Low vitamin D status and cardiometabolic trait clustering in a primary-care adult cohort: association analysis and exploratory explainable machine-learning classification.

BACKGROUND: Low serum 25-hydroxyvitamin D (25(OH)D) commonly co-exists with obesity and adverse cardiometabolic traits. However, its cross-sectional association with dyslipidaemia, body composition and glycaemic markers in primary-care adults remains incompletely characterised. This study examined these associations as the primary objective and, as a secondary exploratory objective, assessed whether routine clinical variables could classify concurrent vitamin D deficiency using interpretable machine-learning models. METHODS: In this cross-sectional study, adults attending routine health check-up services underwent anthropometric and body-composition assessment (BMI, body fat mass, body fat percentage and lean mass), metabolic-profile screening and vitamin-status assessment. Vitamin D status was classified as deficient (< 20 ng/mL), insufficient (20-29.9 ng/mL) or sufficient (≥ 30 ng/mL). Group comparisons, correlation analyses and multivariable regression models were performed. In a secondary exploratory analysis, logistic-regression and random-forest models were used to classify vitamin D deficiency from routine clinical variables after internal cross-validation. RESULTS: In the overall analytic sample, the mean serum 25(OH)D level was 20.3 ± 8.0 ng/mL, and 50.3% of participants were obese. Vitamin D deficiency was associated with significantly higher BMI, body fat mass, body fat percentage, fasting blood glucose, HbA1c, total cholesterol (TC), low-density lipoprotein cholesterol (LDL-c), triglycerides (TG) and TG-derived very-low-density lipoprotein cholesterol (VLDL-c) and with significantly lower high-density lipoprotein cholesterol (HDL-c) (all p < 0.05). Serum 25(OH)D was inversely correlated with adiposity indices and TG and positively correlated with HDL-c. In the secondary exploratory classification analysis, the dimension-reduced random-forest model showed higher internal discrimination than logistic regression for classifying vitamin D deficiency in this dataset. CONCLUSIONS: Primary-care cohort with a high obesity burden, lower vitamin D status was associated with greater adiposity, less favourable glycaemic indices and a more atherogenic lipid profile. The exploratory machine-learning analysis suggests that routine nutritional, body-composition and biochemical variables may help classify concurrent vitamin D deficiency. However, these findings require external validation and should not be interpreted as evidence of causality or immediate clinical utility.