| Summary: | 碩士 === 國立雲林科技大學 === 工業工程與管理研究所碩士班 === 100 === Over the years, the Bureau of National Health Insurance costs have continued to rise. In 2010, the hospitalization costs accounted for 32.9% of the NHI medical expenses. The length of hospitalization has a major impact on medical costs. The purpose of this study is to find a way to predict the length of hospital stays and find ways to reduce the length of these stays to lower overall costs to hospitals. Hospitalization costs of catastrophic illness in 2009, cancer treatment up to 29.3 billion (42.9%), therefore, this study uses the 2009-2010 National Health Insurance database for lung, liver and colorectal cancer patients using data mining techniques to predict the number of days patients will remain in hospitals. With a model that can predict the length of patient stays in hospitals, hospitals can make better decisions about allocation of resources. We constructed a model using C5.0 decision tree, the back-propagation neural and C5.0 decision trees combined with back-propagation nerve to predict hospitalization lengths. The results showed that the C5.0 decision tree was the best forecasting model, and the accuracy rate was 78.13%. In the decision tree rules, the hospital surgical volume of services, hospital ownership and whether patients had other diseases were the key variables to assess the length of hospital stays. The results of this study may be of value to hospitals and to the BNHI as a reference.
|
|---|
