Biostatistics Journal Club: Adjustment for Baseline Response in Longitudinal Studies

Historically, two main analytic approaches have been proposed for adjusting for baseline response in longitudinal studies. The first approach is based on the very simple and intuitive idea of “change scores”. The second approach is based on an analysis of post-baseline responses where an adjustment for baseline response is made by regarding it as a covariate (often this approach is referred to as an “ANCOVA method”). Although much has been written about these two approaches, research scientists seem to be confused about how to choose between them. This may be due in part to a number of statistical myths and misconceptions surrounding the two alternative methods.

In this presentation, Garrett Fitzmaurice, ScD, McLean Hospital, will review these two methods for adjusting for baseline response in longitudinal studies. For ease of exposition, this presentation will consider the two-group study design and focus on change between two time points; however, the main results apply to more general longitudinal study designs with multiple time points. Fitzmaurice will try to dispel some of the statistical folklore surrounding these two approaches and make some practical recommendations concerning their use in applications.

Speaker:

Garrett Fitzmaurice, ScD, McLean Hospital

Required Reading:

None required (Caveat: The statistical literature is replete with papers providing conflicting recommendations).