General Departmental Seminar Series
Should slope estimates from clinical trials be adjusted for baseline values?
David Reboussin, Associate Professor of Public Health Sciences, Wake Forest University
Friday, October 11, 2002
G5/142 Clinical Sciences Center, 600 Highland Ave.
Many clinical trials are designed to collect repeated measurements on participants over time in order to assess rates of change. Mixed effect analysis of variance models as described by Laird and Ware are a popular and flexible approach to the estimation and testing of treatment effects in this setting. There are a variety of choices for model specification for such an analysis; for example, which effects should be considered random and which fixed, and whether any pre--randomization value should be used as a covariate or a dependent variable. It has been widely recommended, and it is intuitively appealing, to compute slope estimates in repeated measures data using baseline values as a covariate on grounds that there is a gain in precision. In certain cases, however, it can be shown that adjusting for the baseline produces a less efficient estimate than using all measurements as dependent variables.
Back to General Departmental Seminar Series