Linear dose-response relationship between TyG-WC index and diastolic dysfunction: insights from restricted cubic spline and machine learning analysis

Figures

• 

  Figure 1. Receiver operating characteristic (ROC) curves for TyG-WC predicting diastolic dysfunction. ROC curves with 1,000 bootstrap resamples illustrating the discriminative performance of TyG-WC for diastolic dysfunction (DD) across three nested logistic regression models. Model 1 (TyG-WC only); Model 2 (TyG-WC + age + sex); Model 3 (TyG-WC + age + sex + SBP + TC + LDL-C + HDL-C + TG + current smoking + alcohol use + hypertension + diabetes + CHD + stroke). The red dot on the Model 1 curve denotes the optimal cutoff determined by the Youden index. Light blue lines represent individual bootstrap iterations; the dark blue line represents the original ROC curve. The red dashed diagonal line indicates the reference line of no discrimination (AUC = 0.5).

• 

  Figure 2. Restricted cubic spline (RCS) analysis of the association between TyG-WC and diastolic dysfunction. (A) Overall dose-response relationship between TyG-WC and the predicted probability of DD, adjusted for age, sex, SBP, TC, LDL-C, HDL-C, TG, current smoking, alcohol use, hypertension, diabetes, CHD, and stroke. The solid blue line represents the fitted RCS curve with 3 knots placed at the 10th, 50th, and 90th percentiles of TyG-WC; the shaded area represents the 95% confidence interval. The gray histogram depicts the distribution of TyG-WC. The dashed horizontal line indicates the overall prevalence of DD. P-overall (< 0.001) and P for non-linearity (0.088) are annotated. (B) Age-stratified RCS curves: < 65 years (blue) vs. ≥ 65 years (orange). (C) Sex-stratified RCS curves: male (teal) vs. female (coral). (D) Combined comparison of age and sex subgroup RCS curves.

• 

  Figure 3. Scatter-density plots of the association between TyG-WC and echocardiographic parameters of diastolic function. Two-dimensional density scatter plots with LOESS smoothing (black line) showing the relationship between TyG-WC and four echocardiographic indicators: (A) lateral mitral annular e′ velocity, (B) septal mitral annular e′ velocity, (C) average mitral annular e′ velocity, and (D) mitral E/e′ ratio. Color gradients represent local point density from low (dark blue) to high (yellow). Marginal histograms along the top and right axes display the univariate distribution of each variable. All analyses are based on complete cases (N = 1,413).

• 

  Figure 4. Subgroup analysis and interaction assessment for the association between TyG-WC and diastolic dysfunction. Forest plots displaying adjusted odds ratios (OR) and 95% confidence intervals for the association between TyG-WC and DD across prespecified subgroups. Panel A1: OR per 1-unit increase in continuous TyG-WC. Panel A2: OR for high vs. low TyG-WC. Subgroups include sex (male/female), age (< 65/≥ 65 years), current smoking (no/yes), alcohol use (no/yes), hypertension (no/yes), diabetes (no/yes), coronary heart disease (no/yes), and stroke (no/yes). P for interaction was derived from likelihood ratio tests comparing models with and without the interaction term. The dashed vertical line at OR = 1.0 indicates the null effect. Lower panels display predicted probability curves illustrating interaction effects between TyG-WC and key continuous covariates (SBP, DBP, LDL-C, HDL-C, TC, TG) across quartiles.

• 

  Figure 5. Machine learning analysis: LASSO feature selection, random forest prediction, and SHAP interpretation for TyG-WC and diastolic dysfunction. (A) LASSO regression coefficient profile. Variables with non-zero coefficients were selected by the optimal lambda (lambda.1se) in a least absolute shrinkage and selection operator (LASSO) model with TyG-WC forced into the model (penalty factor = 0). Bar color indicates the direction of association (blue/red = negative/positive coefficient). (B) Receiver operating characteristic curve for the random forest model constructed using LASSO-selected features, evaluated on the held-out test set. AUC is annotated on the plot. (C) SHAP (SHapley Additive exPlanations) bee swarm plot showing the distribution of SHAP values for each feature in the random forest model. Features are ranked by mean absolute SHAP value (descending). Each point represents one observation; color encodes the SHAP value (blue = low, red = high), with higher SHAP values indicating a greater contribution to the predicted probability of DD.