Generative AI-Based Immersive Training for Endodontic Risk Disclosure and Informed Consent Among Dental Students (GAI-ENDORDIC)

BACKGROUND AND RATIONALE:

Endodontic treatments-including root canal therapy, retreatment, and periradicular surgery-are technically demanding procedures that carry inherent risks of irreversible complications, such as instrument separation, canal ledging, pulp floor perforation, and acute flare-ups. Effective risk disclosure and informed consent (RD-IC) processes are essential ethical and legal safeguards that ensure patient autonomy, foster collaborative doctor-patient trust, and minimize medicolegal hazards. However, traditional dental education relying on passive lectures or case-based learning (CBL) fails to recreate the real-world psychological pressure and conversational fluidity needed during challenging consultations. Generative AI large language models (LLMs), such as Gemini, offer an innovative, highly scalable "safe-to-fail" platform capable of simulating adaptive patient personas with diverse emotional profiles and health literacy levels.

STUDY DESIGN AND PARTICIPANT RECRUITMENT:

This prospective, parallel-group, 1:1 randomized controlled trial was conducted at the Department of Stomatology, The First Affiliated Hospital of Hunan University of Chinese Medicine, from March to May 2026. The trial protocol strictly adhered to the CONSORT 2010 statement and CONSORT-EHEALTH 2023 extension guidelines, with ethical approval granted by the Institutional Review Board (Ethics File No. MR-43-26-046512).

A total of 96 eligible fifth-year undergraduate dental students in their clinical clerkship who had completed core endodontic didactic courses were enrolled. Participants were randomly assigned via a computer-generated random sequence to either:

1. Experimental Group (n=48): Generative AI-based Immersive Training.
2. Control Group (n=48): Conventional Case-Based Learning (CBL).

EDUCATIONAL INTERVENTIONS:

• Experimental Group: Participants completed a 1-month simulation program using the Gemini platform. The curriculum comprised six standardized, high-stakes endodontic clinical scenarios (including acute pain management, instrument separation disclosure, and pulp floor perforation). Students engaged in 10-15 minute adaptive dialogue sessions every 3-4 days. The platform utilized standardized prompt templates to deliver real-time, objective feedback on both technical completeness (risk disclosure) and humanistic communication (empathy, tone, and clarity).
• Control Group: Participants underwent conventional CBL targeting the exact same six endodontic scenarios over the same 1-month duration. Learning involved structured group case discussions, faculty lectures, and standard informed consent document reviews. Both groups had equal access to standard clinical guidelines.

OUTCOME MEASURES AND DATA COLLECTION:

Assessments were conducted at four distinct time points: Baseline (T0), Immediately Post-Intervention (T1), 1-Month Follow-Up (T2-1m), and 2-Month Follow-Up (T2-2m).

1. Primary Outcome: RD-IC Completeness Score, measured via a validated 10-item checklist covering core indications, intraoperative complications, post-op pain, prognosis, alternative plans, and compliance.

2. Secondary Outcomes:

   • SEGUE Framework Scores: Assessing systematic communication competence (Set the stage, Elicit info, Give info, Understand perspective, End encounter).
   • Communication Self-Efficacy Scale (SCS) Scores: Evaluating confidence across clinical communication domains.
   • Communication Anxiety Scale (CAS) Scores: Assessing psychological tension and situational anxiety.
   • Blinded Standardized Patient (SP) Global Ratings: External validation of clinical communication performance.
   • Training Satisfaction Questionnaire: Evaluating platform authenticity, feedback quality, and user experience.
   • Longitudinal Retention: Skill retention and clinical transferability evaluated at T2-1m and T2-2m.

STATISTICAL ANALYSIS:

Data analysis was conducted using SPSS 26.0 following Intention-To-Treat (ITT) and Per-Protocol (PP) principles. Group differences were evaluated using t-tests, Mann-Whitney U tests, chi-square tests, and repeated-measures ANOVA, with Bonferroni corrections applied for multiple comparisons.