Comparison of two discrimination indexes in the categorisation of continuous predictors in time-to-event studies

Abstract

The Cox proportionalhazards model is the most widely used su rvival prediction model for analysing time-to-event data. To measure the discrimination ability of a survival model the concordance probability index is widely used. In this work we studied and compared the performance of two different estimators of the concordance probability when a continuous predictor variable is cate- gorised in a Cox proportional hazards regression model. In p articular, we compared the c-index and the concordance probability estimator. We evaluated th e empirical performance of both es- timators through simulations. To categorise the predictor variable we propose a methodology which considers the maximal discrimination attained for th e categorical variable. We applied this methodology to a cohort of patients with chronic obstructiv e pulmonary disease, in particular, we categorised the predictor variable forced expiratory volu me in one second in percentage