醫療相關手術傷口感染輔助監測與決策支援系統

Abstract

手術傷口感染在醫院感染控制與品質管理中一直都是中要的議題，會增加病患殘疾和死亡的機率。此外，手術傷口感染還會延長住院時間及增加醫療費用。故感染控制中預防手術傷口感染為重點之一。由於現階段人工監測的方法過度依賴感控護理師的判斷，可能造造成品質變動與耗費人力與時間。故本研究目的為其開發監測系統以提高手術傷口感染監測效率，並評估系統中的不同檢測邏輯的效能 (生理及症狀評估紀錄、抗生素處方籤紀錄、微生物檢驗紀錄、手術紀錄、出院診斷碼等)。 由本系統檢測12100位住院病患來自國立臺灣大學醫學院附設醫院，2013年九月至2014二月的電子病歷。2013年資料設為訓練資料集 (建置訓練模型用)，2014年資料則為測試資料集 (驗證模型)。由訓練模型中得知生理及症狀評估紀錄、抗生素處方籤紀錄、微生物檢驗紀錄等邏輯佔有較高的比重。由2014測試資料集驗證的結果得，敏感性是100%，特異性是78.38%，陽性預測值是1.4%，陰性預測值為100%，ROC曲線下面積0.957。 本醫療相關手術感染監測輔助系統減少感控人員所需審核的病歷數量並輔助感控人員判斷，明顯提高手術傷口感染監測效率。

Surgical site infection (SSI) is a long-term issue in the hospital and contributes to patient deaths and disability. Also healthcare-associated SSI (HASSI) increases healthcare costs and extends hospitalization stay, so prevention of HASSI is one of the most important issues in infection control. However, current manual surveillance method might be error prone, labor-intensive and time-consuming. To improve the efficiency of HASSI surveillance and evaluate different detection rule using various data sources (antibiotic prescription, microbiology laboratory, discharge diagnosis codes, surgical report and symptoms & physical examination), the HASSI surveillance system was developed. Electronic medical records from September 2013 through February 2014 of 12,100 inpatients at the National Taiwan University Hospital were screened by five defined rules of HASSI. The data from 2013 was allocated to a train data set (for developing training model), and the rest, to a test data set (for testing model validity). The rule of antibiotic prescription, microbiology laboratory and symptoms & physical examination has higher odds in training model. The validation in training model by the test data set from 2014 has sensitivity of 100%, specificity of 78.38%, positive predictive value of 1.4%, negative predictive value of 100% and the area under receiver operating characteristic (ROC) curve is 0.957. The HASSI system decrease the number of patients need to evaluate by infection control personnel (ICP) and significantly improves the efficiency of HASSI detection.