This systematic review provides an overview of machine learning (ML) approaches for predicting preeclampsia.
This review adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyzes (PRISMA) guidelines. We searched the Cochrane Central Register, PubMed, EMBASE, ProQuest, Scopus, and Google Scholar up to February 2023. Search terms were limited to "preeclampsia" AND "artificial intelligence" OR "machine learning" OR "deep learning." All studies that used ML-based analysis for predicting preeclampsia in pregnant women were considered. Non-English articles and those that are unrelated to the topic were excluded. The PROBAST was used to assess the risk of bias and applicability of each included study.
The search strategy yielded 128 citations; after duplicates were removed and title and abstract screening was completed, 18 full-text articles were evaluated for eligibility. Four studies were included in this review. Two studies were at low risk of bias, and two had low to moderate risk. All of the study designs included were retrospective cohort studies. Nine distinct models were chosen as ML models from the four studies. Maternal characteristics, medical history, medication intake, obstetrical history, and laboratory and ultrasound findings obtained during prenatal care visits were candidate predictors to train the ML model. Elastic net, stochastic gradient boosting, extreme gradient boosting, and Random forest were among the best models to predict preeclampsia. All four studies used metrics such as the area under the curve, true positive rate, negative positive rate, accuracy, precision, recall, and F1 score. The AUC of ML models varied from 0.860 to 0.973 in four studies.
The results of studies yielded high prediction performance of ML models for preeclampsia risk from routine early pregnancy information.