A censoring-aware target interface for tabular foundation models in survival prediction

2026-07-10

AI Summary

Right-censored data hinder the direct application of general-purpose tabular foundation models in survival analysis. This work proposes SurvFM-RMST, a novel framework that leverages restricted mean survival time (RMST) derived from Jackknife pseudo-observations as a transferable regression target, enabling modern tabular foundation models to predict time-specific RMST without architectural modifications or task-specific customization for survival analysis. The method is validated on both simulated data and 36 real-world datasets from SurvSet, demonstrating accurate estimation of conditional RMST and significantly outperforming baselines that rely solely on raw observed times or binary event indicators. Furthermore, SurvFM-RMST facilitates effective patient risk stratification, highlighting its practical utility in clinical and biomedical applications.

Abstract

Time-to-event prediction from tabular patient data is central to prognosis and biomedical decision support, but right-censored follow-up prevents direct use of ordinary regression labels. Tabular foundation models offer reusable prediction machinery for modest heterogeneous datasets, yet they generally assume fully observed outcomes. We introduce SurvFM-RMST, a censoring-aware target-interface framework that converts survival outcomes into jackknife pseudo-observation targets for restricted mean survival time, enabling multiple tabular backbones to perform horizon-specific RMST regression without survival-specific fine-tuning. In controlled simulations with known conditional RMST, SurvFM-RMST recovered restricted event-free time accurately, and pseudo-RMST targets outperformed naive restricted observed-time and event-only targets. Across 36 eligible static SurvSet datasets, SurvFM backbones were competitive with established survival and RMST-regression comparators, though relative performance varied by endpoint, horizon and practical constraints. Predicted RMST further stratified held-out patients into groups with ordered observed event-free time and event enrichment. Overall, the results support pseudo-RMST target construction as a portable interface between censored survival data and tabular foundation-model prediction.

Problem

• survival prediction
• right-censoring
• tabular foundation models
• restricted mean survival time
• pseudo-observations

Innovation

• censoring-aware
• restricted mean survival time
• pseudo-observations
• tabular foundation models
• survival prediction