Artificial Intelligence Simulation, Diagnostic Communication, Medical Education, Type 2 Diabetes Mellitus
Conditions
Keywords
Generative AI, Simulation-Based Education, Medical Students, Diagnostic Communication Skills, Artificial Intelligence in Healthcare, DIALOGUE-DM2
Brief summary
This randomized controlled trial evaluates the effectiveness of a generative artificial intelligence (AI)-based simulation program in improving diagnostic communication skills among medical students. The study is conducted at the Faculty of Higher Studies Iztacala, National Autonomous University of Mexico (UNAM). A total of 120 medical students are randomized to either an intervention group using the DIALOGUE-DM2 AI simulation platform or a control group following traditional educational methods. Participants complete a pre-test, receive training according to group assignment, and then undergo a post-test evaluation. The primary outcome is improvement in diagnostic communication skills, measured by standardized patient scenarios and validated rubrics. Secondary outcomes include self-reported confidence, communication domains, and inter-rater agreement between faculty evaluators and AI scoring. This trial aims to provide high-quality evidence on the potential of generative AI to enhance communication training in medical education, specifically in the context of type 2 diabetes diagnosis.
Detailed description
This study builds on a prior pilot trial (published in 2024) that demonstrated the feasibility of using generative artificial intelligence (AI) to train medical students in diagnostic communication. The current trial extends that work with a randomized, blinded, controlled design and a larger sample size. Design: The study is a randomized, blinded, parallel-group, controlled trial conducted at the Faculty of Higher Studies Iztacala (FES Iztacala), UNAM. A total of 120 medical students are enrolled and randomized (1:1) into either the intervention group (AI-based simulation training) or the control group (traditional training with standardized patients and faculty feedback). Intervention: * Intervention group: Students interact with the DIALOGUE-DM2 platform, which provides generative AI-driven simulated patients. They complete multiple diagnostic disclosure scenarios and receive immediate feedback on performance, based on standardized communication rubrics. * Control group: Students receive standard training, including lectures and supervised practice with peer role-play and faculty-guided feedback. Assessments: * Pre-test: All students complete one standardized patient scenario with faculty and AI evaluation prior to intervention. * Training phase: Participants complete their assigned training (AI vs. standard). * Post-test: Students complete a standardized diagnostic disclosure scenario. Independent faculty evaluators (blinded to group assignment) and the AI platform score performance. Outcomes: * Primary outcome: Change in diagnostic communication performance score from pre-test to post-test, measured by validated rubrics (Kalamazoo framework, MRS). * Secondary outcomes: * Student self-assessment of communication confidence. * Domain-specific improvements (information delivery, empathy, risk explanation, shared decision-making). * Agreement between human evaluators and AI scoring. Ethics and Oversight: The study has been reviewed and approved by the Research Ethics Committee of FES Iztacala, UNAM (Approval Number CE/FESI/042025/1915). Risks are minimal, as the intervention is educational and non-invasive. Significance: This is the first randomized controlled trial in Mexico to evaluate a generative AI-based simulation for diagnostic communication. Results will inform the integration of AI-driven training tools into medical education curricula and could contribute to scalable innovations in the training of healthcare professionals for chronic disease management, starting with type 2 diabetes.
Interventions
BEHAVIORALAI-Based Simulation Training (DIALOGUE-DM2)
Medical students interact with the DIALOGUE-DM2 platform, a generative AI-based simulation system. The platform delivers virtual patient encounters focused on type 2 diabetes diagnostic disclosure. Students complete multiple simulated scenarios and receive immediate AI-generated feedback aligned with standardized communication rubrics (Kalamazoo, MRS). Training aims to enhance diagnostic communication skills prior to post-test evaluation.
BEHAVIORALTraditional Training
Medical students receive traditional training in diagnostic communication. This includes lectures, peer role-play, and faculty-supervised feedback sessions covering diagnostic disclosure in type 2 diabetes. The training duration and number of sessions are matched to the intervention group.
Sponsors
Universidad Nacional Autonoma de Mexico
Study design
Allocation
RANDOMIZED
Intervention model
PARALLEL
Primary purpose
HEALTH_SERVICES_RESEARCH
Masking
TRIPLE (Subject, Investigator, Outcomes Assessor)
Masking description
Participant, Investigator, Outcomes Assessor
Intervention model description
Two-arm randomized, blinded, controlled trial comparing AI-based simulation training with traditional training in medical students facing type 2 diabetes disclosure scenarios.
Eligibility
Sex/Gender
ALL
Age
18 Years to 29 Years
Healthy volunteers
Yes
Inclusion criteria
* Medical students currently enrolled in the Faculty of Medicine (Medical Surgeon Program), UNAM-FES Iztacala. * Age between 18 and 30 years. * Able to provide informed consent. * Willing to participate in all study phases (pre-test, intervention, post-test).
Exclusion criteria
* Prior participation in the DIALOGUE pilot study. * Previous formal training in diagnostic communication beyond the standard medical curriculum. * Incomplete availability for scheduled sessions. * Refusal or inability to provide informed consent.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Change in Diagnostic Communication Performance Score | Approximately 12 weeks (from pre-test to post-test per participant). | Improvement in diagnostic communication skills, measured using validated rubrics - the Kalamazoo Essential Elements Communication Checklist and the Medical Communication Rating Scale (MCRS) - applied to standardized patient scenarios. Independent blinded faculty evaluators and AI scoring will be used. Scores range from 0 to 100, with higher values indicating better diagnostic communication performance. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Change in Student Self-Reported Confidence in Diagnostic Communication | Approximately 12 weeks (from pre-test to post-test per participant). | Change in students' self-reported confidence when disclosing a diagnosis of type 2 diabetes, measured through a structured questionnaire using a 5-point Likert scale (1 = very low confidence, 5 = very high confidence). Higher scores indicate greater self-perceived confidence in diagnostic communication. |
| Change in Domain-Specific Diagnostic Communication Scores (Kalamazoo Framework and Medical Communication Rating Scale) | Approximately 12 weeks (from pre-test to post-test per participant). | Improvement in specific communication domains - information delivery, empathy, risk explanation, and shared decision-making - evaluated using the Kalamazoo Essential Elements Communication Checklist and the Medical Communication Rating Scale (MCRS). Each domain is scored from 0 to 100, with higher scores indicating better performance. |
| Agreement Between Human Evaluators and AI Scoring | Assessed at post-test, approximately 12 weeks after baseline per participant. | Level of concordance between blinded human evaluators and AI-based scoring of diagnostic communication performance, assessed using Cohen's kappa coefficient (κ). Scores range from -1.0 to +1.0, where values closer to +1.0 indicate stronger agreement between evaluators. |
| Student Satisfaction With the Assigned Training Method | Assessed immediately after completion of the post-test, approximately 12 weeks after baseline per participant. | Satisfaction with the assigned training method (AI-based simulation vs. traditional training), measured using a structured 5-point Likert satisfaction survey (1 = very dissatisfied; 5 = very satisfied). Higher scores indicate greater satisfaction with the training method. |
Countries
Mexico
Outcome results
None listed