Bridging clinical knowledge and machine learning: Leveraging large language models to predict in vitro fertilization outcomes

Introduction: In vitro fertilization (IVF) and frozen-thawed embryo transfer (FET) are vital components of assisted reproductive technology. However, predicting pregnancy outcomes remains challenging due to various biological and clinical factors. Recent advances in artificial intelligence (AI) and machine learning (ML) have shown the potential in offering innovative solutions for forecasting reproductive success. Objective: This study explores the use of large language models, specifically ChatGPT-4o, to optimize ML models for predicting pregnancy outcomes in IVF. Methods: The clinical dataset comprised 1061 IVF patients who underwent FET from 2014 to 2017, including variables such as age, body mass index, infertility duration, endometrial thickness, and serum beta-human chorionic gonadotrophin (β-HCG) levels on the 7th day after FET. ChatGPT-4o was tasked with preprocessing the data, evaluating several ML models, and optimizing performance. Results: The random forest model emerged as the best-performing model, achieving an accuracy of 85.45% and an area under the receiver operating characteristic curve of 0.8287 after applying the optimal threshold of 0.548, indicating strong predictive capability. Feature importance analysis revealed that serum β-HCG levels on the 7th day after FET were the most influential predictor of pregnancy outcomes. Despite these promising results, the study noted potential overfitting, likely due to the limited training dataset, a constraint largely attributable to the computational limitations of ChatGPT-4o. Conclusion: ChatGPT-4o shows potential in enhancing ML models in IVF outcome prediction. While AI-driven models can significantly aid clinical decision-making, clinicians should maintain a central role in patient outcome predictions. Future work will focus on improving model generalization with larger datasets and enhanced computational resources.