Serum 25-hydroxyvitamin D levels and prevalent diabetic retinopathy in adults with type 2 diabetes: a retrospective hospital-based matched case-control study

Background Diabetic retinopathy (DR) is a major microvascular complication of type 2 diabetes mellitus (T2DM). This study examined the adjusted association between serum 25-hydroxyvitamin D [25(OH)D] and prevalent DR. Methods This retrospective, single-center matched case–control study was conducted at a tertiary hospital in China among adults with T2DM treated between January 2023 and March 2025. Serum 25(OH)D testing was not routine for all patients and was ordered at the treating clinician’s discretion. Participants were eligible only when a 25(OH)D result was available within ±7 days of bilateral fundus photography. Cases had prevalent DR and controls had no DR; individual 1:2 matching used sex, age within ±3 years, and diabetes duration within ±2 years. The prespecified primary analysis modeled 25(OH)D continuously in conditional logistic regression, and missing covariates were handled by multiple imputation. Results The final sample included 399 selectively tested participants (133 with DR and 266 without DR) in 133 matched sets. Mean serum 25(OH)D was 18.6 ± 7.4 ng/mL in the DR group and 23.4 ± 8.1 ng/mL in the non-DR group. In the prespecified adjusted model, each 10 ng/mL higher serum 25(OH)D level was associated with lower prevalence odds of DR (OR, 0.62; 95% CI, 0.48–0.81; p < 0.001). Additional adjustment for continuous age and diabetes duration yielded a similar estimate (OR, 0.63; 95% CI, 0.48–0.83; p = 0.001). In an exploratory unadjusted analysis, mean 25(OH)D decreased across ICDR categories from no DR to proliferative DR (23.4 to 15.2 ng/mL; P for trend < 0.001). Conclusion In this selectively tested, single-center tertiary-hospital sample, lower serum 25(OH)D was concurrently associated with higher prevalence odds of DR. Given the retrospective cross-sectional case–control design, this finding should be interpreted as an association only; it does not establish temporality or causality, and selection bias, residual confounding, and reverse causality remain possible.