578 – Survival Analysis and Semiparametic and Nonparametric Models
Assessing the Performance of Predictive Scores: A Unified Copula-Based Framework and Algorithm for Numerical Evaluation
Yilong Zhang
Merck Research Laboratories
Yongzhao Shao
New York University School of Medicine
In current biomedical studies involving time to event outcomes, it is of great interest to evaluate the prognostic or predictive accuracy of markers or combination of markers. Concordance probability has been widely used for measuring prognostic accuracy in survival setting (Harrell et al 1982 and Gonen & Heller 2005). The copula is an important method to investigate the dependency of multiple random variables. We showed that concordance probability can be represented by copula and independent of its marginal distribution. Moreover, we used the relationship to propose a unified algorithm for generating multiple correlated concordance probabilities under different censoring types. The algorithm largely extended the commonly used simulation method based on multivariate normal distribution. Simulation studies are conducted to illustrate the algorithm performance with limited sample size.