高血压危险因素logistic回归与分类树分析

Logistic regression and classification tree analysis on risk factors of hypertension

目的应用分类树模型分析广东省居民高血压的危险因素,并与logistic回归分析结果做比较。方法利用广东省2002年居民营养与健康状况调查数据,用分类树分析和logistic回归分析15 343名常住居民高血压的危险因素。结果高血压的主要危险因素为年龄、血脂及肥胖。分类树学习集的灵敏度为91.0%,特异度为81.6%,总正确率为83.3%,约登指数为0.726;检验集的灵敏度为83.8%,特异度为80.5%,总正确率为81.2%,约登指数为0.643。logisic回归分析的灵敏度为31.6%,特异度为95.7%,总正确率为84.5%,约登指数为0.273。结论对高血压患者,尤其是老年人,要调节血脂,减轻体重。分类树分析较logisic回归分析分类效果好,结果直观,便于解释。

Objective To analyse the risk factors of hypertension using the methods of classification tree and logistic regression analysis and to compare the results of these two methods. Methods A crosssectional analysis on hypertension was conducted using the data of 15 343 inhabits from the Inhabits Nutrition and Health Survey in Guangdong province in 2002. Results The important risk factors of hypertension were age, cholesteri and obesity. In classification tree analysis on hypertension, Youden＇s index for learning samples was 0. 726 （sensitivity was 91.0% and specificity was 81.6% ）, and 0. 643 （sensitivity was 83.8% and specificity was 80.5% ） for testing samples. Youden＇ s index of logistic regression analysis was 0. 273 （sensitivity was 31.6 % and specificity was 95.7 % ）. Conclusions For hypertensives （especially the elder）, keeping proper cholesteri concentration and weight were necessary. Compared with logistic regression model, classification tree model could help to judge the controls and patients better, and present the results more easily and understandably.