Introduction: This systematic review evaluates the methodological rigor and risk of bias of machine learning-based predictive models developed to estimate the success of assisted reproductive technologies in cases of unexplained infertility, using the Prediction Model Risk of Bias Assessment Tool.
Materials and Methods: A systematic review was conducted to assess predictive modeling studies focused on unexplained infertility and based on machine learning, guided by the framework of the Prediction Model Risk of Bias Assessment Tool. After rigorously screening 912 records, only three studies met the inclusion criteria. While limited in number, these studies highlight emerging evidence in this underexplored area.
Results: The included studies applied supervised machine learning algorithms, such as Random Forest, Support Vector Machines, Partial Least Squares Discriminant Analysis, and neural networks, across various biomedical data types. Reported predictive performance varied by data modality: spectroscopy-based models demonstrated high classification accuracy, ranging from 92% to 100%, while a couple-based metabolic model with external validation achieved an accuracy of 73.8%. According to the PROBAST assessment, two studies were rated as low risk of bias, whereas one study exhibited an unclear risk, primarily due to limitations in external validation and analytical transparency.
Discussion and Conclusion: This systematic review demonstrates the potential of machine learning-based models to enhance clinical decision-making in the context of unexplained infertility.