AI News AI资讯 3h ago Updated 1h ago 更新于 1小时前 46

Doctors' AI scribes get names of drugs and diagnoses wrong, NHS watchdog warns 医生AI病历记录员药物名称和诊断出错,NHS监管机构警告

Healthwatch England warns that AI scribes used in NHS consultations are making serious errors, including misidentifying drug names and fabricating diagnoses such as "demyelination" (a condition linked to multiple sclerosis) Patients are frequently catching errors that doctors miss, raising concerns about patient safety and the potential for incorrect information to persist in medical records The Medicines and Healthcare products Regulatory Agency has declined to classify AI scribes as medical de AI听写工具在药物名称和诊断记录中出现严重错误,患者比医生更容易发现此类错误 Healthwatch England警告这些错误可能进入医疗记录并影响后续治疗,构成患者安全风险 英国NHS已部署27种不同AI听写工具,但MHRA未将其归类为医疗设备进行监管 医生需花费额外时间审核AI记录,实际并未节省时间,反而可能增加工作量 专家建议AI错误率在复杂病例中更高,但医生手写记录同样存在错误风险

65
Hot 热度
62
Quality 质量
68
Impact 影响力

Analysis 深度分析

TL;DR

  • Healthwatch England warns that AI scribes used in NHS consultations are making serious errors, including misidentifying drug names and fabricating diagnoses such as "demyelination" (a condition linked to multiple sclerosis)
  • Patients are frequently catching errors that doctors miss, raising concerns about patient safety and the potential for incorrect information to persist in medical records
  • The Medicines and Healthcare products Regulatory Agency has declined to classify AI scribes as medical devices, meaning there is no England-wide regulatory oversight for their safety and effectiveness
  • NHS policy expects AI scribes to reduce administrative burden and free up clinical time, but doctors report that reviewing transcripts for errors may negate any time savings
  • Experts warn that error rates increase with multi-person consultations, complex medical histories, and patients whose first language is not English, though some GPs still believe AI records are more accurate than handwritten ones

Why It Matters

This article highlights a critical tension in healthcare AI deployment: the push for efficiency through automation versus the imperative of patient safety. For AI practitioners and health tech developers, it underscores that accuracy in clinical settings—particularly with ambient voice technology and medical terminology—remains insufficient, and that hallucinations and transcription errors can have serious real-world consequences. The lack of regulatory classification as medical devices also signals a governance gap that could expose healthcare systems to legal liability.

Technical Details

  • AI scribe technology in the NHS operates as ambient voice transcription tools that listen to doctor-patient consultations and generate summary notes, with 27 different AI scribe products currently in use across GP and hospital settings in England
  • Documented failure modes include hallucinations (inventing drug names and diagnoses not mentioned in the consultation), misrecognition of similar-sounding drug names, omission of critical instructions (e.g., repeat prescriptions), and misinterpretation of medical terminology such as "null demyelination" versus "demyelination"
  • Error rates are reported to be higher in consultations involving multiple participants, patients with complex medical histories, and non-native English speakers, suggesting limitations in the speech recognition and natural language understanding capabilities of current systems
  • A survey of 1,003 UK GPs found that more than half believed their AI-generated records were more accurate than their own, indicating a perception gap between clinician confidence and documented error rates
  • The regulatory framework currently does not classify AI scribes as medical devices under the Medicines and Healthcare products Regulatory Agency, meaning no mandatory safety or efficacy standards apply at the national level

Industry Insight

  • AI developers in the healthcare space must prioritize clinical-grade accuracy, particularly in medical terminology and drug name recognition, and invest in robust error-detection and correction mechanisms before large-scale deployment
  • Healthcare organizations adopting AI scribes should establish clear patient reporting pathways for errors and implement mandatory human review protocols, as relying on patient vigilance alone is neither sustainable nor ethical
  • Regulators and policymakers should address the current oversight gap by defining clear classification criteria for AI scribing tools, potentially requiring medical device certification to ensure patient safety standards are met before widespread NHS adoption

TL;DR

  • AI听写工具在药物名称和诊断记录中出现严重错误,患者比医生更容易发现此类错误
  • Healthwatch England警告这些错误可能进入医疗记录并影响后续治疗,构成患者安全风险
  • 英国NHS已部署27种不同AI听写工具,但MHRA未将其归类为医疗设备进行监管
  • 医生需花费额外时间审核AI记录,实际并未节省时间,反而可能增加工作量
  • 专家建议AI错误率在复杂病例中更高,但医生手写记录同样存在错误风险

为什么值得看

本文揭示了医疗AI部署中常被忽视的患者安全维度,为AI医疗应用提供了重要的风险案例。对从业者而言,展示了技术落地时监管框架缺失的潜在后果,强调了患者参与在错误检测中的关键价值。

技术解析

  • AI听写工具通过环境语音技术转录医患对话,但存在药物名称混淆(如将相似名称药物误记)、诊断错误(如将"null demyelination"误记为"demyelination")及关键医嘱遗漏等问题
  • 错误检测呈现"患者优先"特征:多起案例中患者首先发现错误,而医生未能识别,表明当前AI系统在医疗场景的准确性存在明显缺陷
  • 监管框架存在空白:MHRA未将AI听写工具归类为医疗设备,导致缺乏全国性安全有效性监管,27种不同工具并行使用进一步加剧监管难度
  • 技术部署与实际效益脱节:医生需额外时间审核AI记录,抵消了预期效率提升,且可能因"AI替代笔记"预期导致门诊量增加压力

行业启示

  • 医疗AI部署需建立患者参与的错误反馈机制,将患者视为安全验证的关键环节而非被动接受者
  • 技术落地前应进行严格的临床场景验证,特别是针对复杂病例、多语言患者及药物名称等高风险领域
  • 监管框架需与技术发展同步,明确AI医疗工具的分类标准和安全责任归属,避免"快速部署-事后补救"的被动模式

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

Healthcare AI 医疗AI Security 安全 Regulation 监管