Association of the Triglyceride–Glucose Index with Premature Myocardial Infarction in a Propensity Score-Matched Non-Diabetic Population

Background/Objectives: Evidence regarding the triglyceride–glucose (TyG) index and premature myocardial infarction (MI) in non-diabetic adults remains limited. We evaluated whether pre-event TyG values were associated with premature MI in a propensity score-matched population. Methods: We retrospectively compared premature MI cases with non-diabetic controls undergoing coronary computed tomography angiography (CCTA) showing either no plaque or mild/non-obstructive plaque. Propensity-based 1:1 matching on age, sex, body mass index (BMI), smoking, and hypertension resulted in 114 matched pairs. Conditional logistic regression, dose–response analyses, restricted cubic spline (RCS) modeling, and incremental discrimination analyses were performed. Results: After propensity score matching, 228 patients (114 premature MI cases and 114 controls) were analyzed. The TyG index was independently associated with premature MI in conditional logistic regression (adjusted OR 2.704, 95% CI 1.408–5.192; p = 0.003). Compared with the lowest TyG tertile, the odds of premature MI increased progressively across higher tertiles (T2 vs. T1: adjusted OR 2.859, 95% CI 1.376–5.940; p = 0.005; T3 vs. T1: adjusted OR 4.424, 95% CI 1.859–10.531; p < 0.001). Restricted cubic spline analysis demonstrated a significant overall association between the TyG index and premature MI (p = 0.003), with no evidence of nonlinearity (p = 0.286). Adding the TyG index to the clinical model increased the AUC from 0.647 to 0.696 (ΔAUC = 0.050; DeLong p = 0.046). Matched-pair bootstrap analysis also supported an improvement in discrimination (95% CI for ΔAUC 0.002–0.096; p = 0.034). Conclusions: Higher pre-event TyG values were independently associated with premature MI in non-diabetic adults, with a graded association across TyG levels and no significant evidence of nonlinearity. Incremental discrimination beyond the clinical model was modest but statistically supported by both DeLong and matched-pair bootstrap analyses. Prospective multicenter validation is warranted.