Application of Machine Learning on Health Examination Data for Predicting the Decrease of Bone Mineral Densit

Background: Early detection and preventive measures for reduced bone density can greatly improve patients' quality of life and reduce economic burdens. This study aimed to develop machine learning algorithms that can accurately predict the risk of bone mineral density loss. Methods: The study included participants aged 40 years and older who underwent health evaluations at an affiliated institution from January 2022 to January 2024. Five machine learning algorithms were used to predict the risk of osteoporosis: k-nearest neighbor (KNN), random forest (RF), support vector machine (SVM), artificial neural network (ANN), and logistic regression (LR). The performances were evaluated based on accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUROC). Results: This study included 11132 patients, of whom 3568 had decreased bone density. The initial dataset contains 17 variables. After the data screening, 13 variables were included in the machine learning model. The AUROC for ANN, KNN, LR, RF, and SVM were 0.882, 0.906, 0.684, 0.918, and 0.896 for males, and 0.881, 0.843, 0.784, 0.922, and 0.872 for females, respectively. The accuracies of ANN, KNN, LR, RF, and SVM were 0.83, 0.86, 0.75, 0.88, 0.82 for males, and 0.81, 0.77, 0.74, 0.85, 0.79 for females. Conclusion: In this study, we developed five machine learning models to predict bone density reduction accurately. The RF model performed best in both male and female populations, with the highest AUROC. Application of machine learning models in clinical settings can help improve the prevention, detection, and early treatment of bone density reduction.