Abstract
OBJECTIVE: To determine whether a low ejection fraction artificial intelligence electrocardiogram (AI-ECG) algorithm predicts incident heart failure with reduced ejection fraction (HFrEF) in patients with atrial fibrillation (AF) independently of known HF risk factors. PATIENTS AND METHODS: A validated AI-ECG algorithm for low ejection fraction (EF) detection was applied to a cohort of patients aged 18 years or older from Olmsted County, Minnesota, with incident AF between January 1, 2000, and December 31, 2014. Risk of HFrEF by tertiles of AI-ECG score were calculated, and model performance was assessed using C-statistic estimates derived from Cox proportional hazard regression models. RESULTS: Among 2569 patients with AF, 248 (9.7%) developed incident HFrEF (EF<50%) over an average follow-up of 7.0 years. Patients in the highest risk AI-ECG tertile had an increased risk of developing HFrEF in comparison with those in the lowest risk tertile after adjusting for common HF risk factors (hazard ratio [HR], 3.50; 95% CI, 2.51-4.87). The C-statistic was significantly higher in a model combining the AI-ECG + HF risk factors (0.76; 95% CI, 0.73-0.80) than a model with HF risk factors only (0.67; 95% CI, 0.63-0.71; P<.0001) and a model with the AI-ECG only (0.73; 95% CI, 0.69-0.76; P=.019). Similar results were observed for HF with EF of 35% or less (highest vs lowest risk AI-ECG tertile: HR, 5.20; 95% CI, 3.03-8.91). CONCLUSION: Incorporation of the AI-ECG algorithm into routine clinical care may provide enhanced ability to identify patients with AF at risk of developing HFrEF with predictive performance that is superior to current clinical models.