Not a Doctor, but I Write Like One: ChatGPT in General Surgery Informed Consent

Informed consent is essential to surgical care yet often burdens clinicians. Recent studies suggest that large language models can streamline this process. This study evaluates ChatGPT-generated consents for common general surgery procedures and compares between surgeon and resident perceptions of their clinical utility.

In this cross-sectional, survey-based feasibility study, board-certified surgeons and general surgery residents at a community-based training program reviewed ChatGPT-generated consents for four common procedures. Each document included indications, risks, benefits, and alternatives. Participants completed structured surveys assessing perceived accuracy, completeness, and appropriateness. Interrater reliability was assessed using Cohen's kappa. Agreement proportions were compared using one-way ANOVA. Free-text responses were thematically analyzed.

Twenty-one reviewers (11 surgeons, 10 residents) evaluated ChatGPT-generated consents. Mean agreement across consent components ranged from 94.5% to 96.4%, with no significant differences by procedure (P = 0.79). Interrater reliability was low and nonsignificant (P > 0.5), indicating limited agreement beyond chance. When responses were pooled by question type, agreement differed significantly (P < 0.01): all reviewers unanimously agreed on the appropriateness of listed benefits (100%), while agreement on risks (93.2%) and overall comprehensiveness (94.3%) was slightly lower. Qualitative responses highlighted the need for physician oversight, concerns about missing clinical nuance, and the importance of preserving patient rapport.

Surgeons and residents rated ChatGPT-generated consents as accurate and appropriate for routine general surgery procedures. With clinician oversight, integrating large language models into the consent process appears feasible and may reduce documentation burden without compromising accuracy.