Estimation of covariate-dependent Markov transition probabilities from nested case-control data

Abstract

Multi-state models are used to describe situations where individuals may move among a finite number of states defined by specific conditions of health, including death. The transition intensities of the models are described by proportional hazards models, and it is reviewed how estimation of the regression parameters and the baseline transition intensities may be performed when only nested case-control data are available for all or some of the transitions. The regression parameter estimates and the estimates of baseline transition intensities are combined to give estimates of the integrated transition intensities for specified covariate histories, and from these estimates covariate-dependent Markov transition probabilities are derived.