Predicting recurrent gestational diabetes mellitus using artificial intelligence models: a retrospective cohort study.

Autor: Chen M; Department of Obstetrics and Gynecology, College of Clinical Medicine for Obstetrics & Gynecology and Pediatrics, Fujian Medical University Fujian Maternity and Child Health Hospital, Fuzhou, China., Xu W; Zhangzhou Health Vocational College, Zhangzhou, China., Guo Y; Department of Obstetrics and Gynecology, College of Clinical Medicine for Obstetrics & Gynecology and Pediatrics, Fujian Medical University Fujian Maternity and Child Health Hospital, Fuzhou, China., Yan J; Department of Obstetrics and Gynecology, College of Clinical Medicine for Obstetrics & Gynecology and Pediatrics, Fujian Medical University Fujian Maternity and Child Health Hospital, Fuzhou, China. yanjy2019@fjmu.edu.cn.
Jazyk: angličtina
Zdroj: Archives of gynecology and obstetrics [Arch Gynecol Obstet] 2024 Sep; Vol. 310 (3), pp. 1621-1630. Date of Electronic Publication: 2024 Jul 30.
DOI: 10.1007/s00404-024-07551-w
Abstrakt: Background: We aimed to develop novel artificial intelligence (AI) models based on early pregnancy features to forecast the likelihood of recurrent gestational diabetes mellitus (GDM) before 14 weeks of gestation in subsequent pregnancies.
Methods: This study involved a cohort of 588 women who had two consecutive singleton deliveries and were diagnosed with GDM during the index pregnancy. The least absolute shrinkage and selection operator (LASSO) regression analysis were used for feature selection. 5 AI algorithms, namely support vector machine (SVM), extreme gradient boosting (XGB), light gradient boosting (LGB), decision tree classifier (DTC), and random forest (RF) classifier, and traditional multivariate logistic regression (LR) model, were employed to construct predictive models for recurrent GDM.
Results: 326 (55.4%) experienced GDM recurrence in subsequent pregnancy. In the training set (67% of the study sample), 13 features were selected for AI models construction. In the testing set (33% of the study sample), the AI models (LGB, RF, and XGB) exhibited outstanding discrimination, with AUROC values of 0.942, 0.936, and 0.924, respectively. The traditional LR model showed moderate discrimination (AUROC = 0.696). LGB, RF, and XGB models also demonstrated excellent calibration, while other models indicated a lack of fit. All AI models showed superior overall net benefits, with LGB, RF, and XGB outperforming the others.
Conclusions: The proposed LGB model demonstrated exceptional accuracy, excellent calibration, and superior overall net benefits. These advancements have the potential to assist healthcare professionals in advising women with a history of GDM and in developing preventive strategies to mitigate the adverse effects on maternal and fetal well-being.
(© 2024. The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.)
Databáze: MEDLINE
Externí odkaz: