Because I review manuscripts and protocols for clinical trials, I observe a number of recurring issues. These issues mostly relate to the statistical analysis but also include appropriate characterization of the trial itself.
Many reports of randomized controlled trials contain a table showing the baseline characteristics for each of the treatment groups, and some journals advocate use of significance tests to compare the groups at baseline. Indeed, the CONSORT (consolidated standards of reporting trials) statement supports inclusion of a baseline table; it also warns of the inappropriateness of using significance tests for comparison1. In statistics, hypothesis/significance tests concern population variables, and if indeed the allocation was randomized, a null hypothesis of no difference is true. Moreover, as Senn points out, an imbalance does not necessarily imply a problem with randomization; nor does a lack of imbalance prove randomization was successful2. Often, the results of baseline testing are used to decide which variables, if any, to include in an adjusted analysis of the outcome. Pocock, et al point out that baseline imbalance does not dictate the need for adjustment but rather it is the strength of the relationship between a baseline variable and the outcome3. Instead of p values, baseline comparability should be considered from the standpoint of clinical significance. Even then, imbalance should not be the criterion for inclusion in an adjusted model. In …
Address correspondence to K.E. Thorpe, Dalla Lana School of Public Health, 155 College St., 6th floor, Toronto, Ontario M5T 3M7, Canada. E-mail: kevin.thorpe{at}utoronto.ca