Poster CC3 - Appearance-related questions about breast reconstruction generated by ChatGPT: Potential to contribute to informed consent and shared decision making

Introduction: Artificial intelligence (AI)-powered tools based on large language models (LLMs) have been leveraged to answer patient questions about breast cancer care, including topics on breast reconstruction. In contrast, prior work has not investigated the quality of questions generated by LLMs for breast reconstruction patients to ask their healthcare team during medical consultations. This study assessed the potential for ChatGPT-generated questions about appearance after breast reconstruction to contribute to informed consent and shared decision making.

Materials and

Methods: ChatGPT was prompted to elicit appearance-related questions representative of what breast cancer patients seeking breast reconstruction may ask during consultations with their healthcare team. US board-certified surgeons with expertise in reconstructive plastic surgery were recruited for the study. Blinded to the source of the questions, the surgeons were prompted to assess the acceptability of each question, its potential to contribute to the informed consent process, and its potential to contribute to the shared decision-making process. The acceptability of each question was rated on a Yes/No basis. Surgeons were prompted to rate a question for its contribution to informed consent and shared decision making only if they indicated the question met the acceptability criterion. A question’s potential contribution to informed consent and shared decision making was rated on a 4-point Likert scale (1 – Strongly disagree to 4 – Strongly agree).

Results, Conclusions, and Discussions: Five surgeons participated in our study and assessed sixteen ChatGPT-generated questions. Fifteen of sixteen (15/16, 94%) questions assessed were considered acceptable by at least 4 of the 5 surgeons (Figure 1). One question was rated as not acceptable by three of five surgeons. This question was: How can I ensure that my reconstructed breasts have a natural-looking texture and skin tone?. One surgeon commented that, “this is a leading question that imparts all responsibility on the surgeon, I would be hesitant about this one especially in radiated patients.”

Twelve of the fifteen (12/15, 80%) acceptable questions assessed were rated as positively contributing to the informed consent process by at least 4 of 5 surgeons (agree or strongly agree). Three questions each received two disagree or strongly disagree ratings from three of five surgeons. The three questions were (a) How long will it take for the reconstructed breast to settle into its final shape and size?; (b) Will my reconstructed breasts be able to support the weight of implants or natural tissue without sagging?; and (c) How will my reconstructed breasts look and feel during intimate moments with a partner?.

All fifteen of the acceptable questions (15/15, 100%) assessed were rated as positively contributing to the shared decision-making process by at least 4 of 5 surgeons (agree or strongly agree).

There was a range in the assessments made by different surgeons in this study. For example, surgeon P5 rated all 16 (100%) ChatGPT-generated questions as acceptable whereas surgeon P2 considered 12 of the 16 (75%) to be acceptable. Differences of opinion were most pronounced in the perceived potential for the questions to contribute to the informed consent process. For example, surgeon P1 did not rate any questions as disagree or strongly disagree for potential contribution to the informed consent process, whereas surgeon P2 rated 75% of the questions as disagree or strongly disagree for potential contribution to the informed consent process.

Acknowledgements (Optional): XTG is a Ruth L. Kirschstein National Research Service Award fellow under National Institutes of Health (T32 EB007507).