A new tabular foundation model outperforms traditional methods in predicting surgical risks for emergency patients, improving decision-making.

• In a study of 7,281 emergency surgery patients, the model demonstrated a mortality prediction accuracy of 0.89 and a morbidity accuracy of 0.82.
• It achieved better calibration scores (0.04 for mortality and 0.15 for morbidity) than logistic regression and XGBoost models.

This model may enhance patient selection and outcomes by enabling personalized risk stratification in varied clinical settings.

• Performance held strong even with smaller site-specific data, indicating flexibility for surgeons in diverse environments.