Work

ECG-Derived Synthetic Left Ventricular Mechanics Predict Mortality After Acute Myocardial Infarction and Fatal Myocardial Infarction in the General Population

Year: 2026

Type: preprint

Abstract: Background: Left ventricular (LV) mechanical function predicts outcomes after acute myocardial infarction (AMI). Recovering these parameters from the 12-lead electrocardiogram (ECG) could enable point-of-care risk assessment wherever an ECG is acquired. Objectives: To determine whether ECG-derived synthetic LV mechanics predict all-cause mortality after AMI and myocardial infarction (MI)-specific mortality in the general population and add information beyond the GRACE score and ejection fraction. Methods: Synthetic peak systolic (s′) and early diastolic (e′) mitral annular velocities and mitral annular plane systolic excursion (MAPSE) were derived from the 12-lead ECG by a previously validated generative adversarial network, applied without retraining and blinded to outcome. Associations were assessed in an AMI cohort (n=1,065) by Cox regression and a population cohort (n=219,564) by Fine-Gray models. Incremental value was tested against GRACE, three AI-ECG reduced ejection fraction (EF) models, and measured LVEF. Results: In the AMI cohort (109 deaths; median follow-up 1.44 years), all three parameters predicted all-cause mortality (each P<0.001). Synthetic e′ was the sole independent predictor after adjustment for clinical covariates (HR=0.90 per 1 cm/s; 95% CI 0.84-0.98) and for GRACE (HR=0.90; 0.83-0.97; P=0.006). In the post-discharge cohort (n=1,017; 61 deaths), synthetic e′ improved discrimination beyond GRACE (C-index 0.67 to 0.70; HR=0.84 per 1 cm/s; 95% CI 0.76-0.93; P=0.001) and remained independent of AI-ECG-predicted and measured LVEF (HR 0.84-0.86 per 1 cm/s; all P≤0.004), contributing 28-46% of each model's prognostic information. In the population cohort (817 MI deaths; median follow-up 3.42 years), all three predicted MI-specific mortality after adjustment for age, sex, and ECG abnormalities, MAPSE showing the largest effect (sHR=0.49; 95% CI 0.37-0.65), with s′ (sHR=0.78; 0.72-0.84) and e′ (sHR=0.87; 0.82-0.93) also predictive (each P<0.001). Conclusions: ECG-derived synthetic LV mechanical parameters predict all-cause mortality after AMI and fatal MI in the general population, adding information beyond the GRACE score and LVEF. Keywords: electrocardiogram; tissue Doppler imaging; generative adversarial network; myocardial infarction; mortality prediction; risk stratification.

Source: medRxiv

DOI: 10.64898/2026.10.21.26366159