An Exploratory Evaluation of GPT-5 in Periodontitis Staging and Grading Using Published Clinical Cases

Area of Science:

• Artificial Intelligence in Dentistry
• Natural Language Processing Applications
• Periodontal Disease Diagnostics

Background:

• Periodontitis affects 42% of US adults aged 30+, necessitating accurate diagnosis and management training.
• Large language models (LLMs) present novel opportunities for dental education and clinical care.
• The diagnostic performance of advanced LLMs, like GPT-5, in periodontitis remains largely unassessed.

Purpose of the Study:

• To evaluate the diagnostic capabilities of Generative Pre-trained Transformer 5 (GPT-5) in staging and grading periodontitis.
• To compare GPT-5's performance against established diagnostic criteria for periodontitis.

Main Methods:

• Utilized 25 public clinical cases of periodontitis sourced from Google and PubMed.
• Employed a zero-shot prompting approach, inputting case descriptions into GPT-5.
• Assessed GPT-5's predictions against reference diagnoses using accuracy and Cohen's kappa.

Main Results:

• GPT-5 demonstrated class-dependent performance, with a tendency to overestimate disease severity.
• Achieved 68.0% accuracy for staging (kappa=0.432) and 77.3% for grading (kappa=0.179).
• Staging showed fair agreement, while grading indicated poor reliability and limited clinical utility.

Conclusions:

• GPT-5 offers improved performance over prior models but has limitations, especially in grading periodontitis severity.
• Current diagnostic accuracy restricts GPT-5's immediate application in clinical assessment and dental education.
• Substantial advancements in reliability and validation are crucial for integrating LLMs into periodontal diagnosis and training.