Can machine learning models predict oocyte yield during assisted conception?: a systematic review.

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Biostatistics

Background:

• Accurate prediction of oocyte yield is crucial for optimizing gonadotrophin dosing in assisted reproduction.
• Excessive or inadequate responses to stimulation carry risks and can impact treatment success.
• Machine learning (ML) models are emerging tools to assist in predicting oocyte yield.

Purpose of the Study:

• To evaluate the accuracy and clinical readiness of ML models for predicting oocyte yield.
• To assess the quality and risk of bias in existing studies on ML for oocyte yield prediction.

Main Methods:

• A systematic literature search was conducted across OVID MEDLINE, EMBASE, and Cochrane Library.
• Nine studies (eight retrospective, one prospective) involving 62,354 cycles were included.
• Study quality and risk of bias were assessed using TRIPOD and PROBAST criteria.

Main Results:

• Reported accuracy varied, with mean absolute error ranging from 0.62 to 4.13.
• Neural network models generally demonstrated superior performance.
• No studies reported external validation, limiting generalizability; significant heterogeneity precluded meta-analysis.

Conclusions:

• Current ML models for oocyte yield prediction show promising internal performance but have a high risk of bias.
• Lack of transparency in data handling and absence of external validation hinder reproducibility and clinical application.
• Clinicians should exercise caution; further development must prioritize external validation and transparent reporting for reliable clinical integration.