Key factors influencing clinical pregnancy rates in frozen-thawed single euploid embryo transfers: an artificial intelligence-based approach

• Published in: Reproductive biomedicine online p. 104860
• Main Authors: Ozer, Gonul, Duzguner, Ipek, Ozmen, Sevinc, Akca, Aysu, Bakir, Lale, Ozkara, Gulcin, Yelke, Hakan, Colakoglu, Yesim Kumtepe, Aygun, Tutku Melis, Cetinkaya, Murat, Deniz, Eylem, Pehlivanli, Ayca Cakmak, Kahraman, Semra
• Format: Journal Article
• Language: English
• Published: 2025
• Subjects: Obstetrics and Gynecology; Euploid embryo; Clinic pregnancy rate; Artificial intelligence; Machine Learning
• ISSN: 1472-6483
• DOI: 10.1016/j.rbmo.2025.104860

AbstractResearch questionCan artificial intelligence (AI) models accurately predict variables that affect clinical pregnancy rates in single-euploid embryo transfer cycles? DesignThis retrospective cohort study was conducted at Sisli Memorial Hospital, Assisted Reproductive Technology (ART), and Reproductive Genetics Centre between October 2011 and February 2023. It involved 4300 frozen-thawed single euploid embryo transfer cycles. Twenty-six variables, including clinical, demographic, and embryological characteristics, were investigated, which may affect clinical pregnancy outcomes in single euploid embryo transfers. This dataset was evaluated using various machine learning (ML) methods, including AdaBoost, Random Forest, XGBoost, LightGBM, and ExtraTree. Model performance and comparative effectiveness were assessed using 5-fold cross-validation, F1-score, recall, and the area under the receiver operating characteristic curve (AUROC). Furthermore, the SHapley Additive exPlanations (SHAP) values were used to analyse the direction and magnitude of the factors influencing clinical pregnancy. ResultsUsing various ML algorithms, seven key factors influencing clinical pregnancy rates were identified based on their level of importance. These factors included the number of previous cycles, anti-Müllerian hormone (AMH) levels, endometrial thickness, post-thaw embryo grade, maternal age, number of frozen embryos, and endometrial preparation method. XGBoost demonstrated promising performance in predicting clinical pregnancy, achieving a sensitivity of approximately 0.74 and a specificity of 0.70. Additionally, the AUROC value of 0.78 further highlighted the superiority of the XGBoost algorithm compared to the other methods ConclusionML algorithms have successfully identified factors influencing clinical pregnancy in euploid embryo transfer cycles. Understanding these factors could potentially assist physicians in optimising in vitro fertilisation (IVF) treatments and customising patient treatment regimens.