Variable Selection for Illness-Death Processes Under Dual Observation Schemes

The classical illness-death process offers a useful framework for studying the progression of chronic disease while jointly modeling death. In many settings the time of disease progression is not observed directly, but progression status is recorded at intermittent assessment times. This creates a dual observation scheme where progression times are interval-censored and death is subject to right censoring. We present a penalized observed data likelihood for variable selection in multiplicative intensity-based models for the joint process, involving different penalty functions for different sets of regression coefficients. Optimization is carried out based on an innovative expectation-maximization algorithm that can be implemented using existing software. Simulation studies demonstrate the finite sample performance of the method, and an application to a dementia study from the National Alzheimer's Coordinating Center (NACC) illustrates the insights that can be gained.