TABLE OF CONTENTS It cannot be overemphasised that statistical analyses of clinical trials should be no more complex than necessary. So even when repeated measures data have been collected it is not always essential to apply a formal repeated measures analysis. We have already described analyses of individual summary measures such as individual mean response scores and proportions that may often not only be statistically and scientifically adequate for the testing of simple treatment differences, but can also be more persuasive and easier to communicate than some of the more ambitious analyses that we now turn to. Nonetheless, a typical trial does not take place in a scientific vacuum in which a simple treatment difference is the only question of interest. It is more usual that, in addition to being used as the basis of formal evidence for efficacy or equivalence, even a phase III trial will gather data that may be informative as to the mode of action of the treatment, dose-response relationship, response heterogeneity, side effects and so on. In addition, attrition and variation in compliance may need to be more carefully examined than can be done within, for example, a simple summary statistics approach.
In this chapter, models suitable for analysing longitudinal data from clinical trials where the response variable can be assumed to have a conditionally normal distribution will be described. Models suitable for non normal responses will be the subject of Chapter 7.
Diggle (1988) lists a...
