Assessment of the risk of osteoporotic bone fracture in postmenopausal women using machine learning methods

Abstract

The main objective of osteoporosis management is to prevent osteoporotic fractures. Using machine learning methods, new risk variables can be identified to enhance the ability to identify women with osteoporosis who are at an increased risk of bone fracture. A multicenter study using machine learning-based methods was conducted in two independent cohorts of postmenopausal women (HURH and Camargo Cohorts), with clinical follow-up periods ranging from 8 to 10 years. The prediction models were developed in the HURH Cohort and validated using the Camargo Cohort, an independent external group of postmenopausal women. This study developed machine learning models to predict the risk of osteoporotic bone fractures. One is for postmenopausal women with osteoporosis, and the other is for general postmenopausal women. For each of these, two variable grouping options were used. The aggregation with the most predictive power included variables that are generally most accessible in medical practice. For postmenopausal women with osteoporosis, the AUC was 0.92, and for general postmenopausal women, it was 0.88. The results highlighted the significance of the previous fracture, DXA data, vitamin D levels, and PTH levels in predicting future fractures. 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.