Machine learning and statistical models for prediction of pregnancy outcomes in In vitro fertilization

This chapter explores the development and validation of prediction models under statistical and machine learning (ML) frameworks to determine pregnancy outcomes in in-vitro fertilization (IVF). Clinical prediction models, including statistical techniques like logistic regression and Cox regression, and ML techniques like random forests and neural networks, estimate individualized probabilities of success. Advances in embryo transfer and cryopreservation have shifted our focus from per-cycle success rates to cumulative live birth rates. These models help inform patients and guide clinical decision-making. Statistical models focus on inference and linear relationships between predictors and outcome, while ML prioritizes accuracy whilst handling complex, high-dimensional data. Key predictors of IVF success include female age, duration of infertility, ovarian reserve, and embryo quality. Many models lack external validation, handle missing data poorly (or not at all), and require updates due to changes in clinical practice over time. This review identifies the best-performing IVF prediction models and highlights the need for improvements in validation, feature selection, and calibration to enhance their reliability in real-world applications.