Machine learning models predicted osteoporotic fracture risk with high accuracy in postmenopausal women followed for 8 to 10 years, according to a study published in Scientific Reports.
Researchers evaluated 576 postmenopausal women across two independent cohorts with prolonged clinical follow-up. The HURH cohort included 276 postmenopausal women diagnosed with osteoporosis, among whom 72 experienced osteoporotic fractures. The Camargo cohort comprised 300 postmenopausal women from the general population, with 91 fractures observed during follow-up.
Prediction models were developed in the HURH cohort and externally validated in the Camargo cohort. The investigators tested two variable groups: one incorporating all available clinical and densitometric variables and another limited to measures more readily available in general medical consultations.
Across tested approaches, Extreme Gradient Boosting demonstrated the strongest predictive performance. In postmenopausal women with osteoporosis, the area under the curve (AUC) reached 0.88 using all variables and 0.92 using the streamlined variable set. During external validation in the general population cohort, performance remained stable at 0.88 for both variable groups, noted Ricardo Usategui-Martín, PhD, of the University of Valladolid, Spain, and colleagues.
Among postmenopausal women with osteoporosis, the most influential predictor of future fracture was a previous fracture. Other influential variables included parathormone levels and lumbar spine T score. When the model was restricted to more accessible clinical measures, the most influential predictors were previous fracture, parathormone, lumbar spine T score, and vitamin D levels.
More complex models incorporating trabecular bone score and three-dimensional dual-energy x-ray absorptiometry did not improve predictive performance compared with simpler models based on commonly available clinical variables.
The researchers noted that parathormone is not included in widely used fracture risk algorithms for postmenopausal women despite its role in bone metabolism. Vitamin D levels were also identified as important contributors to fracture risk prediction in both cohorts.
Bone mineral density measurements at the spine, femoral neck, and total hip contributed to the simplified models. Although bone mineral density is commonly used in diagnosis and treatment decisions, fractures frequently occur in patients with osteopenia or normal bone mineral density.
The researchers acknowledged several limitations. The cohorts were recruited in Spain, which may limit generalizability to other populations. Fracture occurrence was modeled as a binary outcome at the end of follow-up without incorporating fracture timing, meaning the models did not capture temporal dynamics such as imminent fracture risk. The sample size also did not permit detailed analysis by fracture type or location.
"Machine learning should be used to identify postmenopausal women at increased risk of fractures. This study summarizes that previous fractures, DXA, PTH, and vitamin D play crucial roles in identifying these women," noted Dr. Usategui-Martín, and colleagues.
The researchers reported no competing interests