Prognostic Assessment of Tooth Avulsion Using Conversational Artificial Intelligence: A Guideline-Based Comparative Evaluation

Background: Traumatic dental avulsion is a severe dental injury that requires immediate management to preserve the periodontal ligament and ensure tooth survival. Because many clinical factors interact, predicting the prognosis is difficult. This study evaluated the performance of 5 conversational artificial intelligence (AI) models—ChatGPT-4o, Google Gemini, Microsoft Copilot, Perplexity AI, and Grok—in predicting avulsion prognosis compared with a guideline- based reference derived from the International Association of Dental Traumatology (IADT).

Methods: A total of 120 synthetically generated avulsion scenarios were created by combining clinical parameters including extraoral dry time, storage medium, and root development stage. Each model predicted both numerical (0-12) and categorical (good/guarded/poor) prognosis twice (T1 and T2, 48 hours apart) under deterministic conditions. Agreement with the guideline- based reference was assessed using accuracy, macro-F1, and Cohen’s κ, while temporal stability was analyzed via McNemar’s test. Consensus accuracy was determined through majority voting.

Results: None of the models achieved full agreement with the guideline-based reference. Perplexity demonstrated the highest κ value at T1 (0.575), while ChatGPT-4o achieved the highest accuracy at T2 (0.783). Gemini showed the greatest temporal stability (92.5%), and the majority-vote consensus improved overall reliability (κ = 0.698). Mean absolute errors ranged between 1.56 and 2.65, indicating moderate numerical concordance.

Conclusion: Conversational AI models exhibited moderate agreement with IADT-based prognosis predictions but differed in temporal consistency. The consensus approach yielded superior reliability, underscoring the potential of ensemble AI strategies as supportive—not substitutive— tools in clinical decision-making for dental trauma management.