Hi, I'm Eric Prostowski, and welcome to another episode of EP on EP. With me today is a friend and a wonderful investigator in our field, Dr. Sumit Choo, who is the director of the Heart Rhythm Center at Cedars-Sinai. Welcome to the show. I'm glad you're here today. Thank you very much, Dr. Prostowski. So I know you've had no interest in the prevention of sudden death in anywhere in your career, but I'm going to force you to talk about it today. No, Sumit has really been one of the leaders in this area. So let's start with changes. So Sumit, have there been changes in the epidemiology of sudden death? So let me first say that I revel in the fact that I was brought up on what you taught us, Dr. Prostowski. Well, that's very kind. Thank you. I hope I taught you correctly. You absolutely did. That's a really important question, because we still have a lot of sudden cardiac deaths in the US, 1,000 Americans a day, 360,000 a year. But the epidemiology has changed in another very important way. The proportion of shockable sudden cardiac arrests has dwindled, and the proportion of non-shockable cardiac arrests has increased significantly. So 30 or 40 years ago, you'd say 75% of people have VF, or pulse-less VT, and 25% have PEA and ACEs delay. It's been reversed. So that obviously has a lot of implications for how we in the office as clinicians try to figure out who needs an ICD. So give us your thoughts on that. So now you have that piece of information in hand, you're seeing or I'm seeing a patient in the office. How does that change how you plan who gets an ICD? I think at the present time, we can't do very much with this directly. But I would say there are certain directions in which the field needs to go so we can change the way we practice in our office. The first piece of this is that the ICD is a marvelous invention based on established randomized clinical trials. We know that the ejection fraction is an important predictor, meaning an ejection fraction less than 35%. So following the guidelines, that's what I do in my office. There's been recent data about ischemic versus non-ischemic, et cetera, so we incorporate that into our daily practice. But I think what needs to change are two aspects. One is we need to figure out some guidelines for the people who don't have the ejection fraction today that buys them an ICD, i.e. the ones that have an EF over 35%. And the second part is we need to work harder to figure out who's most likely to get a rhythm with a cardiac arrest that's going to be shockable or amenable to anti-tachycardia pacing. So that leaves a wide opening for me to ask you some additional questions. I guess the first one is how. I am no longer a fan of the God ejection fraction. We all grew up with it. It was okay. I mean, I was involved in a lot of MUST and a lot of those early trials. That's what we had, but that was years ago, right? So what would you propose to us now are better differentiators? What are some of the things you think we should be using, not necessarily instead of the EF, but in addition to? Yes. Excellent question. And I think we have some work to do before we start using these novel factors. What happens is that if you look at the ejection fraction and you look at the predictive ability of it, just using a simple C statistic, it's 0.64. Right. I knew it was low, right? Yeah. But we didn't have anything better. Right. So what we must do is not think of the vulnerable ejection fraction, but think of the vulnerable patient. Okay. So, whether it's an ejection fraction below 35% or it's an ejection fraction over 35%, where we have a lot more work to do, we have to think of other predictors that could help us improve on candidate selection for the primary prevention ICD. Right. So we took a tack on this about 20 years ago, and it only took us that amount of time. 20 years ago? Yes. Okay. And we started out in a way that numbers are important. So you call up the Framingham Heart Study and you say, can I study sudden cardiac death? And they said, absolutely. It's taxpayers' money, they're funding the grants, right? Okay. But then they told me, this is 21 years ago, that they have, I asked them, how many sudden cardiac deaths do you have every year in the Framingham Heart Study? They said, well, maybe two, maybe three, and I did a back of the envelope calculation. Yeah, right. It would take me 60 years to write my first paper. So at least, and others have done this too. What we started doing was, we converted a community into a heart disease laboratory. Okay. So one million people in Portland, Oregon, where we followed them. Now we're in our 22nd year of following them. And so we prospectively ascertain cardiac arrest. And then we went back to the time that the head circumference was measured to when they had their cardiac arrest, and looked at their lifetime clinical records, which included risk factors, it included often an ECG, often an echo that was done, and tried to look at all the factors that you thought would predict cardiac arrest, but were too scared to ask. Okay. And so I think that what we have come up with is a more comprehensive risk score, where the ejection fraction is at 0.64 for a C-statistic. Last year, we were able to come up with a risk score that has 13 variables, the ejection fraction's not in there. Okay. Really? And the C-statistic is 0.81. So that's what we're trying to do. So one of my favorite risk predictors is cardiac MRI. I'm a huge fan of that, as most people know. Not as much a fan in ischemic, because I think there it's a little harder, but in non-ischemic, I've tracked that literature for years, and I think it's fairly impressive. Was that, I mean, it's so long ago you started your study. Is that part of your risk calculator? No, it isn't. And you have a very good point, that we started early, and there was not a lot of cardiac MRI. Right. But I agree with you, actually. I think there's really good data that in the mid-myocardium, especially with moderately reduced LV ejection fraction, you can do a decent job of predicting future sudden cardiac arrest risk. So I think that's one way. Let me interrupt you for a second. It makes sense to electrophysiologists, right? Because we know scar can be involved with re-entry loops and that sort of thing. So it sort of has a pathophysiologic correlate, don't you think? It certainly does. The downside is, it's expensive. Well, it's expensive to save a life, is that what you're saying? Look at it the other way. I'm going to challenge you, though. So what if I said to you, and I don't know this to be the case, we take your hard work for two decades, we add CMR, and we find out now we're maybe 0.9. Well, theoretically, that may be 10% of people that you would no longer consider for a defibrillator. That's a lot of money saved. How can I ever disagree with you? No, I'm just asking, right? So both sides of the coin. I do understand. But it's very, I mean, I've been doing this. I put in defibrillators when they first started, you know, I've been around a long time. And over the years, I've been discouraged with seeing patients come in for one, two, three generator changes and never have need for their device. I don't feel bad that I put it in. But it's always made me a little cautious about proceeding in the future with a similar patient. So I think we desperately need the kind of work you're doing, and maybe even more so. Have you done anything with AI and ECGs? Because I find that work really exciting. Yes. So I agree with you. I think that complex problems have complex solutions. Yes. And if we add... Actually, can I challenge you? Sometimes complex problems have a simple solution. If you really understood the key of that. But I don't know that key of that. So I admit. Well, that would be lovely if we had a simple solution. But the simple solution we have used, which is ejection fraction less than 35%. Which we both agree is probably not what we should do. It's really giving us diminishing returns. So I think coming to AI, the model that we used, that we have published recently, it's called VF-Risk because it predicts shockable cardiac arrest. Has used the oldest form of machine learning. And that is backwards, stepwise, logistic regression. We have 13 variables. And there are eight clinical, four on the ECG, one on the echo. And that's where you get your C-statistic of 0.81. Now moving forward, we have what we call our phenomapping project, where the larger the numbers, the better the ability to do machine learning, number one. Number two, everyone knows GEIGO, garbage in and garbage out. So we have been really working to clean this phenotype, make it homogeneous definitions of cardiac arrest. And secondly, if we discover in one population, we replicate in the other. So now we have a second Southern California population. So we're getting towards the end, but I want to ask you just two quick things. In your risk calculator, is it regardless of ischemic or non-ischemic, or is that all part of the mix? We didn't look at ischemic versus non-ischemic. However, it's regardless of age, it's regardless of ejection fraction, and it's regardless of sex. So people need to see this. Would you mind telling the listening audience where it's published? It's published in JAC, Clinical Electrophysiology, in April of 2022. And of course, it's still in the research domain. So we are hoping to do clinical trials, especially focusing on ejection fraction over 35%. Well, this has been a great discussion. You've added so much to our field, and we're all indebted to the work you've done for sudden death. I used to read your papers from Oregon, and then next thing I knew, you're in California. So I guess, but you're still continuing your work with Oregon. Yes. Sweet. Thank you so much for being on this show, and thanks for all the hard work you've done in the field. Thank you, Dr. Priskoski. You're very kind. Thank you.

epidemiology

sudden death

cardiac arrests

implantable cardioverter-defibrillator

risk prediction