Two-Sample Multistate Accelerated Sojourn Times Model

Abstract

Many medical studies involve observations of patients experiencing multiple consecutive states during their follow-up. Since treatments may have differential effects on these states, comparison is of interest with respect to individual sojourn times. In this article I generalize the univariate accelerated failure time model for multistate processes and model the treatment effects as state-specific time scale changes. I propose a two-sample inference procedure that accommodates incomplete follow-up data. This semiparametric procedure, based on estimating equations, yields a class of estimators that are consistent and asymptotically normal. Sample-based consistent variance estimates are derived. Numerical studies demonstrate that the procedure performs well for practical sample sizes. Application to a cancer clinical trial is presented for illustration.

Key Words:

• Dependent censoring
• Estimating equation
• Multistate process
• Multivariate failure time
• Semiparametric inference
• Survival analysis