Electrical risk score and electrocardiographic markers in acute cholecystitis: a prospective cross-sectional study using explainable machine learning.

backgroundThis study investigated the association between the Electrical Risk Score (ERS), derived from 12-lead electrocardiography (ECG), and electrocardiographic markers related to ventricular arrhythmia risk in patients with acute cholecystitis (AC), using conventional statistical analyses and machine learning–based exploratory analyses. Given the cross-sectional study design, clinical cardiovascular outcomes and longitudinal prognostic effects were not assessed.

methodsIn this prospective cross-sectional study, 153 patients diagnosed with AC and 100 age- and sex-matched healthy controls were enrolled. The ERS was calculated as the sum of six binary ECG parameters (range 0–6), and its individual components were compared between groups. The Tp–e interval (reflecting transmural dispersion of ventricular repolarization) and the frontal QRS–T angle (representing spatial heterogeneity between depolarization and repolarization vectors) were also analyzed. Six ensemble ML models including CatBoost and XGBoost were trained to estimate ERS variability using demographic, laboratory, and ECG variables. Model interpretation was performed using SHAP (Shapley Additive Explanations) analysis to improve transparency of model behavior.

resultsERS was significantly higher in the AC group (2.1 ± 1.0) compared with controls (1.36 ± 0.71; p < 0.001). QT and QTc intervals, frontal QRS–T angle, and delayed QRS transition zone were significantly increased in AC patients. Laboratory markers including glucose, AST, ALT, hs-CRP, and WBC counts were also elevated in the AC group. Multivariate regression analysis identified AC, age, and AST as independent factors associated with ERS. Among ML models, CatBoost demonstrated the highest internal predictive performance for ERS estimation (R² = 0.9929, RMSE = 0.0578). Because ERS is derived from electrocardiographic parameters that were also included among model inputs, these predictive metrics should be interpreted cautiously. SHAP analysis indicated that heart rate, Tp–e interval, QTc interval, and frontal QRS–T angle were the most influential contributors to ERS estimation. Sensitivity analyses excluding ERS component variables suggested that systemic factors such as electrolyte balance and inflammatory markers may contribute to ERS variability, although their influence was weaker than that of electrophysiological parameters. Clinical endpoints such as arrhythmia, sudden cardiac death, or mortality were not evaluated.

conclusionsERS values were significantly elevated in patients with AC and were associated with ECG markers reflecting ventricular repolarization heterogeneity. These findings indicate a cross-sectional electrophysiological association rather than evidence of prognostic utility. The integration of explainable ML techniques provided an interpretable exploratory framework for examining ERS variability, although no inference regarding prediction of future cardiovascular events can be drawn from the present data. Prospective longitudinal studies are required to determine whether ERS has independent clinical or prognostic relevance in this population.