Multidimensional longitudinal data: estimating a treatment effect from continuous, discrete, or time to event response variables.

Multidimensional data arise when a number of different response variables are required to measure the outcome of interest. Examples of such outcomes include quality of life, cognitive ability, and health status. The goal of this article is to develop a methodology to estimate a treatment effect from multidimensional data that have been collected longitudinally using continuous, discrete, or time-to-event responses or a mixture of these types of responses. A transformation of the time scale that does not depend on the units of the response variables is used to capture the effect of treatment. This allows information about the treatment effect to be combined across response variables of different types. The model is specified using a pair of regression models for the first two moments, and generalized estimating equations are used for parameter estimation. The methodology is applied to quality-of-life data from an AIDS clinical trial and health status data from an Alzheimer's disease study.