Background: Careful consideration of site effects is important in the analysis of multi-site clinical trials for drug abuse treatment. The statistical choices for modeling these effects have implications for both trial planning and interpretation of findings.
Objectives: Three broad approaches for modeling site effects are presented: omitting site from the analysis; modeling site as a fixed effect; and modeling site as a random effect. Both the direct effect of site and the interaction of site and treatment are considered.
Methods: The statistical model, and consequences, for each approach are presented along with examples from existing clinical trials. Power analysis calculations provide sample size requirements for adequate statistical power for studies utilizing 6, 8, 10, 12, 14, and 16 treatment sites.
Results: Results of the power analyses showed that the total sample required falls rapidly as the number of sites increases in the random effect approach. In the fixed effect approach in which the interaction of site and treatment is considered, the required number of participants per site decreases as the number of sites increases.
Conclusions: Ignoring site effects is not a viable option in multi-site clinical trials. There are advantages and disadvantages to the fixed effect and random effect approaches to modeling site effects.
Scientific significance: The distinction between efficacy trials and effectiveness trials is rarely sharp. The choice between random effect and fixed effect statistical modeling can provide different benefits depending on the goals of the study.