建構肝臟移植病患術後急性排斥反應預警模型

Abstract

肝臟移植手術後病患病情十分不穩定，病患最常遇到的問題是排斥反應的威脅，身體的免疫系統會將移植過來的新器官視為外來物而產生排斥效應，排斥反應不僅可能導致肝功能衰竭，更是增加病患的生命危險，因此，接受器官移植的病患在手術後必須接受醫護人員的嚴密監控，並使用免疫抑制劑來減緩身體的排斥反應，臨床上，病患排斥反應嚴重程度的判斷以及用藥時機，倚賴醫師經驗。 本研究透過資料探勘技術中的關聯分析方法，從病患抽血檢驗中挖掘肝臟移植術後急性排斥的危險變化趨勢，以規則導向的型態呈現，不僅能增加解釋性，也較符合醫護人員規則導向的診斷邏輯，本研究透過醫師經驗找出關聯分析法的結果中，具有醫學意義的變化趨勢，並將變化趨勢轉換成新特徵，透過CART決策樹分析法來建構病患手術後急性排斥反應的預警模型，本研究探討使用兩種連續數值類別化方法來建構研究資料庫所得到的危險趨勢，以及兩種抽血檢驗連續數值的資料型態所建構出的決策樹分析績效與模型解釋力，挑選績效較高的方法來結合專家經驗，並使用交叉驗證來建立模型的AUC、敏感度、特異度與準確度，期望能透過預警模型來輔助醫師進行決策，減輕醫護人員的工作負荷，提升醫療品質，增加病患手術後安全性，未來亦可作為發展電子加護病房使用。

Acute allograft rejection is a common problem following liver transplantation. It makes allograft damage and threatens survival of liver transplant patient under immunosuppressive therapy. The diagnosis of acute allograft rejection and the time of taking immunosuppressive therapy often depend on expert experience. Therefore, physicians and nurses need to look after liver transplant patient and monitor the change of biochemistry closely and regularly. It takes medical staff work under high workload environment. Hence, our main purpose is developing an early warning system to detect the subtle change of biochemistry, providing additional information to medical staff and reducing their workload when monitoring liver transplant patient’s situation. In this study, we use data mining skill to develop early warning model. Data mining is a data driven method to find out the potential relation among data. We use association rule mining to dig out the patterns of biochemistry especially for acute allograft rejection patient as new medical knowledge. Then, we use these patterns to build decision tree as our early warning model to identify the acute allograft rejection patient. The result shows that our model has excellent discrimination.