Editing Physicians' Responses Using GPT-4 for Academic Research.

ABSTRACT

The integration of Artificial Intelligence (AI) into digital healthcare, particularly in the anonymisation and processing of health information, holds considerable potential. OBJECTIVES: To develop a methodology using Generative Pre-trained Transformer (GPT) models to preserve the essence of medical advice in doctors' responses, while editing them for use in scientific studies. METHODS: German and English responses from EXABO, a rare respiratory disease platform, were processed using iterative refinement and other prompt engineering techniques, with a focus on removing identifiable and irrelevant content. RESULTS: Of 40 responses tested, 31 were accurately modified according to the developed guidelines. Challenges included misclassification and incomplete removal, with incremental prompting proving more accurate than combined prompting. CONCLUSION: GPT-4 models show promise in medical response editing, but face challenges in accuracy and consistency. Precision in prompt engineering is essential in medical contexts to minimise bias and retain relevant information.