Sex-Specific Biological Predictors for Identifying Individuals With Low Bone Mineral Density Using the Taiwan Biobank...

Osteoporosis has seen a rapid rise in Taiwan due to its aging population. The pathogenesis of low bone mineral density (BMD) differs between women and men, highlighting the critical need to explore sex-specific biological predictor variables-an area with significant knowledge gaps. This study aimed to identify sex-specific predictors using a population database through machine learning. We conducted a retrospective cross-sectional study of 19,868 individuals aged 40 and above from the Taiwan Biobank database who had no cancer diagnosis and underwent BMD examinations between 2020 and 2022, with available demographic, laboratory, and physical examination data. The Extreme Gradient Boosting (XGBoost) model was employed to identify patterns of predictor variables for low BMD. SHapley Additive exPlanations (SHAP) were applied to visualize and interpret the contribution of each predictor to low BMD in women and men. Among participants, 63% (n = 12,565) were women and 44% had low BMD (35% osteopenia and 9% osteoporosis), with a mean age of 54 years (SD = 9.6; range 40-88). The area under the curve values were 0.75 for women and 0.63 for men. Age, postmenopausal status, and BMI were the most significant predictors of low BMD in women, whereas age, BMI, and waist circumference were the strongest predictors in men. In addition, hematological factors (WBC, RBC, platelet count, hemoglobin), metabolic indicators (body fat percentage, triglycerides, and blood pressure), uric acid, and liver (gamma-GT) and renal (eGFR) function markers were also associated with BMD. These predictors differed in their relative importance between women and men, highlighting clear sex-specific patterns in factors associated with low BMD. Recognizing these determinants is essential for prevention and early intervention strategies, while further research is needed to better understand the complex mechanisms underlying bone metabolism.