Noseworthy, P A

Abstract Introduction Despite its prevalence, risk stratification for tachycardia-induced cardiomyopathy remains poorly understood. A deep learning model developed at Mayo Clinic uses standard 12-lead electrocardiograms (ECGs) to predict reduced LVEF (≤35%) in the general population with high accuracy (AUC 0.93) Purpose We investigated the performance of AI-ECG in predicting systolic dysfunction among patients with atrial fibrillation and rapid ventricular response, and compared it with NT-proBNP. Methods We conducted a retrospective study (IRB approval #25-000904) of adult patients (age ≥ 18 years) evaluated across our Clinic enterprise with atrial fibrillation and rapid ventricular response (heart rate >100 bpm), including AF occurring in-hospital and AF at presentation, between January 2017 and May 2025. We excluded patients with a known history of heart failure with reduced EF (<50%). We also collected data on NT-proBNP levels when available within one week of the ECG and echocardiogram. Results A total of 11,801 patients were included, with a mean age of 73.2 +- 12.6 years, 48.6% female and 94.8% White. A combination of AI ECG low EF probability cutoff of 75% had specificity of 96% and 98% NPV in ruling out LVEF < 35%. A combination of AI ECG low EF probability cutoff of 17% and LVEF <35% provided the best balance of sensitivity and specificity, 66% and 78%, respectively with an accuracy of 78%. NT-proBNP data was available on 5,279 patients (48%). Mean NT-proBNP level was 4090 pg/mL. Based on ROC curve analysis, AI ECG was superior to NT-proBNP in predicting LVEF <35% (AUC 0.72 vs 0.59), with accuracy of 77% versus 47% at Youden’s cutpoints of AI-ECG probability of 17% and NT-proBNP level of 1767 pg/mL, respectively Conclusions Our study demonstrates the utility of AI-ECG in predicting reduced LVEF among patients with AF RVR. AI-ECG demonstrated excellent specificity and NPV in ruling out reduced LVEF, serving as a powerful clinical screening tool to assist with the pharmacologic choice of rate control and could change the timing of further echocardiographic evaluation. Our study offers novel insights paving the way for future investigations to help fine-tune and further delineate the role of AI ECG in risk stratifying patients with suspected tachycardia-induced cardiomyopathy, guiding therapeutic and diagnostic clinical decisions

Abstract Introduction Atrial fibrillation (AF) is a major contributor to stroke risk, yet many cases remain undiagnosed due to its silent and paroxysmal nature. To address this, multiple risk scores have been developed to predict incident AF. This study aimed to evaluate the predictive accuracy of five widely used clinical AF prediction tools (i.e., CHARGE-AF, CHA2DS2-VASc, mC2HEST, HATCH, and HAVOC) among patients undergoing long-term continuous monitoring with implantable loop recorders (ILRs). Methods We conducted a retrospective cohort study across three major US academic medical centers, including patients with ILRs and no prior diagnosis or ECG evidence of AF. Clinical variables used to calculate each risk score were extracted from electronic health records. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC). Kaplan-Meier survival analysis and Cox proportional hazards models were used to evaluate the association between each score’s risk categories and incident AF (defined as any episode lasting ≥6 minutes). Calibration was assessed using calibration plots and Brier scores. Results A total of 3,895 patients were included (mean age 61 ± 17 yrs.; 51% male). Incident AF was detected in 879 patients (22%) over a median follow up of 17 months (IQR 8-35). Table 1 presents baseline characteristics and device detected AF event data during follow up. CHARGE-AF had the highest AUC (0.68) but also the highest Brier score (Fig. 1). Failure curves showed clear risk stratification across all tools (Fig. 2). Calibration plots (Fig. 3) demonstrated modest agreement between predicted and observed risk across all models. Using CHARGE-AF as the reference, pairwise comparisons of Brier scores via the Wilcoxon signed-rank test showed lower calibration accuracy than CHA2DS2-VASc (p = 0.03), mC2HEST (p = 0.001), and HAVOC (p = 0.001), with no significant difference compared to HATCH (p = 0.15). Conclusions In patients undergoing continuous rhythm monitoring with ILRs, commonly used AF prediction scores demonstrated modest discriminatory ability and similar calibration performance. These findings suggest that existing risk models, which were largely developed in cohorts with clinically diagnosed AF, may not optimally translate to continuously monitored populations. Future work is needed to develop and validate AF prediction models specifically in cohorts undergoing long-term continuous cardiac monitoring.Figure 1Figure 2-3