10.3 🩺 內科專科考前版

10.3.0.1 📌 一頁重點整理

  • IOM 2015 「Improving Diagnosis in Health Care」 是 modern diagnostic safety 起點
  • Cognitive + System dual failure 是大多 diagnostic error 的根源
  • Big 3 (cancer/vascular/infection) 占 high-stakes errors 多數
  • Open notes、AI augmentation、closed-loop tracking 是 emerging tools
  • 📍 台灣：醫策會病安通報、TJCHA 病人安全目標含 diagnostic accuracy
  • Just culture + second opinion 是 diagnostic safety 文化基石

10.3.0.2 🧠 深度概念

10.3.0.2.1 Diagnostic Error Epidemiology

22E 引述： - ~10-20% of medical errors are diagnostic - Autopsy studies show 5-10% missed diagnoses - Outpatient diagnostic errors：~5% of US adults annually - 每人一生 ≥ 1 次 diagnostic error - Big 3 占 majority of harm

10.3.0.2.2 Singh’s Taxonomy of Diagnostic Errors
  • Missed: 漏診
  • Delayed: 延誤
  • Wrong: 錯誤
  • Failed communication: 診對但沒傳達
10.3.0.2.3 Cognitive Bias 詳解（連 Ch 4）

從多到少 implicated： 1. Premature closure (~50% of cognitive errors) 2. Anchoring 3. Availability 4. Confirmation bias 5. Search satisficing 6. Representativeness 7. Framing effect 8. Diagnostic momentum

10.3.0.2.4 System Failures 詳解
  1. Test result not followed up：~7% of abnormal labs
  2. Critical lab not notified
  3. Imaging report not read
  4. Handoff information loss
  5. EHR alert fatigue
  6. Specialist recommendation not acted
  7. Patient lost to follow-up
10.3.0.2.5 IOM 2015 Recommendations 詳細
  1. Communication + collaboration culture
  2. EHR support for diagnostic process
  3. Patient + family engagement
  4. Diagnostic safety reporting systems
  5. Research investment
  6. Clinician training in diagnostic reasoning
  7. Reimbursement reform for cognitive work
  8. Continuous process evaluation
10.3.0.2.6 Cognitive Forcing Strategies

22E 提到的 specific strategies： - Diagnostic time-out：30s pause before commit - Differential diagnosis discipline：強迫考慮 ≥ 3 alternatives - Cognitive autopsy：每天回顧 difficult cases - 「What else could this be?」：每個 dx 都問 - Metacognition training：educate trainees on biases

10.3.0.2.7 System Solutions

EHR-based: - CDS rules - Closed-loop test tracking - Critical result alerts - Risk prediction models（sepsis、AKI） - Differential diagnosis suggestion AI

Workflow: - Second opinion culture - Multidisciplinary team review - Tumor board for complex cancer - Morbidity & mortality conferences

Patient Engagement: - Open notes（OpenNotes initiative） - Patient portals - Health literacy programs - Encourage “second opinion seeking”

10.3.0.2.8 AI in Diagnosis

22E 提到： - Imaging AI：mammography、retinal、CT - ECG AI：AF detection - Sepsis prediction：Epic、UCSF models - Differential diagnosis support：emerging - 限制：bias、generalizability、over-reliance、alert fatigue - Khera et al. 2023 JAMA：「Automation bias and assistive AI」警告

10.3.0.3 🌟 Clinical Pearls (8 條)

  1. 「Don’t anchor」：每次新資訊重新 evaluate
  2. 「Diagnostic time-out」：commit 前 30 秒
  3. DDx ≥ 3 alternatives：強制 discipline
  4. 「What if I’m wrong?」：常自問
  5. Patient correctness：尊重病人 “this doesn’t feel right”
  6. Closed-loop tracking：每個 critical result 追到 acknowledge
  7. Second opinion 不是 weakness：是 quality
  8. Open notes 改善 diagnostic accuracy（多項研究）

10.3.0.4 🔍 特殊情境

10.3.0.4.1 1. Hospital-Acquired Diagnostic Errors
  • ICU patients 特別 risky（multiple comorbidities）
  • Daily round 必含 DDx review
  • Discharge summary 寫 working dx + uncertainty
10.3.0.4.2 2. Outpatient Settings
  • Limited time、less follow-up
  • Telephone/messaging diagnosis 高 risk
  • Need closed-loop tracking specifically
10.3.0.4.3 3. ED 環境
  • High-volume + time-pressured
  • “Treat-and-street” risk
  • Big 3 高頻
  • Discharge instructions 重要
10.3.0.4.4 4. Vulnerable Populations
  • Health literacy 低
  • Language barrier
  • Mental health comorbidity
  • Older adults
  • → Tailored communication
10.3.0.4.5 5. Atypical Presentations
  • Older adults often atypical
  • Female MI 不典型
  • Diabetic neuropathy 影響 pain
  • → 高警覺、broad DDx
10.3.0.4.6 6. AI Misclassification Trust
  • Clinicians 即使知 AI 錯仍 follow（Khera 2023）
  • 對應：強化 critical thinking、AI as second opinion 不是 final

10.3.0.5 📍 台灣 Context 專區

10.3.0.5.1 台灣 diagnostic safety
  • 醫策會（TJCHA）病人安全 8 大目標含 diagnostic accuracy
  • 病人安全通報系統 (TPR)：含 diagnostic errors
  • 各醫院 M&M conference 制度
10.3.0.5.2 健保資料庫研究
  • 台灣 NHI claim data 大規模 → diagnostic accuracy research
  • 例：sepsis、ACS missed diagnosis rate 研究
10.3.0.5.3 台灣訴訟環境
  • 醫療糾紛調解先行（2017）
  • Defensive medicine 同樣存在
  • Diagnostic error 是 malpractice 主要 cause
10.3.0.5.4 TJCHA 評鑑要求
  • M&M conference 機制
  • Critical result notification protocol
  • Test result follow-up tracking

10.3.0.6 ⚠️ 老闆地雷區

  1. Anchoring on first dx
  2. Premature closure 在 looks-well 病人
  3. Test result 沒 follow up：order 後不 track
  4. EHR alert fatigue 自動 dismiss
  5. Big 3 不警覺：cancer/vascular/infection 任何 atypical 都應 broaden DDx
  6. Diagnostic momentum：盲信 referral letter 上的診斷
  7. AI 過度信任：即使資訊明顯錯
  8. Open notes 抗拒：失去 patient feedback

10.3.0.7 🎓 內科專科考重點預測

10.3.0.7.1 高機率題型
  1. IOM 2015 report
  2. Big 3 內容
  3. Cognitive bias 辨識
  4. Premature closure 概念
10.3.0.7.2 跨章節整合
  • 配 Ch 4 Decision-Making：cognitive theory
  • 配 Ch 7 Safety：system failures
  • 配各疾病章 atypical presentation

10.3.0.8 📖 延伸閱讀

  • IOM. Improving Diagnosis in Health Care, 2015.
  • Singh H et al. The frequency of diagnostic errors in outpatient care. BMJ Qual Saf 23:727, 2014.
  • Khera R et al. Automation bias and assistive AI. JAMA 330:2255, 2023.
  • Newman-Toker D et al. Diagnostic errors. Diagnosis (Berl).
  • Croskerry P. From mindless to mindful practice. NEJM 368:2445, 2013.

10.3.0.9 📚 三階段教材索引

⚠️ 本 md 為 claude-opus-4-7 撰寫（2026-05-07），未經盧醫師驗證。