General Departmental Seminar Series
Modeling Longitudinal Bivariate Ordinal Outcomes subject to Informative Dropout
David Todem, Ph.D. candidate, Departments of Statistics, UW-Madison
Friday, February 28, 2003, 12-1 p.m.
6225 Medical Sciences Center (MSC), 1300 University Avenue
Analysis of clinical trials of antidepressant drugs is typically based on a bivariate ordinal vector comprising information on efficacy and safety parameters. Additionally, it is often the case that subjects drop out prematurely from study, yielding unbalanced data. A statistical analysis of these data raises a number of challenging issues. The clustering due to repeated observations from the same subject and the multiplicity of outcomes necessitate the use of methods for correlated data. We use the concept of a latent variable to derive the joint distribution of bivariate ordinal outcomes, and then extend the model to allow for longitudinal data. Equal emphasis is put on the dropout process. We establish necessary and sufficient conditions for the existence of the maximum likelihood estimates for the model parameters. Desirable statistical properties such as consistency and asymptotic normality are also studied and established under mild conditions. Data from a psychiatric study, the Fluvoxamine trial, illustrate the proposed model. Selection models, specifically the shared random effects models, are used for studying the sensitivity of the results to dropout. Generalization from bivariate to multivariate models is also discussed.
Back to General Departmental Seminar Series