@article{article_1649690, title={Predictive Modeling of Diabetes Status Based on Vitamin D Levels and Clinical Parameters: A Machine Learning Investigation}, journal={Black Sea Journal of Engineering and Science}, volume={8}, pages={1985–1997}, year={2025}, DOI={10.34248/bsengineering.1649690}, author={Sertbakan, Kübra and Balcıoğlu, Yavuz Selim and Sezen, Bülent and Garip, Tayfun and Aslanbaş, Çağlayan}, keywords={Vitamin D deficiency, Diabetes mellitus, HbA1c, Machine learning, Predictive modeling}, abstract={Vitamin D deficiency has been associated with impaired glucose metabolism, but its predictive value for diabetes status remains incompletely characterized. We applied machine learning methodologies to investigate this relationship and develop predictive models based on vitamin D levels and clinical parameters. This cross-sectional study analyzed data from 817 patients with concurrent measurements of 25-hydroxyvitamin D and HbA1c. Patients were classified as having normal glucose metabolism (HbA1c <5.7%), prediabetes (HbA1c 5.7-6.4%), or diabetes (HbA1c≥6.5%). Logistic regression models of increasing complexity were developed to predict diabetes status, and various vitamin D thresholds were evaluated to determine the optimal cutoff for diabetes prediction. Mean vitamin D levels differed significantly across glycemic categories (normal: 24.57 ng/mL, prediabetes: 25.41 ng/mL, diabetes: 21.85 ng/mL; ANOVA: F(2,814)=4.68, P <0.01). Model-specific performance analysis revealed limited discriminative ability across all models: basic vitamin D model (AUC: 0.55, 95% CI: 0.51-0.59), demographic model incorporating age and gender (AUC: 0.58, 95% CI: 0.54-0.62), and comprehensive model with additional biomarkers (AUC: 0.62, 95% CI: 0.56-0.68). In the logistic regression model, each 1 ng/mL increase in vitamin D was associated with a 4% decrease in diabetes odds (OR: 0.96, 95% CI: 0.93-0.99). The statistically optimal vitamin D threshold for diabetes prediction was 14.0 ng/mL (sensitivity: 31.9%, specificity: 89.3%), differing from the conventional clinical cutoff of 20 ng/mL. However, the weak discriminative ability indicates that vitamin D has limited standalone predictive value for diabetes status and should be considered only as part of comprehensive risk assessment frameworks. The cross-sectional design precludes causal inferences, and the modest performance suggests limited immediate clinical applicability as an isolated predictor.}, number={6}, publisher={Karyay Karadeniz Yayımcılık Ve Organizasyon Ticaret Limited Şirketi}