Triglyceride-to-HDL Ratio Outperforms Individual Lipid Parameters and HbA1c in Predicting Diabetes Complications: Machine-Learning Analysis

Introduction and Objective: Individual lipid parameters and HbA1c show modest predictive value for complications, yet optimal combination remains unclear. We employed machine learning to identify the strongest predictor of multi-system complications in community-based type 2 diabetes, comparing 15 parameters including novel lipid ratios. Methods: Machine learning analysis of 400 T2DM patients using Random Forest with SHAP importance. Input: HbA1c, glucose, duration, age, BMI, lipids (TG, TC, LDL, HDL, VLDL), plus ratios (TG/HDL, TC/HDL, LDL/HDL, non-HDL). Outcome: multi-system complications (≥2 of bone loss, neuropathy, dyslipidemia). 10-fold cross-validation, AUROC assessment, optimal cutpoints via Youden index. Results: TG/HDL ratio showed highest importance (0.224) versus individual TG (0.168), BMI (0.160), HbA1c (0.085). AUROC: 0.82 for TG/HDL alone versus 0.61 for HbA1c (p<0.001). Optimal cutpoint TG/HDL >3.5: sensitivity 79%, specificity 76%, OR 4.82 (CI 2.89-8.04, p<0.001). Among HbA1c<7% patients (n=168), TG/HDL >3.5 identified high-risk with 58% complications versus 24% (p<0.001). Among HbA1c≥8% (n=72), TG/HDL ≤3.5 showed 42% complications versus 81% (p<0.001). Individual complications: osteoporosis OR 3.9 (CI 2.1-7.2), neuropathy OR 4.1 (CI 2.3-7.4). Mean TG/HDL: no complications 2.1±0.8, single 3.2±1.1, multiple 4.8±1.6 (p<0.001). Conclusion: Triglyceride-to-HDL ratio is the strongest single predictor of complications, outperforming HbA1c and individual lipids. Simple cutpoint (TG/HDL >3.5) identifies high-risk patients despite optimal glycemic control. Findings support incorporating TG/HDL ratio into routine risk assessment and targeting TG/HDL reduction as primary therapeutic strategy. Disclosure: A. Bishnoi: None. S. Patil: None. S. Singla: None. A. Raha: None. S.D. Sonawala: None. R. Viegas: None. A. Maheshwari: None.