Evaluating chat generative pre-trained transformer responses to common patient questions on temporomandibular disorders

Unuvar, Bayram Sonmez; Unuvar, Elif Esra Ozmen

doi:10.1590/1806-9282.20251888

Tabla de contenido Anterior Actual Siguiente

Tabla de contenido

ORIGINAL ARTICLE • Rev. Assoc. Med. Bras. 72 (3) • 2026 • https://doi.org/10.1590/1806-9282.20251888 linkcopy

Evaluating chat generative pre-trained transformer responses to common patient questions on temporomandibular disorders

Authorship SCIMAGO INSTITUTIONS RANKINGS

OBJECTIVE: Temporomandibular disorders are among the most common causes of orofacial pain, often leading patients to seek information online. The increasing use of large language models such as chat generative pre-trained transformer in healthcare communication has raised questions about the reliability and readability of artificial intelligence-generated patient information. The aim of this study was to evaluate the accuracy, comprehensiveness, readability, and inter-rater reliability of chat generative pre-trained transformer-generated responses to common patient questions regarding temporomandibular disorders.

METHODS: ChatGPT (version 4.0) was prompted to generate 50 potential patient questions about temporomandibular disorders. Ten representative questions were selected and independently evaluated by five experts (two oral and maxillofacial surgeons, two physiotherapists, and one physical medicine specialist). Responses were rated using a four-point quality scale assessing accuracy and completeness. Readability was calculated using the Flesch-Kincaid method, and inter-rater reliability was assessed using the Intraclass Correlation Coefficient.

RESULTS: The responses demonstrated variable but generally acceptable quality. The overall Intraclass Correlation Coefficient value was 0.862, indicating good inter-rater agreement. Readability levels ranged from grade 6.2–10.7 (mean 8.0), corresponding to middle-to-high school comprehension. While most responses were rated satisfactory, several lacked sufficient clinical detail, particularly in differentiating professional consultation pathways.

CONCLUSION: chat generative pre-trained transformer provides moderately reliable and readable information about temporomandibular disorders, supporting its potential role in patient education. However, reliance on artificial intelligence-generated frequently asked questions introduces methodological limitations and authority bias. Future studies should incorporate real patient data and external fact-checking to enhance clinical relevance.

KEYWORDS:
Temporomandibular disorders; Artificial intelligence; Patient education; Health communication; Large language models

Questions	Flesch-Kincaid grade level	Quality of answer					ICC
Questions	Flesch-Kincaid grade level	Expert 1	Expert 2	Expert 3	Expert 4	Expert 5	ICC
1. What are the most common causes of jaw pain?	7.7	2.00	2.00	2.00	2.00	2.00	0.862
2. Can jaw pain cause neck pain?	7.9	2.00	1.00	2.00	1.00	2.00
3. Which specialist should I see forjaw pain?	7	1.00	1.00	1.00	1.00	1.00
4. Should I apply heat or cold for jaw pain?	9.4	3.00	2.00	3.00	2.00	3.00
5. Do massage or exercises help with jaw pain?	7.2	2.00	2.00	2.00	2.00	2.00
6. What exercises are recommended for jaw pain?	6.2	2.00	1.00	2.00	1.00	2.00
7. Is Botox treatment for jaw pain effective?	8.8	1.00	3.00	2.00	3.00	2.00
8. Is a night splint effective for jaw pain?	7.6	2.00	2.00	2.00	2.00	2.00
9. How long does it take to treat jaw pain?	8.7	3.00	3.00	3.00	2.00	3.00
10. Do muscle relaxants relieve jaw pain?	10.7	2.00	2.00	2.00	2.00	2.00

location_on

Associação Médica Brasileira R. São Carlos do Pinhal, 324, 01333-903 São Paulo SP - Brazil, Tel: +55 11 3178-6800, Fax: +55 11 3178-6816 - São Paulo - SP - Brazil
E-mail: ramb@amb.org.br

rss_feed Acompañe los números de esta revista en su lector de RSS

[1] ICC indicates inter-rater reliability. Quality scores were based on a four-point scale (1=insufficient, 2=satisfactory, 3=good, 4=excellent) as described by Mika et al.¹³. ICC: Intraclass Correlation Coefficient (two-way random, absolute agreement). ICC=0.862 indicates good reliability.