by Harlan M. Krumholz, MD

What is a clinician to do when a clinical trial produces an inconvenient result?

We all appreciate a trial that conforms to expectations and produces results that are strong, clear, and consistent across subgroups. Such findings are easy to interpret, to incorporate into guidelines, and to adopt into practice. However, there are many instances of trial findings that defy easy explanation and complicate decisions about drug approval and application.

A prime example, which has perplexed many experts, is the PLATO (PLATelet Inhibition and Patient Outcomes) study, a double-blind, randomized trial sponsored by AstraZeneca to evaluate the efficacy of ticagrelor compared with clopidogrel in the long-term treatment of patients hospitalized with an acute coronary syndrome. Ticagrelor, a non-thienopyridine antiplatelet agent that reversibly blocks the adenosine diphosphate (ADP) receptors of subtype P2Y12, was developed to prevent thrombotic events.

Inconvenient Truth in the U.S.?

The positive results of PLATO, released to great fanfare at the Congress of the European Society of Cardiology and published on September 10, 2009,1 fulfilled the hopes of the sponsor. After a year of follow-up, the primary endpoint of death from vascular causes, acute MI, or stroke occurred in 9.8% of the 18,624 patients randomized to the ticagrelor group and 11.7% of those randomized to the clopidogrel group, producing a number needed to treat for benefit of about 50. The result was equivalent to a 16% reduction in risk and was highly significant.

In secondary analyses, there were significant reductions in the risk of death from vascular causes and acute MI. Notably, ticagrelor reduced thrombotic events without increasing the risk of bleeding.

So what is the inconvenient finding? Subjects from the United States appeared not to share the benefit from ticagrelor. Elsewhere, ticagrelor was associated with a 19% significant reduction in risk of vascular death, acute MI, or stroke compared with a 27% nonsignificant increase in the risk of the primary outcome in the United States. Of the 1,413 American participants, 11.9% in the ticagrelor group experienced an event compared with 9.5% in the clopidogrel group. Statistical tests showed that the variation in effect by patients’ geographic location was significant, with a p value of 0.01.

How did they end up looking specifically at the experience in the United States? The regional analysis was part of a pre-specified assessment of 31 variables. Among those was an analysis of whether the drug’s effect varied by the patients’ geographic region (North America, the Middle East and Africa, Asia and Australia, and Central and South America). The drug’s effect did vary significantly by region, with ticagrelor having a nonsignificant trend toward a worse result for those from North America. The p value for this pre-specified analysis was 0.045, indicating the possibility of an interaction. Further analysis showed this difference to principally result from the experience of patients in the United States, which led to the post-hoc analysis of the patients in the United States compared with the rest of the world.

The memo from the FDA that conveyed the recommendation of the Division of Cardiovascular and Renal Products to approve ticagrelor stated that “the U.S. results are entirely consistent with there being no effect whatsoever of ticagrelor in the United States. No single or combination of baseline covariates was found to explain the U.S.-foreign differences in outcome.”

Teasing Out the Answers

So we have a large, international trial of a drug shown to reduce the risk of thrombotic events without an excess of bleeding that seems to do no better than clopidogrel in the United States, and may even do worse. It is well known that subgroup analyses can be misleading. They are prone to false-positive findings and should be considered exploratory. And yet, they have occasionally yielded important insights that were subsequently validated and incorporated into practice.

What could explain a difference in the trial result according to whether the patient was from the United States or not? There are at least three possibilities: the play of chance; systematic errors in the conduct of the study; and differences between the United States and other populations in patient characteristics, prognoses, or clinical management.

With a subgroup analysis, chance is always a concern as an explanation for differences across groups of subjects. With so many pre-specified analyses, chance could be the reason. The investigators calculated that the probability of finding a result favoring clopidogrel in the United States and ticagrelor in the rest of the world is 1 in 10. Nevertheless, the concern was substantial enough for the FDA to postpone approval of the drug and request the conduct of further analyses.

The role of the study sponsor is an intriguing possibility, but one that is thus far not supported by the evidence. AstraZeneca played a prominent role in the trial, contributing to the design and overseeing its conduct. The organization coordinated the data management and hired a contract research organization to perform the statistical analyses. AstraZeneca reported that it could not find any systematic errors in study conduct.

Finally, the explanation could lie in true differences that result from something unique about the population of the United States or in aspects of its healthcare. Based on a post-hoc analysis, the suggestion emerged that the finding in the United States resulted from the common practice of using higher doses of aspirin. The ticagrelor group tended to do better in all areas when paired with lower-dose maintenance aspirin (≤100 mg) compared with doses ≥300 mg. This analysis, however, was based on only 280 subjects outside the United States who took higher-dose maintenance aspirin. The investigators termed this finding, which was far from conclusive, as a “possible explanation.” They advise that the findings should “be considered cautiously” because of the limitations of their exploratory analyses. The FDA reviewer stated that “the aspirin hypothesis (that aspirin dose accounts for regional differences in outcome) is not highly persuasive—by mechanism or analyses as a factor in the study outcomes, and no version of it predicts adverse effects of high-dose aspirin.”

So what did the FDA do? After a 6-month delay, it approved the drug. The label did not convey the uncertainty about the role of aspirin and states that in PLATO, maintenance doses of aspirin >100 mg decreased the effectiveness of ticagrelor. Moreover, it warns that maintenance doses >100 mg should be avoided and there should be a Risk Evaluation and Mitigation Strategy (REMS) that compels the company to communicate this concern. The Medication Guide tells patients that they should not take a dose of aspirin >100 mg because it can affect how well the drug works.

Where do we stand, now? The trial result was positive, but there remains some uncertainty about the effect in the United States and the reason for it. Chance cannot be confirmed, nor can true effect. The role of aspirin is speculative. Anyone who expresses a strong opinion about the true effect in the United States is stating a point of view and reasonable people could disagree. Another trial is unlikely to be conducted. Science can only take us so far with the available data and it is difficult to know with certainty if ticagrelor is more effective than clopidogrel when used in the United States. In the end, it is an inconvenient result for doctors and patients and policy-makers.

Maybe we should just admit that.

References

1. Wallentin L, et al. N Engl J Med. 2009;361:1045-57.
   Harlan M. Krumholz, MD, is a cardiologist and the Harold H. Hines, Jr. Professor of Medicine and Epidemiology and Public Health at Yale University School of Medicine.
   
   

Keywords: Uncertainty, Acute Coronary Syndrome, Stroke, Follow-Up Studies, Drug Approval, United States, Thienopyridines

< Back to Listings