Constructing a Prognosis Warning Model for the Hemodialysis Patients

• Main Authors: Yu-Ming Chen, 陳育民
• Other Authors: Bor-Wen Cheng
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/45652686019009593033

碩士 === 國立雲林科技大學 === 工業工程與管理研究所碩士班 === 95 === According to the statistics from the Department of Health, nephritis, renal syndrome and renal diseases is one of the leading causes of death during the years of 1996-2005 in Taiwan. The number of annual deaths due to the above causes is increasing. In Taiwan, about 93% of uremic patients are using hemodialysis. Therefore, hemodialysis has become the major issue of uremic patients. The main purpose of this study is to apply C5.0 algorithm, and classification and regression trees (CART) to construct a warning model for long-term hemodialysis patients to help medical staffs understand their patients overall regimen and to decrease the death due to hemodialysis. The data of the study were collected from a regional hospital located in central Taiwan. The long-term hemodialysis patients take blood test for analysis every three months. After the analysis, we discuss with renal physicians to delete non-essential biochemical parameters and use factor analysis to reduce the indicators. Through C5.0 algorithm and CART, we build a warning model for long-term hemodialysis patients. The results of the study showed that CART model has the highest accurate rate, 89.75%, for training and 97.73% for predicting. Finally, Albumin, UIBC and nPCR, TACurea, cholesterol, and Ca were found to be the most important biochemical indicators to affect long-term hemodialysis patients.