Rationale and ObjectivesOver the past year, studies have been conducted to evaluate the performance of Large Language Models (LLMs), such as ChatGPT, in the fields of gynecologic oncology. This review aims to analyze the applications and risks associated with using LLMs in this specialized field.Materials and MethodsThis systematic review was performed in adherence to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, incorporating elements from the diagnostic test accuracy extension and the CHARMS checklist for reviews of prediction models. A systematic literature search was executed on July 17, 2024, across PubMed, Web of Science, and Scopus databases. We focused on identifying original research that integrates LLMs with gynecologic oncology. We assessed the risk of bias using the adapted QUADAS-2 criteria.ResultsOur search identified eight studies that met our criteria, focusing on healthcare education, clinical practice, and medical code generation. These studies revealed variability in ChatGPT’s performance across different applications. It excelled in genetic testing and counseling, achieving 97% accuracy rate. However, its performance in cervical cancer prevention was less robust, with an accuracy of 83%. While one study demonstrated ChatGPT’s high adherence to quality guidelines, another noted that established guidelines significantly outperformed ChatGPT’s outputs. Additionally, code generation using tools like Google Bard and RoBERTa have shown potential to improve accuracy in clinical predictions and quality assurance. For example, Natural Language Processing (NLP) assisted by RoBERTa (based on Google’s BERT model) has improved the prediction of residual disease in women with advanced epithelial ovarian cancer following cytoreductive surgery. Despite these advancements, challenges related to consistency, specificity, and personalization persist, underscoring the necessity for continuous enhancement of these technologies.ConclusionLLMs demonstrate inconsistent performance in gynecologic oncology. These findings emphasize the need for continuous evaluation of these models before they are implemented clinically.
