## JACC: Cardiovascular Interventions

# What Do Noninferiority Trials Say About Coronary Stents?

## Author + information

- Published online February 6, 2017.

## Article Versions

- previous version (January 18, 2017 - 11:00).
- You are viewing the most recent version of this article.

## Author Information

- John A. Bittl, MD
^{∗}(jabittl{at}mac.com)

- ↵∗
**Address for correspondence:**

Dr. John A. Bittl, Interventional Cardiology Group, Munroe Regional Medical Center, 1221 SE 5th Street, Ocala, Florida 34471.

## Corresponding Author

- Bayesian analysis
- clinical trials
- p value
- percutaneous coronary intervention
- randomized controlled trials
- stent polymers

When investigators are trying to prove that a new treatment is superior to an existing one, they commonly use statistical methods that prevent them from saying what they want to say (1). Before showing that a new treatment is better than an old one, investigators must first fabricate a null hypothesis that says there is no difference between the new and old treatments. After a clinical trial has been completed, a p value is calculated from a mathematical simulation that represents how frequently the observed data would have occurred if the trial could have been repeated a large number of times and the null hypothesis were true. In the frequentist framework, the p value represents the proportion of trials that would have produced similar results—based on a large series of repetitions that does not occur, conditional on a null hypothesis that no one truly believes (2). A p value of less than an arbitrary cutoff of 0.05 allows investigators to conclude that, “Either an exceptionally rare chance has occurred or the [null hypothesis] is not true,” (3) although in fact there may be no relation between the probability of the null hypothesis and the plausibility of an alternative hypothesis. Generating a point estimate and 95% confidence interval helps to define the treatment effect but does little to reverse the inverted logic of frequentist statistics (4,5).^{†}

When investigators are trying to prove that a new treatment is noninferior to a preexisting therapy, the statistical methods become more tortuous and nonintuitive (5). To compare 2 treatments using a marginal analysis (6), investigators postulate a “null” that describes a difference between the 2 treatments, as defined by a noninferiority margin -Δ (5). The goal of a noninferiority trial is to compare the 2 treatments, obtain a 95% confidence interval with a lower bound that does not cross the noninferiority margin, and satisfy the alternative hypothesis that the difference between the new and old treatments is less than -Δ. In the backward world of noninferiority, the null hypothesis contains a treatment difference, whereas the alternative hypothesis seems “null” (5).

Despite the convoluted statistical design, noninferiority trials have been popular for comparing stents. In this issue of *JACC: Cardiovascular Interventions,* 2 reports (7,8) describe several such studies. Yeh et al. (7) summarize the experience with a durable-polymer zotarolimus-eluting stent (DP-ZES) in a series of noninferiority trials and registries known as the Resolute program. The investigators report that the 5-year cumulative incidence of definite or probable stent thrombosis was 1.2%, which consisted of a rate of 0.7% at 1 year and an annualized rate of 0.1% thereafter. Although the Resolute program (7) may have enrolled patients with higher risk than, say, the DAPT (Dual Antiplatelet Therapy) trial (9), a study that used a superiority design, with more patients in the Resolute program having high-risk baseline characteristics such as prior myocardial infarction (29% vs. 22%) or diabetes (37% vs. 31%) than patients in the DAPT trial (9), the annualized rate of very late stent thrombosis was 3 to 10 times lower in the Resolute program than in the DAPT trial. Does this discrepancy reflect differences in stent safety, trial design, or ascertainment of events?

Investigators involved in superiority trials are rewarded for achieving high enrollments and capturing all events to maximize the chance that they will find a difference. Not meant to be a criticism, but investigators involved in noninferiority trials or registries are not penalized for losses to follow-up or incomplete ascertainment of events, because such deficiencies could increase the chance of finding no difference (5). Along these lines, it was reassuring that the Resolute program (7) included studies that reported 2-year follow-up adherence as high as 98.1% and observed 5-year rates of 5.0% for cardiac death and 4.4% for target vessel myocardial infarction that were more in line with the 18-month rates of 0.9% for cardiac death and 3.0% for myocardial infarction in the DAPT trial (9).

In a second paper in this issue of *JACC: Cardiovascular Interventions*, Raungaard et al. (8) report the 3-year results of the SORT OUT (Scandinavian Organization for Randomized Trials with Clinical Outcome) VI trial, a noninferiority trial that compared the DP-ZES with a biodegradable-polymer biolimus-eluting stent (BP-BES) in 2,999 patients. The original publication (10) reassuringly reported follow-up in 99.7% of enrolled patients but found event rates for the primary endpoint of 5.3% in the DP-ZES group and 5.0% in the BP-BES group that were both lower than the expected rate of 6.5% and differed from each other by less than the prespecified absolute risk margin of 2.5% (10). To affirm noninferiority of the DP-ZES compared with the BP-BES, the investigators (8) now report no difference in major adverse cardiac events (8.6% vs. 9.6%), death (2.7% vs. 3.4%), or very late stent thrombosis (0.4% vs. 0.7%) at 3 years, respectively.

Noninferiority trials have become an integral part of the interventional cardiology evidence base and do not necessarily prove that industry has hacked frequentist statistics to promote a raft of “me-too” coronary stents. Although most investigators would prefer to declare that a new stent is better than an old stent, some practitioners might be willing to sacrifice a little efficacy in exchange for improved deliverability, lower cost, or shorter courses of DAPT, although the present studies (7,8) do not show directly whether the DP-ZES or the BP-BES offers such advantages.

A drawback common to both noninferiority and superiority trials involves changes in the standard of care and selection of the wrong comparator. In noninferiority trials this may lead to the problem of “techno-creep” (5), in which progressively inferior devices are shown to be “not unacceptably worse” (5) in sequential comparisons (11). In this regard, it is important to recognize the emerging evidence (12,13) that a different coronary stent, a durable-polymer everolimus-eluting stent containing an antithrombotic fluoropolymer (14) on a thromboresistant platinum chromium metallic backbone (15), is the current standard.

The present studies (7,8), along with a separate noninferiority trial (16), provide a level of reassurance that the DP-ZES is not unacceptably worse than multiple biodegradable-polymer stents. But, given the ever-changing evidence base in interventional cardiology and the convoluted statistics of noninferiority trials, some practitioners may wonder how a DP-ZES would stand up against a durable-polymer everolimus-eluting stent or question whether better methods of comparison might be found. Future efforts could explore the possibility of using Bayesian approaches, which predated frequentist approaches by at least 150 years (1,4) and are firmly rooted in the science of probability (17). A central component of the Bayesian approach is the Bayes factor, which in its simplest form is a likelihood ratio, which has been applied widely in diagnostic testing to calculate the predictive value of a positive result and in its logarithmic form log_{e}(OR) labeled by Alan Turing as the “weight of evidence” to crack the Nazi Enigma codes in the Second World War (4). As compared with p values, which greatly overstate the evidence against the null hypothesis (18), Bayes factors range from 0 to ∞ and constitute a potentially superior approach that weighs the relative likelihood of 2 different hypotheses, with small values close to 0 simultaneously providing strong evidence against a true null hypothesis *and* for an alternative hypothesis (1,4,18). A fresh look at an old approach might allow investigators to make direct probability statements from their data and say precisely what they mean to say.

## Footnotes

↵∗ Editorials published in

*JACC: Cardiovascular Interventions*reflect the views of the authors and do not necessarily represent the views of*JACC: Cardiovascular Interventions*or the American College of Cardiology.↵† The 95% confidence interval, like a p value, is commonly misunderstood as a direct probability statement, but there is not a 95% guarantee that the true treatment effect lies within the confidence interval. Instead, the traditional 95% confidence interval describes the hypothetical performance of a large number of similar trials, and like the theoretical exercise for defining a p value, the process would generate a large number of confidence intervals, of which 95% will contain the true odds ratio but the actual probability that a particular confidence interval contains the true odds ratio cannot be stated (5).

The author has reported that he has no relationships relevant to the contents of this paper to disclose.

- American College of Cardiology Foundation

## References

- ↵
- Jeffreys H.

- ↵
- ↵
- Diamond G.A.,
- Kaul S.

- ↵
- Spiegelhalter D.J.,
- Abrams K.R.,
- Myles J.P.

- ↵
- ↵
- Kaul S.,
- Diamond G.A.,
- Weintraub W.S.

- ↵
- Yeh R.W.,
- Silber S.,
- Chen L.,
- et al.

- ↵
- Raungaard B.,
- Christiansen E.H.,
- Bøtker H.E.,
- et al.

- ↵
- ↵
- Raungaard B.,
- Jensen L.O.,
- Tilsted H.-H.,
- et al.

- ↵
- ↵
- Palmerini T.,
- Biondi-Zoccai G.,
- Della Riva D.,
- et al.

- ↵
- Kang S.H.,
- Chae I.H.,
- Park J.J.,
- et al.

- ↵
- Kolandaivelu K.,
- Swaminathan R.,
- Gibson W.J.,
- et al.

- ↵
- Eppihimer M.J.,
- Sushkova N.,
- Grimsby J.L.,
- et al.

- ↵
- von Birgelen C.,
- Kok M.M.,
- van der Heijden L.C.,
- et al.

- ↵
- Bertsekas D.P.,
- Tsitsiklas J.N.

- ↵