建構智慧化疾病分類編碼輔助系統──以一般外科為例

Development of Intelligent Support Systems for Coding International Classification of Diseases - A Case Study of General Surgery Medical Record

現今疾病與處置編碼的工作，已有電子工具書可輔助疾病分類師使用關鍵字協助編碼。然而，在使用此電子工具書前，仍須自行從主要診斷或主要處置描述中，判斷其關鍵字。對於資淺的疾病分類師，因為經驗不足，常需花費大量的時間進行關鍵字判斷且容易產生錯誤。本研究運用「文字探勘」協助建構主要診斷與主要處置之關鍵字判定系統。首先，以個案醫院在2016與2017年所有主要診斷與主要處置之案例中，整理歸納這些案例中，用來協助判斷關鍵字之判定字以及所對應之關鍵字，用以建立「主要診斷」及「主要處置」關鍵字判定系統之知識庫。此外，並整理出判定字探勘的規則，作為關鍵字判定系統的設計邏輯。本研究以Python進行編碼，並以PyInstaller將此判定系統包裝成執行檔，透過2018年之病歷案例來驗證系統之準確性。本研究最後證實此「主要診斷與主要處置之關鍵字判定系統」，可以有效地降低疾病分類師在進行「主要診斷」及「主要處置」的關鍵字判斷時間。對於主要診斷個案，一位資淺疾病分類師，從原本平均一個主要診斷須花費54秒，降低至17秒，而資深疾病分類師由平均須花費31秒，降低至15秒；對於主要處置個案，一位資淺疾病分類師，從原本平均一個主要處置須花費69秒，降低至19秒，而資深疾病分類師由平均須花費40秒，降低至18秒。以一般外科為案例，藉由「智慧化關鍵字判定系統」來輔助疾病分類師快速查找診斷及處置之「關鍵字」，用以進行疾病分類編碼工作。

Nowadays, disease classifiers can easily make use of key words to do the disease cod-ing tasks via e-books. However, they still need to determine keywords from principal diag-nosis or principal procedures by themselves. It takes time and easily causes mistakes espe-cially for those junior disease classifiers lacking of experience. This study adopts ''Text Mining'' to assist in the construction of keyword determination system for principal diagnosis and principal procedure. First of all, we used the surgery cases in 2016-2017 at the case hospital to build a knowledge base. The knowledge base summarizes all determinant words from the principal diagnosis and principal procedures and correlates these determinant words to keywords. Secondly, this study constructs the rules which explain how this keyword determination system works. Finally, the execution file is encoded in Python and filed with PyInstaller. The verification is done with the data in 2018 from the same case hospital with 100% accuracy. The results of this study show that this ''keyword determination system for principal diagnosis and principal procedure'', can effectively reduce disease classifiers to determinate keywords from ''principal diagnoses'' and ''principal procedures''. For junior disease classi-fiers, it reduces from 54 seconds to 17 seconds on principal diagnostic cases and from 69 seconds to 19 seconds on principal procedures. For senior disease classifiers, it reduces from 31 seconds to 15 seconds on principal diagnostic cases and from 40 seconds to 18 seconds on principal procedures.