Prediction of Pregnancy-Related Cardiovascular Outcomes Using Electrocardiogram-Based Deep Learning Estimation of...

Peak oxygen consumption (peak VO₂), the gold standard measure of cardiorespiratory fitness, may identify women at high risk for pregnancy-related cardiovascular (CV) complications, but ascertainment is not widely scalable. We previously developed and validated a deep learning model to estimate peak VO₂ from the resting 12-lead electrocardiogram (ECG).

The purpose of this study was to examine the association of deep learning ECG-predicted peak VO₂ with incident pregnancy-related CV complications.

We evaluated ECG-estimated peak VO₂ among individuals with a clinical 12-lead ECG from 1 year before pregnancy to 13 weeks of gestation in a multi-institutional electronic health record pregnancy cohort. Multivariable-adjusted mixed-effects logistic regression models examined associations between ECG-estimated peak VO₂ with pregnancy-related CV complications up to 1 year postpartum (severe hypertensive disorders of pregnancy, major adverse cardiac events, and maternal death).

Among 3,650 pregnancies from 3,437 women (mean age at delivery 33 ± 6 years), median ECG-estimated VO₂ was 27.0 mL/kg/min (IQR: 21.9-31.2), and 723 (20%; 95% CI: 19%-21%) experienced a pregnancy-related CV complication over a median follow-up time of 1.7 years (IQR: 1.6-1.8 years). Lower ECG-estimated peak VO₂ was associated with a higher complication risk (adjusted OR: 1.09 per 1-metabolic equivalent lower fitness; 95% CI: 1.03-1.17; P < 0.01). Women in the lowest quartile of ECG-estimated peak VO₂ had 61% greater odds of CV complications than the highest quartile (OR: 1.61; 95% CI: 1.13-2.30; P = 0.008).

Lower ECG-estimated cardiorespiratory fitness was associated with a higher risk of pregnancy-related CV complications, supporting artificial intelligence-enabled ECG analysis as a scalable tool for antepartum risk stratification of pregnancy-related CV complications.