American Journal of Epidemiology Advance Access originally published online on October 16, 2006
American Journal of Epidemiology 2006 164(11):1138-1139; doi:10.1093/aje/kwj359
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
LETTERS TO THE EDITOR |
RE: "WHEN IS BASELINE ADJUSTMENT USEFUL IN ANALYSES OF CHANGE? AN EXAMPLE WITH EDUCATION AND COGNITIVE CHANGE"
Department of Statistics, Radiation Effects Research Foundation, Hiroshima 732-0815, Japan
(e-mail: jcologne{at}rerf.jp)
The article by Glymour et al. (1) provides a lucid and educational description of how to approach an important analysis problem—whether to adjust for the baseline level of the dependent variable in observational studies of a change in that variable. I think it is important to carefully define what is meant by "baseline."
Initial measurements taken at the start of follow-up in observational studies are different from initial conditions or states prior to the application of exposure in experimental studies because exposure typically precedes the observation period in the former, whereas exposure is applied after randomization and initiation of the study in the latter. Baseline measurements in observational studies are therefore subject to spurious effects due to uncontrollable circumstances, such as self-selection of exposure status. With experimental studies, on the other hand, baseline measurements are not subject to confounding with exposure and are unlikely to be confounded with unmeasured factors if randomization is adequate.
Of course, in either type of study, the baseline measurement could be affected by "the horse-racing effect" (1, p. 273), a change already underway at the time the study is initiated, as mentioned by the authors. The simple fact that adjustment for baseline might reduce the power to detect the impact of explanatory variables on change is grounds for concern in general about baseline adjustment. In my opinion, when considering possible adjustment, only an initial state that precedes exposure and exists prior to any onset of change should be considered a true "baseline."
 |
ACKNOWLEDGMENTS |
|---|
Conflict of interest: none declared.
|
References |
|---|
 TOP References |
|---|
- Glymour MM, Weuve J, Berkman LF, et al. (2005) When is baseline adjustment useful in analyses of change? An example with education and cognitive change. Am J Epidemiol 162:267–78.
[Abstract/Free Full Text]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||