate at which patients leave one health state for another and these estimates are synthesized to provide predictions of the probability a patient will be in a given disease state sometime in the future. Markovian models assume that the transition intensity depends only on the patients current health state, while non-Markovian models depend on the duration of time a patient has spent in the current health state. Some transition rates may be proportional to others in which case Cox proportional hazard models are employed to improve the efficiency of the final estimates.