Ranking Weibull Survival Model: Boosting the Concordance Index of the Weibull Time-to-Event Prediction Model with Ranking Losses

The concordance index is a widely used metric for the evaluation of time-to-event prediction models. It describes the proportion of correctly ranked pairs of observations by time to event and hence is closely related to ROC AUC. In this paper, we propose enriching such baseline model as Weibull time-to-event feed-forward network, which optimizes classic in survival analysis log-likelihood with additional concordance-aware loss components. Here we demonstrate that a combination of parametric survival analysis methods with a learning-to-rank approach forces themodel to achieve higher concordance. The experiments over real-world datasets demonstrate the highly competitive performance of the proposed method called RWSM (Ranking Weibull Survival Model) in terms of the concordance index.