(824) Machine Learning-based Pregnancy and Pregnancy Trimesters Prediction Using Electrocardiogram

Pregnancy is a physiologic high-volume state. Physiologic electrocardiogram (ECG) changes in pregnancy have been reported but not well characterized. We hypothesize that the physiologic cardiac volume and compliance changes in pregnancy will result in changes in ECG features that can be used to distinguish pregnant from non-pregnant state. The objective of this study is to establish a machine learning (ML) model to predict pregnant from non-pregnant state based on ECG.

Study Design:

This is an analysis of a database that includes 22,034 ECGs on 7,298 individual patients from 18-50 years of age who were pregnant at least once between 2011 to 2024 in a single tertiary medical center. After excluding patients with pre-existing cardiac morbidity, along with those who developed hypertensive disorders of pregnancy, 3,581 ECGs were included (53% negative versus 47% positive label). A supervised ML method (Random Forest) was compared to multiple logistic regression to predict pregnancy state based on four ECG features (QRS interval, QTc interval, PR interval, and heart rate).

Results:

Random forest ML model performs better than multivariate logistic regression in the prediction of pregnancy state (Table). From the decision tree display of Random Forest ML model, higher QTc, higher heart rate, and lower QRS duration were associated with the classification of pregnancy state (Figure).

Conclusion:

Four features that are commonly reported in clinical assessment of ECG can assist in the prediction of pregnancy state, especially when ML methods are used. By establishing ECG features associated with pregnancy by trimester, we can begin to characterize patterns associated with abnormal pregnancies.