本研究回溯運用2004-2010年間,桃園地區某區域教學醫院醫療資訊系統之電子病歷內,所有最後一次住院日與死亡日,二者間隔小於或等於30天之末期病人基本資料及此住院期間,病人部份之實驗室檢查數據作為輸入變項,並使用資料分類技術,包括決策樹(C5.0、C&RT)、支援向量機(Support Vector Machines, SVM)及類神經網路(Neural Networks, NN)等,發展建構末期病人短期存活時間之預測模型。研究比較各組模型之正確率,非癌症病人組以NN、C&RT、C5.0,結合癌症與非癌症病人組以C5.0分類技術進行預測時,預測正確率可達九成以上;同時決策樹分類技術(90.22%)應用於末期住院病人30天以內存活時間之預測平均正確率優於類神經網路(88.93%)及支援向量機(88.03%)。在屬性敏感度分析結果,末期病人在第二、三週檢驗血清白蛋白(Albumin)、第三週檢驗C反應蛋白(CRP)之異動情形,在預測其30天以內之存活機率上具有指標性意義。病人入住醫院後,其常規執行之一般血液及生化檢驗值可發現與末期病人短期存活時間具有顯著之關聯性,若能利用臨床醫療常規檢查作業之可得性,自動載入資料進入預測模式並有效的轉變為有用的資訊及知識,將可正確協助臨床工作人員,早期推論末期病人此次入住後之存活機率及時間,以作為醫療照護時支援決策之有利工具。期望此篇研究結果除能提供臨床工作人員有用資訊外,亦能夠激發出政府及醫療政策主導者對末期照護與臨終關懷等相關議題的重視。
This research tracks back and uses among 2004-2010 years, in the electronic charts of one medical information system of teaching hospital in Taoyuan, all is in hospital day and death day for the last time, the interval of two is less than or equal to 30 days in hospital, check the data as inputting changing one in the laboratory of patient's component, and use the materials to classify technological C4.5, C&RT, SVM and NN, etc., development build construct to a short-term survive prediction model of terminal stage. Research compares the accuracy and precision of every group of models, in non-cancer patient's group with NN, C&RT, C5.0, while combining cancer and non predicting with C5.0 categorised technology by cancer patient's groups, predict the accuracy can be more than 90%; Categorised technology (90.22%) of the decision tree at the same time Survive prediction of time apply will it be 30 day such as inpatient terminal stage to, accuracy superior to kinds of neural network (88.93%) on average And support vector machine (88.03%) . In the attribute sensitivity analysis result, patient in second and third weeks, examine Albumin, third weeks examine CRP, have indicator nature meanings on the surviving probability within predicting its 30 days. After the patient moves in the hospital, its regular general blood and biochemistry examining value that carry out can find the patient with terminal stage the time of surviving has apparent concerned withing nature in a short time, can utilize clinical medical the intersection of routine inspection and homework can get, write into materials, enter, predict way and changing useful information and knowledge into effective automatically, can assist clinical staff members, patient's surviving probability and time after moving in this time in early inference terminal stage correctly, in order to be regarded as medical treatment to support the favorable tool of decision while looking after. Expect this result of study besides can offer clinical staff member's useful information ing, can also stimulate government and medical policy leader out to the attention in relevant topics such as looking after and approaching one's end of life care in terminal stage.