Artificial intelligence predicts pregnancy complications based on cytokine profiles

Abstract

Background: Early prediction of pregnancy complications is important for adequate and timely prevention, management, and reducing maternal/fetal pathogenesis. Objective: To study the prognostic value of cytokines as predictors of pregnancy complications using unbiased artificial intelligence/machine learning (AI/ML) methods. Methods: For this study, we used our previously published data on 127 women with pregnancy complications and 97 women with a history of normal delivery and undergoing a normal delivery. A panel of seven cytokines were analyzed from activated peripheral blood mononuclear cells (PBMC). AI/ML methods such as kNN, SVM, decision tree, and ensemble classification were applied to explore the possible use of AI/ML to compare and predict normal gestation and normal delivery as opposed to different pregnancy complications such as recurrent spontaneous miscarriage (RSM), preterm delivery (PTD), pregnancy-induced hypertension (PIH), and premature rupture of fetal membranes (PROM). Results: The study examined cytokine levels in various pregnancy conditions, revealing significant differences, particularly in the levels of IL-2 and IFN-γ, across age-matched comparisons. Additionally, binary classification tasks demonstrated notable accuracies and f-measures for methodologies such as Ensemble (Bagged), QDA, and SVM (Cubic), showcasing their effectiveness in distinguishing between normal delivery and different pregnancy complications. Conclusion: The study provides a machine learning-based methodology for the prediction of pregnancy complications based on levels of cytokines produced by peripheral blood cells.