A New Biomarker of Aging Derived From Electrocardiograms Improves Risk Prediction of Incident Cardiovascular Disease

Abstract

Background

A biomarker of cardiovascular aging, derived from a deep learning algorithm applied to digitized 12-lead electrocardiograms, has recently been introduced. This biomarker, δ-age, is defined as the difference between predicted electrocardiogram age and chronological age.

Objectives

The purpose of this study was to assess the potential value of δ-age in enhancing the performance of primary prevention models for cardiovascular disease that incorporate traditional cardiovascular risk factors.

Methods

In this cohort study, we included 7,108 men and women from the Norwegian Tromsø Study in 2015 to 16, with follow-up through 2021 for incident fatal and nonfatal myocardial infarction (MI) and hemorrhagic or cerebral stroke. We used Cox proportional hazards regression models, Harrell's concordance statistic (C-index), and the net reclassification improvement.

Results

During a median follow-up of 5.9 years, we observed 155 cases of MI and 141 strokes. In men and women combined,HR per SD increment in δ-age, after adjustment for traditional risk factors included in the Norwegian risk model for acute cerebral stroke and myocardial infarction (NORRISK 2) score, was 1.24 (95% CI: 1.09-1.41) for the combined outcome, with similar HRs for MI and stroke. In men, the HR was significant for MI and in women for stroke. The C-index increased significantly but modestly when δ-age was added to a model with traditional risk factors. The net reclassification improvement was 26.0% (95% CI: 13.3%-38.1%) for the combined outcome, 17.5% (95% CI: 0.6%-33.5%) for MI, and 37.2% (95% CI: 20.1%-53.0%) for stroke.

Conclusions

Incorporating δ-age into primary prevention risk prediction models significantly improved performance beyond traditional cardiovascular risk factors for the combined outcome and separately for MI and stroke.