| Summary: | 碩士 === 國立雲林科技大學 === 工業工程與管理研究所碩士班 === 99 === According to the statistics of the Kidney Foundation ROC(2008), total of 45,894 patients receive hemodialysis treatment, and about 8000-9000 new patients appears each year in our domestic. The rate of occurrence ranks first in the world.
Arterio-venous fistula (AVF) is the second life of hemodialysis. In clinical, dialysis patients implement "ostomy " or " catheter implantation " which is one kind of fistula established. It can be used 1 to 2 years in normal state. However, some patients have poor blood vessels, those vessels are easy to plug. If vessels occlude, we must rework "ostomy " or " catheter. If the rate of fistula reestablish is too higher in the dialysis hospital, it means that failure rate needs to improve. Dialysis hospitals need to achieve the rate of reestablish 100 person-months≦2.
The case is the largest hospital dialysis center in Yunlin County. There are total about 260 hemodialysis patients, and dialysis attendance about 3500 people each month. The renewal rate of arterio-venous fistula was 2.4/100 person-months, it who over the threshold, 2.0/100 person-months, which who set by National Health Insurance Bureau. Therefore, we use the back-propagation neural network to predict AVF occlusion, C5.0 decision tree and Variable Precision Rough Set (VPRS) to find out the main factor rules of AVF occlusion, and use confusion matrix to evaluate performances. We discuss the AVF to predict occlusion or not, the causation of occlusion and the number of occlusion times expect to find the rule for health care workers.
The results show that back-propagation neural network performance of predict best than others. But we are unable to know the operation process in neural network. The results of C5.0 decision tree are more fit with practical situation. Those rules from decision tree are easy to understand and can be a reference for health care workers. Occlusion factors include Percutaneous Transluminal Angioplasty (PTA) or not, types of vessels, reinserted times, the years of hemodialysis. The results of VPRS are too many rules and the accuracy of predict are too low, it does not apply in our cases.
|
|---|
