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 February 14, 2025, John Mandrola, MD, comments on the following topics: Silent cerebral embolism after left atrial appendage closure (LAAC), artificial intelligence (AI) helping in ECG rhythm analysis, anti-thrombotic strategies in patients with atrial fibrillation (AF) and coronary artery disease (CAD), and more on subclinical AF decision-making.
Silent Cerebral Embolism after LAAC
The Journal of the American Heart Association published a case series of 75 patients who had LAAC and follow-up MRI scans as well as neurocognitive testing.
The study comes from Nanjing, China.
This was a pretty simple study looking at pre- and post- MRI studies as well as neurocognitive testing within the first year after implant. One cool aspect is that the patients had lots of MRIs at each visit post-procedure — 45 days, 3 months, 6 months, and 1 year.
As background, we should set out that either all or most left sided (arterial) procedures can leave “white” spots on the brain. The Chinese team call these silent cerebral embolisms (SCE).
One interesting thing about baseline characteristics in this series: Unlike the United States, the average age of these patients was young at 67 years. This will figure in when I let you know my thoughts.
75 patients had LAAC. Of these, 29 patients or 39% had a SCE post procedure. Most of these resolved by the 45-day visit.
However, 11/75 (15%) patients developed new-onset SCE during the follow-up, with a total of 16 new lesions.
The biggest finding was that neurocognitive testing tracked lower in patients with SCE. The cognitive defects did not seem to reverse over the 1-year follow-up.
The small number of patients who had late SCE, also had significant cognitive decline.
Neurocognitive testing was performed with both the Mini-Mental State Examination and Montreal Cognitive Assessment.
Comments. This is a small study, but the number MRIs and cognitive tests were striking.
The decreases in the two cognitive tests were small in absolute terms. I asked my AI helper about the minimally important difference and Claude said a 2 point decline was considered relevant.
In this paper, it was about 3-point difference between the positive SCE vs no SCE.
I cite a paper from JACC in 2019, which also correlates SCE with cognitive decline.
I made a search for such studies and found only very small series. I worry we have not looked at the risk of SCE and cognitive decline from this elective procedure.
Maybe this paper is an outlier, but I think we ought to have more data. As I noted in the earlier comments, these were 67-year-old patients. Cognitive decline is a big deal in someone that age.
AI Transforming Rhythm Monitor Reading
Nature Medicine has published a nice randomized controlled trial (RCT) comparing an AI model, called DeepRhythm AI ECG rhythm monitoring, and ECG technicians.
It’s an important study because more and more ambulatory ECG monitors are being ordered. I order a ton of them. Also not well known to the average person is that when I read the monitor, it has already been screened to find the places to look at. A reading doctor cannot scan 2 weeks, 24 hours per day, of data.
In the past, ECG technicians screened the important parts out. The promise of AI is that it may be able to do the work of the technician, and it may be more accurate.
This, of course, would require a proper trial.
Led by Dr. Linda Johnson in Norway and a huge team of cardiologists and technicians, they tested the AI algorithm for direct-to-MD reporting of ambulatory ECGs.
They had 167 ECG technicians as a control arm. They used a random sample of 5200 rhythm events identified by the AI model or the technicians.
About half of these 5200 rhythm events were identified as “critical” arrhythmias.
The mean sensitivity of the AI model for the identification of critical arrhythmias was 98.6% compared with the ECG technicians’ 80.3%.
False negatives were observed in 3.2/1000 patients for the AI model vs 44.3/1000 for the ECG techs.
The relative risk of a missed diagnosis was 14 times higher for the technicians.
In terms of specificity, the AI model had more false positives (12/1000) than the ECG techs (4/1000) patient days.
The authors concluded that the DeepRhythmAI model has excellent negative predictive value for critical arrhythmias, substantially reducing false-negative findings, but at a modest cost of increased false-positive findings.
And, AI-only analysis to facilitate direct-to-physician reporting could potentially reduce costs and improve access to care and outcomes in patients who need ambulatory ECG monitoring.
Comments. First, this is a really nice study. The DeepRhythm AI model is US Food and Drug Administration (FDA) approved and CE marked by the European Union but I laud the authors for formally studying it.
It did well. In relative terms the AI model was 14 times more likely to correctly identify a critical arrhythmia. That’s a lot. Though there was a higher rate of false positives, these can be sorted out with either ECG technicians or more likely the reading doctor.
Many of these monitors are done for critical arrhythmias, say in the evaluation of syncope. Thus, false negatives are a big problem, and that is where the AI model shines.
To me, this seems like a clear area where AI is an advance, which is different from the large language model (LLM) studies. For instance, social media is full up of studies wherein a LLM looks better than a doctor for diagnostic dilemmas. To me, these are curious findings, but I don’t see AI helping in this area that much. Why? Because 90% of proper diagnosis happens when you pull up a stool, sit at the bedside, look the patient in the eye, and listen to the patient and/or family.
But this is a clear advance. It’s pattern recognition. It may not seem like a lot to reduce ECG technician burden, but in many places, ECG technicians are scarce. What’s more, the model was better, and faster, and it never fatigues. What’s more, I am sure there will be many entries into the AI-ECG-reading market. This is probably the beginning.
I realize AI isn’t a click getter, like anticoagulation or mitral clips or TAVI, but it’s coming and it is going to help us. In a few years, it will be hard to remember medicine without it, just as it is hard for me to remember medicine without a smartphone and Internet.
Anticoagulation Alone or plus Antiplatelets in Patients with CAD and AF
JACC has published a meta-analysis of four trials comparing efficacy and safety of oral anticoagulation (OAC) monotherapy vs OAC plus single antiplatelet therapy (SAPT) in patients who have both AF and stable CAD.
It feels like I have covered this before, but I could not find any of my reports. Anyway, I strongly feel we overuse combination therapy with OAC and SAPT.
Four trials randomly assigned patients — EPIC-CAD, PRADO-AF, AFIRE, and OAC-Alone. All were done in South Korea or Japan.
Two trials used edoxaban, one trial used rivaroxaban, and the fourth used any direct OAC. All were OAC alone vs OAC + SAPT. Most of the trials used both aspirin (ASA) and a P2Y12 inhibitor. ASA predominated, I estimate by about two-thirds to one-third.
Median follow-up was 22 months. The primary efficacy outcome of the meta-analysis was myocardial infarction (MI), stroke, systemic embolism, or death. Primary safety outcome was major bleeding.
Just to be clear about what it meant to have stable CAD — patients had to be more than 6 months post a percutaneous coronary intervention (PCI) or coronary artery bypass graft if it was done for stable CAD, but more than 1 year out if the revascularization was done for acute coronary syndrome or in those patients who had more than 50% stenosis of a major coronary who were not having PCI.
Main findings:
The rate of the primary efficacy outcome did not statistically differ. Combining trials led to a hazard ratio (HR) of 0.90 but a confidence interval (CI) that went from 0.72 to 1.12 and a P = 0.34. This result stood up whether it was random or fixed effects model used.
Looking at the combined efficacy components, there was no diff in MI, stroke, death, cardiovascular (CV) death or unplanned revascularization.
Stent thrombosis occurred in two of 2000 patients in the OAC arm vs zero in the combined arm. Obviously, that was not enough events to make a comparison.
Bleeding outcomes were clearly better for the OAC alone group. Rates of major bleeding were 3.3% vs 5.7% in the four RCTs (HR: 0.59; 95% CI: 0.44 to 0.79; P < 0.001; I2 = 0%).
Two things that are important in meta-analyses are, one, the sensitivity analyses wherein the authors leave out a trial at time. This tells you if one trial dominated the results. Here, there was no change in the significance of the pooled estimates.
Also important are subgroups, as you always wonder whether there are heterogeneous treatment effects for one group or another. Here, there were no obvious subgroup effects for efficacy. For instance, those with previous PCI did not benefit from combination therapy with SAPT.
For major bleeding most subgroups showed similar results as the main finding — less with OAC monotherapy. However, there was more of a protective effect with OAC alone in men vs women, and in those with diabetes vs those without diabetes.
You may wonder whether the choice of SAPT made a difference. The answer was no. Rates of major bleeding were similar whether or not the single antiplatelet was aspirin or something else.
Comments. This is the third meta-analysis of these trials. Even if I have covered it previously, this was a strong effort on the part of the authors. As they prespecified development of their protocol and registration, they collaborated with the principal investigators of included studies to harmonize data elements and obtain relevant, otherwise unpublished information within key subgroups.
Guidelines recommend OAC monotherapy in these patients who are out a year from PCI, but it’s not what I see. Maybe your neighborhood is different but in mine, we almost never stop the antiplatelet.
But this combined data surely suggests we should consider OAC alone much more than we do.
The meta-analysis found no difference in the efficacy endpoint, no difference in any of the components of the efficacy composite, though it was underpowered to detect differences. For example, two vs zero stent thromboses in a group of 4000 patients.
The meta-analysis also finds a super-clear signal of increase in major bleeding in the combined group. Two subgroups, men and those with diabetes, may benefit most from OAC monotherapy.
My friends, we should be much more open to OAC monotherapy. Of course there are caveats, one is that all four trials were done in East Asia and, as the authors write, “Racial differences in both thromboembolism and bleeding have been reported.” The authors seem to downplay this caveat because prior studies are “directionally reproducible in European/North American cohorts”.
Also excluded from these trials were patients scheduled for coronary revascularization and other patient subgroups at high risk of ischemic events (such as those with prior stent thrombosis). So don’t translate this data to super-high-thrombotic risk CAD patients.
The authors also say there is a European trial called AQUATIC that is recruiting patients and will help inform this question.
Bottom line — we need to be less afraid of dropping the antiplatelet therapy in patients with stable CAD and AF.
DOAC for Subclinical AF – A Subgroup Analysis of ARTESIA
Once again, friends, we come to the most difficult question in electrophysiology (EP) right now: When to start OAC in patients who have short-duration asymptomatic AF on their monitor or pacer or implanted cardioverter-defibrillator (ICD) or loop recorder.
We have NOAH (rivaroxaban vs placebo) and ARTESIA (apixaban vs ASA) and both showed a reduction in stroke and systemic embolism but also a corresponding increase in major bleeding (Editor’s Note: The NOAH trial compared edoxaban, not rivaroxaban, vs placebo).
The main message I think from these trials is that patients with subclinical AF have a very low stroke rate per year, despite their high CHADSVASC score. And, since it is so low, it’s hard to make it much lower with a DOAC.
ARTESIA was a bigger trial than NOAH; 4000 vs 2500. Apixaban vs ASA led to a 37% reduction in stroke and systemic embolism. This was highly significant statistically but resulted in only a 0.5% per year risk reduction in events. And this was countered by an 80% statistically significant increase in major bleeding.
So, the main results make it hard to know what to do.
The most recent subgroup analysis, in Lancet Neurology poses an interesting question. What if we take people from the trial who had a history of stroke. Surely, these patients are enriched to be at higher risk and may benefit more. And indeed, they did. But there are caveats.
The rate of stroke in patients with a history of stroke was 7% vs 1% in those without a history of stroke.
In ARTESIA, for patients who had a history of stroke, the yearly rate of stroke/systemic embolism was 1.2% on apixaban vs 3.1% on ASA. That had an HR of 0.40 (CI 0.17 to 0.95) so statistically significant. These are yearly figures. If you look over the entire 3.5 years and consider absolute risk reductions for those with a prior stroke it is 1% on apixaban vs 7% on ASA.
By contrast, for patients in ARTESIA without a prior stroke, the yearly numbers were 0.74% on apixaban vs 1.07% on ASA. And the HR was nonsignificant at 0.69.
What about major bleeding? The yearly rate of major bleeding in patients with a history of stroke was 2.26% with apixaban (n=13; 95% CI 1.21 to 3.87) vs 1.16% with ASA (n=7; 0.47 to 2.39; HR 1·94, 95% CI 0.77 to 4.87). The absolute risk increase for apixaban in patients with prior stroke at the end of 3.5 years was 3% vs 1%.
The authors conclude — as you might if I leave out the key caveat — treatment with the apixaban in people with subclinical AF and a history of stroke or transient ischemic attack (TIA) led to a 7% absolute risk reduction in stroke or systemic embolism over 3 to 5 years, compared with a 1% absolute risk reduction for individuals without a history of stroke. And the corresponding absolute increase in major bleeding was 3% and 1%, respectively.
Thus, apixaban could be considered for secondary stroke prevention in people with subclinical atrial fibrillation and a history of stroke or TIA
Now I will tell you the problem — and it is a huge problem — that should have been set out in the abstract.
The subgroup of patients with a prior stroke or TIA was tiny. Only 8.6% of the total 4000 patients in the trial. Recall that ARTESIA required 4000 patients to be powered to detect differences.
This subgroup analysis included only 170 patients per group. And it gets worse.
The number of primary outcome events was also tiny — 7 vs 18. That’s only a small fraction of the total number of primary outcome events, which were 36 and 65 respectively.
Same with bleeding. The total number of bleeds were also only 13 vs 7.
So, while it makes sense that patients with a prior stroke and subclinical AF may benefit more from apixaban vs ASA, I think it is hard to make clinically actionable decisions on such a small number of strokes.
What’s more, the authors tell us that they did not make adjustments for multiplicity. That sounds kind of like jargon, but the gist of it is that when you make multiple comparisons, you increase the chance of finding a positive finding on chance alone. Like the famous ISIS 2 subgroup where ASA effect varied significantly by astrological sign.
The authors tell us that these findings may be related to chance and should be considered to be hypothesis generating, but this sentence is buried far into the discussion. Such a statement is not seen in the abstract nor in the first paragraph of the discussion. A reader would easily get the feeling that this is enough data to support use of apixaban in patients with subclinical AF who have a history of stroke.
I disagree with this conclusion. Yes, it’s plausible that apixaban is more likely to benefit this group but remember that these were patients with pacers and ICDs. They could easily have had a vascular stroke due to atherosclerosis. And apixaban may not be beneficial.
Recall also, that we have two negative direct OAC trials in ESUS. NAVIGATE ESUS and RE-SPECT ESUS both showed that the direct OAC did not benefit those with embolic stroke of unknown source, which was likely AF in many cases.
To me, the next step is not to over interpret a subgroup analysis that included less than 10% of the total trial cohort, and perhaps do trial of apixaban vs ASA for patients who had prior stroke and subclinical AF.
Maybe you all disagree with me on this. I realize it’s a plausible subgroup finding, and I would not be surprised if it were true, but I still think the authors were too aggressive in their conclusions.
© 2025 WebMD, LLC
Any views expressed above are the author's own and do not necessarily reflect the views of WebMD or Medscape.
Cite this: Feb 14, 2025 This Week in Cardiology Podcast - Medscape - Feb 14, 2025.
Comments