Neurogastroenterology and motility · 2026
Gut Instincts, Machine Decisions: Evaluating AI Accuracy in the Diagnosis and Treatment of Disorders of Gut-Brain Interaction.
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.
Ahn Matthew, Parker Colleen H
Аннотация
BACKGROUND: Disorders of Gut-Brain Interaction (DGBIs) are common yet challenging diagnoses for gastroenterologists and primary care clinicians. Large language models (LLMs) may support clinical management, but their accuracy remains poorly defined. This study evaluated the diagnostic and treatment accuracy of five LLMs using a convenience sample of clinical scenarios from the Rome IV Multidimensional Clinical Profile (MDCP).
METHODS: Sixty eight cases representing the DGBIs were entered into commonly accessible, untrained LLMs: ChatGPT 4.0, Google Gemini 2.5 Pro, Microsoft Copilot, OpenEvidence, and Perplexity. A standardized prompt elicited a diagnosis, treatment options, and confidence ratings for each diagnosis. Outputs were assessed against the MDCP recommendations and expert opinion.
KEY RESULTS: Diagnostic accuracy across models demonstrated Perplexity at 74%, ChatGPT and Google Gemini at 72% respectively, and Microsoft Copilot and OpenEvidence at 65% respectively. Treatment accuracy across models demonstrated ChatGPT, Google Gemini, and Microsoft Copilot generating accurate treatment options between 53% and 54% of cases each, compared with 41% for OpenEvidence and Perplexity each. No statistically significant differences were observed between AI models for diagnostic or treatment accuracy. Confidence ratings were uniformly high regardless of accuracy, with the mean scores ranging from 93% to 96% and standard deviation ranging from 1% to 4%.
CONCLUSIONS AND INFERENCES: Untrained LLMs demonstrated promising but imperfect diagnostic and treatment performance in this study of a convenience sample of DGBI cases. High confidence despite diagnostic errors and inconsistent treatment recommendations highlights the need for cautious clinical integration and validation.
METHODS: Sixty eight cases representing the DGBIs were entered into commonly accessible, untrained LLMs: ChatGPT 4.0, Google Gemini 2.5 Pro, Microsoft Copilot, OpenEvidence, and Perplexity. A standardized prompt elicited a diagnosis, treatment options, and confidence ratings for each diagnosis. Outputs were assessed against the MDCP recommendations and expert opinion.
KEY RESULTS: Diagnostic accuracy across models demonstrated Perplexity at 74%, ChatGPT and Google Gemini at 72% respectively, and Microsoft Copilot and OpenEvidence at 65% respectively. Treatment accuracy across models demonstrated ChatGPT, Google Gemini, and Microsoft Copilot generating accurate treatment options between 53% and 54% of cases each, compared with 41% for OpenEvidence and Perplexity each. No statistically significant differences were observed between AI models for diagnostic or treatment accuracy. Confidence ratings were uniformly high regardless of accuracy, with the mean scores ranging from 93% to 96% and standard deviation ranging from 1% to 4%.
CONCLUSIONS AND INFERENCES: Untrained LLMs demonstrated promising but imperfect diagnostic and treatment performance in this study of a convenience sample of DGBI cases. High confidence despite diagnostic errors and inconsistent treatment recommendations highlights the need for cautious clinical integration and validation.