Prediction for the development of preeclampsia through non-invasive hemodynamics using machine learning,...

Area of Science:

• Cardiovascular Physiology
• Maternal-Fetal Medicine
• Biomedical Engineering

Background:

• Preeclampsia (PE) and hypertensive disorders of pregnancy (HDP) are significant causes of maternal and fetal complications, including prematurity.
• Classification into early- and late-onset PE is based on gestational timing and impacts management.
• Non-invasive hemodynamic assessment offers a promising avenue for early risk stratification.

Purpose of the Study:

• To evaluate impedance cardiography (ICG) for non-invasive hemodynamic assessment in predicting early- versus late-onset PE risk.
• To develop a machine learning model for identifying novel hemodynamic patterns associated with PE.
• To differentiate between early- and late-onset PE using advanced predictive analytics.

Main Methods:

• Utilized the J48 classification tree algorithm, a machine learning technique, to build a predictive model.
• Assessed hemodynamic parameters in 405 high-risk pregnant patients (17-33 weeks gestation) using ICG.
• Explored non-traditional hemodynamic variables beyond conventional metrics for PE prediction.

Main Results:

• The machine learning model achieved 95% accuracy in correctly classifying high-risk pregnancies that developed PE.
• The model successfully distinguished between early-onset and late-onset PE cases.
• Identified key hemodynamic variables related to contractility, cardiovascular performance, and afterload.

Conclusions:

• Non-invasive hemodynamic assessment using ICG, coupled with AI, shows significant potential for early PE detection in high-risk pregnancies.
• The developed model offers improved risk prediction and could advance clinical management strategies for PE.
• External validation is recommended to further confirm the model's robustness and generalizability.