A reader raises questions about a large cardiology trial. Are his concerns valid?
Dr. Oransky: I read your column on equivalence studies (The moral problem of equivalence studies. Gastroenterology & Endoscopy News 2006;57:38) in detail. I have tried to use this information to evaluate the non-inferiority OASIS-5 (Organization to Assess Strategies in Acute Ischemic Syndromes Investigators) trial.1 Based on previous information, the added risk of placebo over heparin or low molecular weight heparin was 1.89 (CI, 1.37 to 2.63) of suffering myocardial infarction (MI) or death in patients with unstable angina or MI without ST segment elevation. The non-inferiority margin was set at 1.185, which is 50% of the lower half of the CI [(1.37-1)/2 = 0.185]. The results showed a hazard ratio of 1.01 (CI, 0.90 to 1.13). This was well below the limit of 1.185, and the authors concluded there was no additional risk of death or MI at day 9 when using fondaparinux over enoxaparin for treatment of unstable angina or MI without ST segment elevation.
However, the authors did not include these details for primary or secondary safety, nor the secondary efficacy results. Can their secondary results be valid without this detail? Although the sample size (enoxaparin 10,021; fondaparinux 10,057) appears high, could bias somehow be introduced here by this omission? I could find no obvious flaws with the primary efficacy results of OASIS-5. However, I am not a statistician and am wondering if you had any comments.
I want to thank you for your column. Your approach makes the practical understanding of statistics much easier.Tim Clark, PharmD, BCPS
Pharmacy Clinical Coordinator
Mount Clemens Regional Medical Center
Mt. Clemens, Mich.
Dr. Clark’s letter is quite timely. The FDA is now reviewing an application for an acute coronary syndrome indication for fondaparinux (Arixtra, GlaxoSmithKline), an anticoagulant it has already approved for prevention of deep vein thrombosis and pulmonary embolism following surgery. It also follows, as he notes, my column on equivalence trials, and raises similar issues. So should we ignore the OASIS-5 data?
Let’s start by describing the trial to which Dr. Clark refers.
OASIS-5 was actually quite straightforward. Researchers wanted to see whether fondaparinux would offer the benefits of enoxaparin (Lovenox, Sanofi-Aventis) in acute coronary syndromes, with fewer bleeding complications. As Dr. Clark notes, they randomized 20,078 patients with acute coronary syndromes to either fondaparinux (2.5 mg daily) or enoxaparin (1 mg/kg bid).
Bleeding was the primary outcome of this study, a fact that the study’s authors are quite honest about. The Cox proportional hazards conducted in the study found the fondaparinux group to have a 2.2% risk of bleeding compared with a 4.1% risk in the enoxaparin group (P<0.001). Dr. Clark describes the results for secondary outcomes of the study correctly. The non-inferiority margins are appropriate using the 50% rule or 95-95 method—for more on how to calculate those, see the column to which Dr. Clark refers—and he is correct that the OASIS-5 investigators didn’t include a power analysis on the secondary target. But they did run the same Cox proportional hazards analyses for death or MI, the combined measure they used as their secondary target. The outcome rate in the fondaparinux and enoxaparin groups was identical (4.1%), and the confidence interval for this comparison did not overlap the non-inferiority boundary. All of this makes me fairly unconcerned about the lack of a power analysis.
Study Design Problematic
There is a trend toward lower rates of ischemia in the fondaparinux group, but the authors don’t make very much of this, and you know from previous columns how I feel about trends anyway. As Dr. Clark points out, the authors didn’t power each of the comparisons for all of the end points. The authors say they compared their secondary outcome with its non-inferiority boundary, but they do not include that boundary per se in their paper.
That points up an important problem in study design. Some argue that you should power your study for the end point or comparison that is most important in the study. Others say you should power it for the one that will show the biggest difference. But hardly anyone says you should power it for every single end point or comparison. Doing so creates the problem of multiple comparisons and a need to adjust significance levels when multiple comparisons are made with a single factor—here, the drugs tested. If you want to be very conservative, you can power it for the result that will be toughest to get, and that’s what they did here.
I would raise one issue about the OASIS-5 study: They only report a bivariate Cox proportional hazards model, with the treatment group as the only covariate. They could have easily controlled for other risk factors, such as the baseline creatinine level, prior use of unfractionated heparin, age, or the presence or absence of a catheterization laboratory at the facility. Instead, many of these factors were examined in individual subgroup analyses. I don’t know that including these factors in a multivariate analysis would have changed the results, but I would have suggested putting a multivariate analysis in the main analysis. This would have prevented giving ammunition to critics who love to question subgroup analyses.
But that’s quite minor, and I’d say Dr. Clark’s criticisms, while well taken, are more of a technicality than something that will discredit the OASIS-5 data. That being said, I’d much rather that everyone in Dr. Clark’s position subject every study that might have prescribing or formulary implications to this kind of skeptical inquiry. Let drug reps, or the investigators themselves, answer these kinds of questions.
A Question of Context
What does all this mean for fondaparinux? It always helps to look at the context, so I looked for other studies involving fondaparinux or enoxaparin and bleeding. In a somewhat related study (N Engl J Med 2006;355:2203-2216), patients with acute coronary syndromes who were receiving heparin or enoxaparin plus glycoprotein IIb/IIIa inhibitors had a 5.7% risk of major bleeding complications compared with 3.0% bivalirudin, a thrombin inhibitor.
More to the point, the publication of the OASIS-5 data actually followed the release of data from OASIS-6 (JAMA 2006; 295:1519-1530). That study compared fondaparinux to either placebo or unfractionated heparin, depending on the patient’s particular clinical scenario, in patients with ST-segment elevation MI (STEMI). It showed that in patients with STEMI, particularly those who didn’t undergo primary stenting, fondaparinux reduced mortality and subsequent MIs within 30 days without an increase in bleeding and strokes.
So, I’d say the news is pretty good for fondaparinux, and wouldn’t be surprised if the FDA approved it for use in acute coronary syndromes fairly soon. Thanks for the question, Dr. Clark—and if anyone has others, send them along!
1. Yusuf S, Mehta S, Chrolavicius S, et al. Comparison of fondaparinux and enoxaparin in acute coronary syndromes. N Engl J Med 2006; 354:1464-1476.
Ivan Oransky, MD, is deputy editor of The Scientist and a frequent contributor to The Lancet. He also holds appointments at New York University as clinical assistant professor of medicine and adjunct professor of journalism. Dr. Oransky thanks Howard Barkan, a biostatistician and research methodologist in the Department of Surgery, Kaiser Permanente, Oakland, California, for his assistance.