Does Meta-Regression Have Enough Power? An Empirical Study of 500 Meta-Analyses of Randomized Controlled Trials

• Activity Number: 667
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Stephanie Ann Kovalchik*+
• Companies: University of California at Los Angeles
• Keywords: meta-analysis ; meta-regression ; subgroup analysis ; effect modifier ; randomized controlled trials

Abstract

When assessing treatment interventions, meta-analysts are interested in using meta-regression to identify subgroups of patients with a different treatment response. Simulation studies have suggested that, owing to loss of within-study information, meta-regression might not have sufficient power to detect covariate-treatment interactions for patient-defined covariates. To determine whether this generally holds for current quantitative reviews, I assessed the power of meta-regression for 500 recently published meta-analyses of randomized controlled trials, considering a range of subgroup scenarios--varying ratio of within- to between-study information. In each case, comparison was made to the power of an individual patient data (IPD) analysis. I also summarize the findings from a survey of meta-analysts on the process and success of collecting IPD. Implications for future multitrial subgroup analyses are discussed.