| 摘要: | 背景：臨床人員需要花費大量精力，並在時間壓力下閱讀病歷，以及比對治療檢查表，以確認患者是否有禁忌症。電子病歷（Electronic Medical Record, EMR）的出現帶來了自動化的可能，然而，自動化所需之醫學規則的編修門檻高，需要技術人員的協助，造成高昂的溝通與維護系統的成本。
目標：開發基於電子病歷和自然語言處理技術（Natural Language Processing, NLP）的臨床決策輔助系統，結合積木式視覺化臨床規則設計界面，試圖減少檢核檢查表的工作負荷，並提高時間效率及正確率。
資料與方法：本研究開發之系統採用MetaMapLite，從非結構化的EMR記錄中識別與核查表相關的醫學概念，以及結合結構化的檢驗資料，自動判讀檢查表之規則成立與否。另外，受積木式程式設計啟發，以Blockly開發視覺化界面協助臨床人員透過較簡易的方式創作規則。本研究邀集十二位臨床人員進行模擬任務實驗，請受試者使用本研究所開發之系統與傳統EMR界面，分別對四位患者檢核tPA (tissue plasminogen acti-vator) 和Factor XI (plasma thromboplastin antecedent) 檢查表，紀錄完成時間、填答結果，並由完成任務後填寫之系統可用性量表（System Usability Scale, SUS）、NASA心智負荷指標量表（NASA-Task Load Index, NASA-TLX）衡量可用性及工作負荷。
結果：根據衡量指標的評估結果，本研究所開發之電子病歷臨床決策輔助系統優於傳統EMR介面。檢查表的準確率從0.89提高至0.97（p = 0.046），檢核時間減少了23%；新系統的SUS分數優於傳統EMR介面（67.08對45.62，p = 0.005），NASA-TLX的工作負荷分數也較低（38.65對56.42，p = 0.046）。
討論與結論：本研究結合Blockly積木式視覺設計界面，以降低培訓成本，使臨床人員能在無技術人員的協助下設計規則，也透過tPA和Factor XI檢查表揭示了該方法的泛用性。儘管目前尚無法理解臨床筆記的上下文語意，與傳統的EMR界面相比，檢核錯誤和審查時間皆顯著減少。;Background: The widespread adoption of electronic medical records (EMR) has created oppor-tunities for automating clinical checklists. Previously, when determining whether to administer tPA (tissue plasminogen activator) to patients with acute ischemic stroke, clinicians faced signif-icant time pressure and effort to verify if the patient had any contraindications. Additionally, the technical barrier for creating medical rules is substantial, often necessitating the help of technical personnel, which results in high maintenance costs for clinical decision support system (CDSS).
Objective: To develop an EMR-based CDSS utilizing natural language processing (NLP) tech-niques aimed at reducing the workload for clinicians reviewing checklists, and to improve both time efficiency and accuracy.
Materials and Methods: Our system utilizes MetaMapLite to extract medical concepts from un-structured EMR notes, integrating these with structured laboratory data to evaluate the comple-tion of checklist criteria. Additionally, this study introduced a Blockly-based visual program-ming interface to facilitate rule creation by clinicians, reducing the need for technical expertise and training costs. Twelve clinicians were enlisted by us to perform simulated tasks comparing our system with a traditional EMR interface. The testers reviewed tPA and Factor XI (plasma thromboplastin antecedent) checklists for four patients, during which completion time, checklist accuracy, usability (via the System Usability Scale, SUS), and cognitive workload (via the NASA-Task Load Index, NASA-TLX) were measured.
Results: The proposed system demonstrated statistically significant improvements over the tradi-tional EMR interface. Checklist accuracy increased from 0.89 to 0.97 (p = 0.046), and the time to complete checklists decreased by 23%. Additionally, our system’s SUS score is better than the traditional EMR interface (67.08 vs. 45.62, p = 0.005), and the NASA-TLX workload score is lower than the traditional EMR interface (38.65 vs. 56.42, p = 0.046).
Discussion and Conclusion: This study successfully integrated a block-based visual design in-terface to facilitate independent rule design by clinical specialists, significantly reducing both training costs and the reliance on technical personnel. Although the current implementation can-not fully interpret the context of clinical notes, it significantly reduced checklist errors and re-view time compared to traditional EMR interface. |
