Machine learning evaluation of TyG-based metrics for arteriosclerosis progression

Insulin resistance (IR) is a significant risk factor for arteriosclerosis. The triglyceride-glucose (TyG) index and its obesity-related derivatives (TyG-BMI, TyG-WC, and TyG-WHtR) have emerged as reliable markers of IR. While diabetes, a consequence of IR, is a primary risk factor for arteriosclerosis, it is important to note that arteriosclerosis may develop prior to the onset of diabetes. Therefore, this study utilized machine learning to assess the relationship between cumulative exposure to these IR indicators and the development of arteriosclerosis in non-diabetic populations. This cohort study enrolled 4,160 participants from the health management departments of two tertiary general hospitals in southern China between 2017 and 2024, with an average follow-up of 3.49 visits per participant. Multivariable linear regression models and restricted cubic splines were employed to investigate the association between cumulative average exposure to TyG and obesity-modified TyG indicators with arteriosclerosis progression and explore potential nonlinear relationships. Machine learning models were utilized to assess the predictive value of cumulative average exposure to the TyG index and its derived metrics for arteriosclerosis.its performance was evaluated in both the development and validation cohorts using the area under the receiver operating characteristic curve (AUC), calibration curves, and the area under the precision-recall curve (PR-AUC). Cumulative average exposure to the TyG index and TyG-WHtR demonstrated a strong association with the progression of arteriosclerosis, whether analyzed as continuous variables or as quartile variables (p < 0.05). A potential U-shaped relationship was observed between cumulative average exposure to TyG-WHtR and both changes in baPWV and its annual rate of change. Validation in an independent cohort demonstrated that incorporating the TyG index or its derived metrics into predictive models moderately enhanced their performance in predicting arteriosclerosis progression, with the AUC increasing from 0.726 for the baseline model to a range of 0.737–0.744. The cumulative average exposure to the TyG index and its derivative indicators is a significant predictor of arteriosclerosis progression in non-diabetic populations. Integrating these low-cost and readily available metrics into predictive models enables the early identification of high-risk individuals within this cohort.