Association between cardiometabolic index and incident electrocardiographic abnormalities in older adults

Abstract

Electrocardiographic (ECG) abnormalities are important predictors of cardiovascular events in older adults, yet their early identification remains challenging. The cardiometabolic index (CMI), which integrates central obesity and lipid metabolism, has emerged as a potential marker of cardiometabolic risk, but its association with incident ECG abnormalities has not been well established. In this prospective cohort study, 10,028 adults aged ≥ 60 years from Nan'an Hospital (2022–2024) were followed for two years. Incident ECG abnormalities—including ST-T changes, conduction block, atrial fibrillation, premature ventricular contractions, and other abnormalities—were identified by comparing baseline and follow-up ECGs. Six machine learning models (GBDT, LightGBM, XGBoost, random forest, decision tree, and logistic regression) were developed using 14 features retained after multicollinearity screening. Model performance was assessed by discrimination, calibration, and decision curve analysis, and SHapley Additive exPlanations (SHAP) were applied for interpretability. The association between CMI and incident ECG abnormalities was further examined using quartile and subtype analyses. During follow-up, 14.19% of participants developed ECG abnormalities. The gradient boosting decision tree (GBDT) model achieved the best overall discrimination (validation AUC 0.902, 95% CI 0.886–0.917; training AUC 0.929) and was well calibrated (calibration slope 0.98; Brier score 0.070). SHAP analysis identified hypertension as the most influential predictor, followed by CMI, alanine aminotransferase, and glucose. CMI showed an independent, graded association with incident ECG abnormalities: compared with the lowest quartile, the fully adjusted odds ratio rose progressively to 4.08 (95% CI 3.29–5.05) in the highest quartile ( P for trend < 0.001), and this association was consistent across all ECG subtypes rather than driven by a single phenotype. Higher CMI is independently associated with incident ECG abnormalities in older adults, showing a dose–response relationship that is consistent across ECG subtypes. Machine learning models incorporating CMI, particularly GBDT, provided good discrimination and calibration for these outcomes. As these findings are based on internal validation only, external validation in independent cohorts is needed before clinical application can be considered.