Evaluating the Accuracy of Large Language Model (ChatGPT) in Providing Information on Metastatic Breast Cancer.

Abstract

Purpose: Artificial intelligence (AI), particularly large language models like ChatGPT developed by OpenAI, has demonstrated potential in various domains, including medicine. While ChatGPT has shown the capability to pass rigorous exams like the United States Medical Licensing Examination (USMLE) Step 1, its proficiency in addressing breast cancer-related inquiries-a complex and prevalent disease-remains underexplored. This study aims to assess the accuracy and comprehensiveness of ChatGPT's responses to common breast cancer questions, addressing a critical gap in the literature and evaluating its potential in enhancing patient education and support in breast cancer management.

Methods: A curated list of 100 frequently asked breast cancer questions was compiled from Cancer.net, the National Breast Cancer Foundation, and clinical practice. These questions were input into ChatGPT, and the responses were evaluated for accuracy by two primary experts using a four-point scale. Discrepancies in scoring were resolved through additional expert review.

Results: Of the 100 responses, 5 were entirely inaccurate, 22 partially accurate, 42 accurate but lacking comprehensiveness, and 31 highly accurate. The majority of the responses were found to be at least partially accurate, demonstrating ChatGPT's potential in providing reliable information on breast cancer.

Conclusion: ChatGPT shows promise as a supplementary tool for patient education on breast cancer. While generally accurate, the presence of inaccuracies underscores the need for professional oversight. The study advocates for integrating AI tools like ChatGPT in healthcare settings to support patient-provider interactions and health education, emphasizing the importance of regular updates to reflect the latest research and clinical guidelines.