人工智慧輔助雙軌病歷量審效益評估──出院病歷審查流程為例

An Evaluation of an AI-Assisted Dual-Track Medical Record Quantitative Auditing System: Evidence from Discharge Record Review

研究背景：隨著電子病歷制度普及，病歷內容與管理規範日益複雜，傳統以人工為主之病歷量審流程，在效率、一致性與治理穩定性方面逐漸顯露侷限。如何透過制度化設計提升病歷審查效率與品質，已成為病歷資訊管理與醫院醫療數位轉型治理的重要課題。
研究目的：本研究旨在評估人工智慧（AI）輔助雙軌病歷量審制度之實務成效，並從內部控制、風險管理與品質治理角度，分析其對病歷管理之影響。
研究方法：本研究採準實驗設計比較AI導入前後住院病歷之量審作業效率。以「需審查張數」及「審查效率」衡量人工實際工作負荷。AI輔助量審採雙軌設計，第一階段由AI進行預先篩檢，第二階段由人工專業進行覆核，以比較紙本審查與AI輔助模式之差異。研究期間為114年10月1日至10月15日某區域醫院出院病歷共計300本。
研究結果：研究結果顯示，導入AI輔助雙軌病歷量審制度後，每案平均（±標準差）需人工審查之病歷張數由30.90±18.98張下降至15.34±19.89張，平均減少15.56張，整體審查效率提升50.36%（p-value < 0.001）。AI預先篩檢可有效將人工審查工作由全面逐本翻閱，轉為重點式覆核，降低人工遺漏風險，並使審查流程更具一致性與穩定性。隨著制度運作成熟，人工覆核比例呈現下降趨勢，顯示病歷紀錄品質與系統規則穩定性同步提升。
結論：AI輔助雙軌病歷量審制度不僅能提升審查效率，亦透過人機分工重構病歷管理流程，強化內部控制、風險預防與品質治理。其制度意義已超越行政流程優化，亦是醫院數位治理與永續經營之重要策略。

Background: With the widespread adoption of electronic medical records, the volume and complexity of clinical documentation have increased substantially, placing growing pressure on traditional manual medical record auditing processes. Audit efficiency, consistency, and governance stability have become critical challenges for health information management.
Objective: This study aimed to evaluate the effectiveness of a human–AI collaboration–assisted dual-track medical record auditing system using a page-based efficiency perspective, and to examine its implications for internal control, risk management, and quality governance.
Methods: A quasi-experimental pre–post design was employed to compare medical record auditing processes before and after Human–AI Collaboration. In addition to audit turnaround time, page-based workload, defined as the number of pages requiring manual review, was used as the primary efficiency indicator. The dual-track model consisted of AI-based pre-audit screening followed by human professional oversight. Page-based efficiency outcomes were compared between traditional paper-based auditing and the human–AI collaboration approach. The study period was from October 1 to October 15, 2025, and a total of 300 discharged medical records were collected from a regional hospital.
Results: After implementation of the Human–AI Collaboration dual-track auditing system, the mean number of pages requiring manual review per medical record decreased from 30.90 to 15.34 pages, representing a reduction of 15.56 pages and an overall efficiency improvement of 50.36%, p-value < 0.001. AI-based pre-audit screening effectively transformed the auditing process from comprehensive page-by-page review to exception-based review of flagged items, substantially reducing manual workload while maintaining audit sensitivity. As the system matured, the proportion of cases requiring human review declined, indicating concurrent improvement in documentation quality and rule stability.
Conclusions: From a page-based efficiency perspective, the AI-assisted dual-track medical record auditing system significantly reduces manual auditing workload while preserving professional oversight. Beyond operational efficiency, this model strengthens internal control mechanisms, enhances risk prevention, and supports defensible documentation quality. The findings suggest that AI-assisted auditing represents a governance-oriented transformation rather than a mere process optimization.