Copyright 2011 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2011
The letter by Claggett and Wei focuses on the statistical methods used to perform our meta-analysis of the cardiovascular (CV) safety of rosiglitazone.1 Specifically, these authors question the robustness of the P values and propose alternative statistical methods that yield higher P values. The purpose of any meta-analysis is not to perform every possible statistical approach to the data, but rather to characterize the pattern of outcomes associated with the studied treatment. We believe strongly that authors should prespecify the methods used and stick closely to the selected approach, which we did. The Peto and Mantel Haenszel methods are time-honored analytic approaches that have been used by many other authors. Regarding the robustness of the data, it must be understood that rosiglitazone has been marketed for 11 years. During that period, the drug maker carefully avoided performing any high-quality CV outcomes studies, thereby limiting the availability of CV safety data. Importantly, the majority of clinical trials used in our analysis were unpublished and available only because of a court settlement requiring the company to publicly disclose clinical trial findings. We were forced to use study-level data, not patient-level data, in our analyses. Accordingly, the most important limitations of our meta-analysis relate to the behavior of the drug maker in avoiding CV safety studies, not the statistical methods used to analyze the available clinical trials. The use of study-level data is a more serious limitation than the precise statistical methods used to calculate P values. Fortunately, the Food and Drug Administration (FDA) had access to patient-level data, enabling use of more robust time-to-event methods. The FDA analysis of patient-level data reported a hazard ratio (HR) for myocardial infarction of 1.80 (95% CI, 1.03-3.25).2 Sensitivity analyses using the end point of myocardial ischemia yielded a significant HR of 1.34 (95% CI, 1.07-1.70). Strikingly, the FDA reviewer also reanalyzed the RECORD trial, reporting an HR for myocardial infarction of 1.38 (95 CI, 0.99-1.93), which was very similar to our findings using the Peto odds ratio.2 Given the FDA findings, we believe that our reported odds ratios and P values accurately represent the hazards of rosiglitazone. In 2010, the FDA's July 14 Advisory Committee agreed, voting overwhelmingly that rosiglitazone increases the risk of ischemic CV events and recommending either withdrawal of the drug from the market or severe restrictions on its availability.
Nissen SE, Wolski K. Analytical Issues Regarding Rosiglitazone Meta-analysis—Reply. Arch Intern Med. 2011;171(2):179-180. doi:10.1001/archinternmed.2010.509