Analyzing Ordinal Repeated Measures Data Using SAS®

نویسندگان

  • Bin Yang
  • Eli Lilly
چکیده

This paper provides a brief review of commonly used statistical methods for analyses of ordinal response data. Generalized CMH Score Tests of Marginal Homogeneity, GEE, and random-intercepts logistic regression ordinal model for analysis of repeated ordinal response data will be particularly discussed. SAS procedures, Proc NLMIXED, Proc GENMOD, Proc IML, and Proc FREQ for categorical ordinal analysis, are described and illustrated with data from a clinical trial. A SAS macro to produce estimated marginal Probabilities will be presented. INTRODUCTION Ordinal variables are common in clinical research studies. Ordinal variables have a hierarchical ordering, Such as Severity score (none, mild, moderate, severe). Repeated measures refer to multiple measurements taken from the experimental unit over time. For example, Many clinical studies require patients to return to the clinic at several times and observe the response variable at each visit. The layout for the sample data with two treatments is shown in the following: PATINET TREATMENT VISIT Y1 Y2 Y3 1000 A 1 111 y 112 y 113 y ... ... ... ... 6 161 y 162 y 163 y 1001 B 1 211 y 212 y 213 y ... ... ... ... 6 261 y 262 y 263 y These longitudinally repeated measures can be used to characterize a response profile over time. Repeated measures within a subject are usually positively correlated. The correlation structure of the longitudinal data can be described by random subject effects. Random subject effects indicate the degree of subject variation that exists in the population of subjects. Data from studies with repeated measurement in general are incomplete due to drop out. We will use terminology of little and Rubin (1987, Chapter 6) for the missing-value process. A non-response process is said to be missing completely a random (MCAR) if the missing is independent of both unobserved and observed data and missing at random (MAR) if, conditional on the observed data, the missing is independent of the unobserved measurements. When either of these is plausible, with a likelihood-based analysis it is not necessary to model the missingness mechanism (Agresti A. 2002). For ordinal response data, we can compare mean response score between treatments by assigning numerically score which are the ranks of the categories when ordered from smallest to largest. This continuous model does not take into account the ceiling and floor effects of the ordinal outcome. The result can be biased when the ordinal variable is highly skewed. Here we first use a generalized CMH test for testing marginal homogeneity at the end of the study. Next, we use the GEE method implemented in the GENMOD procedure to fit a marginal model. The mean response of the marginal model depends only on covariates variable, and not on random effects. GEE uses quasi-likelihood estimation and assumes that the missing data are MCAR. To accommodate random effects, we describe a mixed-effects proportional odds model using the NLMIXED procedure. The NLMIXED procedure relies on approximating the marginal log likelihood by integral approximation through Gaussian quadrature. The objective function for the NLMIXED procedure is the marginal log likelihood obtained by integrating out the random effects from the joint distribution of responses and random effects using quadrature techniques. Although these are very accurate, the number of random effects that can be practically managed is limited (SAS/STAT). Since the estimation of model parameters is based on Maximum likelihood approach, the missing data are assumed to be MAR.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Statistical Methods for the Analysis of Repeated Measurements

The best ebooks about Statistical Methods For The Analysis Of Repeated Measurements that you can get for free here by download this Statistical Methods For The Analysis Of Repeated Measurements and save to your desktop. This ebooks is under topic such as statistical methods for the analysis of repeated measurements statistical methods for the analysis of repeated measurements statistical method...

متن کامل

Mixed Effect and Spatial Correlation Models for Analyzing a Regional Spatial Dataset

Understanding the spatial variability of snow at many different scales is critically important for avalanche forecasting. This research uses new techniques to reanalyze an existing spatial dataset collected in southwest Montana’s Bridger Range during the 1996-97 winter. Recent developments in statistical software have greatly increased the ease with which mixed effects and spatially correlated ...

متن کامل

Estimating Ordinal Reliability Using SAS

In evaluation instruments and tests individual items are often collected using an ordinal measurement or likert type scale. Typically measures such as Cronbach’s alpha are estimated using the standard Pearson correlation. Gadderman, Guhn and Zumbo (2012) illustrate how using the standard Pearson correlations may yield biased estimates of reliability when the data are ordinal and present methodo...

متن کامل

Different Methods of Analyzing Multiple Samples Repeated Measures Data

Three methods of analysis viz. Standard ANOVA, Repeated Measures ANOVA and Linear Mixed Model were used to analyze two sets of data due to Cole and Grizzle (1966) and Crowder and Hand (1990) by using SAS 9.2. Four groups of dogs (dataset 1) were found to differ significantly with respect to blood histamine levels under all the three methods of analysis except linear mixed approach under H-F cov...

متن کامل

197-30: Analyzing Incomplete Binary Repeated Measures Data Using SAS®

This paper presents a case study in longitudinal data analysis where the goal is to estimate the efficacy of a new drug for treatment of MDD. Data characteristic indicate: 1. Subjects from different treatment groups drop out differentially across time. 2. There are a high proportion of subjects who never experience any response. To overcome these challenges, we developed a logistic random-effec...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2006