MétaCan
Menu
Back to cohort
Record W4416208872 · doi:10.2196/73857

Enhancing Large Language Models With AI Agents for Chronic Gastritis Management: Comprehensive Comparative Study

2025· article· en· W4416208872 on OpenAlexvenueno aff
Qing Ye

Bibliographic record

VenueJMIR Medical Informatics · 2025
Typearticle
Languageen
FieldMedicine
TopicArtificial Intelligence in Healthcare and Education
Canadian institutionsnot available
Fundersnot available
KeywordsWorkloadChronic gastritisChronic diseaseQuality (philosophy)MEDLINEValue (mathematics)Chronic care

Abstract

fetched live from OpenAlex

Background: The prevalence of chronic gastritis is high, and if not intervened in a timely manner, it may eventually lead to gastric cancer. Managing chronic gastritis essentially requires comprehensive lifestyle changes. However, the current health care environment does not support continuous follow-up by professional health care providers, making self-management a key component of postdiagnosis care. Increasingly, researchers are exploring the use of large language models (LLMs) for patient management. However, LLMs have limitations, including hallucinations, limited knowledge scope, and lack of timeliness. Artificial intelligence (AI) agents may provide a more effective solution. Nevertheless, it remains uncertain whether AI agents can effectively support postdiagnosis self-management for patients with chronic gastritis. Objective: The purpose of this study was to explore the effectiveness of AI agents in the postdiagnosis management of patients with chronic gastritis from different perspectives. Methods: In this study, we developed an agent framework for the health management of patients with chronic gastritis based on LLMs in conjunction with retrieval-augmented generation and a search engine tool. We collected real questions from patients with chronic gastritis in clinical settings and tested the framework's performance across different difficulty levels and scenarios. We analyzed its safety and robustness and compared it with state-of-the-art models to comprehensively evaluate its effectiveness. Results: Using a dual-evaluation framework comprising automated metrics and expert manual assessments, our results demonstrated that AI agents substantially outperformed LLMs in addressing high-complexity questions (embedding average score: 82.849 for AI agents vs 77.825 for LLMs) and were particularly effective in clinical consultation tasks. Clinical evaluation of safety based on a 5-point Likert scale by physicians indicated that the safety of the agents was 4.98 (SD 0.15; 95% CI 4.96-4.99). After 30 repeated experiments, the mean absolute deviation of the AI agents in the embedding average score and BERTScore metrics were 0.0167 and 0.0387, respectively. Therefore, the safety and robustness analysis confirmed that the AI agents can produce safe, stable, and minimally variable responses. In addition, comparative results with those of advanced medical-domain LLMs (Baichuan-14B-M1 and MedGemma-27B) and general-domain LLMs (Qwen3-32B) also demonstrated that the AI agents in this study performed outstandingly in the field of chronic gastritis. Our findings underscore the superior reliability, interpretability, and practical applicability of AI agents over conventional LLMs in chronic gastritis management, offering a robust foundation for their broader adoption in health care settings. Conclusions: AI agents based on LLMs have high application value in the management of chronic gastritis. They can effectively guide patients with chronic diseases in addressing common issues, which may potentially reduce the workload of physicians and improve the quality of patient home care.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame machine prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.

metaresearch head score (Codex)0.007
metaresearch head score (Gemma)0.017
Version: metacan-v3-hybrid-931329e0061cValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: Simulation or modeling
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.018
Threshold uncertainty score0.039

Distilled classifier scores by category (both heads)

CategoryCodexGemma
Metaresearch0.0070.017
Meta-epidemiology (narrow)0.0020.000
Meta-epidemiology (broad)0.0010.001
Bibliometrics0.0010.001
Science and technology studies0.0000.001
Scholarly communication0.0020.002
Open science0.0020.001
Research integrity0.0010.002
Insufficient payload (model declined to judge)0.0020.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.094
GPT teacher head0.461
Teacher spread0.366 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designSimulation or modeling
Domainnot available
GenreEmpirical

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations2
Published2025
Admission routes1
Has abstractyes

Explore more

Same venueJMIR Medical InformaticsSame topicArtificial Intelligence in Healthcare and EducationFrench-language works237,207