Colorectal disease : the official journal of the Association of Coloproctology of Great Britain and Ireland · 2026
Using large language models to integrate international IBD guidelines: A retrieval-augmented generation approach.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Salahi-Niri Aryan, Safavi-Naini Seyed Amir Ahmad, Devi Jalpa, Naderi Nariman, Sebastian Shaji, Adamina Michel, Nadkarni Girish, Soroush Ali, El-Hussuna Alaa
Аннотация
BACKGROUND AND AIMS: Clinical guidelines for inflammatory bowel disease (IBD) are essential for standardizing care, but synthesizing recommendations from multiple, often conflicting guidelines is a laborious task for clinicians. We developed and evaluated a proof-of-concept tool using a large language model (LLM) with retrieval-augmented generation (RAG) to help clinicians navigate this complexity by harmonizing guidelines, identifying consensus and controversy, and generating actionable statements.
METHODS: An LLM-driven RAG pipeline (GPT-4o) was designed to segment guideline content and compare recommendations across four international guidelines (ACG, ECCO, BSG, ACPGBI). This tool was evaluated on eight common clinical questions in Crohn's disease and ulcerative colitis. Outputs were assessed against expert-generated references by four independent reviewers using five-point Likert scales for completeness, accuracy, relevance, coherence and conciseness.
RESULTS: The tool reliably identified similarities and differences across guidelines, with mean scores of 4.34 (95% CI, 4.20-4.48) for consensus recognition and 4.61 (95% CI, 4.46-4.77) for disagreement detection. Completeness, accuracy and relevance consistently scored >4.0, while conciseness was lower (3.84, 95% CI, 3.50-4.19). Outline generation performance was moderate (3.25, 95% CI, 2.85-3.65). In 87.5% of cases, tool-generated recommendations aligned with expert conclusions.
CONCLUSIONS: This proof-of-concept study demonstrates that an LLM-RAG framework can systematically integrate IBD guidelines with high fidelity. This approach has the potential to improve guideline usability and support decision-making at the point of care, though further refinement is needed for conciseness and comprehensive outline generation.
METHODS: An LLM-driven RAG pipeline (GPT-4o) was designed to segment guideline content and compare recommendations across four international guidelines (ACG, ECCO, BSG, ACPGBI). This tool was evaluated on eight common clinical questions in Crohn's disease and ulcerative colitis. Outputs were assessed against expert-generated references by four independent reviewers using five-point Likert scales for completeness, accuracy, relevance, coherence and conciseness.
RESULTS: The tool reliably identified similarities and differences across guidelines, with mean scores of 4.34 (95% CI, 4.20-4.48) for consensus recognition and 4.61 (95% CI, 4.46-4.77) for disagreement detection. Completeness, accuracy and relevance consistently scored >4.0, while conciseness was lower (3.84, 95% CI, 3.50-4.19). Outline generation performance was moderate (3.25, 95% CI, 2.85-3.65). In 87.5% of cases, tool-generated recommendations aligned with expert conclusions.
CONCLUSIONS: This proof-of-concept study demonstrates that an LLM-RAG framework can systematically integrate IBD guidelines with high fidelity. This approach has the potential to improve guideline usability and support decision-making at the point of care, though further refinement is needed for conciseness and comprehensive outline generation.