Survival Models With Time-Varying Coefficients: A Flexible Approach to the Analysis of Psychiatric Survival Data

Survival data from psychiatric studies are typically modeled using the Cox proportional hazards model. While the Cox model tends to provide “average” effects of covariates, these effects on the hazard may vary with time. We describe a flexible approach to examine the time-dependence of the covariate effect using the Cox proportional hazards model and Gray’s piecewise constant time-varying coefficients model in an analysis of time to recurrence of major depression in an elderly population in relation to two covariates: medication (nortriptyline) use and log-transformed total rapid eye movement (REM) activity counts. Results show that the effect of medication use is constant over the follow-up period and decreases the risk of recurrence. A time-varying effect of total REM activity counts on the risk of recurrence is indicated in Gray’s model, with higher risk seen in the later period of follow-up. The Cox model fails to discern the effect of total REM activity count in this data set. We recommend the use of Gray’s model for the modeling of time-varying effects of covariates. Psychopharmacology Bulletin. 2002;36(4):84-91