Loneliness severity and heart disease risk in older Chinese rural adults: a machine learning-based cross-sectional study

Figure 1. Odds ratios (ORs) for heart disease by loneliness severity from covariate-adjusted logistic models stratified by residence (rural vs. urban); reference = no loneliness. Adjusted for age, sex, current smoking, current drinking, hypertension (HTN), diabetes mellitus (DM), dyslipidemia, and stroke. Points denote ORs; error bars denote 95% confidence intervals.

Figure 2. Model performance comparison (heatmap) showing area under the receiver operating characteristic curve (ROC-AUC), area under the precision-recall curve (PR-AUC), accuracy, precision, recall, and F1 on the held-out test set.

Figure 3. SHAP-based interpretability for the top two models (Panel A and Panel B: LR; Panel C and Panel D: GBDT). Beeswarm panels display the distribution and direction of SHAP values; bar panels show mean absolute SHAP values for aggregated feature importance.