Construction of a clinical prediction model for osteoporosis in asymptomatic elderly population based on machine learning algorithm.

Background: Osteoporosis is a metabolic bone disease characterized by a decrease in the amount of bone per unit volume. It is highly prevalent and has a harsh impact on patients' lives. The development of accurate predictive models for osteoporosis is beneficial in helping physicians improve the accuracy of clinical diagnosis and provide a high-quality treatment experience for older adults. Method: In this study, a robust and accurate prediction model for osteoporosis was developed and validated based on machine learning and SHAP techniques. We validated the model using ROC, calibration, and DCA curves. The data in this paper were obtained from elderly participants in several communities in Beijing from June 2021 to May 2022, including 161 (27.6%) males and 423 (72.4%) females, 248 (42.47%) with osteoporosis and 336 (57.53%) without osteoporosis. Results: Upon comparing and assessing the predictive outcomes of 135 models utilizing a combination of 10 machine learning algorithms, we found that the KNN+RF combination algorithm performs the best in terms of prediction performance. The Sensitivity, Specificity, PPV, NPV, Precision, Recall, F1, Detection Prevalence, AUC, and Brier metrics of this combined algorithm are 0.7500, 0.6634, 0.6136, 0.7614, 0.6136, 0.7200, 0.6626, 0.5000, 0.904, and 0.1601. Calibration and decision curve analyses further demonstrated the model's potential clinical utility. Ultimately, we created the Shiny web application for osteoporosis diagnosis. Conclusions: The osteoporosis prediction model is readily generalizable and can aid physicians in efficiently screening for osteoporosis in the broader older demographic. This will facilitate rapid detection and diagnosis of the disease, as well as the formulation of improved therapeutic treatment strategies for patients.