External Validation of an In-Hospital Mortality Prediction Model Using Comprehensive Hospital Medical Data

This study evaluates an AI-based 30-day in-hospital mortality prediction model initially developed at Nagoya University Hospital when applied to a different university hospital. AUROC values for the Nagoya dataset were 94.1 (lung), 99.8 (liver), and 97.3 (colorectal), while the external facility showed AUROCs of 75.6, 77.6, and 85.2, respectively. Conventional cancer staging had much lower AUROCs. These findings suggest that AI-based models can outperform traditional methods across different facilities, despite regional practice variations and data discrepancies.