Prediction of clinical pregnancy outcome after single fresh blastocyst transfer during in vitro fertilization: an ensemble learning perspective

To establish a predictive model for clinical pregnancy outcomes following the transfer of a single fresh blastocyst in vitro fertilization (IVF). 615 patients (492 in training set and 123 in test set) who underwent the first single and fresh blastocyst transfer in the first IVF or intracytoplasmic sperm injection cycle performed in fertility centre of Shenzhen Zhongshan Obstetrics & Gynecology Hospital from July 2015 to June 2021 were enrolled in this study. Conventional method such as logistic regression (LR), individual machine learning methods including naive bayesian (NB), K-nearest neighbours (K-NN), support vector machine (SVM), decision tree (DT), and ensemble learning methods including random forest (RF), XGBoost, LightGBM, Voting were used to establish the clinical pregnancy outcome prediction model, and the efficacy among different models was compared. Three major types of prediction models, including conventional method: LR (AUC = 0.707), individual machine learning classifiers: NB (AUC = 0.741), K-NN (AUC = 0.719), SVM (AUC = 0.761), DT (AUC = 0.728), ensemble models: RF (AUC = 0.790), XGBoost (AUC = 0.799), LightGBM (AUC = 0.794), Voting (AUC = 0.845) were established. It was found that the performance of the voting model was best. This study revealed that a voting classifier can provide a more accurate estimate of IVF outcome, which can assist clinicians to make individual patient counselling.