| Summary: | 碩士 === 輔仁大學 === 資訊管理學系碩士班 === 106 === The demand for emergency medical services has increased in recent years. The number of ED visits has increased from 1,157,189 in 2005 to 1,164,340 from 2011 to 2016 in the great Taipei based on statistics published by Taiwan’s Ministry of Health and Welfare. Accordingly, ED crowding is a national problem that requires a solution. In response, we applied data mining techniques to analyze the relationship between characteristics of patients and ED overcrowding conditions mainly from the peseptives of legth of stay (LOS), triage-levels, and ICD-9 CM codes.
This research adopted C4.5 algorithm (J48 in Weka) to build the LOSs’ prediction model for explainating patients’ ED visists behaviors under various triage levels. In this research we mainly selected invidual attributes, e.g., age, arrival way, and medical diagnosies attributes, e.g., ICD codes, triage, frequency of X-Ray, to predict the pateints’ LOSs under different triage levels. We then interpreted results by dection trees (DT) with assocated profiles to measure the the consistency and communicability of the DT and ultimately scored the tree according to clinical values and clinical correctness after consulting the attending physician in the ED.
The preliminary major findings of our research as follows are several. First, ICD-9-CM codes, frquency of laboratories, and patient arrival times are important attributes for explaning the LOSs of patients. Second, the patients belongs to the third level of triage is the major groups of ED visits. The results show that intergrating ICD codes into the mining process can achive high clinical values and clinical correctness. Thrid, the results confirmed that exceptional group of pattines are hard to induce rules to explan the characterisits of pateints after confirming with the attending physicians. Our results can serve as a reference for understanding the overcrowding conditions in efforts to provide higher-quality care.
|
|---|
