New printed evidence reignited the controversy over the EXCEL affair, exposing more
important flaws, both in the trial’s statistical analysis and the revelation that
the previously concealed myocardial infarction data has now been made available.
Further to the extensive disclosure of flagrant bias introduced into the design, analysis,
and publication of the 5-year follow-up of the EXCEL trial[1], the June edition of
the Journal of American Medical Association (JAMA) Internal Medicine revealed and
highlighted that the formal hypothesis testing, which in the EXCEL trial was prespecified
as a demonstration of noninferiority at 3 years, was without discussion or explanation
switched to a superiority test in the 5-year analysis[2]. If as prespecified, the
same noninferiority analysis had been conducted at 5 years it would not have met the
criteria for statistical significance, and the printed conclusion would not have been
possible. This is a classic case of interpretation bias, otherwise known as spin.
The newly identified violation follows in the wake of the controversy over data manipulation
in the EXCEL trial. The article authored by Professor James M. Brophy, who is a professor
of Medicine, Epidemiology, and Biostatistics at McGill University in Montreal, Canada,
re-examines the design and results of the EXCEL trial and other recent randomized
clinical trials related to the left main coronary artery disease (LMCAD) using Bayesian
methods[2].
The article shows furtive aspects and details of the statistical analysis, not previously
acknowledged, that in effect concealed and circumvented the real results of the EXCEL
trial to attain the desired conclusion. The method of statistical analysis used to
interpret the 5-year data was different from that used in the 3-year publication,
and in direct contrast to the original protocol, without explanation or justification.
If EXCEL’s 5-year results had been interpreted in alignment with its original primary
non-inferiority design of a pre-specified 4.2% primary result margin, a different
conclusion would have been reached, even within the context of a standard analysis.
Instead, the EXCEL authors construed the 5-year results as a superiority analysis,
in which the null hypothesis did not consist of any differences so that the observed
data was not extreme enough to reject that null hypothesis. The Bayesian analysis
of the EXCEL primary outcome estimated with 95% probability that the 5-year primary
outcome difference was increased with PCI compared with CABG and 87% probability that
this difference was greater than one extra event per 100 patients treated. Given that
in EXCEL these were relatively young patients with a mean age of 66 years (and therefore
a potential life expectancy of 15-20 years) and that the statistically significant
observed increase in mortality of 38% was actively accelerating raises real clinical
concern. For the secondary composite endpoint, which also included repeated revascularizations,
there were estimated probabilities of 98% for at least 4 extra events and of 90% for
at least 5 extra events per 100 patients treated with PCI. Therefore, PCI was associated
with less favorable results for all events, including mortality, compared to CABG
in patients with LMCAD.
Professor Brophy, although recognizing that these analysis failures by the authors
of the EXCEL trial may have occurred due to carelessness and/or incompetence of the
authors, questions if sponsorship by the company that produced the stent used in the
study led to more favorable results and conclusions. He reinforces that fourteen of
34 EXCEL trial authors, including the first and last authors, had a relationship with
the study sponsor (the stent maker), and that 8 authors listed an affiliation with
the Cardiovascular Research Foundation (CRF) which oversaw the study. CRF received
a donation of US$ 937,000 from the sponsor during the study and Professor Brophy infers
that these factors may have contributed to the entirely speculative hypothesis that
these conflicts of interests contributed to the biased interpretation of EXCEL[2].
In the same issue of JAMA Internal Medicine, a commentary by Professor Sanjay Kaul
(Professor of medicine at Cedars-Sinai Medical Center in Los Angeles) raises the same
concerns about the inappropriateness of changing the protocol specified statistical
analysis from non-inferiority at 3 years to superiority at 5 years[3]. The reason
for this change is not justified or even explained in the final EXCEL trial manuscript.
However, it emphasizes that, if the authors had maintained the non-inferiority analysis
that was performed in 3 years (with a pre-specified upper margin of 4.2%), it would
not meet those existing criteria at 5 years because the upper limit of the 95% difference
at 6.5%, now exceeds that non-inferiority margin. He reiterates that the conclusion
of EXCEL that PCI and CABG were similar, despite a 3.1% difference in mortality that
favors CABG, is a classic example of interpretation bias, also known as spin. Spin
is recognized to occur frequently when associated with commercial study sponsorship[3,4].
The disclosure of alteration in the statistical methodology in the 5-year EXCEL analysis,
making possible the printed conclusion, is very worrisome in view of its potentially
adverse implications for patient safety. Likewise, the earlier claim from the EXCEL
trial authors that the per-protocol Universal Definition of Myocardial Infarction
data analysis, which was prespecified as a secondary endpoint, had been deemed ‘impossible’
because troponin was not properly collected, was overly refuted by the unveiling of
the definition of the Third Universal Definition of Myocardial Infarction, which clearly
states that if a cTn assay is not available, the best alternative for analysis is
CKMB (measured by mass assay)[1,5].
Despite the prior assertions of the EXCEL authors that no such data was available,
in a new publication from the EXCEL trial authors, compulsorily replying to letters
to New England Journal of Medicine (NEJM), now reported the protocol specified analysis
of Third Universal Definition of Myocardial Infarction, based on CKMB data. This revealed
a stark difference from the original paper, by now demonstrating the higher rate of
periprocedural myocardial infarction related to PCI. And therefore, nullifying the
original conclusion published in the NEJM[6].
The authors claim that the difference in all-cause death was driven by an increase
in non-cardiovascular death, due to late malignancies and sepsis, and in the absence
of any obvious biologic plausibility is therefore likely due to chance. However, this
denotes unfamiliarity with the basic principles of pathophysiology of coronary artery
disease and the repercussion of distinctive revascularization therapies[7]. Mounting
evidence implicates inflammation as a key player in many, if not all, diseases of
the human body. The presence of a metallic intracoronary stent, acting as a foreign
body inducing formation of a cardiac granuloma, elicits a local and systemic chronic
inflammatory reaction, leading to a perennial low-grade inflammation with persistently
higher blood TNF-alpha and interleukin-6 expression[8-10]. Inflammation has long been
recognized as a key component of carcinogenesis and chronic low-grade inflammation,
characterized by elevated concentrations of IL-6, TNF-α, and C-reactive protein, increase
susceptibility to cancers and infections[11-14]. Additionally, drug-eluting stents
actively releasing antiproliferative drugs promote chronic inflammation and determine
a greater endothelial dysfunction and neoatherosclerosis[15], which is associated
with an long-term increased risk of solid-tumor cancer[16].
As external and independent denouncements of the conduct and interpretation of the
EXCEL trial are escalating, why do the EXCEL authors, of whom many are distinguished
doctors, still remain silent over these crucial facts and the implication of potential
real harm to patients? And what about the implications for guidelines that may have
been or could in the future be based on such a flawed trial and conclusion?
Furthermore, the EXCEL trial has generated dozens of offspring manuscripts. Can we
now trust these manuscripts if the primary manuscript was so severely flawed? And
what about the patients enrolled in the EXCEL trial, who committed their trust and
safety, to participate in a trial professedly created to help save lives? And how
they might be now frustrated and even feeling betrayed by outcomes skewed for not
confessed reasons?
Until these issues are addressed the EXCEL trial has the potential to become a deplorable
example of distrust as a distorted form of science, leaving both the public and the
medical profession to question how this happened and how to ensure that it cannot
happen again.
While it would be hoped that the serious flaws of the EXCEL trial occurred through
ignorance and/or incompetence it still raises the question of whether the EXCEL affair
may have been driven by other conflicts both by the investigators and the company.
The data generated by the EXCEL trial, the largest ever trial on this subject, is
crucial for science and medical practice to understand the impact of the competing
therapies of PCI and CABG in this deadly condition. Accordingly, and to restore trust
both for patients and the medical profession the EXCEL data must be made open for
unbiased and truthful thorough scrutiny, managed by independent reviewers. In the
meantime, and given these very serious flaws the NEJM should issue a temporary retraction
to prevent patients from being submitted to serious potential harm.