Subscribe
Share
Share
Embed
Welcome to EP Edge Journal Watch, where cardiac electrophysiology meets evidence, precision, and perspective.
Hosted by Dr. Niraj Sharma, this weekly podcast distills high-impact cardiovascular and Cardiac Electrophysiology and arrhythmia research into clear, clinically meaningful insights. Each episode goes beyond headlines and abstracts to uncover what new studies actually mean for patient care, decision-making, and the future of electrophysiology.
What EP Edge Journal Watch stands for:
Evidence-based practice
Precision electrophysiology, arrhythmias analysis
A forward-thinking, edge-driven approach to how we interpret and apply data in real-world clinical settings.
Whether you’re an electrophysiologist, cardiologist, researcher, trainee, or allied health professional, EP Edge Journal Watch brings you the signal not the noise. Expect sharp summaries, thoughtful commentary, and practical takeaways designed for the busy clinician who wants to stay ahead of the curve
Welcome to epedge with HRS. I'm Mike Lloyd, and this is Niraj Sharma. We're recording from HRX this year in Atlanta. Both of us are local boys here, so it was an easy commute. We decided to pick three abstracts from the abstract session of this year's meeting.
Dr Michael Llyod:We tried to get you some variety and cover a breadth of topic. And, Niraj, thank you again, as always, for commentating on these. Abstract one, I thought was fairly interesting. Evaluating zero shot, which I didn't really understand the term, evaluating zero shot capabilities of multimodal large language mod models for single lead ECG interpretation. Tell us what that is.
Dr Michael Llyod:The the lead author is Lert Avadjaraj in New York, New York. I'm sorry if I mispronounced your name. Tell us about what this what this was doing.
Dr Niraj Sharma:Great. Thanks. Thanks again, Mike. As usual, pleasure to be collaborating with Heart Rhythm Society at HRX here in Atlanta. It's been a phenomenal conference thus far.
Dr Niraj Sharma:Yeah. Great abstracts, and I'm looking forward to the breaking trial session, which is happening tomorrow. Is that right? Tomorrow. Alright.
Dr Niraj Sharma:So we have three abstracts. The first one, I don't know whether you've done this or not, but I have done this where I put an EKG strip into, one of the commercially available models to see if they interpret it. And, you know, it's been a mixed bag, and this is what this trial actually, showed was a mixed bag. So what these authors did, their their objective was to use commercially available large language models and put single lead ECG strips into it from loop recorder. This is actually implantable loop recorder strips and see what the outcomes would be.
Dr Niraj Sharma:So what they studied was they looked at a total of 100 well, they looked at a thousand strips evaluated by an electrophysiologist physician agreement was threefold, I guess, in that sample of a a thousand strips. And then they land they ran these strips through chart GPT 5.4, Claude Sonnet, 4.6, Gemini, 2.5, and Gwen, 3.7. And then they also had a medical specific, a l sorry. Medical medical specific LLM called MedGemma 1.5. So those were the agents that were used to analyze this.
Dr Niraj Sharma:And what did they find? They found a mixed bag. It wasn't a a clean win for one model over another. To give you some examples, the overall best performance was by Charge GPT, and this was model 5.4. Now the models since then have changed.
Dr Niraj Sharma:We have more sophisticated models now. I think the the most recent one is Astra six. So the models have actually changed significantly since this trial was was carried out. Anyway, nonetheless, using the ChartGPT model that was available at that time, it had the highest rhythm accuracy of 52.7. And
Dr Michael Llyod:That's pretty bad. Yeah. Accuracy of 52?
Dr Niraj Sharma:That's pretty bad. Yeah. It was pretty bad. And MedGemma, the one that was designed specifically for medical input was the worst at 28 So, point yeah, it was not very good, and it failed failed predominantly, you know, in many facets. For atrial flutter, for example, it tended to call it sinus more often than atrial flutter.
Dr Niraj Sharma:It tend to undercall atrial fibrillation. That was another problem. The, the other issues were with PVC detection that was, variable across different models. So my take on this is that it's unreliable. At least the commercially available models are unreliable.
Dr Niraj Sharma:Now there's proprietary models that we have built into our loop recorder model that companies have. Sure. That's a different model, but the commercial commercial ones are based on on this abstract, not reliable.
Dr Michael Llyod:I think that it's it's a nice piece of data that we can hang on to because I believe our patients are doing this at this point. And and who's to blame them? I mean, they're there, they're gonna analyze, and they and they report with with, a confidence that probably is not founded. And I think our patients need to know this, when they just dump medical information into currently available large language models.
Dr Niraj Sharma:So I think Absolutely.
Dr Michael Llyod:I think it's a great, a great little nugget and a and a good idea.
Dr Niraj Sharma:Right. You know, one of the things that was lacking is, I mean, maybe the the whole article, if it's published, we'll we'll deep dive into it. But what I wanna know is the inconclusive that, smartwatches have, and, that's the problem that I think majority of patients would be putting it into this the commercially available models to what what is it? What's going on? And we really don't have data on the inconclusive ones.
Dr Michael Llyod:So people that get inconclusive readings on their cardiac, say, let's say, well, let's see if chat GPT figures it out. How would it how would it perform? I see what you're saying. Yeah. Well, atrial flutter was the big the big foiler for all of these.
Dr Michael Llyod:It would it did the worst. They could not figure that one out.
Dr Niraj Sharma:Yeah. So per you know, I mean, I I guess you could have atrial flutter with variable block. Yeah. But it tended to caught its sinus. Right.
Dr Niraj Sharma:Maybe it didn't see one of the flutter waves buried in there somewhere. Yeah. Maybe that was a bit caught. We this happens in our toilet EKGs quite often. Right.
Dr Niraj Sharma:We see that it's a flutter and it's labeled as sinus Yeah.
Dr Michael Llyod:Alright. VT incidents during exercise and activities of daily living. Primary author, Niall Ahmed. This is out of research groups in India and in Macon, Georgia, actually.
Dr Niraj Sharma:Yeah.
Dr Michael Llyod:Tell me about this one.
Dr Niraj Sharma:Okay. I think the premise here by by in this abstract was we we, EPs, read event monitors all the time, and we see I mean, our population, our clientele are are, I guess, patients who have higher risk for rhythm problems. But what what is the what is the rhythm what is the, I guess, prevalence of ventricular tachycardia in the general population? So that's what they were trying to figure out in this abstract. What they did here was they looked at 3,502 22.
Dr Niraj Sharma:3,522 randomized monitors. They were two week monitor duration, and then they had an electrophysiologist look at these monitors and and determine whether this was VT or not. So it was a retrospective analysis using ambulatory ECGs over a duration of fourteen days. And what they found was the percentage of ventricular tachycardia. Of is an abstract.
Dr Niraj Sharma:They don't really go into too much detail. I'm assuming it was the conventional three beat definition of VT, and the percentage that they found was one point zero five percent.
Dr Michael Llyod:Which is a lot higher than national. Right?
Dr Niraj Sharma:Yeah. Well, you know, it depends on which population you're looking at. Here, I mean, there must have been a reason why these patients got an event monitor. This was a retrospective analysis. What prompted the the monitor to be placed?
Dr Niraj Sharma:This was not just a generic population with no symptoms. So these are probably symptomatic patients without structural disease or no structural disease, we don't know, based on the abstract.
Dr Michael Llyod:It wasn't clear to me. They said 3,500 random patients, and I don't know if they actually went and just picked people to have a monitor or they selected random people out of out of a group of already monitored. So your your selection bias
Dr Niraj Sharma:Biases is is
Dr Michael Llyod:might be a problem.
Dr Niraj Sharma:Yeah. And, you know, I mean, there's data out there. You know, there's data from, I guess, Solomon. This was published in BNC, cardiovascular disorder, and they found a VT rate in patients with structural disease as high as eighteen point three percent. So the bias becomes important.
Dr Niraj Sharma:What what what population are you studying?
Dr Michael Llyod:Assuming it's general population, I found this piece of information kind of affirming because it's unusual for me to to see a a a holter, especially two week holter without any twinge of nonsyst AVT. I see it all the time. It's way higher than 0.14, which is the standard that they give in the abstract. So I I I I think our view should change on what what the real overall pop overall incidence or prevalence of VT is
Dr Niraj Sharma:in the fall. So too and and, of course, the population that you're studying. The other interesting finding that I found in this, think, which was was was striking to me, the PVC burden was 90 per hour. That correlated with more instance of, non sustained VT versus four per hour. So you if you have a patient who has lots of PVCs but does not have VT at least on that two week monitoring window, it's probably a longer term monitoring may pick up these episodes of non sustained VT.
Dr Niraj Sharma:Yeah. But that's one of the findings that came out that I took away from this abstract.
Dr Michael Llyod:Last but not least, attempted conduction system pacing and mortality insights from conspire registry with Core Lab adjudication. The primary author is Gaurav Apadhyay from Chicago, a big, conduction system pacing researcher. What what did this abstract tell us?
Dr Niraj Sharma:Right. So the COSPIRE is a was the first international multicenter prospective registry for patients. So they have a huge database of patients. And what they what they tried to do was and this was very interesting. They tried to look at con conduction system pacing versus deep septal pacing.
Dr Niraj Sharma:And they were trying to figure out whether deep septal pacing was bad versus just conduction system pacing. And the way they did that was they they looked at criteria for conduction system pacing at the time of implant, so based on the implanter's judgment. And then they sent it to a core lab to see whether it matched the the implanting physician's criteria versus the independent blinded practice. So they did that. And then once they had this, they they got that piece of data, and then they followed these patients for mortality, benefit or not.
Dr Niraj Sharma:So that that was the premise of this, of this trial, of this randomized, registry, I guess. And
Dr Michael Llyod:seven point five versus zero point eight percent mortality in people with the Core Lab now. This is not what the doctors in whole said, but the core lab, if you were adjudicated as deep septal versus left bundle branch, we're talking seven point five percent mortality versus zero point eight percent mortality. That was significant. Does that surprise you?
Dr Niraj Sharma:Yeah. I'm actually I don't I I'm not really sure. It does surprise me because it really didn't meet the criteria for condition system pacing. So the benefit that you would expect was not there with deep septal pacing. I'm not that surprised, but the question I have is this.
Dr Niraj Sharma:You are doing an implant, and you are getting criteria that doesn't really match clearly, conduction system pacing, maybe more criteria towards deep septal pacing. Do you, at that point, try different areas going deeper into the septum, or do you, at that point, switch over to a CS lead? And I assume that these folks here that were implanting, they had already attempted, you know, try to try and get, like, the, you know, adequate criteria for septal pacing. So do you leave what what would you do? Would you just I mean, based on this, I would not.
Dr Michael Llyod:Well, I think this tells me that if you don't have a nice set of crike first of all, we're still not there in terms of standardizing the definition of left bundle branch area pacing. And if you don't have good criteria for that, if it's a left bundle, then, yes, add the LV lead. Yeah. If not, you know, then then it's a little bit more muddy. But I I feel like the take homes for me for this was, people that do this still can't get what they think is is left on the branch area is not according to the core lab who who look at, you know, thousands of these.
Dr Michael Llyod:And there seemed to be an association with a very hard outcome if you kinda half did it versus you really did it.
Dr Niraj Sharma:Right. Yeah. Here, they found a 19% reclassification. So that was a nearly 20% reclassification. And and the the they had six deaths.
Dr Niraj Sharma:That was the mortality of seven point five versus zero point eight. And out of six people that died, five had indications for heart failure. The people that died, you know, maybe there's a subset of people that maybe we should be if you're not getting good criteria for CSP and the indication is for heart failure, maybe just go with the CS Yeah. At that time. Right.
Dr Niraj Sharma:This was a good and another another point that they mentioned in the abstract was the the time to peak r in lead v six. The criteria that they use at the time to peak r wave in lead v six was more than eighty five milliseconds, and that was associated with a bad outcome. Yep. Now the other criteria, I mean, one that I've used is usually 75. This was a little bit more than Little more liberal.
Dr Niraj Sharma:A little bit more liberal. Yeah. I thought. But but, anyway, I thought this was a this was a good trial. I in my mind, I I will probably not accept deep septal pacing if the indication is hard for you.
Dr Michael Llyod:I think that's the message here. You've been listening to epedge with HRS at HRX. Niraj, tell us where where we can hear more about it.
Dr Niraj Sharma:Right. So thank you again. It was a a pleasure recording this session at HRX. There's more to come. We have more trials lined up.
Dr Niraj Sharma:Look out for the weekly podcast. Look out for the weekly newsletter, and then go to epedge.org, which is our website that should be coming shortly.
Dr Michael Llyod:Nice. Thank you.
Dr Niraj Sharma:Thanks, Mike. Thank you.