Predicting intensive care need in women with preeclampsia using machine learning - a pilot study

Area of Science:

• Obstetrics and Gynecology
• Maternal-Fetal Medicine
• Biomedical Data Science

Background:

• Predicting severe preeclampsia and the need for intensive care remains a significant clinical challenge.
• Identifying high-risk pregnancies is crucial for preventing adverse outcomes like eclampsia, representing an unmet global health need.
• Current prediction methods may not fully capture the complexity of severe disease progression.

Purpose of the Study:

• To develop a predictive model for severe preeclampsia outcomes.
• To utilize routine biomarkers and clinical characteristics for improved risk stratification.
• To identify patterns indicative of severe disease using advanced machine learning techniques.

Main Methods:

• Employed machine learning models on data from an intensive care cohort with severe preeclampsia (n=41) and a preeclampsia control cohort (n=40).
• Compared machine learning approaches against traditional logistic regression models to uncover complex disease patterns.
• Focused on identifying predictive biomarkers and clinical factors associated with severe disease progression.

Main Results:

• The optimal prediction model incorporated aspartate aminotransferase (ASAT), uric acid, and body mass index (BMI).
• This model achieved a cross-validation accuracy of 0.88 and an area under the curve (AUC) of 0.91.
• Internal validation on a test set yielded an accuracy of 0.82 and an AUC of 0.85.

Conclusions:

• Routine clinical parameters including ASAT, uric acid, and BMI are highly indicative of severe preeclampsia.
• ASAT reflects liver involvement, uric acid may play a role in preeclampsia pathophysiology, and BMI is a known risk factor.
• These findings suggest a potential for improved prediction of severe preeclampsia using readily available clinical data.