JMIR Publications

ABSTRACT

Background:

Shared decision making (SDM) is a key element of patient-centered care; however, opportunities for structured and scalable SDM training remain limited in both medical education and clinical practice. Recent advances in artificial intelligence (AI) have enabled the development of chatbot-based simulations that may provide learners with repeated practice opportunities and individualized feedback on SDM-related communication behaviors.

Objective:

This study aimed to (1) compare performance and attitudes towards SDM between medical students and physicians, (2) investigate changes in communicative self-efficacy following chatbot-based SDM practice and (3) assess the validity of AI-generated feedback on shared decision-making performance using the OPTION-12 scale.

Methods:

A pre-post intervention design was employed. Medical students and licensed physicians participated in a 20-minute text-based consultation with an AI-simulated patient, followed by automated feedback on their performance. Prior to the interaction, participants completed measures assessing attitudes toward SDM (IcanSDM) and communicative self-efficacy (SEPCQ-24-GER). SDM performance was evaluated using AI-generated OPTION-12 ratings. Communicative self-efficacy was reassessed post-intervention, followed by demographic data, perceived authenticity of the interaction, and perceived benefits. Quantitative analyses included group comparisons, pre-post analyses, and psychometric evaluation of the OPTION-12 scale in an AI setting.

Results:

Medical students and physicians reported similarly positive attitudes towards SDM, but students achieved higher performance scores on the OPTION-12 scale then physicians (Mstudents = 28.67, Mphysicians = 22.79). Contrary to the expectation, communicative self-efficacy decreased slightly after SDM practice. There was strong agreement between AI and the human rater for the total OPTION-12 score, ICC(2,1) = .86, p < .001, although the AI systematically assigned slightly higher scores. The AI-generated ratings demonstrated good internal consistency across the 12 items, Cronbach’s alpha = .88, with an average inter-item correlation of r = .43. Overall, the results suggest that participants perceive the chatbot as authentic, regard SDM as implementable, and are likely to use it more frequently in practice.

Conclusions:

The present study provides preliminary evidence on the feasibility of AI-supported SDM training for both medical students and physicians. The findings of this study contribute to the contextual validation of the OPTION-12 scale in AI-mediated consultations and provide valuable insights into participant attitudes towards SDM. The findings also highlight the potential of chatbot-based training to practice SDM and offer precise, structured feedback on SDM performance. Overall, these findings provide a solid foundation for the development of scalable, feedback-driven approaches to SDM education and lay the groundwork for future research on AI-assisted communication training in clinical education.