A Comprehensive and Bias-Free Machine Learning Approach for Risk Prediction of Preeclampsia with Severe Features in a...

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Biomarker Discovery

Background:

• Preeclampsia is a leading cause of maternal morbidity, presenting diverse clinical challenges for prediction and management.
• Early identification of preeclampsia with severe features or eclampsia is crucial for improving maternal and fetal outcomes.
• Nulliparous women represent a key demographic for preeclampsia risk assessment.

Approach:

• Developed and applied machine learning models to predict preeclampsia onset in a nulliparous pregnant cohort.
• Utilized maternal serum samples from the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-be (nuMoM2b) cohort.
• Iteratively refined prediction models to address and correct for racial bias, enhancing predictive equality.

Key Points:

• Prediction models achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) ranging from 0.72 to 0.77.
• Initial models exhibited bias towards non-Hispanic Black participants; bias correction reduced the predictive equality ratio from 1.31 to 1.14.
• Key predictive features included biomarkers and ultrasound measurements, with placental analytes being strong indicators for early-onset preeclampsia.

Conclusions:

• Machine learning models demonstrate potential for accurate and equitable early screening of preeclampsia with severe features or eclampsia.
• Correction of racial bias in prediction models is feasible, leading to more equitable risk assessment.
• The findings support the development of bias-free screening tools for at-risk nulliparous pregnant individuals.