Retrieval-aligned Tabular Foundation Models Enable Robust Clinical Risk Prediction in Electronic Health Records Under Real-world Constraints
Researchers have developed a new framework called AWARE to improve clinical risk prediction using electronic health records. This framework addresses challenges like data imbalance and heterogeneity by using supervised embedding learning and lightweight adapters for retrieval-aligned tabular models. AWARE demonstrated significant improvements in predicting rare outcomes, particularly in complex datasets, by focusing on retrieval quality and alignment between retrieval and inference processes. AI
IMPACT Improves accuracy and robustness of AI models in clinical settings, potentially leading to better patient outcomes.