Search papers, labs, and topics across Lattice.
This study introduces a lightweight adaptation method for applying tabular foundation models to clinical survival analysis, specifically targeting right-censored time-to-event outcomes. By integrating a survival-aware head with pretrained representations from models like TabPFN, TabDPT, and TabICL, the authors demonstrate that this approach can outperform traditional baselines in predicting mortality in clinical settings. Notably, the TabDPT-FT-MTLR model achieved a C-index of 0.856 on the MIMIC-IV dataset, surpassing the best non-foundation model by 1.4% and the best zero-shot model by 6.7%.
Tabular foundation models can significantly enhance clinical survival predictions, achieving superior performance with minimal adaptation.
Predicting time-to-event outcomes such as mortality is a fundamental task in clinical decision-making, commonly addressed through survival analysis. While classical statistical and deep learning approaches have been widely studied, they typically require task-specific training and sufficient labeled data. Recent advances in tabular foundation models offer a new paradigm by learning general-purpose representations for structured data. However, their applicability to censored time-to-event prediction in clinical settings remains underexplored, as typical applications are restricted to discrete classification rather than survival analysis tasks. In this work, we propose a lightweight adaptation approach for applying tabular foundation models to clinical survival analysis by directly training a survival-aware head on top of the pretrained representations. We study representative architectures, including TabPFN, TabDPT, and TabICL, and adapt them using a multi-task logistic regression (MTLR) head to model right-censored time-to-event outcomes. We evaluate this approach on a diverse set of public survival benchmarks and two large-scale ICU cohorts, MIMIC-IV and eICU. Our results show that this transfer learning approach achieves competitive or superior performance compared to strong baselines. On MIMIC-IV, TabDPT-FT-MTLR reaches a C-index of 0.856, corresponding to a relative improvement of +1.4% over the best non-FM baseline (DeepSurv, 0.844) and +6.7% over the best zero-shot model (0.802). On eICU, TabICL-FT-MTLR achieves 0.797, yielding gains of +1.7% (DeepSurv, 0.784) and +6.4% (0.749), respectively. These findings highlight the importance of combining pretrained tabular representations with survival-aware objectives and suggest that tabular foundation models provide a practical and effective alternative for clinical survival prediction.
