多反应变量重复测量资料的协方差类型模型应用及SAS实现

目的 探讨协方差类型模型在多反应变量的重复测量资料分析中的应用方法 为了评价盐酸吡格列酮片治疗2型糖尿病的有效性,以安慰剂为对照,对240例2型糖尿病患者的空腹血糖和餐后2 h血糖重复观测数据进行多反应变量的协方差类型模型分析,对模型的固定效应参数矩阵作最小二乘估计并进行组间比较,同时给出误差效应的方差协方差矩阵,利用...

Application of covariance pattern model for multivariate repeated measures data and implementation of SAS

Objective Applying covariance pattern model to analyze the multivariate repeated measurement data. Methods In order to assess the effectiveness of Pioglitazone hydrochloride for Type II Diabetes mellitus, 240 patients with Type II Di- abetes mellitus were arranged to randomly take 2 kinds of medicine, Pioglitazone hydrochloride or placebo. Both fasting plasma glucose and 2 h plasma glucose repeated measurement data were analyzed by covariance pattern model. The fixed effect parameters matrix of model coefficients were estimated by using least squares estimation method, the effects between treatment groups were compared and the variance-covariance matrices was also estimated. Corresponding analysis methods were programmed with MIXED procedure of SAS software. Results There was statistically significant difference between the 2 groups of fasting plasma glucose and 2 h plasma glucose overall （ P〈 0.01 ） ; There was statistically significant difference among different time points （ P 〈 0.01 ） ; The interaction effects of response variables category and time was statistically significant （P〈0.05）, it indicats that the trend changing with time was different between fasting plasma glucose and 2 h plasma glucose. The interaction effects of group and time was statistically significant （P〈0.01）, it indicats that the trend of blood sugar changing with time was different between the 2 groups. Estimated parameters with fixed effect were obtained and graphs were drawn. Both fasting plasma glucose and 2 h plasma glucose changed with time after treatment and the trends between 2 groups were different. Curve of treatment group fell rapidly over time, and curve of placebo group was very gentle changing with time. Conclusion The covariance pattern model for the multivariate repeated measurement data can handle the materials with random missing data, and allow different observed times and observation time for each observation object. It can directly build the model on the correlation structure through specifying the variance-covariance matrices of the dependent data. The model consider not only the correlation between multiple repeated measurements of each response variable, but also the correlation of all the response variables. The model can effectively analyze the dynamic change trend of multivariate repeated measurement data, so the statistical conclusion is more credible.