Introduction An artificial intelligence (AI) algorithm that utilizes data contained within a single 12-lead ECG during normal sinus rhythm has recently shown to identify patients with occult AF. Patients with HFpEF frequently display AF in association with left atrial (LA) dysfunction. Hypothesis Patients with HFpEF and a higher probability of AF according to their AI-enabled ECG display more advanced LA remodeling and dysfunction, putting them at greater risk for AF development. Methods This retrospective cohort study includes 613 patients with definite HFpEF: 184 ambulatory patients diagnosed through invasive hemodynamic testing and 429 inpatients admitted because of decompensated HFpEF. A previously validated AI algorithm that predicts current AF risk based on ECG analysis was applied to estimate future AF risk during long-term follow-up, with patients grouped into quartiles of risk. Cardiac structure and function, as well as incident AF rates were compared. Results AI-predicted AF risk was 42% (14−69%) overall. Patients with higher risk were older, with more comorbidities, lower blood pressure, higher natriuretic peptide levels, greater conduction delays, a higher prevalence of pacing, and more frequent use of diuretics. In addition, underlying structural heart disease was more severe, with more pronounced left ventricular hypertrophy, larger LA volumes, lower LA reservoir and booster strain, higher cardiac filling pressures (confirmed through invasive measurements ambulatory patients), and more tricuspid regurgitation. Echocardiography images from a representative patient within each AF risk quartile are presented in the Figure. Invasive hemodynamic measurements in the ambulatory cohort showed additionally that patients with higher AI-predicted AF risk had lower cardiac output (10.01±2.85L/min and 6.94±2.49L/min in the lowest and highest risk quartile, respectively) and a higher pulmonary vascular resistance at peak effort (121±102dynes/s/cm−5 and220±133 dynes/s/cm−5 in the lowest and highest risk quartile, respectively). Over a median follow-up of 41months (11-71months), AI-predicted AF risk was associated with new-onset AF [HR(95%CI) = 1.31(1.20-1.41); p<0.001] and all-cause death [HR (95%CI) = 1.12(1.08-1.17); p<0.001]. Conclusions An AI-risk score based upon a single 12-lead ECG reflects the severity of underlying LA myopathy in HFpEF and predicts new-onset AF. These data suggest that application of an AI-enabled ECG may be useful to identify patients with the AF/LA myopathy phenotype in HFpEF. An artificial intelligence (AI) algorithm that utilizes data contained within a single 12-lead ECG during normal sinus rhythm has recently shown to identify patients with occult AF. Patients with HFpEF frequently display AF in association with left atrial (LA) dysfunction. Patients with HFpEF and a higher probability of AF according to their AI-enabled ECG display more advanced LA remodeling and dysfunction, putting them at greater risk for AF development. This retrospective cohort study includes 613 patients with definite HFpEF: 184 ambulatory patients diagnosed through invasive hemodynamic testing and 429 inpatients admitted because of decompensated HFpEF. A previously validated AI algorithm that predicts current AF risk based on ECG analysis was applied to estimate future AF risk during long-term follow-up, with patients grouped into quartiles of risk. Cardiac structure and function, as well as incident AF rates were compared. AI-predicted AF risk was 42% (14−69%) overall. Patients with higher risk were older, with more comorbidities, lower blood pressure, higher natriuretic peptide levels, greater conduction delays, a higher prevalence of pacing, and more frequent use of diuretics. In addition, underlying structural heart disease was more severe, with more pronounced left ventricular hypertrophy, larger LA volumes, lower LA reservoir and booster strain, higher cardiac filling pressures (confirmed through invasive measurements ambulatory patients), and more tricuspid regurgitation. Echocardiography images from a representative patient within each AF risk quartile are presented in the Figure. Invasive hemodynamic measurements in the ambulatory cohort showed additionally that patients with higher AI-predicted AF risk had lower cardiac output (10.01±2.85L/min and 6.94±2.49L/min in the lowest and highest risk quartile, respectively) and a higher pulmonary vascular resistance at peak effort (121±102dynes/s/cm−5 and220±133 dynes/s/cm−5 in the lowest and highest risk quartile, respectively). Over a median follow-up of 41months (11-71months), AI-predicted AF risk was associated with new-onset AF [HR(95%CI) = 1.31(1.20-1.41); p<0.001] and all-cause death [HR (95%CI) = 1.12(1.08-1.17); p<0.001]. An AI-risk score based upon a single 12-lead ECG reflects the severity of underlying LA myopathy in HFpEF and predicts new-onset AF. These data suggest that application of an AI-enabled ECG may be useful to identify patients with the AF/LA myopathy phenotype in HFpEF.
Read full abstract