Mapping the sex-dimorphic hormonal ecosystem in osteoarthritis: non-linear interactions between androgen activity and estrogen

Abstract Background The pronounced sex disparity in osteoarthritis (OA) risk implicates sex hormones, yet single-hormone studies have yielded inconsistent results. We hypothesized that androgen activity influences OA not in isolation but within a dynamic network defined by sex, menopausal status, and estradiol. Methods We analyzed 13,848 adults from NHANES (2013–2016, 2021–2023) using an integrated framework combining complex-survey regression, two-way restricted cubic splines, generalized additive models (GAMs), and SHAP-based machine learning to model non-linear interactions across the estradiol-FAI plane. Results Higher free androgen index (FAI) was significantly associated with lower OA risk only in post-menopausal females (OR = 0.85, 95% CI: 0.74–0.97, P = 0.021); associations in pre-menopausal females (OR = 0.75, P = 0.145) and males (OR = 0.94, P = 0.714) were not statistically significant. Non-linear models revealed distinct regional patterns: a “buffering” pattern in post-menopausal females, where the low-estradiol/low-FAI quadrant exhibited the highest risk (GAM P = 0.002; regional synergy mean = 0.077); a “cross-over” pattern in males (GAM P = 0.042); and a flat surface in pre-menopausal females, serving as a negative control. Candidate boundaries (e.g., estradiol ≈ 2.63, FAI ≈ − 0.54 in post-menopausal females) emerged from SHAP dependence plots as hypothesis-generating indicators requiring formal validation. Machine learning confirmed that standalone predictive contributions of sex hormones were minimal (mean |SHAP| ≤ 0.0135), supporting their role as contextual modulators rather than primary drivers. Conclusions Sex hormone effects on OA are context-dependent and non-linear, only discernible when examined in combination. The regional risk patterns—particularly the low-low quadrant in post-menopausal females—represent the robust core of our findings and provide a conceptual framework for understanding sex disparities through steroid interaction. Candidate boundaries are hypothesis-generating and await validation.