Clinical trials: relevance of correlation between treatment responses

Abstract.

Objective: Trials that do not allow rejection of the null hypothesis of no treatment effect may have had an inappropriate design. Trials are virtually never assessed for correlation between responses to different treatment modalities.

Methods:

Using a hypothetical example and several published studies we examine the influence of correlation levels between treatment modalities on the sensitivity of testing.

Results:

The level of correlation between responses to different treatment modalities is a major determinant of the sensitivity both of crossover and parallel group clinical trials.

Conclusions:

It is very relevant to assess a priori correlation levels between responses to the different treatment modalities of a trial. If a negative correlation is anticipated, a crossover design is likely to lack sensitivity. If a positive correlation is anticipated a parallel-group design seems less appropriate, because it would lack the extra sensitivity of accounting for the positive correlation. Both designs would seem suitable for approximately zero correlations (e.g. comparison vs baseline or vs placebo under the assumption that the number of placebo responders is negligible).