Statistical Methods for Ophthalmologic and Cluster Data

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Ophthalmologic data present a data analyst with challenging methodologic problems. First, data are present for two eyes of an individual, which are usually positively correlated, but not the same. The challenge is how to use all the data in an efficient manner that accounts for the correlation between scores on fellow eyes. Substantial progress has been made in this area in the context of normally and binomially distributed outcome variables. SAS procedures PROC MIXED and PROC GENMOD deal effectively with these types of data. However, many scales in ophthalmology are ordinal in nature (e.g. ETDRS diabetic retinopathy scale, LOGS III cataract scale) and some continuous scales (e.g. Humphrey visual field) tend to be skewed and non-normally distributed. Hence, nonparametric analyses are desirable for two-group comparisons. In the previous cycle of this grant, we have extended the Wilcoxon rank sum test and signed rank test to account for clustered data in the setting where the unit of randomization is the subject, but the unit of analysis is the eye, and both eyes of a subject are randomized to the same treatment. In this renewal application, we extend these methods to cover the case when the unit of randomization is the subject, but fellow eyes may or may not receive the same treatment. This occurs commonly in ocular allergy trials. In addition, we propose to extend our methodology to provide estimates of power and sample size for clustered data analyzed by the Wilcoxon rank sum test, and assess efficiencies of different methods of weighting when clusters are of unequal size. Furthermore, we propose to apply our methodology for the Wilcoxon rank sum test for clustered data to the important problem of assessing goodness of fit of GEE models either when fit to longitudinal data or to cross-sectional data with correlated outcomes (e.g. fellow eyes). Finally, in the previous cycle of the grant we have studied methods for assessing whether exposure-disease relationships are the same for each component of a composite endpoint (e.g. different types of AMD), with the restriction that a subject can have only one type present. In this cycle, we propose to extend these methods to allow for the case where a subject can have multiple disease subtypes at the same time (e.g. different types of cataract). The methods developed in this proposal as well as in the previous cycle will be available in the form of SAS macros which can be accessed at the following web site URL:(http://www.qeocities.com/bernardrosner/channina.html).