急性腎衰竭(Acute Renal Failure, ARF)在加護病房中(Intensive Care Unite, ICU)約有5 ~20%的病患會發生的嚴重併發症,死亡率更高達50 % ~ 60 %。醫學預後(Medical Prognosis)是評估病患的存活能力和治療方式。國內急性腎衰竭腎衰竭預後的研究並不多,若能有一套急性腎衰竭預後模型,將可降低死亡的風險並且增加腎臟康復的機率。 本研究資料來源為西元2001年至2006年北部某醫學中心併發急性腎衰竭的病患在入院檢驗階段、加護病房檢驗階段與透析前檢驗階段等三個階段的資料,預後結果分為二階層:第一階層為發病結果,第二階層針對發病結果為存活的病患再探討其康復狀態。本研究分別應用決策樹(Decision Tree)與Logistic迴歸(Logistic Regression)篩選出急性腎衰竭預後因子,再分別應用貝式分類法(Bayesian Classification)和倒傳遞類神經網路(Back-Propagation Network)建構預後模型。 研究結果,第一階層的最佳預後模型為以透析前檢驗資料為自變項,應用決策樹搭配倒傳遞類神經網路。第二階層的最佳預後模型為以入院時檢驗資料為自變項,使用決策樹搭配貝式分類法。本研究結果可提供給醫療院所作為參考,將可以提升醫療品質與降低不必要的醫療糾紛。
Acute Renal Filure (ARF) , patients have incidence serious compication approximately 5%~20% in Intensive Care Unite (ICU). The patients’ death rate approximately 50%~60%. The Medical Prognosis is a way to evaluate survival and treatement. In domestic, the researches of ARF are inadequate, if we have a prognosis model of ARF, we expect to reduce death risk and increase recovery of kidney. This paper’s data is come from someone medical center of north Taiwan since A.D. 2001 to A.D. 2006. The data are definition three stages. There show of Hospital diagnostic test stage, ICU diagnostic test stage and Pre-dialysis diagnostic test stage. The prognosis result have difition two levels: the first level is incidence result, and the second level is focus to survival patients’ recovery stature.This research use Decision Tree & Logistic Regression to picking up Acute Renal Failure Prognosis factor, and then we use Bayesian Classification & Back-Propagation Network to building a model. Accroding to research result, the best prognosis model of first level for pre-dialysis is independent variable with decision tree and back-propagation networks. The best prognosis model of second level for hospital diagnostic test data is independent variable with decision tree and Bayesian classification. This paper’s consultation expect to provide medical agency. Hoping to increase medical quality, and reduce medical dispute unnecessarily.