Research Square · Oct 27, 2026
Serum 25-hydroxyvitamin D and diabetes in US adults: a pooled analysis of NHANES 2011–2018 with dose–response and race/ethnicity-specific associations
By Lan^ Y, Hong Y, Chen Y et al.
Background. Whether low vitamin D status tracks with diabetes remains debated, and most nationally representative evidence rests on a single two-year NHANES cycle, limiting statistical power for dose–response and subgroup analyses. We pooled four consecutive cycles to examine the vitamin D–diabetes association, with attention to dose–response and race/ethnicity differences. Methods. In this cross-sectional analysis we studied 20,232 US adults aged ≥ 20 years with measured serum total 25(OH)D across NHANES 2011–2018. Diabetes was identified through any of three criteria: self-reported physician diagnosis (excluding gestational diabetes), glycated hemoglobin (HbA1c) ≥ 6.5%, or fasting plasma glucose ≥ 126 mg/dL. Vitamin D status used Institute of Medicine thresholds (deficient < 30, insufficient 30–49.9, sufficient ≥ 50 nmol/L; reference) and sample quartiles. Using pooled 2-year examination weights and Taylor-series linearization, we fitted survey-weighted logistic regression models to estimate odds ratios (ORs), adjusted for age, sex, and BMI, with a sensitivity model adding education, smoking, family income-to-poverty ratio, and survey season, plus E-values for unmeasured confounding. Race/ethnicity-specific estimates and a multiplicative interaction test were computed. Results. Weighted diabetes prevalence was 13.5%. Both deficiency (adjusted OR 1.54, 95% CI 1.23–1.93) and insufficiency (1.52, 1.33–1.75) were associated with higher diabetes odds versus sufficiency (both p < 0.001), rising monotonically across 25(OH)D quartiles (Q1 vs Q4: OR 1.78, 1.47–2.14; p for trend < 0.001). A continuous spline confirmed a smooth rise in odds below approximately 75 nmol/L (30 vs 75 nmol/L: OR 1.61, 1.37–1.89). Exploratory stratified analyses showed associations in Mexican American (OR 1.88, 1.25–2.85) and Non-Hispanic White (insufficiency OR 1.55, 1.21–2.00) adults but not in Black or Asian adults (ORs ≈ 0.9–1.0); the interaction test was non-significant (p = 0.45). After further adjustment for income and season the gradient persisted but attenuated (deficiency 1.24, 0.98–1.57; insufficiency 1.37, 1.16–1.61; Q1 vs Q4 1.54, 1.25–1.91); E-values of 2.2–2.6 indicated moderate robustness to unmeasured confounding, and findings were consistent across alternative diabetes definitions. Conclusions. In a large pooled US sample, lower 25(OH)D was dose-dependently associated with diabetes, with exploratory stratified analyses suggesting a stronger association among White and Hispanic adults. Cross-sectional findings require prospective confirmation.