Applying Decision Tree Induction Model to Analyze the Patients’ Behaviors in the Emergency Department

• Main Authors: CHEN,CHI-YUAN, 陳祺元
• Other Authors: WU,I-CHIN
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/04738728706357692691

碩士 === 輔仁大學 === 資訊管理學系碩士班 === 104 === Since the start of National Health Insurance (NHI) in Taiwan in 1995, the outpatient numbers and medical expenses have grown quickly. In addition, compared to the general outpatient visits, the emergency department (ED) provides immediate medical services. Thus, the demand for emergency medical services has increased in recent years. In other words, the ED has become the most important and busiest unit within most hospitals. The disparity in hospital supply and patient demand has caused long-term overcrowding in EDs—also known as the ED overcrowding problem. This research aims to analyze patients’ behaviors through Length of Stay (LOS), Taiwan Triage and Acuity Scale (TTAS), and ICD-9-CM codes by data mining techniques. Basically, we analyze the relationship between various types of patients’ behaviors under various LOS for the cooperating hospitals. Then, we adapt an information systems portfolio prioritization (ISPP) model to build our research model in the ED which is suitable to our research domain. Accordingly, we mainly adopt decision-tree methods (e.g., J48 and CART) to build a prediction model with explanation capability. We attempt to adjust input parameters and attribute to findings about patient behavior and generate behavior trees of patients of various LOS. Then we measure the quality of the tree by communicability and consistency indices proposed by the ISPP model. The major findings in this research are las follows: (1) Adjusting the value scale of the input attribute (i.e., number of LABs) can increase the explanatory capability and accuracy of the decision tree; (2) We confirm that non-urgent patients with short LOS in the ED comprise the major group causing ED overcrowding; and (3) Finally, we find that CD-9-CM codes, age, number of LABs, and arrival time are important attributes to analyze behaviors of the patients who cause ED overcrowding. Finally, the research results can serve as a reference for EDs for investigating crowding problems and, hopefully, help EDs provide a higher quality of medical services.