Circulation July 21, 2020 Issue

This week’s episode of Circulation on the Run features author
Robert Yeh and Associate Editor Brendan Everett as they discuss
the article "Use of Administrative Claims to Assess Outcomes and
Treatment Effect in Randomized Clinical Trials for Transcatheter
Aortic Valve Replacement: Findings from the EXTEND Study."


TRANSCRIPT


Carolyn Lam: Welcome to Circulation on the Run, your weekly
podcast summary and backstage pass to the journal and its
editors. I'm Dr Carolyn Lam, Associate Editor from the National
Heart Center and Duke National University of Singapore.


Greg Hundley: And I'm Greg Hundley, Associate Editor, Director of
the Pauley Heart Center at VCU Health in Richmond, Virginia.
Well, Carolyn this week, we're going to examine outcomes in
patients that have undergone transcatheter aortic valve
replacement or TAVR. I can't wait to get to the results from the
EXTEND study. But before we do that, how about we grab a cup of
coffee and start in with some of the papers and maybe I'll go
first this time.


My paper involves a validated model for sudden cardiac death risk
prediction in pediatric hypertrophic cardiomyopathy. And the
corresponding author is Dr Seema Mital from the Hospital for Sick
Children. Well, Carolyn in this study, the objective was to
develop and validate a sudden cardiac death risk prediction model
in pediatric hypertrophic cardiomyopathy to guide sudden cardiac
death prevention strategies.


To address this, the authors performed an international
multi-center observational cohort analysis. Phenotype positive
patients with isolated hypertrophic cardiomyopathy, who were
under the age of 18 years at diagnosis were eligible. The primary
outcome variable was the time from diagnosis to a composite of
sudden cardiac death events at five years of follow-up. That
included sudden cardiac death, resuscitated sudden cardiac
arrest, and aborted sudden cardiac death, that is, an appropriate
shock following primary prevention ICD.


Carolyn Lam: Nice. What did they find?


Greg Hundley:   Well, overall 572 patients met the
eligibility criteria with 2,855 patient years of follow-up. The
five-year cumulative proportion of sudden cardiac death events
was 9%. Risk predictors included age at diagnosis, documented
non-sustained ventricular tachycardia, unexplained syncope,
septal diameter Z scores, LV posterior wall diameter Z scores, LA
diameter Z scores, peak LV outflow tract gradients, and the
presence of a pathogenic variant.


Now, unlike adults, LV outflow tract gradient had an inverse
association and family history of sudden cardiac death had no
association with sudden cardiac death. The combination of
clinical and genetic data were developed to predict five-year
freedom from sudden cardiac death.


In conclusion, the authors study provides a validated sudden
cardiac death risk prediction model with over 70% prediction
accuracy and incorporates risk factors that are unique to
pediatric hypertrophic cardiomyopathy. These results therefore
raise the possibility that an individualized risk prediction
model has the potential to improve the application of clinical
practice guidelines and shared decision making for these children
prior to an ICD insertion.


Carolyn Lam: Very interesting. Well, Greg, have you ever wondered
what are the temporal trends in the burden of comorbidities and
risk of mortality among patients with heart failure with
preserved ejection fraction or HFpEF and heart failure with
reduced ejection fraction or HFrEF? Well, the next paper comes
from Dr Caughey and colleagues from University of North Carolina
and North Carolina State University who performed an analysis of
the community surveillance component of the atherosclerosis risk
in communities, or ARIC study, and they found a significant
increase in the burden of comorbidities among hospitalized
patients with HFpEF as well as HFrEF across both sexes. Higher
number of comorbidities was associated with higher risk of
one-year mortality with a stronger association noted among
patients with HFpEF compared to HFrEF. The one-year mortality
risk associated with increasing comorbidity burden also increased
over time.


Greg Hundley: Interesting, Carolyn. So more comorbidities in
HFpEF versus HFrEF. How do we use this clinically?


Carolyn Lam: This study demonstrated a shift from ischemic
etiology heart failure to multi morbidity heart failure over
time, particularly among patients with HFpEF. This really
highlights the importance of a holistic approach in targeting
multimorbidity burden and guiding the management of patients with
heart failure.


Greg Hundley: Very interesting. Well, Carolyn, my next paper
comes from Professor Matthias Nahrendorf from Mass General
Hospital and involves the relationship between bone marrow
endothelial cells and myelopoiesis in those with diabetes.


 Carolyn, this study investigated the role of bone marrow
endothelial cells in diabetic regulation of inflammatory myeloid
cell production. The authors utilized three types of mice with
diabetes, including a streptozotocin model, a high fat diet
model, and a genetic induction using leptin receptor deficient
mice. They assayed leukocytes, hematopoietic stem cell and
progenitor cells, and endothelial cells in the bone marrow with
flow cytometry and expression profiling.


Carolyn Lam:  What did they find?


Greg Hundley:   Well in diabetes, they observed
enhanced proliferation of hematopoietic stem cells leading to
augmented circulating myeloid cell numbers. Analysis of bone
marrow niche cells revealed that endothelial cells in diabetic
mice expressed less CXCL-12, a retention factor promoting
hematopoietic stem and progenitor cell quiescence. Transcriptome
wide analysis of bone marrow endothelial cells demonstrated
enrichment of genes involved in epithelial growth factor receptor
signaling in mice with diet induced diabetes.


In summary, Carolyn, in diabetes, bone marrow endothelial cells
participate in the dysregulation of bone marrow haematopoiesis
specifically diabetes reduces endothelial production of CXCL-12,
a quiescence promoting niche factor that reduces stem cell
proliferation. The authors also describe a previously unknown
counterregulatory pathway in which protective endothelial EGFR
signaling curbs hematopoietic stem cell and progenitor cell
proliferation as well as myeloid cell production.


Carolyn Lam:     Wow, thanks for explaining
all of that, Greg. For this next paper, we're going to switch
tracks a little. This comes from Dr Drakos and colleagues from
University of Utah in Salt Lake City. They noted that significant
improvements in myocardial structure and function have been
reported in some advanced heart failure patients. This is they're
going to call responders and the responders improve the
myocardial structure and function following left ventricular
assist device induced mechanical unloading.


This therapeutic strategy may alter myocardial energy metabolism
in a manner that reverses the deleterious metabolic adaptations
of the failing heart. Dr Drakos and colleagues hypothesized that
the accumulated glycolytic intermediates are channeled into
cardioprotective and repair pathways, which may mediate
myocardial recovery in these responders.


To test this hypothesis, they prospectively obtained paired left
ventricular atypical myocardial tissue from non-failing donor
hearts, as well as responders and non-responders at left
ventricular assist device implant and at transplantation. They
conducted protein expression and metabolic profiling and
evaluated mitochondrial structure using electron microscopy.


Greg Hundley:   Interesting. What did they find,
Carolyn?


Carolyn Lam:     The recovering heart appears
to direct glycolytic metabolites into pentose phosphate pathway
and one carbon metabolism, which could contribute to
cardioprotection by generating NADPH to enhance biosynthesis and
by reducing oxidative stress. This new information could redirect
future translational investigations to efforts to identify novel
therapeutic targets for myocardial recovery in patients with
chronic heart failure.


Well, Greg, can I tell you a little bit more about what else is
in this issue? There's a letter by Dr Wang regarding the article,
A Novel Role of Cyclic Nucleotide Phosphodiesterase 10A in
Pathological Cardiac Remodeling and Dysfunction, and there's also
a response by Dr Yan. In Cardiovascular Case Series, there's a
paper by Dr Michelena on the nosology spectrum of the bicuspid
aortic valve condition, the complex presentation of valvular
aortopathy. That's so interesting.


There's a research letter by Dr Gaudino on the response of
cardiac surgery units to COVID-19, an internationally based
quantitative survey. As well as another research letter by Dr
Salem on cardiovascular toxicities associated with
Hydroxychloroquine and azithromycin and analysis of the World
Health Organization pharmacovigilance database.


There is a perspective piece by Dr Jacobs entitled The Temporary
Emergency Guidance to STEMI Systems of Care During the COVID-19
Pandemic: AHA's Mission: Lifeline.


In cardiology news, Tracy Hampton reviews three papers, one,
Video-Based AI for Beat-to-Beat Assessment of Cardiac Function in
Nature, 2020. Two, Dynamic Transcriptional Responses to Injury of
Regenerative and Non-regenerative Cardiomyocytes Revealed by
single Nucleus RNA Sequencing, that is in developmental cell
2020. And three, ATP and Voltage-Dependent Electro-Metabolic
Signaling Regulates Blood Flow in the Heart, the proceedings of
the National Academy of Science, 2020.


Greg Hundley:   Very nice. Well, I've got an in-depth
review from Dr Bin Zhou regarding the heart regeneration by
endogenous stem cells and cardiomyocyte proliferation,
controversy, fallacy, and progress. And then there are three on
my mind pieces. The first is from Dr Rashmee Shah regarding
machine learning and artificial intelligence. Do we need more
data, or do we need the right data? The next one is from Sharon
Reimold and it discusses the importance of gathering historical
information on risk factors when seeing patients with, or
suspected, of COVID-19. And then finally, Dr Prateeti discusses
ethical challenges in cardiology during the COVID-19 pandemic.


Well, Carolyn, what a great review. How about we proceed to that
feature discussion?


Carolyn Lam: Let's go.


Greg Hundley: Well listeners, we are now turning to our feature
discussion and this week we have Dr Robert Yeh, also Bobby Yeh,
from Beth Israel Hospital and our own associate editor, Dr
Brendan Everett from Brigham and Women's Hospital. Welcome
gentlemen. Bobby let's start with you. Can you tell us a little
bit about some of the background related to your study, and then
what hypothesis were you trying to address?


Robert Yeh: The study that we performed is the sub study of what
we're calling the EXTEND study, which is an NIH funded group of
investigations meant to really examine what the value is of real
world data and how it can augment clinical trial evaluations of
medical devices and therapies.


We know that randomized clinical trials remain the gold standard
for therapeutic evaluation, but they are expensive, difficult to
do, and sometimes impractical. Real world data is cheaper, it's
potentially more efficient to do observational research studies,
and in fact the 21st Century Cures Act explicitly asks, among
other things, that the FDA explore the use of real-world data for
regulatory evaluations.


People have problems with real world data, of course, they have
their own inherent challenges which are subject really to
confounding. What we thought about is, well, there's probably
this middle ground that we and others have proposed, which is
can't real world data somehow supplement or augment randomized
clinical evaluations, and in particular, in our question, can
real world data be the provider of outcomes in place of
adjudicated clinical trial outcomes?


What we did is we took two large pivotal randomized clinical
trials of transcatheter aortic valve replacement, namely the
CoreValve, high risk, and intermediate risk trials. Otherwise,
the intermediate risk trials known as the SURTAVI trial. And we
found those patients in those trials and then linked them with
real world data from administrative claims databases in Medicare.


Our hypothesis was that had the trial been evaluated in terms of
outcomes by the Medicare claims instead of the clinical trial
adjudicated outcomes. Our main question was, would we have had
the same findings within those trials? Would the primary
hypothesis of those trials still have been met with this
alternative clinical trial end point ascertainment strategy?


Greg Hundley: In your study design, how did you accomplish the
comparisons? You've told us a lot about the study population. Was
this everyone from those two studies or was this a subgroup of
them? Maybe just expand on that a little bit.


Robert Yeh: Good question. This is a US based comparison, so we
have claims for US patients, and most patients in these trials
were in the United States, but the CoreValve trial and the
SURTAVI trials, we took all of this patients and then tried to
find those patients who we could also find in Medicare claims.


It turns out that in order to qualify for Medicare, you have to
be over 65 years of age, under most circumstances. And so it's
limited to really those patients over age 65, who we could then
search for in Medicare and then within Medicare, there's two
types of insurance, fee for service and Medicare Advantage
managed care. And what we can only find are those patients who
are in Medicare fee for service, which represents somewhere
between two thirds and three quarters of patients age greater
than 65 or older. So it is a subgroup of patients in these two
large pivotal randomized trials.


We've compared those who could not be linked versus those who
could, find large part, from some of the age differences, which
are just inherent in looking at Medicare, there are really not
that many differences between those two groups.


Greg Hundley: Bobby, what did you find?


Robert Yeh: We found those patients. So now we have this
situation where we have patients in trials, and we can look at
them from two lenses. The same group of patients, one lens is
through the clinical trial lens and the second is through the
lens of real-world data, those exact same patients.


We found that whether or not we ascertained their outcomes via
claims or with clinical trial adjudication, essentially the
primary hypotheses were identical, that in both scenarios, the
transcatheter aortic valve was non-inferior to the traditional
surgical aortic valve replacement, that the effect sizes and the
hazard ratios, the confidence intervals, they were roughly the
same. In fact, for the primary endpoint of the high-risk pivotal
study, which was all cause mortality, it was identical. It turns
out that Medicare claims and what we called the denominator file
very accurately identifies exactly when a patient dies. It does
so equally well compared to rigorous clinical trial adjudication.


For SURTAVI the primary endpoint was combined death or stroke in
that case, stroke is reasonably accurate. There were a lot of
deaths that also drove that combined end point. And the net
result was that really very similar, both effect sizes and
primary, P values for those comparisons.


Greg Hundley: Now how about secondary analysis?


Robert Yeh: I think the secondary analysis that's where you had
some variability. There are some types of outcomes in this device
specific trial that are procedure oriented. Pacemakers are a
concern. Aortic valve reintervention is a concern. Those end
points in billing claims turns out are quite accurate. You can
understand why. I think providers and institutions, when they do
a service that requires insertion of a new device, they want to
get paid for those devices and they do that billing accurately.


 But there are other claims which I think are a little bit
more subjective, diagnoses that are more subjective, those like
bleeding or cardiogenic shock, those things actually started to
look different. And in fact, in some cases started to give you
different inferences if you used the clinical trial data versus
the real-world data. And so if I were to summarize it, I would
say that mortality looked absolutely pristine identical between
the two groups that some diagnoses, particularly procedural ones,
looked quite good, sufficient, I think, for an accurate
estimation of the treatment effect size as well as the magnitude
of the risk, but then some end points, I think the softer more
subjective endpoints are slightly different.


Greg Hundley:   Thank you so much, Bobby. Now we're
going to turn to our associate editor, Dr Brendan Everett, who
has helped work this article through the entire editorial process
and is also an expert epidemiologist. Brendan, we have randomized
trial data versus real world data. How do you interpret these
results in the context of how we're conducting studies both now
and then how we will conduct them in the future?


Brendan Everett: If you think of observational research and
clinical research on a spectrum with truly just observational
studies on one end, where you were trying to look at an exposure
and an outcome and adjusting for potential confounders, to
tightly controlled randomized trials on the other, Bobby's group
has managed to create a hybrid, which I think gives us some
opportunity to not have to be either on one and or the other of
the spectrum.


What I mean by that is that there are a couple key features of
trials that are retained in the approach that Bobby used, and his
group used. He mentioned those, but I want to emphasize them. I
think the key thing is that there's a randomization step. From
the perspective of an epidemiologist, that's key because it
balances between the people who get your therapy, in this case a
TAVR and don't get the new therapy…confounders that you can
measure and confounders that you can't measure, the unmeasured
confounders. So it allows some balance between the two treatment
groups so that you can be sure, at least at baseline, that
they're similar groups and what you're measuring after that point
is the effect of the intervention.


The key piece that Bobby replaced is the classical trial
ascertained end points where investigators are asked if their
patient had one of these outcomes such as a stroke or a death,
and then they're adjudicated independently.


As he pointed out, I think there are many of those outcomes, at
least in this particular application, are really well collected
by billing data. And in fact, some might argue that in some cases
they're actually better collected. There's a higher sensitivity,
if a somewhat lower specificity for the events of interest.


I think the key question, and you touched on this, Greg, is what
about the outcomes that maybe are not collected quite as well by
billing data? In particular, remember that any clinical trial is
looking at both the efficacy of a novel treatment as well as its
safety. You ultimately, at the end of the trial, want to be able
to compare efficacy with safety, to make a decision, in this case
from the FDA, a regulatory decision about whether to approve the
device or the drug.


The question becomes, what safety events are you worried about
and how reliably are you going to be able to collect them with
claims data? In this case, I think Bobby mentioned that the
bleeding data maybe was not quite as good as some of the other
safety concerns that are common in TAVR.


I think when you look to apply this approach, which I think is
ingenious, to a different research question, you have to ask
whether or not the end points, the efficacy end points, and the
safety end points that you're collecting will be done in a valid
and consistent and sensitive way with claims data as compared
with the traditional trial ascertainment process. In this case, I
think they were, but that's not always the case. We can all think
of examples where you might run into some trouble depending upon
what your end points are.


Greg Hundley:   Well, gentlemen, this has been really
an informative study to present and talk about in this feature
discussion. I want to ask you both just briefly, in a minute or
so, what do you see as the next step forward in research in this
particular area? Maybe Bobby you first and then we'll follow with
Brendan.


Robert Yeh:  I think that there are a couple of different
areas that really need to be pushed forward. One, and Brendan
alluded to this, is because these studies are really domain
specific this validation does not tell us that all claims can be
used to answer all questions, that in this particular question it
worked, but in others it might not, so more validation work in
different fields, different randomized trials, need to be done.


We're doing some of those, but they need to be done throughout so
we can really get a better sense of what are the types of
questions that are best answered by this type of linkage
approach.


The second that is more operational, we were limited to Medicare
claims data, so for questions for patients who are younger than
65, this approach just doesn't work. Whereas a place like Sweden
can do a large national registry like they did in the TASTE
randomized clinical trial and do this for their entire country.


We do need to develop better systems that have comprehensive
real-world data collection. Maybe those involve more consolidated
electronic health system, health record data that are available
in big integrated health systems, for example, but a better
system needs to be developed that can answer questions among more
than just Medicare fee for service patients.


Greg Hundley:   Very good. And Brendan?


Brendan Everett: Well, I think it's a really promising approach
to trying to lower the cost of clinical trials and to do valid
research on the effect of some treatments as compared to others.
From my standpoint, we have to be careful that we don't try and
shortcut the process too much. In particular, I think the
randomization step, at least for novel treatments, is of
fundamental importance. And of course, to do that, you have to
collect a population that is then willing to be randomized to
option A or B.


There's a lot of upfront work that is not eliminated by looking
solely at the outcomes, using this technique to look at the
outcomes. There's a lot of upfront work to collect the patients
and then randomize them.


I think also it's important, as we saw recently, that the quality
and validity of the database be ascertained and be
well-established both with the investigators and the providers of
the database. We can see that sometimes, if you're not careful,
you can come up with outcomes that are not correct because of the
example. Of course, I'm alluding to is the two papers in Lancet
and the New England Journal that had to be retracted that were
large database studies as well.


The quality of the underlying data remains paramount. That, of
course, is where a lot of the elbow grease comes in. It's not
just in the ascertainment of the events, but a lot of the stuff
that leads up to counting the events at the end of the study.


Greg Hundley:   Well, listeners, this has been just a
superb discussion. On behalf of Carolyn and myself, we wish you
another great week and look forward to catching you on the run
next week. Take care.


 This program is copyright, the American Heart Association
2020.