Abstract Background Type 2 diabetes mellitus (T2DM) exhibits profound geographic inequities across the United States, yet the spatiotemporal dynamics of diet-attributable burden and underlying nutrient-metabolic networks remain poorly characterized. Methods Leveraging Global Burden of Disease (GBD) 2017–2021 data (Release 2023), we quantified subnational T2DM mortality, disability-adjusted life years (DALYs), and years lived with disability (YLDs) across 51 U.S. jurisdictions. Age-standardized rates were analyzed for four dietary risks (whole-grain deficiency, sugar-sweetened beverage overconsumption, vegetable insufficiency, fiber deficiency) using Bayesian meta-regression and ordinary least squares trend modeling. Nutrient biomarker interactions were interrogated via Pearson correlation matrices, geographically weighted regression, and machine learning (elastic net-regularized logistic regression), with hierarchical clustering identifying metabolic modules. Spatial heterogeneity was assessed using choropleth mapping and Cohen’s *d* effect sizes. Results Southern states exhibited 2.3-fold higher mean T2DM burden versus national averages (peak mortality: 20.1/100,000 in West Virginia; DALYs: 1194.1/100,000). While mortality remained stable (Δ − 0.2–0.5%/year), DALYs and YLDs increased non-significantly (1.2–2.4%/year). Dietary risks demonstrated marked geospatial divergence: Southeast states manifested concurrent elevations in all four risk domains (e.g., whole-grain deficiency DALYs: 78.92 in West Virginia vs. 37.81 in Colorado), driving 2.3-fold faster aggregate DALY growth (β = 1.82, SE = 0.21, p < 0.001). Nutrient biomarker networks revealed diabetes-associated metabolic dysregulation, with vitamin B12 deficiency emerging as the strongest independent predictor (standardized β = 0.418, p < 0.001), followed by calcium and lycopene depletion. Hierarchical clustering identified three conserved nutrient modules (B-vitamin complex, antioxidant network, mineral pathway), with perturbation of the antioxidant-mineral supercluster conferring 3.7-fold higher diabetes risk (95% CI: 2.1–6.5). A multivariate biomarker model achieved robust diabetes prediction (AUC: 0.791; accuracy: 82.3%). Conclusion This study uncovers entrenched geographic disparities in T2DM burden driven by modifiable dietary risks and defines nutrient-metabolic networks underpinning diabetes pathophysiology. Our findings advocate for spatially targeted interventions prioritizing micronutrient sufficiency and whole-food accessibility in high-risk regions.
Liu et al. (Sat,) studied this question.