Multilevel models for repeated measurement data
in clinical trials



Simon Thompson
Department of Medical Statistics and Evaluation
Royal Postgraduate Medical School, London


Abstract

Multilevel models, that is hierarchical random coefficient models, have considerable potential for the analysis of repeated measurement data in clinical trials. This is exemplified here by the analysis of quality of life data, which are multivariate and unbalanced. Not only do multilevel models provide a flexible modelling framework for the investigation of underlying average behaviour, for example yielding simple estimates of overall treatment effects, but they also permit a description of the differences between individuals. The assumptions of normality, homogeneity, and independence of the within and between subject variance components can be investigated. Indeed the models can be used to provide explicit modelling of variance heterogeneity. An important extension to multilevel models is for the analysis of binary outcome data. It is concluded that multilevel models, for which software is now available, provide a natural and powerful approach to the analysis of longitudinal data in general, and multidimensional quality of life data in particular.

Reference: Goldstein H. Multilevel statistical models (2nd edition). London: Edward Arnold, 1995.