Bivariate Outcomes in Clinical Trials

Specialeforsvar: Majken Hamann Sey

Titel: Bivariate Outcomes in Clinical Trials

Resume: It is common practise to analyse longitudinal data with models which restrict attention to the analysis of one single outcome, measured repeatedly over time. Multivariate longitudinal data arise when instead of a single outcome, a set of different outcomes on the same unit is measured repeatedly over time. Multiple outcomes are often used to properly characterize an effect of interest. Specifically, the outcome of main interest is often not observable or is difficult to measure. This thesis explore potential advantages of modelling the two depression rating scales HAM-D and MADRS coherently. When only a single outcome is used in the analysis of longitudinal data, the univariate linear mixed model is commonly used. We use mainly the bivariate linear mixed models with various Kronecker covariance structures in order to explore the relation between the two outcomes, and to explore potential advantages of using the bivariate model compared to the univariate model. We do not find that there is much gained by jointly modelling HAM-D and MADRS.

Vejleder: Helle Sørensen
Censor: Søren Andersen