Abstract

Background/Objectives: Hip fractures in older adults are associated with substantial morbidity, functional decline, and high in-hospital mortality. Early identification of patients at increased risk of death may improve clinical decision-making and resource allocation. This study aimed to develop and internally validate supervised machine learning models to predict in-hospital mortality among older adults hospitalized for hip fracture using nationwide administrative data from Chile. Methods: A retrospective cohort study was conducted using anonymized hospital discharge records from the Chilean National Health Fund (FONASA), covering admissions between 1 January 2019 and 31 December 2024, across 72 public hospitals. Demographic, clinical, and care-related variables were included as predictors. Multiple supervised machine learning algorithms were trained and evaluated using stratified train-test partitioning. Model performance was assessed using AUC-ROC, precision, recall, and F1-score. Model interpretability was explored using SHapley Additive exPlanations (SHAP). Results: A total of 40,253 hospitalization episodes were analyzed. The Gradient Boosting model achieved the best overall performance, with an AUC-ROC of 0.885 and a favorable balance between precision and recall. SHAP analysis identified age, comorbidity burden, and surgical treatment as the most influential predictors, revealing nonlinear and clinically meaningful contributions to mortality risk. Conclusions: Supervised machine learning models based on routinely collected administrative data demonstrated strong predictive performance for in-hospital mortality after hip fracture. Interpretable models may support early risk stratification and clinical decision-making at a national healthcare level.