BACKGROUND: Cardiometabolic multimorbidity (CMM) poses a growing global health burden, yet few studies have combined the Triglyceride-Glucose (TyG) index, which reflects metabolic dysfunction, with the Frailty Index (FI), which captures physiological reserve and aging-related vulnerability, to assess CMM risk. Given their complementary biological information, this study examines whether a composite TyG-FI index is associated with incident CMM and whether it improves risk stratification beyond established factors. METHODS: This prospective cohort study analyzed data from Chinese adults aged ≥ 45 years in the 2011-2020 waves of the China Health and Retirement Longitudinal Study (CHARLS). To assess the association between the TyG-FI index and incident CMM, we used Kaplan-Meier survival curves and multivariable Cox proportional-hazards models adjusted for potential confounders; restricted cubic spline analyses were employed to explore non-linear relationships. Predictive performance was evaluated using eight machine-learning algorithms: CatBoost, Extra Trees, Random Forest (RANGER), XGBoost, Recursive Partitioning (RPART), k-Nearest Neighbors (KKNN), Neural Network (NNET), and Support Vector Machine (SVM). Subgroup and sensitivity analyses were conducted to test the robustness of the results across population subgroups and modeling choices. RESULTS: The analytic cohort comprised 2961 adults. Kaplan-Meier curves showed a graded, significant increase in cumulative CMM incidence across TyG‑FI quartiles (log‑rank P < 0.001). In multivariable Cox models, each unit increase in TyG‑FI was associated with a 1.80-fold higher CMM risk (HR = 1.80, 95% CI 1.57-2.05; P < 0.001); participants in the highest quartile had markedly elevated risk versus the lowest (Q4 vs. Q1 HR = 7.86, 95% CI 4.16-14.86). Restricted cubic spline analyses revealed significant non-linear relationships (P for non-linearity < 0.001), showing a J-shaped association between TyG-FI and CMM with threshold effects at TyG-FI ≈ 0.7 and cumulative TyG-FI ≈ 2.7. Subgroup analyses indicated stronger associations in participants < 60 years and in normotensive individuals. TyG-FI demonstrated better predictive performance for CMM than TyG index or FI alone, with improved C-statistic, Integrated Discrimination Improvement (IDI), and Net Reclassification Improvement (NRI). Among machine-learning models, RANGER performed best (AUC ≈ 0.81), and SHAP analysis identified cumulative and baseline TyG-FI as the primary predictors. Findings were robust in sensitivity analyses. CONCLUSIONS: TyG-FI exhibits non‑linear, threshold-defined associations with incident CMM and age‑dependent effect modification. Machine‑learning models incorporating TyG-FI show strong predictive performance. TyG-FI assessment may facilitate cost‑effective risk stratification for CMM and guide targeted prevention.
Hou et al. (Sun,) studied this question.