摘要
目的 探索胃癌的危险因素 ,并探讨研究中存在的多因子共线性的处理方法。方法 采用病例对照方法 ,获得 5 0名胃癌患者和 5 0名对照的流行病学资料 ;PCR方法检测个体基因型 ;应用线性回归中的三个工具 ,对各研究因素进行共线性诊断 :用主成分分析改进的方法 ,得出并解释最终的回归模型。结果 多因素logistic回归结果与单因素分析结果不一致 ,共线性诊断显示方差膨胀因子普遍较大 ,GSTM 1基因型、肿瘤家族史等因素之间存在多因子共线性。应用主成分分析改进后的logistic回归模型拟合数据 ,不仅各回归系数的标准误均有减小 ,而且有更多的因素被选入模型。结论 遗传易感性和环境因素在胃癌的发生中共同起作用。对疾病危险因素进行logistic回归分析时 ,应首先进行原始变量的多重共线性诊断 。
Objective To explore the risk factor of gastric carcinoma,and to discuss the method of handling the problem of multi variable collinearity in logistic regression model.Methods A case control study was design to obtain the epidemiological data of 50 gastric carcinoma patients and 50 controls.Individual genotype was detected by PCR method.Extent three tools for diagnosis colliearity in standard regression models to likelihood estimation for logistic model.Modify the logistic regression by the principle of principal component and get a new estimation for logistic regression coefficients.Results The results of multi variable logistic regression was contrary to that of single variable analysis.Collinearity diagnosis for logistic regression show that many VIF is larger than 9,and variance decomposition proportion of GSTM1 genotype,tumor family history and intercept exceed 50%.Fitting the model by modified logistic regression show that the standard error of regression coefficients decreased,and more factors were selected into the model.Conclusion Genetic susceptibility and environmental factors have mutual effects on the onset of gastric carcinoma.The modified logistic regression model could avoid losing of information and give the data more reasonable explanation.
出处
《中国卫生统计》
CSCD
北大核心
2001年第3期141-144,共4页
Chinese Journal of Health Statistics
