Ignoring Site-Specific Treatment Effects in Multicenter Clinical Trials Increases False Positive Results

• Activity Number: 259
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Eric Macklin*
• Companies: Massachusetts General Hospital/Harvard University
• Keywords: multicenter clinical trials ; randomized block design ; variance estimation ; type I error

Abstract

Multicenter clinical trials with randomizations stratified by site represent a classic randomized block design with sites representing random blocks and treatment effects conditional on block. In traditional ANOVA, the appropriate treatment effect error term is the treatment x block interaction, with degrees of freedom equal to the number of blocks minus one. In modern analysis of multicenter clinical trials with longitudinal follow-up, this is implemented by acknowledging site-specific random effects for an intercept, treatment effect, and treatment x time interaction in a mixed effects model. Nevertheless, most analyses of multicenter clinical trials omit site-level treatment and treatment x time random effects and thus under-estimate the treatment effect standard error and over-estimate significance. In the current paper, I identify the bias in variance estimates obtained by failing to recognize site-specific effects of treatment in multicenter clinical trials and contrast estimates from the original analyses and corrected analyses of several ALS trials, suggesting at least one cause of our common failure to reproduce prior clinical trial results.