📝 SummarXiv

Abstract

Most existing time-to-event methods focus on either single-event or competing-risk settings, leaving multi-event scenarios relatively underexplored. In many real-world applications, the same patient may experience multiple events that are non-exclusive, and sometimes semi-competing. A common workaround is to train separate single-event models, but this approach fails to exploit dependencies and shared structure across events. To address these limitations, we propose MENSA (Multi-Event Network for Survival Analysis), a deep learning model that jointly models flexible time-to-event distributions for multiple events, whether competing or co-occurring. In addition, we introduce a novel trajectory-based likelihood that captures the temporal ordering between events. Across five benchmark datasets, MENSA consistently improves prediction performance over many state-of-the-art baselines. The source code is available at https://github.com/thecml/mensa.