决策树模型与回归模型在天津市某区公务员健康状况分析中的应用与比较  

Comparison between Decision Tree and Logistic Regression Applied in the Study of Health Status and Correlates in the Government Employee in a District of Tianjin

作　　者：魏凤江[1] 崔壮[2] 李长平[2] 宋春华[2] 朱宝[2] 刘媛媛[2] 马骏[2] 

机构地区：[1]天津医科大学基础医学研究中心,300070 [2]天津医科大学公共卫生学院卫生统计学教研室,300070

出　　处：《中国卫生统计》2013年第1期42-45,共4页Chinese Journal of Health Statistics

基　　金：全国统计科学研究计划项目(2009LY048);全国统计科学研究计划项目(2009LY011);教育部人文社会科学研究青年基金项目(11YJCZH080)

摘　　要：目的了解天津市某区公务员健康状况的影响因素,为提高该人群的健康水平提供依据。方法于2008年9～12月,采用整群抽样的方法对天津市某区公务员进行健康状况及影响因素的问卷调查。应用SAS8.2Enter-priseMiner模块建立决策树模型和回归模型,对该区公务员人群的健康状况影响因素进行分析和预测。结果该区公务员总体患病率为47.0%,模型筛检出影响健康状况的因素有:年龄、体质指数、吸烟、被动吸烟、饮酒、睡眠时间、按时吃饭情况、体育锻炼花费时间、文化程度、婚姻状况,亚健康分值、心理健康分值。将logistic回归模型与决策树模型进行预测性能的比较,ROC面积比较结果发现,两者差别无统计学意义(χ2=1.6073,P=0.2049)。结论公务员人群健康状况不容乐观,各种慢性病患病率较高,是今后开展健康管理的重点群体。Objective To comprehend the health status of government employee in a district of Tianjin, and to provide relative guidelines of health management for government employee. Methods A questionnaire survey was conducted within September to December, 2008. All participants were included by cluster sampling. Decision tree model and logistic regression model was conducted using SAS8.2 Enterprise Miner to analyze and predict the influential factors of the health status. Results The total prevalence rate was 47.0% ,Multi-variable analysis disclosed that age, BMI, smoking, passive smoking, drinking, sleep time, regular diet, intensity of physical exercise, education, marital status, sub-health scores, and mental health scores were associated with the health status. Use roc curve to comparise prediction effect between logistic regression model and the decision tree model, The results declosed that their was no statistical significance （X^2 = 1. 6073, P = 0. 2049 ）. Conclusion The health status of government employee was far from ideal for some chronic diseases, the health management should be emphasized in those personals in the future.

关 键 词：公务员 决策树 LOGISTIC回归 健康状况 

分 类 号：R195[医药卫生—卫生统计学]