A Hierarchical Approach to Multi-Event Survival Analysis

In multi-event survival analysis, one aims to predict the probability of multiple different events occurring over some time horizon. One typically assumes that timing is drawn from distribution conditioned on an individual's covariates. However, during training, does not have access this distribution, and natural variation in observed event times makes task prediction challenging, top potential interdependence among events. To address issue, we introduce a novel approach for analysis models occurrence hierarchically at scales, using coarse predictions (e.g., monthly predictions) iteratively guide finer grained scales daily predictions). We evaluate proposed across several publicly available datasets terms both intra-event, inter-individual (global) intra-individual, inter-event (local) consistency. show method consistently outperforms well-accepted commonly used approaches analysis. When estimating curves Alzheimer's disease mortality, our achieves C-index 0.91 (95% CI 0.88-0.93) local consistency score 0.97 0.94-0.98) compared 0.75 0.70-0.80) 0.94 0.91-0.97) when modeling each separately. Overall, improves accuracy by reducing original set nested, simpler subtasks.