Adhoc Analysis of Site By Treatment Interaction

In order to enroll patients more quickly, and to help establish efficacy in a new drug, the sponsor of the drug may have multiple investigators conducting a clinical trial. Each investigator enrolls and treats patients at a particular center or site. Such a clinical trial is considered a multicenter study. In the analysis of results from such a trial, it is expected that sites will contribute a small effect so that including site as a covariate in the analysis of treatment effects is a common means of increasing the power of a multi-center clinical trial. Controlling for Site and Site by Treatment effects in a linear model increases the power of analysis when those factors are found to contribute significantly to reducing the residual error of the model. There may be factors associated with site, such as patient demographics and medical history, as well as site characteristics which may not have been otherwise accounted for, which increase or decrease a response variable for all treatments or specific treatments. ICH Guideline E91 recommends that if treatment effects vary by site, this should be explained, requiring further analyses of an ad-hoc nature. These can follow two possible strategies: looking for sites which are outliers, and looking for sites which have commonalities with other sites. By starting with residuals of the site and treatment-only linear model, we assume that simple treatment effects are removed. The residuals are then averaged by site and treatment and displayed graphically to look for sites which are similar