Abstract
OBJECTIVE: To investigate the association of triglyceride-to-high-density lipoprotein cholesterol (TG/HDL-C) and metabolic score for insulin resistance (METS-IR) with diabetic retinopathy (DR) in patients with type 2 diabetes (T2DM), to identify factors independently associated with DR, and to explore their relationship with DR severity. METHODS: This study enrolled 329 hospitalized patients with T2DM (75 with DR, 254 without). LASSO regression was applied to select factors associated with DR, followed by multivariable logistic regression. Model performance was assessed using area under the ROC curve (AUC), bootstrap internal validation (1,000 resamples) for optimism correction, calibration plots, and decision curve analysis (DCA). For the exploratory analysis of proliferative diabetic retinopathy (PDR, n = 32), Firth's penalized logistic regression was used to identify associated factors. RESULTS: LASSO identified age, fasting C-peptide, TG/HDL-C and METS-IR as factors associated with DR. Multivariable logistic regression confirmed that age (OR = 1.05 per year, 95% CI 1.02-1.08), TG/HDL-C (OR = 1.58 per unit, 1.33-1.88) and METS-IR (OR = 1.08 per unit, 1.01-1.15) were positively associated with DR, whereas fasting C-peptide showed an inverse association (OR = 0.11 per ng/mL, 0.04-0.29). The model achieved an AUC of 0.934 (95% CI 0.905-0.963), with an optimism-corrected C-index of 0.929 (0.887-0.971) after bootstrap validation. Calibration was satisfactory (intercept = 0.00, slope = 1.00, Brier score = 0.08), and DCA indicated positive net benefit within the 5-30% threshold range. Exploratory analysis of PDR using Firth's penalized regression showed no statistically significant association for TG/HDL-C (OR = 3.10, 1.77-7.21, p < 0.001), METS-IR (OR = 1.80, 1.31-3.07), likely due to limited sample size. CONCLUSION: TG/HDL-C and METS-IR are independently associated with DR in hospitalized patients with T2DM. The model incorporating these factors demonstrates good discrimination and calibration, with potential utility for risk stratification. Exploratory analysis of PDR, constrained by small sample size, did not confirm significant associations, underscoring the need for larger studies on progression to proliferative disease.