Abstract
This study aimed to develop and optimize machine-learning models to predict live birth outcomes following single vitrified-warmed blastocyst transfers (SVBT) in infertile couples with advanced paternal age (APA) over 35 years. A retrospective cohort study analyzed 1,044 SVBT cycles from two reproductive centers between June 2016 and December 2022. Data were split into training and validation sets in a 0.75:0.25 ratio, with fourteen clinically relevant variables selected for prediction. Ten machine-learning models were evaluated using 10-fold cross-validation, repeated three times. Among the cycles, 29.5% resulted in live births. Key predictors included trophectoderm, inner cell mass, maternal age at oocyte retrieval, blastocyst origin, total gonadotropin dose, endometrial thickness, number of oocytes retrieved, and maternal BMI. The extra trees and stacking models showed the highest performance with AUC and accuracy scores of 0.813 and 0.812, respectively, followed by the random forest model with an AUC and accuracy of 0.810. These models provide reliable tools for predicting live birth outcomes, highlighting the importance of multiple factors such as maternal age and embryo quality in couples with APA undergoing SVBT.