Improved prediction of sudden cardiac death in patients with heart failure through digital processing of electrocardiography

Abstract

Aims Available predictive models for sudden cardiac death (SCD) in heart failure (HF) patients remain suboptimal. We assessed whether the electrocardiography (ECG)-based artificial intelligence (AI) could better predict SCD, and also whether the combination of the ECG-AI index and conventional predictors of SCD would improve the SCD stratification among HF patients. Methods and results In a prospective observational study, 4 tertiary care hospitals in Tokyo enrolled 2559 patients hospitalized for HF who were successfully discharged after acute decompensation. The ECG data during the index hospitalization were extracted from the hospitals’ electronic medical record systems. The association of the ECG-AI index and SCD was evaluated with adjustment for left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) class, and competing risk of non-SCD. The ECG-AI index plus classical predictive guidelines (i.e. LVEF ≤35%, NYHA Class II and III) significantly improved the discriminative value of SCD [receiver operating characteristic area under the curve (ROC-AUC), 0.66 vs. 0.59; P = 0.017; Delong’s test] with good calibration (P = 0.11; Hosmer–Lemeshow test) and improved net reclassification [36%; 95% confidence interval (CI), 9–64%; P = 0.009]. The Fine–Gray model considering the competing risk of non-SCD demonstrated that the ECG-AI index was independently associated with SCD (adjusted sub-distributional hazard ratio, 1.25; 95% CI, 1.04–1.49; P = 0.015). An increased proportional risk of SCD vs. non-SCD with an increasing ECG-AI index was also observed (low, 16.7%; intermediate, 18.5%; high, 28.7%; P for trend = 0.023). Similar findings were observed in patients aged ≤75 years with a non-ischaemic aetiology and an LVEF of >35%. Conclusion To improve risk stratification of SCD, ECG-based AI may provide additional values in the management of patients with HF.

Sudden cardiac death after heart transplantation: a population-based study

“… 22 Second, new tools such as machine learning may improve statistical performance of models since they have been shown to outperform traditional models in various areas. 23 Third, the integration of signal analysis by artificial intelligence may represent a major advance in this our field. 24 Altogether, these approaches might lead to better risk stratification and therefore earlier detection of high-risk patients.…”

Time-Dependent ECG-AI Prediction of Fatal Coronary Heart Disease

“…In some published AI-driven FCHD or SCD prediction research, very high accuracies (AUC>0.85) were achieved hours before the event, which reduces usability for therapeutic interventions as well as they are not applicable to population level screening 5,15,29 . However, results from our research show sustained high AUC of 0.91 for 2-year risk prediction of FCHD.…”

“…There is potential for the development of AI models for pre-screening during any clinical service to assess the risk of FCHD/SCD at an early enough stage to allow for prevention or timely therapeutics. Current research has tried to predict FCHD risk using clinical risk factors 14 , ECG features as biomarkers 5,15 , cardiac imaging 16,17 and have also made available a 5-year SCD risk calculator which is specific to people with hypertrophic cardiomyopathy 14 but these have limitations attributed to prediction performance, small sample sizes or require additional resources, which might not be accessible to everyone.…”

Implantable Cardioverter Defibrillator Tachycardia Therapies: Past, Present and Future Directions

“…A systematic review and exploratory metanalysis of 46 studies that used electrophysiological signals to predict malignant VAs concluded that the AI models developed achieved a high performance, and that there is potential for the personalized prediction of malignant VA, although significant methodical limitations were identified in multiple studies [ 71 ]. AI-ECG applied to heart failure patients was found to be more discriminatory than conventional methods in determining SCD, especially in patients with an LVEF range of 35–50% and NICMP, and was able to discriminate between SCD and non-SCD [ 72 ].…”