To enhance methodological transparency and reduce potential bias, several strategies were implemented in the study design and analysis. First, this study was based on a single-center hospital electronic health system (HIS) rather than a population-based survey; therefore, survey weights and complex sampling designs were not applicable. To minimize selection bias, all eligible patients with complete exposure, outcome, and covariate data during the study period were consecutively included according to predefined inclusion and exclusion criteria. Second, multiple covariate adjustments were applied in multivariable models, including demographic, lifestyle, reproductive, physiological, and biochemical factors, to reduce potential confounding. Third, consistent diagnostic criteria and validated laboratory assays were used for exposure and outcome measurements to limit misclassification bias. Finally, subgroup and sensitivity analyses were conducted to evaluate the robustness of the findings and explore potential residual confounding.

This study is a retrospective analysis, with data derived from patients at the Affiliated Hospital of Jiangxi University of Traditional Chinese Medicine between June 2014 and December 2024. The study population included individuals with complete records in the hospital electronic health system (HIS), initially screening 4,034 patients. According to exclusion criteria, patients with missing serum VD data (N = 1,990) were removed from the study. In addition, patients were excluded due to missing covariate information, including missing smoking data (N = 323), missing alcohol consumption data (N = 134), and missing BMI data (N = 190). Ultimately, a total of 1,397 participants were included in this study. The complete participant screening flow is shown in Figure 1. This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Ethics Committee of the Affiliated Hospital of Jiangxi University of Traditional Chinese Medicine (Ethics approval number: JZFYLL202402230018). This study did not undertake a formal sample size calculation; all eligible cases were included in the analysis from the existing medical records database. The study population included ALL individuals with complete records in the hospital electronic health system (HlS).

Participants’ baseline characteristics were classified according to the presence or absence of PCOS. Continuous variables were expressed as mean ± standard deviation (mean ± SD), and categorical variables were expressed as percentages (%). Differences between categorical variables were assessed using the chi-square (χ²) test, while differences between continuous variables were evaluated using the Kruskal–Wallis test (11). A two-sided p-value < 0.05 was considered to indicate statistical significance.

Following descriptive analyses, multivariate logistic regression models were used to examine the association between serum VD levels and the risk of PCOS, quantified as odds ratios (ORs) with 95% confidence intervals (CIs). Three models were established to sequentially control for potential confounders: Based on the predefined covariates, three progressively adjusted logistic regression models were constructed. Model 1 was unadjusted. Model 2 adjusted for demographic and lifestyle factors, including age (< 40 or ≥ 40 years), educational level, marital status (married/living with partner, never married, widowed/divorced/separated), smoking status, and alcohol consumption. Model 3 further adjusted for reproductive history (ever pregnant or never pregnant), physiological indicators including body mass index (BMI; < 25, 25–30, or ≥ 30 kg/m²), hypertension, and diabetes, as well as biochemical markers including CRP, total testosterone, and glutathione (GSH), based on Model 2. Serum VD concentration (nmol/L) was analyzed as the exposure variable across all models. Additionally, restricted cubic spline (RCS) models were applied to evaluate the non-linear dose–response relationship between serum VD levels and PCOS risk, adjusting for the same covariates included in Model 3 (11). In addition, subgroup analyses were conducted to assess the robustness of associations and potential heterogeneity, exploring the modifying effects of age, marital status, BMI, smoking, alcohol consumption, and hypertension on the relationship between VD and PCOS risk (12). Furthermore, sensitivity analyses were performed to evaluate the stability of the main findings. These analyses included repeating the multivariable models using alternative categorizations of serum vitamin D status based on different cutoff values reported in previous literature, as well as excluding participants with major metabolic comorbidities. The overall direction and magnitude of the associations remained consistent, indicating the robustness and reliability of the results.

A total of 1,397 women aged ≥ 20 years were included, of whom 590 (42.2%) were diagnosed with PCOS (Table 1). No significant difference was observed in age distribution between the PCOS and non-PCOS groups (P = 0.089). Compared with non-PCOS participants, those with PCOS were less likely to be married or cohabiting (55.1% vs. 68.7%) and more likely to be widowed, divorced, or separated (18.3% vs. 9.7%; P < 0.001). Higher BMI (≥ 30 kg/m²) and smoking prevalence were also more common in the PCOS group (42.9% vs. 34.2% and 46.8% vs. 35.8%, respectively; both P < 0.001), whereas drinking status, hypertension, and diabetes did not differ significantly. Biochemically, women with PCOS had lower serum VD (51.87 ± 25.50 vs. 64.45 ± 26.03 nmol/L; P < 0.001) and glutathione levels (4.63 ± 1.05 vs. 5.01 ± 0.98 μmol/L; P < 0.001), but higher testosterone (0.81 ± 0.36 vs. 0.48 ± 0.21 ng/mL; P < 0.001) and CRP levels (4.4 ± 3.1 vs. 3.5 ± 2.4 mg/L; P < 0.001). Pregnancy history differed significantly, with fewer PCOS participants having ever been pregnant (63.6% vs. 80.6%; P < 0.001). Overall, PCOS was associated with adverse demographic, lifestyle, and biochemical profiles, including higher BMI, smoking, testosterone, and CRP levels, and lower VD and glutathione levels.

The relationship between serum VD levels and the risk of polycystic ovary syndrome (PCOS) is summarized in Table 2. In the unadjusted model (Model 1), each 1 nmol/L increase in serum VD was associated with a 2% reduction in PCOS risk (OR = 0.98, P < 0.001). After sequential adjustment for demographic and lifestyle factors in Model 2 and additional comorbidities in Model 3, the inverse association was slightly attenuated but remained statistically significant (Model 3: OR = 0.99, P < 0.001), suggesting that higher VD levels may independently reduce the risk of PCOS. When serum VD was analyzed as a categorical variable, participants with normal VD levels (≥ 50 nmol/L) had a 42% lower risk of PCOS compared with those with VD deficiency (< 50 nmol/L) in the fully adjusted model (Model 3: OR = 0.58, 95% CI: 0.44–0.76, P < 0.001). Restricted cubic spline (RCS) analysis (Figure 2) revealed no evidence of a significant non-linear association between serum VD levels and PCOS risk (P for non-linearity = 0.120). The results showed a consistent downward trend in PCOS risk as VD levels increased, supporting a dose-response relationship. Overall, these findings indicate that higher serum VD levels are independently associated with a lower risk of PCOS, highlighting the potential protective role of VD in reproductive health. The results are shown in Table 2.

The subgroup analyses demonstrated several patterns of effect modification in the association between serum VD levels and PCOS. A statistically significant interaction was observed for age (P for interaction = 0.042). Among women aged < 40 years, serum VD was not significantly associated with PCOS (OR = 1.00, 95% CI: 0.99–1.01, P = 0.554), whereas a significant inverse association was observed among those aged ≥ 40 years (OR = 0.97, 95% CI: 0.96–0.99, P = 0.001). No significant interaction was detected for marital status (P for interaction = 0.204), although women who were widowed, divorced, or separated showed a marginally stronger inverse association (OR = 0.97, 95% CI: 0.95–1.00, P = 0.035). Similarly, BMI did not exhibit significant interaction with VD levels (P for interaction = 0.870). However, the association appeared strongest in women with BMI ≥ 30 kg/m² (OR = 0.98, 95% CI: 0.97–1.00, P = 0.033), suggesting a potentially greater sensitivity among individuals with obesity. No significant interactions were found for smoking status (P = 0.760), drinking status (P = 0.917), or hypertension (P = 0.726). Across these subgroups, the direction of association remained consistent, though effect sizes were modest and generally did not reach statistical significance. The results are shown in Figure 3.

Sensitivity analyses were conducted to assess the robustness of the findings. First, multivariable logistic regression analyses were repeated after excluding participants with major metabolic comorbidities, including hypertension and diabetes. Second, serum vitamin D was alternatively modeled using quartiles to evaluate whether the observed associations were sensitive to different exposure definitions. The overall direction and magnitude of the associations remained materially unchanged across all sensitivity analyses. The results are shown in Table 3.