Generalizability of a prediction model for walking independence at discharge in older adults after hip fracture surgery: a multicenter study

Abstract This study aimed to evaluate the generalizability of a clinical prediction model for walking independence at discharge from acute-care hospitals after hip surgery in older patients. This retrospective multicenter study was conducted at acute-care hospitals. A total of 2,548 older patients registered in a hip fracture database involving 19 hospitals who were able to walk independently before injury and underwent surgery were included. To evaluate generalizability, we conducted internal–external cross-validation (IECV)across the participating hospitals. The performance of the prediction model was evaluated using the area under the curve (AUC), calibration intercepts, and calibration slopes. The generalizability of the model was assessed using 95% prediction intervals (PIs). Among the 2,014 patients included in the final analysis, 38.2% regained independent walking ability at discharge. Across IECV, the AUC was 0.81 (95% PI: 0.62–0.92). Calibration slopes were 0.96 (95% PI: 0.34–1.58) and calibration intercepts were 0.11 (95% PI: −1.56 to 1.77). Given the marked heterogeneity observed across clusters, our findings indicate that the generalizability of the prediction model for walking independence at discharge after hip fracture surgery in older adults is limited, highlighting the impact of between-hospital heterogeneity on model performance.