Frontiers in Endocrinology · Nov 9, 2026
Central adiposity, insulin resistance, and systemic inflammation in osteopenia and osteoporosis: a hospital-based study using explainable machine learning
By Chen Liang, Zongyao Li, Junlei Yin
Background: Osteopenia and osteoporosis are common in routine health-examination populations, but cross-sectional indices must not be interpreted as predictors of incident fractures. Objective: To compare routinely derived anthropometric, metabolic and inflammatory indices in relation to prevalent lower bone mass and to explore internal classification performance. Methods: In this cross-sectional study, we analyzed 848 adults attending health examinations in 2024–2025. Bone mineral density was assessed by central dual-energy X-ray absorptiometry at the lumbar spine, femoral neck, and total hip. Bone status was classified using the lowest recorded T-score: normal (≥−1.0), osteopenia (−2.5 < T-score < −1.0), or osteoporosis (≤−2.5). Three borderline classifications were retained as recorded and are detailed in Supplementary Table S17. Because Z-scores were unavailable, diagnostic terminology for premenopausal women and men younger than 50 years requires caution. We examined adjusted associations, quartiles, restricted cubic splines, sensitivity analyses, internal model performance, and exploratory clusters. Results: Of 848 participants, 397 (46.8%) had normal bone mass, 351 (41.4%) osteopenia, and 100 (11.8%) osteoporosis. The fully adjusted odds ratios (95% CIs) per 1 SD were 1.61 (1.25– 2.06) for WWI, 1.59 (1.23–2.05) for conicity index, 1.34 (1.01–1.79) for TyG–WHtR, 1.41 (1.12– 1.77) for SII, and 0.71 (0.53–0.96) for BMI. After false-discovery-rate correction across 25 indices, TyG–WHtR was no longer statistically significant (q=0.119). In the held-out test sample, the AUROC was 0.948 (95% bootstrap CI 0.92–0.97) for the demographic model and 0.947 (0.92–0.97) for the novel-index model (paired difference −0.001, 95% CI −0.02 to 0.01). Conclusion: Central adiposity and inflammatory indices were associated with existing lower bone mass, but the new-index model did not improve discrimination over basic demographic characteristics. Prospective and external studies are needed before these measures can inform clinical decisions.