# Machine learning-based identification of determinants of pulse pressure in pregnant women

## TL;DR

This study uses machine learning to identify factors influencing pulse pressure in pregnant women, showing better performance than traditional models.

## Contribution

The study introduces machine learning for modeling longitudinal pulse pressure in pregnancy, revealing nonlinear and time-dependent relationships.

## Key findings

- Tree-based models outperformed mixed-effects models in predicting pulse pressure.
- Maternal age, weight, gestational age, and prior pulse pressure were key predictors.
- Machine learning enhances understanding of cardiovascular risk during pregnancy.

## Abstract

Pulse pressure (PP) is an important marker of arterial stiffness and cardiovascular risk during pregnancy, yet its longitudinal determinants remain insufficiently characterized, particularly in low-resource settings.

To identify determinants of longitudinal pulse pressure among pregnant women using machine learning approaches and to compare their predictive performance with a conventional mixed-effects modeling framework.

We conducted a retrospective cohort study of 549 pregnant women attending public antenatal care services at Bishoftu General Hospital, Oromia region, Ethiopia, comprising 2760 repeated pulse pressure measurements. Pulse pressure was modeled as a continuous longitudinal outcome. Predictors included maternal sociodemographic characteristics, clinical measurements, obstetric history, and gestational age at each visit. A generalized linear mixed model, random forest regression, and XGBoost regression were applied. Participant-level data partitioning was used for model training and evaluation, and predictive performance was assessed using root mean squared error (RMSE) and mean absolute error (MAE).

Tree-based machine learning models showed improved predictive performance compared with the mixed-effects model, indicating the presence of nonlinear and time-dependent relationships between predictors and pulse pressure trajectories. Maternal age, body weight, gestational age, and pulse pressure values from previous visits consistently contributed to pulse pressure prediction.

Machine learning methods applied to longitudinal antenatal data provide a flexible and effective framework for modeling pulse pressure dynamics during pregnancy. This approach enhances understanding of key clinical and temporal determinants and may support improved cardiovascular risk assessment in maternal health care settings.