Machine learning-based prediction model for in vitro fertilization success: Predicting the likelihood of a live birth in subsequent in vitro fertilization cycles following an unsuccessful first cycle

Abstract

Patients undergoing in vitro fertilization (IVF) are often informed about their average chances of achieving a live birth, which is mainly influenced by the woman’s age. This project is part of a larger study and aims to predict individualized chances of success using machine learning (ML), helping patients set realistic expectations and guide their decisions. Three research questions were addressed, each focused on predicting the likelihood of a live birth in one of the following scenarios: the second cycle following an unsuccessful first cycle; the second or third cycle following an unsuccessful first cycle; and the third cycle following an unsuccessful first and second cycle. A complete IVF cycle includes ovarian stimulation, oocyte retrieval, fertilization, embryo culture and all resulting fresh and frozen embryo transfers. The dataset used to train the ML models was collected from Sahlgrenska University Hospital between January 2016 and December 2022 and included 21 features from 3,217 patients undergoing 5,170 cycles. Three ML models were evaluated: logistic regression, random forest (RF) and support vector machines (SVM). The predictive performance of logistic regression and RF was consistent with results from similar studies that used datasets comprising more than 70,000 IVF cycles. This study suggests that comparable results can be obtained with a smaller dataset and with fewer adjustments, such as accounting for non-linear relationships between the features and the live birth outcome in logistic regression, which RF handles naturally.