Top Stories: Cardiac Arrest – Patients at Risk

Lower risk following acute myocardial infarction.¹

Curtain et al performed an analysis of the VALIANT (Valsartan in Acute Myocardial Infarction) and PARADISE-MI (Prospective ARNi vs ACE Inhibitor Trial to Determine Superiority in Reducing Heart Failure Events After MI) trials to evaluate whether risk of sudden cardiac arrest (SCA) following acute myocardial infarction (MI) has changed over time.¹ Patients recruited to these trials had left ventricular ejection fraction (LVEF) <40% following acute MI. Between 1998 and 2020, SCA rates dropped from 7.4% to 2.6% with comparable median follow-up duration (VALIANT 24.7 months and PARADISE-MI 22 months) . The authors attributed the reduction in risk to a higher proportion of patients being treated with primary percutaneous intervention (PCI) and guideline-directed therapy (as currently defined) in recent years. Given this substantial reduction in risk, the authors questioned the absolute benefit of primary prevention defibrillators in the contemporary clinical setting.

Important sex differences.²

Recent studies suggest that compared to men, SCA risk stratification is less effective in women, and women appear to benefit less from the primary prevention defibrillator. Now Krzowski et al report that the same is true for recurrent ventricular tachyarrhythmias (VTs).² In a retrospective post-hoc analysis of RAID (Ranolazine Implantable Defibrillator), recurrent VTs requiring ICD therapies were observed in 38% of men but only 25% of women during follow-up of two years. These findings underscore the importance of sex-specific risk prediction and management of SCA.

How do we improve assessment of risk?³

Peek et al attempted to extend risk prediction beyond the LVEF in a retrospective analysis of >140,000 patients pooled from twenty heterogeneous datasets from Europe, the USA and Israel.³ A model using LVEF alone was compared to a multi-variable survival model and a machine learning model. They reported that none of these approaches identified any significant predictors of SCD. There are several potential explanations for these unexpected findings. The definitions of SCA as well the number and nature of variables available did not appear to be consistent across the various datasets that included patients with and without ICDs. Another weakness was that ICD shocks were considered as equivalent to sudden cardiac arrest. The main lesson from this study is that while sample size of the SCD prediction dataset matters, homogeneity of definitions and quality of the data are equally important.

Concept of dynamic risk.⁴

Current SCD risk prediction is limited to static SCD risk assessment which largely consists of measuring the LVEF at a single timepoint. Pham et al recently reported proof of a novel concept: Instead of taking the usual static approach of considering a baseline ECG for risk prediction, they developed a “dynamic risk” approach using serial archived ECGs obtained prior to SCD events.⁴ Dynamic ECG remodeling was measured as change in a previously validated cumulative 6-ECG variable electrical risk score (ERS). Increase in the ERS over time was independently predictive of SCA and improved the performance of a risk prediction model that included clinical conditions and baseline ECG electrical risk, comorbidities, baseline ERS, and increased ERS (Area under the curve, AUC, increased from 0.770 to 0.869). There is potential for improving risk prediction by combining static and dynamic components of SCA risk.

A role for artificial intelligence tools.⁵

From a unique archive of >3800 digital ECGs, Holmstrom et al trained and validated a 12-lead ECG-based deep learning (DL) algorithm for detecting increased SCA risk.⁵ The DL model achieved an AUC of 0.889 (95% CI 0.861-0.917) for the detection of SCA cases vs. controls in the internal held-out test dataset and was successfully validated in external SCD cases with an AUC of 0.820 (0.794-0.847). The DL model performed significantly better than a conventional ECG model that achieved an AUC of 0.712 (0.668-0.756) in the internal and 0.743 (0.711-0.775) in the external cohort. The authors proposed that this algorithm could potentially be used as an inexpensive and broadly available pre-screening tool to identify individuals at increased risk of SCA.

Summary

Collectively, these recent papers highlight the critical need for a renewed and active investigative focus on the prediction of SCA risk. Over time, the utility of LVEF ≥35% as a single SCA risk predictor has diminished and may no longer be efficient or effective. As we discover and develop multi-marker risk scores across the spectrum of LVEF, it is important to employ universal definitions and pre-specified variables in large datasets that are homogeneous for methodology, but diverse with respect to sex, race and ethnicity. There is room for innovative concepts like dynamic risk of SCA. AI tools will greatly facilitate these goals, especially when findings are validated in external datasets. As high-performing risk prediction scores are discovered and validated, it will be important to design new primary prevention trials as a final step to making a real impact on the burden of SCA.