Sep 13 2024 This Week in Cardiology

Please note that the text below is not a full transcript and has not been copyedited. For more insight and commentary on these stories, subscribe to the This Week in Cardiology podcast, download the Medscape app or subscribe on Apple Podcasts, Spotify, or your preferred podcast provider. This podcast is intended for healthcare professionals only.

In This Week’s Podcast

For the week ending Sept 13, 2024, John Mandrola, MD, comments on the following news and features stories: More from the European Society of Cardiology (ESC) meeting, including the SCOFF trial, MATTERHORN, humbling data with artificial intelligence (AI), and a reflection the work of being a patient.

Fasting Before Cardiac Cath

Coronary angiography isn’t what it used to be. In the old days, patients had large sheaths placed in the femoral artery. The doctor blindly stuck the artery, and an ad-hoc percutaneous coronary intervention (PCI), urgent coronary artery bypass graft (CABG), and too often vascular surgery to repair damaged femoral arteries could sometimes be required. It was a big procedure.

I guess this is the reason that proponents felt patients needed to be in the fasting state. The more likely reason for requiring fasting (NPO) after midnight is because that is what was done. That is, it is habit, or therapeutic fashion.

Patients, therefore, were kept NPO. It’s not the worst thing in the world to be fasting, though it is a buzzkill for many patients, and may cause issues in patients with diabetes.

A group from Newcastle Australia, first author, Daniel Ferreria, decided to test the dogma of fasting before cardiac procedures. The European Heart Journal (EHJ) published the trial manuscript. It wasn’t just before angiography. Slightly less than one in five procedures were for devices (pacers and implantable cardioverter-defibrillators (ICDs).

Patients were randomly assigned 1:1 to fasting as normal (6 hours solid food and 2 hours clear liquid) or no fasting requirements (encouraged to have regular meals but not mandated to do so).

They had about 350 patients in each group. The primary endpoint was composite of aspiration pneumonia, hypotension, hyper- or hypoglycemia assessed with a Bayesian approach. The authors assessed this with non-inferiority (NI).

Secondary outcomes included patient satisfaction score, new ventilation requirement (non-invasive and invasive), new ICU admission, 30-day readmission, 30-day mortality, 30-day pneumonia.

Those in the fasting arm had significantly longer solid food fasting — 13 hours vs 3 hours.
The primary composite occurred in 19% of those in the fasting arm vs only 12% in the no-fasting arm. Obviously significant, and both noninferior and superior.
The nearly 7% higher rate was driven by higher rates of hypotension, high glucose, and low glucose. No patient in either group had an aspiration pneumonia. Secondary outcomes were not different, and there were no severe outcomes related to the procedure.
Patient satisfaction was slightly higher in the no-fasting group, though 15% of patients did not feel out a questionnaire.

The authors concluded that this data “supports removing fasting requirements for patients undergoing cardiac catheterization laboratory procedures that require conscious sedation.”

Comments. I like this study. As I often say, we need to put our dogmas to the randomized controlled trial (RCT) test. And I can’t congratulate the Newcastle team enough for doing this work. Seriously, I highlight this work because de novo non-industry funded trials deserve encouragement.

I also celebrate the Bayesian analytic method. Instead of P-values we get Bayes factor and probabilities. It was beautiful to read, even though in this trial the differences were so obviously higher in the fasting arm, that no statistics were needed.

But I want to push back on the conclusion that “this data” supports removing the fasting requirement.

My reasoning: fasting is not that big of a deal.

Our goal with fasting is to prevent that one rare complication. Each group had about 350 patients per arm. That’s a lot, but our cath labs may do thousands of cases a year. If even one patient has a terrible aspiration event due to a full stomach, then the no-fasting policy fails.

The authors showed that milder events such as blood sugar levels drove the results of this trial. Big events — aspiration, ICU visits, death, etc — were 0 in each arm.
So, I am not sure they can exclude the possibility of terrible but rare events occurring with only 700 patients.
For instance, let’s say a coronary angiogram turns into a complex PCI requiring left ventricular (LV) support. Do you really want that patient to have pancakes and bacon in their stomach?
Also, Australian sedation practices are different from those in Kentucky; the average midazolam dose was 1 mg and average fentanyl was 40 mcgs in this trial. We give tons more for our device procedures. Americans, in general, require lots of sedation.

The authors allude to this in the limitations, writing that devices were only 15% of the cohort and this “limits generalizability” to device surgeries. I agree strongly.

While I like this study and conclude from it that the signal is surely pointing toward no differences in fasting vs no fasting when procedures are done like they are in Newcastle Australia, I would like to see more data before changing a policy. I could see cluster-randomized trials wherein entire cath labs or hospitals are randomized. Show me data in thousands of cases.

Finally, though, one conclusion I would make from this study is that non-thinking cancellation of procedures due to strict fasting rules should be stopped. There is room for clinical judgement. For instance, a patient who traveled 2 hours for his generator change but ate a small breakfast or had coffee with milk; we can still do these procedures especially when light sedation is planned.

Functional Mitral Valve Regurgitation

Large Trials Support Mitral Valve Repair in Heart Failure

I covered the RESHAPE-HF trial of transcatheter edge to edge repair (TEER) vs medical therapy in pts with HFrEF and functional MR last week. Which was positive but driven by hospitalizations for heart failure (HF) and quality of life measures, not cardiovascular (CV) death. This is a problem because it was an open label trial.

This week, let’s review MATTERHORN, an RCT in 210 patients with functional mitral regurgitation (MR). One group got mitral valve surgery, the other TEER. First author was Stephan Baldus And the NEJM published the paper.

These were 70 year-old patients with a mean LV ejection fraction (EF) of 43%. Pause there and note the differences with the TEER vs medical therapy trials: RESHAPE, 32%; COAPT, 31%; MITRA FR,33%.

The primary endpoint of MATTERHORN was a composite of death, HF hospitalizations, mitral valve (MV) re-intervention, assist device implant, or stroke within 1 year. Note here the many components of the primary as well as the 1-year follow-up. Whenever you have short follow-up of a surgery vs something else trial, there is bias against surgery.

This was an NI trial, and it’s important to consider how they set this up. They expected a primary endpoint in 35% of both groups. Remember that number, because event rates that come out lower make it easier to reach NI when absolute differences are chosen. Here the absolute difference chosen as the margin was 17.5%. So TEER could be declared NI if the upper bound of the risk difference was less than 17.5%.

The results were as follows:

16.7% of patients in the TEER arm had an endpoint;
22.5% of patients in the surgery arm had an endpoint.
The risk difference was 6% and 95% confidence intervals (CI) go from -17% better to 6% worse. So, since 6% is less than 17.5%, TEER is declared NI.

But you can see the problem. Instead of getting 35% endpoints, the TEER group had only 16.7%. So, it’s easy to meet NI. Had the expected event rate been estimated more accurately, they would have chosen a much tighter NI margin.

They also report safety endpoints, and it is 15% for TEER vs 55% for surgery. Most of the higher rates of safety endpoints in the surgery arm were bleeding and atrial fibrillation (AF), which is sort of ridiculous, because bleeding and AF are expected after surgery.

NEJM displays a Kaplan Meier curve for the primary endpoint and the curves are coming together at 400 days, so it makes me wonder what a 2-year endpoint would look like.

NEJM also displays a death curve. There were 8 vs 9 deaths at 1 year. And the one death causes separation at the end. It’s a silly looking curve because it makes one death look significant when it is more likely just noise.

For secondary endpoints, they chose recurrence of MR 3+ or 4+, which favored surgery. Surgeon Victor Dyan writes on X that in a young lower risk cohort, it would be more important to look at recurrences of milder forms of MR, because even 2+ MR predicts need for more intervention.
Here, surgery also favored recurrence of 2+ MR, 12% vs 27%. Also notable is that 53% of surgery patients had zero MR vs only 16% of TEER patients.
Another peculiar difference: almost 3 times as many patients in the TEER group had triple medical therapy at discharge from the hospital.

Comments. At first, I thought MATTERHORN was a problem trial because we hadn’t yet established TEER benefit over medical therapy. And I believe that is still true because of the divergent MITRA-FR and COAPT trials, and the fact that RESHAPE-HF enrolled less severely ill patients and had positive TEER outcomes driven by endpoints susceptible to open-label bias.

But MATTERHORN enrolled different patients. The EF was much higher, the EROA (effective regurgitant orifice area) much lower, and half the patients had MR because of ventricular tethering rather than annular dilation. Still, if they included medical therapy, and only had outcomes at 1 year, I wonder how that would have turned out.

A surgery vs TEER comparison was reasonable. The problem here was, it wasn’t a fair comparison.

Endpoints at a year for 70-year-old patients with decent EFs is not only not helpful, but it induces cynicism about industry influence, since industry funded the trial.
Everyone knows that CV surgery makes for tough first year. The STITCH trial of CABG vs meds in patients with severe coronary artery disease and LV dysfunction took 10 years to show a benefit.
When you choose 1 year follow-up, you make users of evidence think you are biased for the device. It may not be true, but it looks it that way.
What’s more, the NI calculations are way off. You have a 16% incidence of events in the TEER arm and a margin of 17%. To me that looks like outcomes could be twice as bad and still be NI.
It’s also cynicism-inducing when you include AF and bleeding in the first year as safety events. Everyone knows you see more of that in CV surgery. Why include it?

A patient with severe MR and relatively preserved LV function has to choose between surgery and TEER. Before they accept the newer but less invasive procedure, we should be able to offer them stronger data at longer follow-up.

I still believe TEER can be a helpful procedure, but I do not see MATTERHORN as a reason to extend its use to patients who are amendable to surgery.

The Rapidx AI Trial

I am an older doctor so I am drawn to studies that show new things aren’t better than current standards.

Dr. Dereck Chew from Australia presented the results of the The RAPIDx AI RCT at ESC. There is no paper so this is a preliminary report based on his presentation and slides.

Here is a brief discussion of the project:

The RAPIDx AI Project aims to test whether using computer programs in hospital Emergency Departments (EDs) can assist doctors in providing better care for patients with symptoms related to chest pain.
The project involves the development of computer algorithms that compare a patient's health information with a vast database of past ED patients. These programs identify similarities in variables like age, symptoms, and health status. By analyzing treatments and recovery patterns of similar patients, the algorithms can potentially support doctors in making improved decisions and suggesting personalized care.
The trial — thank goodness there was a trial — was a cluster randomized trial in 14,000 patients in 12 hospitals (6 city and 6 rural) in South Australia.
ED were randomly assigned, not patients. Control arm was standard practice; active arm was AI-based clinical support.
The primary endpoint was a typical composite of CV death, new/recurrent myocardial infarction (MI), and CV readmission within 6 months
And it was nearly identical between the control and intervention groups, at 26.4% and 26.0%.

One of the discussants pointed out the obvious — that a good doctor armed with current risk calculators is pretty good. Hard to beat. The other discussant noted that this is the beginning of an era of evaluating AI use in medicine.

I look forward to the paper and its discussion. My opinion on AI in medicine is not much more informed than a medical student’s.

I would say that we need to continue to force these new algorithms to face the test of an RCT. So good on Dr Chew and his team.

There are likely places where AI will shine, and places where it will not. Chest pain ED evaluation is not likely one where AI will help. Here is why: the problem with ED chest pain evaluation is identification of 4 to 5 sigma rare events — the one in one thousand patients who have missed MIs. These are so rare that I doubt any system will help.

What would help, I propose, is a more realistic culture wherein we speak plainly about an acceptable miss percentage. For if we keep it at zero, we will continue to overtest, overtreat, and harm people in our zero failure quest.

A place I would expect AI to help is in electrocardiogram (ECG) reading. It boggles my mind that we still have Commodore-64 level technology putting computer reads on ECGs. Someone in AI and machine learning, please develop an algorithm for ECG reading — not just in ST-elevation MI calls, but in regular everyday reads.

Finally, is it not conceivable that machine learning will help us with stroke prediction in patients with AF? We have the CHADSVASC or I guess now, just the CHADSVA score, but to me, it seems that machine learning could incorporate much more data and produce a better predictive score. Maybe I am wrong, let me know.

PROTEUS

By the way, another place AI seems not to work is in its use for interpreting stress echocardiography.

PROTEUS is a UK-based trial in 2500 patients comparing an AI image analysis report (EchoGo Pro, Ultromics Ltd, Oxford, United Kingdom) to use during image interpretation vs standard stress echo interpretation.

Before I tell you the results, let me say that stress echo readings make me grin. I trained at Indiana University and all we did was stress echo, but still, I look at them now and my partner says there is clear hypokinesis in this or that wall, and I am like ok, if you say so. It’s totally different than what we do in the electrophysiology (EP) lab, with direct measurements.

Anyway, it would seem to me that AI would help.

The primary endpoint of PROTEUS was severe CAD or evidence of an event at 6 months. It was assessed by the area under the receiver operator characteristic curve.

In sum, there was no difference in performance. Both approaches, AI vs eyeballing it, had 99% specificity and no significant differences in sensitivity, negative, or positive predictive value.

One interesting subgroup was that centers with lower stress echo rates seemed to do better with the AI tool, but it’s a subgroup of a negative trial.

Comments. Again, I commend the authors for doing this. Presenting author was Ross Upton, from University of Oxford. Imagine a world, where these products are just marketed to gullible docs, perhaps over nice lunches in clinic.

I am not nearly as enthusiastic as some of the trial discussants, who were enthusiastic and hardly deterred by the negative results.

Yes, surely, there is more work to do. More iterations. More testing. But I believe that our Bayesian priors regarding AI disrupting current care default should be pessimistic. And thus, it would take strong data to overcome these priors

In other words, decades of practicing clinical medicine count for a lot.

The Work of Being a Patient

Let me delve outside of pure cardiology for a moment (I will come back to it). This is a paper I saw from my thoughtful list of docs on X.

It was an observational study in that quantified something they called time toxicity among older patients with cancer who are receiving palliative systemic therapy. I know; it’s an oncology paper in the journal Supportive Care in Cancer.

Enrique Soto is the senior author; Joosje Baltussen is first author. The study comes out of Mexico City. Perhaps a first for This Week in Cardiology.

It was a simple but profound study: 158 patients receiving chemotherapy, about three-quarters were frail.

Within the first 6 months, patients spent a mean 21% of days with healthcare contact. One in 5 days.

Comments. I don’t believe cardiologists think enough about time toxicity. It comes up when I discuss rate vs rhythm control. All we hear about in EP meetings is EAST AF net — early rhythm control is better than rate control.

But the thing is that rhythm control comes with a lot of time toxicity. There are days for cardioversion, days for inhospital initiation of antiarrhythmic drugs, days for procedures, days for recovery from procedures, days for dealing with recurrences.

With rate control, the patient goes on a med and is seen once a couple times per year. For younger or highly symptomatic patients, or those with a decent chance of rhythm control success, the time toxicity is worth it. But the degree of symptoms relative to the baseline co-morbidity is often not worth the time toxicity. I discuss how hard rhythm control can be, but many doctors don’t.

Another place time toxicity comes up is guideline directed medical therapy in HF. Again, a 55-year-old with a bad LV and HF and no other co-morbid conditions ought to be encouraged to take aggressive uptitration of meds, no matter how much time in clinic it takes, because outcomes are modifiable.

But it is a different story for older patients with frailty or less than robust social support or multiple conditions. For these patients, get-with-the-guidelines type treatment is a disaster. For these patients, we ought to find the best tolerated minimalist approach. What drug or drugs give this patient the best benefit with the least impingement on living life? For instance, It makes no sense to prescribe an ARNI when a patient cannot afford it. Why not use a basic ACE inhibitor?

Victor Montori has written extensively on helping design the most minimally disruptive therapy. While we work to achieve the best outcomes for our patients, we should always remember that people are supposed to have lives.

Comments

Commenting is limited to medical professionals. To comment please Log-in.

Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.