二次推断函数与广义估计方程缓控释制剂工艺数据建模比较

Comparison of Generalized Estimation Equation and Quadratic Inference Functions in Modeling Study of the Sustained-release and Controlled-release Formulation

目的比较二次推断函数(QIF)及广义估计方程(GEE)对具有重复测量特征的缓控释制剂制备工艺累积释放度数据的建模效果。方法选取三种作业相关矩阵对缓控释制剂制备工艺正交试验设计数据建立QIF模型并进行拟合优度检验,用评价指标AIC、BIC选择模型及选择合理的作业相关矩阵;基于此合理的作业相关矩阵建立广义估计方程(GEE)模型;用均方误差(MSE)比较QIF模型及GEE模型的建模效果。结果累积释放度数据用QIF方法以可交换相关矩阵建立的模型评价指标AIC、BIC最小(AIC=24.000、BIC=26.367),均方误差MSE=0.0123,而基于可交换相关矩阵建立的GEE模型MSE=32.6535,QIF的建模效果要优于GEE。结论在具有重复测量数据特征的缓控释制剂的建模研究中,二次推断函数建模可以用评价指标AIC、BIC选择模型,并且建模效果优于广义估计方程,具有应用价值。

Objective To compare the modeling effects of application of generalized estimation equation （GEE） and quadratic inference functions（QIF） in the modeling of the sustained-release and controlled-release formulation. Methods Em- ploys three different work correlation matrixes to establish equation with QIF and according to the test result of goodness of QIF, obtain a suitable work correlation matrix type by AIC and BIC. Based on the suitable work correlation matrix type, GEE and QIF were applied, and results were compared by MSE. Results According to the test result of goodness QIF, the AIC and BIC are minimal by adopting the exchangeable work correlation matrix type （AIC = 24. 000, BIC = 26. 367 ） and the MSE value is 0. 0123. Based on the same work correlation matrix type, GEE were applied in modeling and the MSE value is 32. 6535, which il- lustrates the QIF modeling effect is better than that of the GEE. Conclusion In the modeling of the sustained-release and con- trolled-release formulation which display repeated and interrelated character, QIF can select the modeling methods by using AIC and BIC criterion,and the QIF modeling effect is better than that of the GEE,so the QIF has value of application.