Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces SurvICL, a novel tabular foundation model designed for in-context survival analysis, addressing the current TFM underperformance in this domain compared to traditional methods like the Cox model. SurvICL leverages a flexible proportional-odds prior for synthetic data generation and a plain quantile regression loss for training, enabling it to predict survival curves effectively even with censored data. The model achieves state-of-the-art performance across a large new benchmark, outperforming 20 baselines, including the Cox model, particularly on smaller datasets.
A foundation model now surpasses the Cox model on survival analysis, even for small clinical cohorts, by predicting survival curves in-context.
Tabular foundation models (TFMs) now outperform gradient-boosted trees in classification and regression. Yet in survival analysis, which predicts when an event occurs from censored data, often on small clinical cohorts, we find that no existing TFM beats the half-century-old Cox model. We address this issue by introducing SurvICL, a TFM that predicts survival curves in-context. SurvICL is pretrained on purely synthetic datasets generated from a novel flexible proportional-odds prior that outperforms other prior types in our ablation study. Unlike most other survival models, SurvICL is trained with a plain quantile regression loss, which is possible because its synthetic pre-training data generator can provide uncensored event times. Additionally, we introduce the largest survival benchmark to date spanning 103 datasets. On the 63 datasets held out from all design decisions, SurvICL outperforms 20 tree-based, deep, and foundation-model baselines without any tuning: it wins 82.9\% of pairwise comparisons in integrated Brier score, against 71.3\% for the runner-up, the Cox model, and is the first model to rank above Cox on small datasets.