Abstract
BACKGROUND: While dual-energy X-ray absorptiometry (DXA) remains the gold standard for osteoporosis diagnosis, its clinical utility is constrained by cost and accessibility challenges. This study aims to develop a predictive model for osteoporosis using routinely available clinical blood biomarkers, thereby providing an innovative and accessible approach for early detection. METHODS: We retrospectively analyzed 8,144 orthopedic inpatients who underwent DXA scans at Panyu Hospital of Guangzhou University of Chinese Medicine between January 2022 and December 2023. Demographic characteristics and first 24-hour admission blood parameters were collected. Potential predictors were identified through univariate analysis, least absolute shrinkage and selection operator (LASSO) regression, and Boruta algorithm. Ten supervised machine learning algorithms were employed to construct predictive models. Model performance was evaluated using area under the curve (AUC), calibration plots, decision curve analysis (DCA), accuracy, sensitivity, and specificity in the test cohort. SHapley Additive exPlanations (SHAP) analysis provided interpretable visualization of feature contributions. RESULTS: The cohort was randomly divided into training (n = 5,702) and testing sets (n = 2,442). Feature selection convergence across three methods identified 11 key predictors. The logistic regression model demonstrated superior performance in the testing set (AUC = 0.800), outperforming other algorithms in calibration and clinical utility assessments. SHAP analysis revealed age, gender, uric acid concentration, alkaline phosphatase levels, hemoglobin levels, and neutrophil count as the six most influential predictors. An accessible web-based risk calculator has been deployed at: https://op-lm.shinyapps.io/osteoporosis/ . CONCLUSION: We developed an easy-to-use online calculator based on machine learning, which outperforms traditional models, enabling patients to preliminarily screen for osteoporosis using routine blood test results from their health check-ups. This interpretable machine learning model demonstrated promising performance and may assist in improving osteoporosis screening and risk stratification in clinical settings. CLINICAL TRIAL NUMBER: Not applicable.