Open Collections

Analysis of longitudinal data from the betaseron multiple sclerosis clinical trial

University of British Columbia

Longitudinal data sets consist of repeated observations for each subject over time; and often a corresponding set of covariates is available. Analysis of longitudinal data is often based on summaries over time. Summarizing the data allows one to use simple techniques for analysis but does not allow analysis of the patterns over time and does not take advantage of the within subject information. In many fields, repeated measures analysis of variance and multivariate analysis of variance are commonly used to analyze longitudinal data on continuous responses. Such analyses are appropriate only when the responses for each subject are multivariate Gaussian with a common covariance matrix for all subjects. In addition, all subjects are required to have measurements at exactly the same times, and no missing values may be present. In many cases, however, the longitudinal response does not satisfy these assumptions. Therefore, application of the traditional methods of analysis is limited even for continuous responses. This thesis discusses and compares several more recently developed methods for the analysis of longitudinal data. One method, the generalized estimating equations approach, requires only minimal assumptions about the true correlation structure in the data for each subject to yield consistent estimates of regression parameters and their standard errors. The method can be applied to binary and count data as well as to continuous data. Another method, the random effects regression model, is limited to the analysis of continuous responses. An advantage of this method is that in addition to estimating population average parameters it also allows estimation of individual parameters for each subject. Finally, the modification of the random effects regression approach for the analysis of ordinal responses, the mixed effects ordinal logistic regression model, is presented. The methods are extensively illustrated using the data from the Betaseron clinical trial in relapsing-remitting multiple sclerosis (MS). These methods facilitated the examination of patterns over time, therefore they not only identified the presence of treatment effect, but also indicated the nature of the effect. Hence, these methods enable much more information to be extracted from the MS data set than the traditional ANOVA-based methods, and therefore provide useful and powerful tools for researchers in this subject area.

Thesis/Dissertation

application/pdf

2009-03-21

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/6342

Statistics

University of British Columbia

UBCV

DSpace