Statistical Modeling and Evaluation of the Survival Data from the Discharge of Hospital Intensive Care Unit
Due to the different health conditions of an increasing number of serious patients, the intensive care unit (ICU) of a hospital has to correctly classify patients according to their conditions so that medical resources could be properly utilized. The seriousness of the illness can be classified based on the significant risk factors and its corresponding impacts on the patients survival. Large numbers of statistical approaches on empirical data have been purposed in the literature to solve this problem; however there is no widely accepted approach due to its difficulty. This paper mainly assesses classification performance of three statistical models of survival ICU data in hospital. Prior to investigate the relationship between the outcome and risk factors, the exploratory data analysis (EDA) of these survival data is conducted to reduce the number of nonsignificant variables. After preprocessing, three commonly used classification strategies, including multiple logistic regression (MLR), classification trees and linear discriminant analysis (LDA) are introduced to classify ICU patients, and the comparison of their performances are carried on towards model robustness by varying testing sample sizes. To evaluate the performance, a cohort of 200 consecutive ICU patients with 19 variables was borrowed for study. The result shows that MLR with EDA provides more satisfactory identification performance in terms of ROC curve and AUC.
ICU data EDA multiple logistic regression linear discriminant analysis classification trees
Lili Chen Xi Zhang Xiaoyun Xu
Department of Industrial Engineering and Management,Peking University, Beijing, China
国际会议
The Institute Industrial Engineera Asian Conference 2011(2011年国际工业工程师协会亚洲会议)
上海
英文
207-215
2011-06-10（万方平台首次上网日期，不代表论文的发表时间）