Developing a diagnostic thinking strategy to maximize the diagnostic accuracy when using AI-based automated medical history taking systems

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K10355
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 58010:Medical management and medical sociology-related
• Research Institution: Dokkyo Medical University
• Principal Investigator: Harada Yukinori 獨協医科大学, 医学部, 講師 (40810502)
• Co-Investigator(Kenkyū-buntansha): 志水 太郎 獨協医科大学, 医学部, 教授 (50810529)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: AI自動問診 / 診断エラー

Outline of Final Research Achievements

In our study, we demonstrated that while the presence of an AI-generated differential diagnosis list does not inherently influence physicians' diagnostic accuracy, the accuracy of the AI's diagnoses does. Additionally, we found that the use of AI for medical history taking in general internal medicine outpatient clinics could marginally decrease diagnostic errors. The diagnostic precision of AI improved with the increase in shared diagnoses across different AI systems, highlighting the benefit of utilizing multiple AI tools. However, we observed that the diagnostic capability of AI has not advanced over time and remains limited for uncommon diseases and atypical presentations. Moreover, we showed that physicians cannot correctly assess the reliability of diagnoses provided by AI-driven medical history taking systems.

Free Research Field

診断エラー

Academic Significance and Societal Importance of the Research Achievements

本研究の結果から、現在日本で利用されているAI自動問診の診断精度は全面的に信用してよいほどの水準にはないこと、特に稀な疾患や非典型的な病像など診断に誤りや遅れが生じる危険性が高い患者においてはさらに低くなることから、そのような患者であると感じた場合にはAIを頼らない方がよいことが示唆されるほか、AIの診断が正しいか否かの判断は医師の直観は当てにはならないため、他のAIを併用して共通する鑑別が多いかどうかで判断する方が安全であることも示唆される。このように、本研究はAI自動問診を安全に使用する際の具体的な方法を推奨することができた。