Secondary fragility fractures after hip fracture surgery in four thousand, four hundred and eighteen older adults: risk factors and internal validation of an interpretable machine-learning model

Abstract

Secondary fragility fractures remain a major complication after hip fracture surgery in older adults, yet individualized postoperative risk stratification remains challenging. We aimed to identify factors associated with secondary fragility fracture and internally validate an interpretable machine-learning model for three year refracture risk. We retrospectively analyzed adults aged ≥ 65 years who underwent surgery for a first low-energy hip fracture between 2014 and 2021. Patients were followed for three years for secondary fragility fracture. Demographic, clinical, functional, and rehabilitation variables were extracted from electronic medical records. Cox regression with multiple imputation was performed to evaluate time-to-event associations. XGBoost was developed as the primary prediction model and compared with SVM and Cox regression. Model performance was assessed using a held-out test set and decision curve analysis. Among 4,418 patients, 594 (13.4%) sustained a secondary fragility fracture. Increased refracture risk was associated with older age, osteoporosis, low BMI, female sex, cardiovascular disease, visual impairment, syncope, non-standard rehabilitation, and postoperative Harris hip score < 80. In the test set, XGBoost achieved an AUC of 0.701 (95% CI, 0.657–0.743), with 61.8% sensitivity and 67.2% specificity at the default threshold. Decision curve analysis demonstrated the highest net benefit for XGBoost across threshold probabilities of approximately 7%–32%. Secondary fragility fractures were associated with skeletal fragility, fall-related vulnerability, and impaired postoperative recovery. The interpretable XGBoost model showed moderate discrimination and potential clinical utility as an adjunctive screening tool to identify patients who may benefit from intensified secondary prevention. External validation and prospective evaluation are required before routine clinical implementation.