An Advanced Two-Stage Model with High Sensitivity and Generalizability for Prediction of Hip Fracture Risk Using Multiple Datasets.

Hip fractures represent a major source of disability, mortality, and healthcare burden in older adults, making early risk assessment essential for effective prevention. However, widely used tools such as the Dual-energy X-ray Absorptiometry (DXA) T-score and Fracture Risk Assessment Tool (FRAX) are limited in sensitivity and often miss individuals who will experience a fracture, particularly those without prior fracture history or with osteopenia. These limitations highlight the need for new predictive approaches that can incorporate both clinical characteristics and imaging features to improve early detection. In this study, we present a novel sequential two-stage model for predicting hip fracture risk, using data from the Osteoporotic Fractures in Men Study (MrOS), the Study of Osteoporotic Fractures (SOF), and the UK Biobank. In Stage 1 of the model (Screening Stage), clinical, demographic, lifestyle, cognitive, and functional factors are used to estimate the baseline risk of hip fracture. In Stage 2 (Imaging Stage) of the model, imaging features from DXA scans are incorporated to further improve the prediction. The model's performance was rigorously evaluated through both internal and external validations, including independent testing in the UK Biobank, demonstrating its adaptability across diverse populations. This stepwise approach improved sensitivity compared to traditional tools like T-score and FRAX, enabling earlier and more accurate detection of individuals at high risk for subsequent hip fracture. By prioritizing early detection, this sequential model reduced missed cases by offering a more cost-effective strategy that reserves imaging for those identified as possibly at high risk in the initial screening evaluation for subsequent personalized fracture risk assessment.