Stratifying risk of heart failure death and arrhythmic events: a ¹²³I-meta-iodobenzylguanidine-based multinomial logistic model

Differentiating heart failure-related death (HFD) from arrhythmic events (ArEs) is clinically important for patients with chronic heart failure (CHF), as they have distinct mechanisms and therapeutic strategies. We developed and validated a multivariable model to predict HFD, ArEs, and survival using clinical parameters and cardiac 123 I- meta -iodobenzylguanidine ( 123 I- m IBG) images. We retrospectively analyzed data derived from 997 patients with CHF (mean age 70 ± 13 years, left ventricular ejection fraction (LVEF) 32% ± 13%) over a mean follow-up of 41 ± 27 months. Outcomes were survival, HFD, or ArEs (including sudden cardiac death). Appropriate implantable cardioverter defibrillator therapy for lethal arrhythmias was included in ArEs. Late heart-to-mediastinum ratios (HMRs) were derived from 123 I- m IBG images. A multinomial nested logistic regression model using 2 years of outcomes was constructed ( N = 854). Internal validation used a 2:1 development-validation split, repeated 3 times. Model performance was assessed by receiver operating characteristic (ROC) analysis, calibration of predicted vs. actual event rates, survival curves, and sex-specific predictive models. Selected variables were age, sex, New York Heart Association (NYHA) functional class, LVEF, hemoglobin, estimated glomerular filtration rate, hypertension, ventricular tachycardia history, and late 123 I- m IBG HMR. Areas under ROC curves for survival, HFD, and ArEs in the final 9-variable model were 0.800, 0.717, and 0.838, respectively. The sex-specific 7-variable models showed comparable AUCs of 0.834/0.827 (male/female) for HFD and 0.714/0.826 for ArEs. Risk groups based on median predicted probabilities of HFD and ArEs separated survival curves and corresponded well with actual outcomes. A practical, interpretable model incorporating clinical and 123 I- m IBG imaging data enabled reliable and separate prediction of HFD and ArEs, supporting personalized risk stratification in CHF.