Incremental Predictive Value of ECG Burden Added to Clinical Risk for Appropriate ICD Therapy Selection: Model Development and Internal Validation
TL;DR
Adding the ECG burden score did not improve discrimination beyond minimal clinical predictors and provided only minimal incremental clinical utility in predicting appropriate ICD therapy selection.
Abstract
Background: Risk stratification for selecting appropriate implantable cardioverter-defibrillator (ICD) therapy remains imperfect, and pragmatic electrocardiogram (ECG)-derived markers may provide incremental information beyond routine clinical variables. Objective: We sought to evaluate whether a prespecified 6-component ECG burden score improves the prediction of appropriate ICD therapy selection beyond a minimal clinical model. Methods: Retrospective prediction study using deidentified data from a parent ICD cohort at King Abdulaziz Medical City, Riyadh (n=236; 35 events; mean [SD] follow-up, 7.3 [2.3] years), analyzed under King Abdulaziz Hospital/KAIMRC governance (Protocol No. NRA 26/004/1). Minimal clinical predictors (age, sex, left ventricular ejection fraction, cardiomyopathy type, and ICD indication) were modeled with ridge logistic regression (clinical only) and with the addition of ECG burden (combined). Internal validation used 5-fold cross-validation with out-of-fold predictions. Results: In complete-case modeling (n=231), out-of-fold discrimination was 0.502 for clinical only and 0.452 for combined (ΔAUC, −0.050; 95% CI, −0.097 to −0.008). In the imputed sensitivity analysis (n=236), discrimination was 0.514 vs 0.485 (ΔAUC, −0.028; 95% CI, −0.066 to 0.008). Decision curve analysis showed minimal separation between the models. Conclusion: In internal validation, adding the ECG burden score did not improve discrimination beyond minimal clinical predictors and provided only minimal incremental clinical utility in predicting appropriate ICD therapy selection.