Development and Validation of a Robust Prediction Model for Postoperative Pneumonia in Elderly Patients with Hip Fracture: Integrating Frailty, Nutrition, and Comprehensive Risk Assessment

Objective: To develop and rigorously validate a multivariable prediction model for postoperative pneumonia (POP) in elderly patients with hip fracture by integrating the five-item modified frailty index (5‑mFI), the Geriatric Nutritional Risk Index (GNRI), and various clinical variables, with comprehensive assessment of model performance. Methods: We conducted a retrospective cohort study of 2183 patients aged ≥ 65 years undergoing hip fracture surgery. Predictors included comorbidities, laboratory values (including partial pressure of oxygen PO 2 , B-type natriuretic peptide BNP, and GNRI), and the 5-mFI. We employed multivariable logistic regression to develop original and extended models, the latter adjusting for functional status and perioperative factors. Model performance was evaluated via area under the curve (AUC), bootstrap-corrected AUC, calibration, and decision curve analysis. Time-to-event and competing risk analyses were performed, and machine learning models (Random Forest, XGBoost) were compared. Results: The extended logistic regression model identified chronic obstructive pulmonary disease (odds ratio OR=2.60), postoperative intensive care unit admission (OR=2.72), lower PO 2 (OR=0.987), lower GNRI (OR=0.871), higher 5-mFI (OR=1.94), and higher BNP (OR=1.000) as independent predictors. The model demonstrated robust discrimination (AUC=0.781; bootstrap-corrected AUC=0.773), good calibration, and clinical utility. Results were consistent in competing risk analysis and robust to multiple imputation of missing data. Machine learning models confirmed GNRI and 5-mFI as top predictors, with comparable yet miscalibrated performance (XGBoost corrected AUC=0.791). Conclusion: We developed and internally validated a robust prediction model for POP that integrates frailty, nutrition, and key clinical variables. The model demonstrates strong, validated performance and clinical utility, providing a practical tool for preoperative risk stratification to guide targeted preventive measures in elderly patients with hip fracture. Keywords: hip fracture, postoperative pneumonia, frailty, nutritional status, risk prediction model, machine learning