Large Language Model-Assisted Shared Decision-Making (SDM) in Rhinology: A Randomized Controlled Pilot Study

Shared decision-making (SDM) integrates patient preferences into evidence-based care. Generative AI tools may offer a novel approach to facilitate SDM in rhinology, but their clinical utility has not been assessed.

In this prospective randomized pilot study conducted at a single tertiary academic rhinology clinic, English-speaking adult patients with medically refractory chronic rhinosinusitis (CRS) and/or nasal obstruction were randomized to use either ChatGPT-4 or Google to ask treatment-related questions. The primary outcome was change in Decisional Conflict Scale (DCS) score. Secondary outcomes were changes in a treatment-related knowledge questionnaire (MCQ), system usability scale (SUS) scores, and accuracy ratings of ChatGPT responses.

A total of 57 participants completed the study (ChatGPT arm = 29; Google arm = 28). Both groups showed significant reductions in DCS scores with a mean difference of -4.3 in the ChatGPT arm (95% CI: -7.1 to -1.5, p = 0.004) and -2.7 in the Google arm (95% CI: -4.3 to -1.1, p = 0.002), with no significant difference between groups. Participants uncertain about treatment at baseline had higher decisional conflict both pre-and post-intervention, and demonstrated less improvement than those who preferred surgery. Treatment preferences and MCQ performance remained unchanged post-intervention in both groups. SUS scores were comparable between platforms, and ChatGPT's responses were rated as moderately accurate (7.2/10, SD = 2.0).

SDM using ChatGPT and Google may reduce decisional conflict in patients considering sinus surgery, though patients with baseline uncertainty may require additional support. Larger, multi-center trials are needed to validate these tools and explore their integration across diverse settings.

Level II.