A common objective in longitudinal studies is to characterize the relationship between a failure time process and time-independent and time-dependent covariates. Often, the proportional hazards model is used for this purpose, which necessitates knowledge of the time-dependent covariates for each individual at each failure time. In most studies, the longitudinally-measured covariate is only collected at a finite set of time points and may be subject to measurement error and biological variation. "Naive" methods, such as last-value-carried-forward, lead to biased estimators for the underlying relationship of survival to the time-dependent covariates. Recently, joint modeling of both the survival distribution and the longitudinally-measured time-dependent covariates as a function of a common set of random effects has been advocated for this problem. We consider such models, give the rationale for their use, and trace some of the analytic methods proposed for estimation.
Return to Biostatistics Working Group