CORRESPONDENCE: LETTER TO THE EDITOR
Statistical Methodology Used in the TAXUS ATLAS Trial
Mark A. Turco, MD, FACC, FSCAI* and
John A. Ormiston, MBChB
* Center for Cardiac and Vascular Research, Washington Adventist Hospital, 7600 Carroll Avenue, 6th Floor, Takoma Park, Maryland 20912 (Email: MTurco{at}AHM.com).
In a story entitled "Boston Scientific Stent Study Flawed" that appeared in the August 14, 2008, issue of the Wall Street Journal (1), reporter Keith J. Winstein questions the statistical methodology used in the TAXUS ATLAS trial to show noninferiority of target vessel revascularization (TVR) for the newer thinner-strut Taxus Liberté stent relative to a historical Taxus Express control stent, as published in the Journal (2).
Although the Taxus Liberté patients in the study proved to be significantly more complex than the historical Taxus Express study control patients, the unadjusted rates of the clinical primary end point showed a 0.94% difference (7.95% for Taxus Liberté and 7.01% for the Taxus Express control) (2). To show noninferiority required the 1-sided upper 95% confidence interval of this small difference to not exceed the pre-specified delta of 3.0%, using the methods as defined in both the study protocol and the statistical analysis plan approved by the U.S. Food and Drug Administration (FDA) before the study began.
Because the methods for data analysis in the statistical analysis plan were not detailed in the published article, we have reproduced them herein (Fig. 1). As applied to source-verified patient outcomes adjudicated by an independent Clinical Events Committee and reviewed by the FDA, the upper 1-sided 95% confidence interval for the unadjusted TVR difference was 2.98% (i.e., not exceeding 3.0), with a p value of 0.0487 (sufficient to reject the null hypothesis and accept the alternate hypothesis of noninferiority).
The point raised by Mr. Winstein centers on the mathematical precision of this specific statistical method used for noninferiority trials in general. He suggests that, had alternative mathematical tests been used, p values up to 0.051 might have been calculated. Although this may be mathematically interesting, it is clinically irrelevant to use other post hoc methods of calculation. An acceptable method (3) was pre-specified in the FDA-approved statistical analysis plan. To apply other alternative statistical methodologies would be grossly inappropriate, a violation of the fundamental scientific contract to adhere to the original trial design and the associated statistical analysis plan specified before the study commenced.
Should different types of analysis be mandated by the FDA in the future, investigators and industry sponsors will design trials to meet these requirements.
Accordingly, the authors stand firmly behind the findings from the TAXUS ATLAS workhorse trial that the unadjusted TVR for the Taxus Liberté stent is noninferior to that of the earlier Taxus Express stent. This is expected for 2 stent designs that use identical drug and polymer coatings and provide identical reduction in excess tissue growth within the stent (late loss 0.41 mm for Taxus Liberté and 0.42 mm for Taxus Express) (2). The small apparent difference in unadjusted TVR thus likely reflects the significantly greater complexity of the TAXUS ATLAS patients compared with the historical control subjects; the TVR rates become nearly identical (adjusted TVR 7.82% for Taxus Liberté and 7.48% for Taxus Express, noninferiority p value 0.0276) when corrected for the measurable baseline differences between the respective study populations. Additionally, results from the real-world OLYMPIA (Global TAXUS Liberté Registry Program to Support Worldwide Commercialization) registry in more than 7,000 Taxus Liberté patients treated without routine angiographic follow-up shows a 1-year target lesion revascularization rate of 2.5% (4).
 |
Footnotes
|
|---|
Please note: The authors thank the following individuals at Boston Scientific: Lazar Mandinov, MD, PhD, for valuable assistance with medical monitoring of the trial, and Peter Lam, PhD, for heading the statistical analyses.
|
References
|
|---|
1. Winstein KJ. Boston Scientific Stent study flawed Wall Street Journal August 14, 2008:B1B6.2. Turco MA, Ormiston JA, Popma JJ, et al. Polymer-based, paclitaxel-eluting TAXUS Liberté stent in de novo lesions: The pivotal TAXUS ATLAS trial J Am Coll Cardiol 2007;49:1676-1683.[Abstract/Free Full Text] 3. Agresti A. Categorical Data AnalysisNew York, NY: Wiley; 1990. 4. Mendiz O, Thomas MR, Ahmed W, et al. OLYMPIA Global TAXUS Liberté Registry: 12-month results in high-risk subsets in the first 7,000 patients enrolled (abstr) Am J Cardiol 2007;100(Suppl 1):96L.
|
Top
References