Age-platelet-to-lymphocyte ratio phenotypes and prevalent osteoporotic fracture in postmenopausal women: cross-cohort analysis with interpretable machine learning

Fragility fractures in postmenopausal women often occur before the bone mineral density (BMD) T-score reaches − 2.5. Chronic low-grade inflammation contributes to postmenopausal bone loss, but inflammatory markers have not entered fracture-risk assessment. We applied Gaussian mixture modeling to age and platelet-to-lymphocyte ratio (PLR) in 1,000 women aged ≥ 50 years from NHANES 2005–2018, using age as a proxy for postmenopausal status and osteoporosis as the outcome. Three age–PLR subgroups were identified and mapped by standardized Euclidean distance to 572 inpatients assessed for bone health at Deyang People’s Hospital (2023–2025), where prevalent osteoporotic fracture was the outcome. An NHANES XGBoost model generated a source-derived osteoporosis probability feature for the hospital fracture model; it was not interpreted as a direct fracture probability. We performed exploratory age-stratified PLR analyses. In NHANES, the High-Risk and Elderly-Low subgroups had higher covariate-adjusted odds of osteoporosis than Low-Risk (odds ratio [OR] 3.64, 95% confidence interval [CI] 1.94–6.84, P < 0.001; and OR 3.97, 95% CI 2.37–6.65, P < 0.001, respectively). Robust regression down-weighting outliers attenuated the High-Risk estimate to 1.21 (95% CI 0.69–2.14, P = 0.565). In the hospital cohort, High-Risk was associated with prevalent osteoporotic fracture after covariate adjustment (OR 2.20, 95% CI 1.15–4.22, P = 0.017) and BMD adjustment (OR 2.09, P = 0.027). Adding the source-derived feature improved discrimination compared with the hospital-only model (area under the curve 0.716 vs. 0.675; DeLong P = 0.025), but provided limited incremental net benefit. PLR was associated with osteoporosis among women aged 50–64 years (OR 1.09 per 10-unit increase, P = 0.013), but not in older groups. Source-derived age–PLR phenotypes were associated with osteoporosis in NHANES and prevalent fracture in a selected hospital cohort. The feature modestly improved discrimination but showed limited incremental clinical utility. These cross-sectional findings do not establish temporality, prediction, or independence from BMD and require prospective external validation.