以監督式學習技術預測骨科手術病人之輸血需求性

Predicting Blood Transfusion for Orthopedic Surgery Patients:A Supervised Learning Approach

張嘉美(Chia-Mei Chang)

指導教授 ： 胡雅涵

國立中正大學/管理學院/資訊管理系醫療資訊管理研究所/碩士(2015年)

摘要

輸血在手術過程中是常見的醫療行為，而且常常是必要的治療手段，但研究發現國內的醫師普遍是以自身所受到的醫學教育訓練及臨床經驗來預測病人是否需要輸血，且多數醫師都有未雨綢繆心理，造成許多骨科手術前都要常規備血，導致60-70%以上備血未輸用。而過多的備血未輸所產生的結果，即是增加了醫院人力、物力及時間的壓力與成本。
本研究收集南部某區域教學醫院2011年7月至2013年12月間共1396筆骨科病患開刀備血歷史資料，運用資料探勘中的決策樹(J48、M5)、支援向量機(SVM)和羅吉斯迴歸等技術，預測骨科手術病人之輸血需求性。結果預測效能最佳的是CfsSubsetEval模組搭配GeneticSearch所建構的羅吉斯迴歸分類器(AUC：78.7%)。
本研究探勘出影響輸血的因子，可幫助臨床醫師在術前評估時作為參考依據，以提昇病人輸血之安全。另也提供臨床骨科醫護人員術前備血評估時有更明確的參考準則，即當病患術式為脊椎、股骨、髖部或膝、髖關節置換術患者，只要有年紀大於75歲之老年人、手術前血色素(Hb)<12mg/dl、血小板(PLT)<8萬、有肝臟疾病及凝血功能(PT)異常疾病患者，以上只要符合其中一項則可備血，而依此準則即可減少許多不必要的備血作業，減少病人被反覆抽血、抱怨，也減少醫護人員工作負荷，更能減少醫事成本。

關鍵字

羅吉斯迴歸 ； 資料探勘 ； 備血 ； 輸血

並列摘要

Blood transfusion during operation is common in medical practice. Previous studies found that physicians generally determine the needs of blood transfusion based on their training and clinical experience. Furthermore, most physicians have a proactive attitute, resulting in many unnecessary routine preparation of blood products before orthopedic surgery. In clinical practice, 60 to 70% of blood products were not actually transfused during operation. The unnecessary preparation of blood products not only increases the pressure and workload of hospital workers but also wastes manpower, resources and time.
The study enrolled a total of 1396 patients undergoing orthopedic surgery between July 2011 to December 2013 in a regional teaching hospital in southern Taiwan. Data on preparation of blood products were investigated by using data mining techniques, including decision tree, support vector machines, and logistic regression to predict the need for transfusion. The model constructed by using logistic regression with GeneticSearch and CfsSubsetEval modules performed best with an area under curve of 78.7%.
The study identified several critical factors for blood transfusion, including patients undergoing operation on the spine, femur, hip or knee, hip replacement surgery, patients older than 75 years, preoperative hemoglobin <12 mg/dl, platelet counts <80x103/ul, medical history of liver disease, and prolonged prothrombin time. There findings can help clinicians evaluate the need for transfusion preoperatively and enhance patient safety. Preparation of blood products might be unnecessary if none of these factors exist. As a result, patient complaints of frequent blood withdrawals, workloads of health care providers, and medical costs could be reduced.

並列關鍵字

logistic regression ； data mining ； preparation of blood ； transfusion

參考文獻

邱詩涵 (民102 )。「應用時間性摘要預測慢性腎藏病患者病程惡化時間之研究」，中正大學資訊管理系碩士論文。

Abdel, M.P., & Morrey, B.F. (2010). Implications of revision total elbow arthroplasty on blood transfusion. Journal of Shoulder and Elbow Surgery, 19(2), 190-195.

Ahmadi, S., Lawrence, T.M., Sahota, S., Schleck, C.D., Harmsen, W.S., Cofield, R.H., & Sperling, J.W. (2014). The incidence and risk factors for blood transfusion in revision shoulder arthroplasty: our institution's experience and review of the literature. Journal of Shoulder and Elbow Surgery, 23(1), 43-48.

Arora, R.C., Légaré, J.-F., Buth, K.J., Sullivan, J.A., & Hirsch, G.M. (2004). Identifying patients at risk of intraoperative and postoperative transfusion in isolated CABG: toward selective conservation strategies. The Annals of thoracic surgery, 78(5), 1547-1554.

Bierbaum, B.E., Callaghan, J.J., Galante, J.O., Rubash, H.E., Tooms, R.E., & Welch, R.B. (1999). An Analysis of Blood Management in Patients Having a Total Hip or Knee Arthroplasty*. The Journal of Bone & Joint Surgery, 81(1), 2-10.