腎臟移植手術之住院天數與醫療費用評估研究

A Study on Length of Stay and Expenditure on Surgery of Kidney Transplantation

陳冠廷(Kuan-Ting Chen)

指導教授 ： 張俊郎

國立虎尾科技大學/管理學院/工業工程與管理研究所/碩士(2012年)

摘要

末期腎臟疾病在台灣之發生率排行全球第二名，而盛行率則位居全球之冠，伴隨如此之高的發生率與盛行率，加上人口結構快速老化，隨之而來的是患病人數與日劇增，造成龐大的醫療費用支出，使健保財務上產生重大的問題與負擔，因此如何有效的評估住院天數與醫療費用對於醫療資源的分配與控制是有相當幫助的。而常見的末期腎臟疾病治療，以血液透析、腹膜透析與腎臟移植等三種為主，其中透析治療俗稱洗腎，只能控制病情卻無法改善，因此病患必頇長期洗腎，故考量病患長期醫療費用支出以及病情改善等諸多因素，腎臟移植被視為病患之治療首選。

本研究經由文獻蒐集，彙整其相關影響因子，並以全民健保資料庫四年中進行腎臟移植之病患做為研究對象，透過決策樹、類神經網路、支援向量機以及其相互結合等方法建構出六種住院天數評估模型，再以案例式推理技術建構住院天數與醫療費用評估系統，以提供醫療管理機構、醫療服務提供者、病患與家屬做為評估整合醫療資源之參考依據。由本研究所建構的住院天數評估模型中，以倒傳遞類神經網路、決策樹結合倒傳遞類神經網路以及倒傳遞類神經網路結合支援向量機三種模型具有較佳的評估能力，其帄均準確度與ROC曲線下面積，分別為77%與0.645以上，且經統計檢定結果顯示與其它模型皆有顯著差異；在案例式推理系統上，住院天數與醫療費用兩種指標皆以倒傳遞類神經網路進行權重值分析結合案例式推理系統，有較好的案例比對結果，其準確度皆有80%以上。

關鍵字

腎臟移植 ； 住院天數 ； 決策樹 ； 倒傳遞類神經網路 ； 支援向量機 ； 案例式推理 ； 粒子群演算法

並列摘要

Taiwan ranks the 2nd place in the world in its end-stage kidney incidence rate and 1st place in the prevalence rate respectively. Such the amazingly incidence and prevalence rates coupled with rapid aging population structure lead to increasingly patients and immense hospital and healthcare costs, deteriorating the financial burden for the national health healthcare. It’s be pretty helpful if it can evaluate effectively the hospitalization days and medical cost distribution and control. The frequently-occurred end-stage kidney diseases can be cure mainly by hemodialysis, peritoneal dialysis and kidney transplant. The peritoneal dialysis is so-called the kidney dialysis which can only control but can't improve the disease deterioration. Therefore, the patients shall take the kidney dialysis treatment for long term. For many factors consideration, such as the patients’ long-term medical costs and disease improvement, the kidney transplant is regarded by the patients as the most favorite.

There have conducted the reference documents collection this research for summarizing the influential factors and adopted the patients who have taken the kidney transplant surgery as the research objects from the national healthcare database, in conjunction with the decision tree, artificial neural network, support vector machine and the mutually integrating methods for constructing 6 models for appraising the hospitalization days. Further, there applied the empirical reasoning for constructing the evaluation system for the length of stay and medical expenditure that can be referred by the medical institutes, medical service providers, patients’ families for evaluating the medical resources. Among all the models for evaluating the length of stay established by this research, such the 3 models as the back propagation network, decision tree combined with back propagation network and the back propagation combined with support vector machine can reach better appraisal efficiency. Their average accuracy and the areas under the ROC (receiver operative characteristic) curve are over 77% and 0.645. The statistics testing indicates that the results are proven to have substantial difference with other models. In term of the empirical case reasoning system, when there applies the back propagation network in the weight coupled with the case-based reasoning, such two indicators as the length of stay and medical expenditure can demonstrate better comparison results of which the accuracy are more than 80%.

並列關鍵字

Kidney Transplantation ； Length of Stay ； Decision Tree ； Back Propagation Neural Network ； Support Vector Machines ； Case-Based Reasoning ； Particle Swarm Optimization

參考文獻

1. 中央健康保險局，取自http://www.nhi.gov.tw/，參考日期：2012/5/22。

2. 內政部統計處，取自http://www.moi.gov.tw/stat/，參考日期：2012/5/22。

3. 王孚玠、吳明儒(2004)，「腎臟移植病患之惡性腫瘤概論」，腎臟與透析，第十六卷，第二期，第81-90頁。 4. 王派洲(2008)，資料探勘的概念與方法，滄海書局，台中。

5. 王庭荃、陳長興 (2008)，「醫師年資、醫療服務量與消化性潰瘍治療效果之相關研究」，台灣衛誌，第二十七卷，第一期，第57-66頁。

6. 台北榮總血液透析室，取自http://homepage.vghtpe.gov.tw/~neph/hdu/edu-hd.htm，參考日期：2012/4/18。