Avaliação da idade-eletrocardiográfica determinada através de inteligência artificial como preditor de eventos cardiovasculares em adultos brasileiros

Abstract

Introduction: Cardiovascular diseases (CVD) are the leading cause of death globally. Chronological aging is a key risk factor for CVD, but biological aging varies among individuals. Electrocardiographic interpretation models based on deep neural networks (DNN) can estimate electrocardiographic age (ECG-age), a possible surrogate of biological and cardiovascular aging.

Aims: To evaluate the association between accelerated ECG-age and cardiovascular outcomes in Brazilian adults, using data from the ELSA-Brasil cohort. Methods: This analysis included 12,854 participants from the ELSA-Brasil cohort (2008–2010), after excluding individuals with prevalent CVD and incomplete outcome data. ECG-age was derived from a DNN model trained on the Brazilian CODE cohort, which contains a database of 1,558,415 digital electrocardiograms. ECG-age was categorized as accelerated based on the difference (Δage) between the DNN-predicted ECG-age and chronological age, using the mean absolute error (MEA) from the derivation study in ELSA-Brasil (MEA=8.45). A Δage greater than 8.45 years above chronological age was classified as “accelerated”. Cox proportional hazards models, adjusted for age, sex, and traditional cardiovascular risk factors (e.g., hypertension, diabetes, smoking, BMI, lipid profile), were used to examine the association between accelerated ECG-age and composite and specific cardiovascular outcomes, including stroke, acute coronary syndrome (ACS), heart failure (HF), and cardiovascular mortality. Fine-Gray models with the same covariate adjustments were applied to account for competing mortality risks. We also assessed whether ECG-age could improve the WHO HEARTS risk prediction equation performance.

Results: Accelerated ECG-age was significantly associated with increased risk of stroke (HR=3.04; 95% CI: 1.54–6.00; p=0.001), HF (HR=5.13; 95% CI: 2.07–12.74; p<0.001), and cardiovascular mortality (HR=3.82; 95% CI: 1.59–9.17; p=0.003). Heart failure showed the strongest association, with a more than five-fold increase in risk. No significant association was observed for ACS. The WHO HEARTS score alone had an AUC of 0.76, and adding ECG-age did not significantly improve the score’s discrimination (AUC 0.76; p>0.05).

Conclusion: Accelerated ECG-age is a significant predictor of specific cardiovascular outcomes, notably stroke and heart failure, in Brazilian adults. However, its integration into the existing WHO HEARTS risk prediction score did not enhance its discriminatory power. Further studies are needed to determine how ECG-age can be effectively utilized in clinical practice. Opportunistic cardiovascular risk screening through ECG presents a viable possibility, considering the broad accessibility and low cost associated with ECG use.