Heitjan D F, Sharma D
Division of Biostatistics, Columbia University School of Public Health, New York, NY 10032, USA.
Stat Med. 1997 Feb 28;16(4):347-55. doi: 10.1002/(sici)1097-0258(19970228)16:4<347::aid-sim423>3.0.co;2-w.
We present a model for describing repeated-series longitudinal data, that is, longitudinal data where each unit may yield multiple series of the same variable. Such data arise commonly in ophthalmologic studies, where one obtains measurements on the same variable for the right and left eyes at each clinic visit. We model the mean as a linear function of predictors, and assume that the error term is a sum of a random subject effect and a vector AR(1) process. We fit the model by maximum likelihood and assess the adequacy of the error assumptions by an extension of the empirical semivariogram. We apply our model for data from a clinical trial comparing two treatments for ocular hypertension and glaucoma, with intra-ocular pressure as the primary endpoint. Results suggest that autocorrelation within and between eyes is a significant feature of the variance model. Standard errors depend critically on the variance assumption.
我们提出了一个用于描述重复序列纵向数据的模型，即每个单元可能产生同一变量的多个序列的纵向数据。此类数据常见于眼科研究中，在每次临床就诊时，会对左右眼的同一变量进行测量。我们将均值建模为预测变量的线性函数，并假设误差项是随机个体效应与向量自回归（AR(1)）过程之和。我们通过最大似然法拟合模型，并通过经验半变异函数的扩展来评估误差假设的充分性。我们将模型应用于一项比较两种治疗眼压过高和青光眼的方法的临床试验数据，以眼压作为主要终点。结果表明，眼内和眼间的自相关性是方差模型的一个显著特征。标准误严重依赖于方差假设。
