广义估计方程与多水平logistic回归模型在临床纵向数据分析中的应用及比较

目的探讨广义估计方程和多水平模型的应用与临床纵向研究以解决个体重复观测数据内部的相关性问题。方法根据临床纵向实例数据的特点，拟合因变量为二分类的广义估计方程和多水平模型，并与一般logistic模型比较。结果广义估计方程和多水平模型的分析结果与一般logistic模型不同。由于未能考虑个体内重复观测数据的相关性，一般logistic模型错误显示临床分期与近期疗效相关，而广义估计方程和多水平模型分析结果则显示相关无统计学意义。经分层分析也未发现临床分期与近期疗效的关联。结论广义估计方程和多水平模型都能有效地考虑重复观测数据内部相关性并能处理有缺失值的资料。与多水平模型相比，广义估计方程的参数估计较为稳定，可有效的估计各解释变量的效应。

Application and comparison of generalized estimation equation and multilevel logistic regression model in analyzing clinical longitudinal data

Objective To explore the application of Generalized Estimation Equation （GEE） and multilevel model in clinical longitudinal data to solve the problem of internal correlation of repeated measurement data. Methods Based on the characteristic of a clinical repeated measurement study, GEE and Multilevel logistic Model were applied and results was compared to that of a general logistic model. Results The results of GEE and multilevel model are different from that of general logistic mod- el. False positive results of logistic model are caused by ignorance of intra-individual correlation of repeated measurement. Con- clusion Both the GEE and Muhilevel Model efficiently considers the intra-individual correlation and also can deal with the missing value. The parameter estimation in GEE was more robust than multilevel model, and it can supply the effect size of covariates.