Title: Statistical boosting for time-to-event data: An overview on recent developments
Authors: Matthias Schmid - University of Bonn (Germany)
Elisabeth Waldmann - Friedrich-Alexander-Universitaet Erlangen-Nuernberg (Germany)
Andreas Mayr - University of Bonn (Germany) [presenting]
Abstract: Statistical boosting algorithms combine a powerful machine learning approach with classical statistical modelling, offering various practical advantages like automated variable selection and implicit regularization of effect estimates. In the context of time-to-event data, the general boosting concept was also extended beyond the classical Cox framework. For example, we developed a boosting approach to directly estimate prediction models that are optimal with respect to the concordance index (Harrell's C). Another recent approach focuses on joint models for longitudinal and time-to-event data. The aim is to provide a short introduction to the concept of boosting and its application for the analysis of survival data, highlighting both advantages and limitations for practical data analysis.