A machine learning algorithm for predicting maternal readmission for hypertensive disorders of pregnancy

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Maternal postpartum hypertensive emergencies contribute significantly to maternal mortality and hospital readmissions.
• Current prediction methods for readmission are limited, leading to inaccuracies.

Purpose of the Study:

• To develop and validate a machine learning-based predictive algorithm for maternal postpartum readmission.
• To identify clinical features at discharge that predict readmission for hypertensive disorders of pregnancy.

Main Methods:

• A cohort study involving over 20,000 delivering women.
• Development of a predictive algorithm using machine learning on prospectively collected electronic medical record data.
• Validation of the algorithm in an independent cohort.

Main Results:

• The predictive algorithm achieved an area under the curve of 0.85 in the derivation cohort and 0.81 in the validation cohort.
• Both derivation and validation models utilized 31 comparable clinical features.
• Readmission rates were 1.2% and 1.4% in the derivation and validation cohorts, respectively.

Conclusions:

• Machine learning can predict maternal postpartum readmission for hypertensive disorders with reasonable accuracy using discharge clinical data.
• Further validation in diverse healthcare settings is required to confirm the algorithm's utility.