Rethinking prediction of sudden cardiac arrest: The role of electrocardiography in forecasting low-incidence, high-consequence events

Sudden cardiac arrest (SCA) remains a leading cause of mortality, accounting for 300,000–400,000 deaths annually in the United States. Despite advances in device therapy, current approaches to risk stratification remain limited in both sensitivity and specificity. This reflects a broader challenge in medicine: predicting low-incidence, high-consequence events, where traditional statistical frameworks often fail to achieve meaningful clinical utility. In this review, SCA is examined as a model problem highlighting key conceptual and methodological challenges, including class imbalance, heterogeneity of mechanisms, ambiguity in defining cases and controls, and temporal variability in risk. Electrocardiography (ECG) is emphasized as a scalable modality capable of capturing important components of the substrate-trigger-autonomic triad. However, existing ECG-based markers have not translated into robust clinical tools and recent machine learning approaches have not yet overcome this translational gap. We argue that the central translational gap is not the absence of stronger predictors, but insufficient specification of the clinical decisions, target populations, and performance thresholds against which model utility should be evaluated. Within this framework, ECG feature selection should be matched to the mechanistic target: depolarization markers index structural substrate, repolarization markers capture dynamic electrical instability, and autonomic markers reflect modulatory state, each operating on distinct timescales. Population decomposition into mechanistically coherent subproblems, rather than pursuit of a single overarching prediction model, is likely to accelerate both performance and clinical translation. Lessons learned from SCA may extend broadly to other high-impact, low-frequency conditions in medicine.