B1020
Title: Federated regression analysis of heterogeneous data with competing risks
Authors: Bella Vakulenko-Lagun - University of Haifa (Israel) [presenting]
Abstract: A privacy-preserving federated learning approach is presented for regression analysis of heterogeneous survival data with competing risks. The approach extends a semi-parametric additive hazards model to a federated learning setting, allowing for different types of heterogeneity across participating sites, e.g., heterogeneous population compositions (or covariate shifts) or site-specific baseline hazards. The development of the method was motivated by a drug repurposing study that combines information from diverse patient populations across the US and UK electronic health records systems, with the aim of finding treatment for Alzheimer's disease.
