Author + information
- Ori Ben-Yehuda, MD∗ ()
- ↵∗Reprint requests and correspondence:
Dr. Ben-Yehuda, Cardiovascular Research Foundation, 111 East 59th Street, New York, New York 10022.
Subgroup analyses play an important role in the interpretation of clinical trials. Consistency of a treatment effect across different demographic and baseline characteristics is of the greatest importance, as an overall positive result may not translate to be of benefit in all subgroups. The most important subgroups, such as those categorized by sex, age, diabetes status, and prior myocardial infarction (MI), are usually reported in the primary publication, followed by a host of other publications, with only the imagination and academic stamina limiting the number of possible analyses.
But subgroup analyses are fraught with pitfalls (1,2). Both false positive (type 1) and false-negative (type 2) errors are more likely in subgroup analyses, the former due to multiple testing and the latter due to small sample size. The problem with subgroup analyses was memorably highlighted 25 years ago when the ISIS-2 (International Study of Infarct Survival-2) trial investigators showed that aspirin therapy in the treatment of suspected MI appeared to be harmful in the subgroup born under the astrological sign of Gemini or Libra, although it was beneficial in lowering mortality and reducing recurrent MIs in the overall study (3). No such astrological effect was seen with streptokinase. More recently, a subgroup analysis of the ATHENA (A Placebo-Controlled, Double Blind, Parallel Arm Trial to Assess the Efficacy of Dronedarone 400 mg bid for the Prevention of Cardiovascular Hospitalization or Death from any Cause in Patients with Atrial Fibrillation/Atrial Flutter) trial suggested a benefit in patients with permanent atrial fibrillation (4). Subsequently, however, the PALLAS (Permanent Atrial Fibrillation Outcome Study Using Dronedarone on Top of Standard Therapy) trial (5) not only failed to show any benefit, but was prematurely stopped due to a statistically significant increase in mortality, stroke, and heart failure, highlighting the potential danger of subgroup analyses influencing clinical care and forming the basis of phase 3 studies.
How, then, should we approach subgroup analyses, particularly when one subgroup is at odds with the overall findings of the study? Beyond a hefty grain of salt, certain subgroup analyses are more believable. The larger the subgroup, the more believable the result; indeed, we should be looking at the power of the subgroup itself. A rule of thumb is to limit subgroup analyses to those that maintain at least 40% to 50% power for the endpoint of interest (2). In a study with 90% power, this rule would preclude analyzing subsets comprising fewer than 30% of the original cohort. Additional questions include whether the analysis was pre-defined, whether a formal test of interaction is significant, and whether there is biological plausibility for the divergent findings. And finally, even if all these conditions are met, subgroup analyses should be viewed as hypothesis-generating and not overinterpreted.
It is with these caveats and limitations in mind that we should examine the subgroup analysis by Whellan et al. (6) from the TRACER (Thrombin Receptor Antagonist for Clinical Event Reduction in Acute Coronary Syndrome) trial published in this issue of the Journal. The TRACER trial (7) was a large (n = 12,944) phase 3 study of vorapaxar in acute coronary syndrome. Vorapaxar, a protease-activated-receptor 1 antagonist, inhibits the activation of platelets by thrombin. Thrombin is considered the most potent platelet agonist. Vorapaxar showed promise in phase 2 trials in patients with stable percutaneous coronary intervention and acute coronary syndromes. Importantly, in the phase 2 trials (8,9), there was no increased bleeding as assessed by the Thrombolysis In Myocardial Infarction (TIMI) bleeding score, despite concomitant dual antiplatelet therapy in the majority of patients. Vorapaxar was therefore touted as a potential blockbuster with anti-ischemic effects, but no increase in bleeding. Two large phase 3 trials were subsequently carried out, the TRACER trial (8) in patients with acute coronary syndrome and the TRA 2P–TIMI 50 (Thrombin Receptor Antagonist in Secondary Prevention of Atherothrombotic Ischemic Events–Thrombolysis In Myocardial Infarction 50) trial (10) in patients with prior MI, ischemic stroke, or peripheral arterial disease.
The TRACER trial failed to meet its primary quadruple endpoint (cardiovascular death, MI, stroke, or ischemia-driven revascularization or hospitalization), but did meet the secondary triple endpoint of cardiovascular death, MI, or stroke with a hazard ratio (HR) of 0.89 (p = 0.02). Most importantly, in divergence with the phase 2 data, there was a significant increase in major bleeding, including intracranial hemorrhage, which was increased from 0.2% in the placebo group to 1.1% in the vorapaxar group (HR: 3.39).
The subgroup analysis by Whellan et al. (6) looked at 1,312 patients from the trial who underwent coronary artery bypass graft (CABG) surgery during the index hospitalization and showed a 45% reduction in the primary quadruple endpoint for the TRACER trial. Most important, there was no significant increase in major CABG-related bleeding. These findings (part of a pre-specified subgroup analysis) differ significantly from the non-CABG group, with a p value for interaction of 0.012.
Based on the relatively small number of patients in the CABG subgroup, the skeptic would be inclined to dismiss the findings, despite the significant p value for interaction, as this subgroup appears a priori to be underpowered, with only 10.8% of the study population. Even the marked benefit (seen with an HR of 0.55) should be viewed with caution. Indeed, to be statistically significant, effects in small-sized subgroups have to be of greater magnitude (2).
Is there biological plausibility in the results? Thrombin generation is increased during surgery in general and on-pump bypass surgery in particular. At the site of wounds, such thrombin generation is beneficial and necessary for hemostasis, but when there is systemic generation of thrombin, such as that during bypass, a host of coagulation disorders may ensue, with both increased bleeding and thrombosis (11). It is therefore plausible that in the setting of CABG surgery, vorapaxar would reduce thrombin-related platelet activation and reduce ischemic events, which in turn may reduce perioperative MIs and thrombotic graft failure.
The most important finding of the CABG subgroup analysis was not, however, that it met the primary endpoint of the TRACER trial. After all, the main trial was stopped prematurely, showed a trend toward a reduction in the primary endpoint with a p value of 0.07, and met the similar secondary composite endpoint, consistent with an anti-ischemic effect. An anti-ischemic effect was also shown in the even larger trial, the TRA 2P–TIMI 50 trial, in patients post-MI. Rather it is the lack of major CABG-related bleeding and a lower, albeit still increased rate over placebo of 2-year TIMI major bleeding that is remarkable. One possibility is of a type II error (false negative) as the CABG subgroup was not powered for this important safety endpoint. A more intriguing possibility is that the CABG subgroup had a lower risk of bleeding due to the expectedly lower use of clopidogrel in this group. At discharge, clopidogrel use in the patients undergoing CABG was only 18% versus 84% in the non-CABG group. Clopidogrel was also held prior to CABG, with only 39% receiving the drug within 5 days of surgery, and probably a smaller percentage closer to surgery.
Withholding concomitant clopidogrel treatment, however, is unlikely to prevent the most dreaded risk of treatment with vorapaxar, namely intracranial hemorrhage in all patients. In the TRA 2P–TIMI 50 trial, there was a marked increase in intracranial hemorrhage in subjects with a history of stroke, leading the data and safety monitoring board to discontinue study treatment in this group 2 years into the study. This group, as with the CABG subgroup in the TRACER trial, had a low use of thienopyridines (mostly clopidogrel) at 24%. It would appear that with vorapaxar, intracranial bleeding is also dependent on the patient's underlying substrate. Thrombin generation, which is triggered by release of anionic phospholipids in brain injury, may be particularly important in preventing intracranial hemorrhage.
Is there a second act in store for vorapaxar, once considered a potential blockbuster? Is there a future for the concept of protease-activated receptor 1 inhibition, whether in a surgical population or in a wider acute coronary syndrome or percutaneous coronary intervention population? Many uncertainties remain, including the correct dose of the drug, either as monotherapy, in addition to aspirin or in addition to dual antiplatelet therapy. Interestingly in the phase 2 study of vorapaxar, both the 1 mg and 2.5 mg maintenance doses led to at least 80% inhibition of thrombin receptor agonist peptide (TRAP)-induced platelet inhibition in 100% of patients, both at 30 and 60 days (9). Based on the phase 2 data, the 2.5 mg dose was chosen for the phase 3 program in anticipation that major bleeding would not be increased, an expectation which was not confirmed. The phase 2 program also indicated that the drug would not lead to increased bleeding, even in the presence of dual antiplatelet therapy. Given the clear finding of increased bleeding in the phase 3 program, thereby changing the anticipated benefit/risk ratio for the drug, the possibility of using the drug in the absence of a P2Y12 blocker is intriguing.
Decades after the introduction of aspirin into routine use, and more than 15 years after the availability of clopidogrel, we are still defining the proper dose and use of these important antiplatelet drugs. Vorapaxar is presently undergoing evaluation by the Food and Drug Administration for the treatment of patients with a history of prior MI and no prior stroke and transient ischemic attack (12). Given the issues of bleeding associated with the drug, even if approved, its proper use will entail a difficult task of identifying the patient in whom the benefit outweighs the risk. The study by Whellan et al. (6) adds an interesting hypothesis to the inevitable debate.
↵∗ Editorials published in the Journal of the American College of Cardiology reflect the views of the authors and do not necessarily represent the views of JACC or the American College of Cardiology.
Dr. Ben-Yehuda has reported that he has no relationships relevant to the contents of this paper to disclose.
- American College of Cardiology Foundation
- Brooks S.T.,
- Whitley E.,
- Peters T.J.,
- et al.
- Wittes J.
- Nieuwlaat R.,
- Hohnloser S.H.,
- Connolly S.J.
- Whellan D.J.,
- Tricoci P.,
- Chen E.,
- et al.
- ↵Merck & Co. press release. Merck Announces FDA Acceptance of New Drug Application for Vorapaxar, Investigational Anti-Thrombotic Medicine. July 24, 2013. Available at: http://www.mercknewsroom.com/press-release/research-and-development-news/merck-announces-fda-acceptance-new-drug-application-vora. Accessed October 6, 2013.