Large language models for periodontitis staging and grading according to the 2018 classification: a scoping review

Abstract Background Large language models (LLMs) are increasingly explored as clinical decision-support tools, but their ability to apply the 2018 classification of periodontitis is unclear. This scoping review mapped the evidence on LLM performance in periodontitis staging and grading, the evaluation methods used and the limitations reported. Methods Following PRISMA-ScR and a registered protocol, PubMed, Web of Science and Scopus were searched on 19 August 2026 for English-language articles published from 2023 onwards, supplemented by citation searching. Original studies comparing LLM-generated stages and/or grades with a reference standard were eligible. Two reviewers independently selected studies and charted data; QUADAS-2 was used for descriptive appraisal and findings were synthesised narratively. Results Six studies (2024–2026), comprising 400 patients, cases or vignettes in total, were included. They used real patient records ( n = 2), published or author-developed cases ( n = 2) or simulated vignettes ( n = 2). All provided clinical information as text only, and all evaluated OpenAI GPT models. Staging accuracy ranged from 36.2% to 70.5% and grading accuracy from 50.5% to 100%; reported agreement with the reference standard was slight to moderate (κ = 0.18–0.48). The one model evaluated in both settings performed better on clear-cut simulated vignettes than on real records with incomplete input. Errors clustered between adjacent categories, and models differed in whether they underestimated or overestimated disease severity. Comparisons favouring a task-specific neural network and retrieval-augmented platforms each came from a single study with important caveats; in the latter, between-system differences were smallest for staging and grading. Risk of bias related to patient selection was high or unclear in all studies. Conclusions The evidence is limited, recent and heterogeneous. LLMs show modest to moderate, context-dependent accuracy in staging and grading periodontitis from textual data, and the available studies do not support their use as autonomous diagnostic tools. Future studies should use consecutive real-world cases with complete clinical and radiographic information, independent reference standards and transparent reporting of model versions and settings. Whether multimodal models that interpret radiographs directly perform better remains a question for future research. Trial registration Open Science Framework, DOI https://doi.org/10.17605/OSF.IO/WYCDH .

Authors

Institutions

Publication Details

Journal
BMC Oral Health
Published
2026-11-13
DOI
https://doi.org/10.1186/s12903-026-10429-9
Primary Topic
Artificial Intelligence in Healthcare and Education
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Large language models for periodontitis staging and grading according to the 2018 classification: a scoping review

Ilkim Karadag, Adnan Tezel
7 links
BMC Oral Health
Artificial Intelligence in Healthcare and Education
article

Large language models for periodontitis staging and grading according to the 2018 classification: a scoping review

Ilkim Karadag, Adnan Tezel
article en

Abstract

Abstract Background Large language models (LLMs) are increasingly explored as clinical decision-support tools, but their ability to apply the 2018 classification of periodontitis is unclear. This scoping review mapped the evidence on LLM performance in periodontitis staging and grading, the evaluation methods used and the limitations reported. Methods Following PRISMA-ScR and a registered protocol, PubMed, Web of Science and Scopus were searched on 19 August 2026 for English-language articles published from 2023 onwards, supplemented by citation searching. Original studies comparing LLM-generated stages and/or grades with a reference standard were eligible. Two reviewers independently selected studies and charted data; QUADAS-2 was used for descriptive appraisal and findings were synthesised narratively. Results Six studies (2024–2026), comprising 400 patients, cases or vignettes in total, were included. They used real patient records ( n = 2), published or author-developed cases ( n = 2) or simulated vignettes ( n = 2). All provided clinical information as text only, and all evaluated OpenAI GPT models. Staging accuracy ranged from 36.2% to 70.5% and grading accuracy from 50.5% to 100%; reported agreement with the reference standard was slight to moderate (κ = 0.18–0.48). The one model evaluated in both settings performed better on clear-cut simulated vignettes than on real records with incomplete input. Errors clustered between adjacent categories, and models differed in whether they underestimated or overestimated disease severity. Comparisons favouring a task-specific neural network and retrieval-augmented platforms each came from a single study with important caveats; in the latter, between-system differences were smallest for staging and grading. Risk of bias related to patient selection was high or unclear in all studies. Conclusions The evidence is limited, recent and heterogeneous. LLMs show modest to moderate, context-dependent accuracy in staging and grading periodontitis from textual data, and the available studies do not support their use as autonomous diagnostic tools. Future studies should use consecutive real-world cases with complete clinical and radiographic information, independent reference standards and transparent reporting of model versions and settings. Whether multimodal models that interpret radiographs directly perform better remains a question for future research. Trial registration Open Science Framework, DOI https://doi.org/10.17605/OSF.IO/WYCDH .

BMC Oral Health
Ankara University (TR)
Good health and well-being
Artificial Intelligence in Healthcare and Education
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon