A Positive Trial That Fails to Inform Clinical Decisions

The EARLY TAVR trial compared transcatheter aortic valve replacement (TAVR) with clinical surveillance in patients with severe but asymptomatic aortic stenosis (AS). Its results strongly favored TAVR, but two glaring design flaws remove nearly all its clinical value. 
The question being asked in EARLY TAVR is when to intervene on patients with severe AS. In days of old, we deferred aortic valve surgery until major symptoms occurred: syncope, angina, or heart failure symptoms. 
Two surgical trials — RECOVERY and AVATAR — found that early surgical aortic valve replacement (SAVR) in patients with severe asymptomatic AS proved beneficial. The core problem with these trials were the small numbers of patients and events. 
EARLY TAVR was designed to enroll more patients and to test TAVR, not SAVR, in asymptomatic patients. To be enrolled, patients had to prove they were asymptomatic. The vast majority had negative stress tests. 
Patients also had to have suitable anatomy for the balloon-expandable transcatheter valve used in the trial — because the trial was sponsored by the manufacturer, Edwards Lifesciences. I offer this quote from the manuscript: 
The sponsor funded all trial-related activities, participated in site selection, oversaw data collection and monitoring, and performed analyses according to the statistical analysis plan (available with the protocol).
The planners chose a composite primary endpoint of death, stroke, and unplanned hospitalization for cardiovascular causes. EARLY TAVR was not a blinded trial. Patients knew that they had severe valvular heart disease, and one group would have a valve intervention while the other group would have “clinical surveillance” — a euphemism for no procedure. 
Unlike the prior surgical trials, which enrolled only 150 patients each, EARLY TAVR screened 1600 patients and enrolled 900. 
Patients were 76 years of age on average; 30% were female; and most notably, they had low surgical risk, with a mean STS-PROM score of 1.8. 
Over a median follow-up of 3.8 years, the primary endpoint occurred in 27% of the TAVR group vs 45% of the clinical surveillance group (hazard ratio, 0.50; 95% CI, 0.40-0.63; P <.001). 
Individual components of the composite tell an important story. Death was not significantly different (8.4% vs 9.2% for surgery). Stroke was also not significantly different, though it was numerically lower in the TAVR arm (4.2% vs 6.7%). 
The driver of the lower incidence of the primary endpoint was lower rates of unplanned hospitalization for cardiovascular reasons in the TAVR arm (21 vs 42%; hazard ratio, 0.43; 95% CI, 0.33-0.55). In the paper’s main figure, the Kaplan-Meier curves for unplanned hospitalization and the primary endpoint were superimposable. 
In the surveillance arm, conversion to TAVR occurred rapidly and steadily. More than 1 in 4 surveillance patients had TAVR at 6 months, nearly half at 1 year, and 71% had TAVR at 2 years. 
The reasons for conversion to TAVR in the surveillance arm was almost always progressive symptoms and signs of AS, not acute decompensation. 
The fatal flaw in EARLY TAVR occurred before the first patient was enrolled. Namely, all patients knew they had severe valvular heart disease. One group gets early surgery; the other group gets no treatment. This, of course, is a reasonable comparison. But the endpoints must be resistant to bias. That was not the case in EARLY TAVR. 
Christopher Rajkumar and colleagues at Imperial College called this bias “faith healing” and “subtraction anxiety.” The group who gets the active treatment is faith healed. The group who gets no intervention develops anxiety. 
They cite the FAME-2 trial as a classic example. In FAME-2, patients knew they had a serious coronary lesion, as documented by fractional flow reserve measurements. Then, one group got stented: the faith-healed group. The control arm group got medical therapy. The coronary lesion was not stented; these patients were the subtraction anxiety group. 
FAME-2 delivered positive results for stenting, but the three-component endpoint was driven by urgent revascularization. Rates of death and myocardial infarction were similar. 
In the stent arm, patients were revascularized in equal proportions because of ECG changes, troponin rises, and symptoms. But in the group who knew they had a severe and “unfixed” lesion, urgent revascularization was almost always due to symptoms rather than objective findings. This is basic human nature. No matter how skillfully it was explained, patients in the control arm believed they were left unfixed. Practicing clinicians know what happens next: Patients will be anxious and hypervigilant about symptoms. The second they present with any symptom, they go to the cath lab for revascularization. 
The exact same thing happened in EARLY TAVR. No matter the skill of the clinicians explaining the problem, patients in the control arm understood that they had severe valvular heart disease and were unlucky not to be in the treatment arm. They may have believed they had a ticking time bomb in their chest. 
In EARLY TAVR, the active arm had faith healing. The control arm had subtraction anxiety. 
Two observations provide near-proof of this proposal: First was the almost immediate separation of the Kaplan-Meier curves for unplanned hospitalizations. This is unexpected because these were asymptomatic patients who just passed an exercise test. The second was that most patients converted to TAVR not because of acute problems but because of progressive signs and symptoms — just as occurred in FAME-2. 
EARLY TAVR also found no significant differences in death or stroke. If not for the inclusion of unplanned hospitalization in the primary endpoint, clinical surveillance would have performed similarly to TAVR. There would be little enthusiasm for early intervention. The “old way” of active surveillance would have been confirmed. 
But now, solely because of a biased endpoint in an industry-sponsored trial, guidelines will probably change, key opinion leaders will lecture, review articles will be published, and cardiologists the world over will begin sending patients for early TAVR. 
Sadly, bias is not the only problem with EARLY TAVR. The other glaring problem was the lack of a surgical arm. The trial enrolled relatively young patients with low comorbidity. 
I do not understand, nor did I read an explanation in the manuscript, for why a surgical valve replacement arm was not considered in these ambulatory and relatively young patients. In fact, the two previous trials that established the equipoise underpinning EARLY TAVR were surgical trials. 
Longer-term results (5 years) of the PARTNER-3 trial comparing balloon-expandable TAVR vs SAVR clearly show a catch-up benefit for SAVR. Surgical valves had the advantage of lower rates of aortic insufficiency and valve thrombosis. 
My point is not to relitigate TAVR vs SAVR, but to question the trialists decision not to include surgical valve replacement in these patients. 
I wish it weren’t so, but the design and choices made in EARLY TAVR tempt me to be cynical about the scientific enterprise. Choosing unplanned hospitalization in this trial guaranteed a “positive” result. The purpose of medical science is not to design positive trials, but to answer important clinical questions.
The endpoint choice and the omission of a SAVR arm means the trial doesn’t fully inform the decision regarding the timing of aortic valve intervention in patients with asymptomatic but severe AS. 
Patients were experimented on, money and time were spent, and little was learned. Yet practice patterns will probably change. 
John Mandrola practices cardiac electrophysiology in Louisville, Kentucky, and is a writer and podcaster for Medscape. He espouses a conservative approach to medical practice. He participates in clinical research and writes often about the state of medical evidence.