Machine Learning-Driven Prediction of Low BMD in Postmenopausal Women Using Cytokine, RANKL/OPG, and Oxidative Stress...

Aim: To evaluate cytokine, RANKL/OPG, and oxidative stress biomarker profiles in postmenopausal women with differing bone mineral density (BMD) and to apply machine learning (ML) models for early identification of low BMD. Materials & Methods: Seventy-one postmenopausal women were classified as normal BMD (N), osteopenia (OSN), or osteoporosis (OSR) using ISCD/WHO criteria. Ten cytokines, RANKL, OPG, and five oxidative stress markers were quantified. Group differences were assessed using nonparametric statistics and rank-biserial correlation. ML classifiers were developed within the PyCaret automated machine learning framework and evaluated using repeated stratified cross-validation to distinguish N vs. low BMD (L = OSN + OSR) and OSN vs. OSR using cytokine-only, bone/oxidative, and integrated biomarker panels, while SHapley Additive exPlanations (SHAP) were employed to interpret feature contributions. Results: Low BMD was associated with elevated pro-resorptive cytokines (TNF-α, IL-6, IL-12) and reduced anti-resorptive cytokines (IL-4, IL-10, IL-23). OPG and antioxidant enzymes (catalase, SOD2, PRX2) were significantly lower in L. Effect-size findings confirmed strong associations of IL-4, IL-10, IL-23, and OPG with normal BMD, and IL-6, IL-12p70, and TNF-α with low BMD. Logistic regression integrating cytokines with OPG and SOD2 achieved the best N vs. L performance (mean F1 ≈ 0.90, SD = 0.09). Biomarkers showed limited ability to discriminate OSN from OSR (maximum F1 ≈ 0.70, SD = 0.18). Conclusions: Integrating cytokine, bone-regulatory, and oxidative stress markers enhances ML-based prediction of low BMD and supports improved early osteoporosis risk stratification.