Hello and welcome. On behalf of HRS, I am pleased to welcome you to HRS 2020 Science Young Investigator Competition. Today, we will highlight our clinical finalists, Edwil Present, followed by a Q&A session. Before we get going, I'd like to introduce our judges, Dr. Vasaghi, Dr. Sundaram, and Dr. Dastow. Let's get started. Our first presenter is Dr. Nong. Dr. Nong, you may load your slides now and you will have 10 minutes for your presentation. Thank you. Hello everyone. Today, I will present about the gender differences in complications following catheter ablation of atrial fibrillation. I have no conflicts of interest. This is the outline of my presentation. Now, let's start with some background information. As you all know, catheter ablation of atrial fibrillation is rapidly disseminating worldwide, but there are concerns about its safety as this procedure can cause serious complications such as pericardial effusion, major bleeding, or stroke. There are even more concerns in females as they are consistently underrepresented in AF ablation studies and have higher risk profiles such as older age and more comorbidities than males. In the literature, however, there are only few studies that have examined this gender gap, most of which were from selected cohorts and the few population studies, on the other hand, reported in-hospital complications only. Therefore, it is important to have a population study that provides unbiased estimates and fully captures both in-hospital and post-discharge complications to provide a better understanding about this gender gap. Accordingly, we used nationwide data to evaluate the gender differences in risk of complications following AF ablations. We also sought to determine whether there are gender differences in specific complications and whether they occurred in-hospital or post-discharge. So how could we achieve those aims? Our study used hospitalization data from all public and most private hospitals in Australia and New Zealand from 2010 to 2015. For each hospital encounter, a standard set of variables was collected, including patient demographics, the primary and secondary diagnosis, all procedures performed, and patient status at discharge. All disease were coded per ICD-10 and all procedure was coded per Australian classification of health interventions with good coding accuracy, and we captured post-discharge events by linking to death registry and subsequent remissions to any hospitals. The primary outcome in our study was any procedure complications that occurred during the hospital stay and up to 30 days after discharge. In-hospital complications were identified based on the secondary diagnosis and procedure codes of the index hospitalization. Post-discharge complications included post-discharge death or remissions for complications, and there were a total of 14 complications in our study which were identified based on a literature review, expert opinions, and empirical analysis of patient records. To evaluate the association between female gender and the risk of complications, we used multivariate logistic regression to adjust for gender differences in baseline characteristics, including age, ablation of both atria, AF hospitalization and ablation, and 180 comorbidities. And to evaluate the robustness of any observed association, we also performed a sensitivity analysis using propensity score matching to identify two matched cohorts with balanced covariates. Now I will talk about what our study found. So our study included more than 27,000 patients having a primary diagnosis of AF and a procedure code of catheter ablation. More than 6,000 patients were excluded mainly due to having a previous or current cardiac device implantations or unplanned hospitalizations. The final study cohort included 21,309 patients, among which 29% were females. In our study, we also observed some gender differences in baseline characteristics. Specifically, females were significantly older at higher rate of hypertension, valvular heart disease, but lower rate of coronary artery disease than males. Females also had higher rate of AF hospitalizations in the preceding year than males. Our unadjusted analysis showed that females had significantly higher rate of overall and in-hospital complications, while rates of post-discharge complications were comparable between genders. When individual complications were considered, females had significantly higher rates of certain complications, namely pericardial fusion, bleeding, vascular injury, and complete heart block, while rates of other complications such as death, stroke, or cardiac surgery were comparable between genders. After adjustment, females were still associated with a 28% increase in risk of complications, which occurred exclusively during the hospital stay, and this gender gap was driven by the increase in risk of certain complications, specifically pericardial fusion, bleeding, vascular injury, and complete heart block. As I mentioned in the method section, we used propensity score matching to evaluate the robustness of the observed gender gap in our study. Propensity score is the probability of being female conditioned on the measured covariates. Each female was matched with one male patient with similar propensity score, and there were a total of 5,736 matched pairs with balanced characteristics. As you can see in this graph, there was a good balance in the distribution of propensity score between females and males after matching, and the results from propensity score matching were fairly consistent with those of logistic regression. In conclusion, in this large nationwide cohort study, we found that females undergoing AF ablations experienced a 28% increase in risk of complications, which occurred exclusively during the hospital stay, with females having a 42% increase in risk of in-hospital complications, and this gender gap was driven by the increase in risk of certain complications, specifically vascular injury, complete heart block, pericardial fusion, and bleeding. There are several possible mechanisms behind this gender gap. The first possible explanation is the higher risk profile of females, but in our study, the gender gap persisted even after adjustment with both logistic regression and propensity score matching. The anatomical differences, such as the smaller femoral vessel size, thinner atrial wall, and the higher rate of septal aneurysm in females may also contribute to this gender gap. However, the timing and nature of these gender differences suggest that they are potentially related to procedure techniques, such as vascular access or catheter manipulation techniques. There are, however, several limitations in our study that need to be considered. Firstly, we used coded data that are generally considered less granular than data collected specifically for research purposes, but validation studies have shown good coding accuracy. There are some unmeasured confounders, like the types of energy used or ablation religions that were not captured in routinely collected data and were not adjusted for in our study, and there are some complications that do not have specific code and were not captured in our study, including atrial esophageal fistula, phrenic nerve injury, and pulmonary vein stenosis, but these are rare complications that are either rarely required treatment or were captured under other types of complications, like sepsis or stroke. So, collectively, our findings suggest that proceduralists need to be aware of the increased risk of complications in females undergoing AF ablations, and there are several strategies that can improve procedure safety in females, and we should focus on reducing the rates of pericardial fusion, vascular injury, and bleeding, as these complications were the main drivers of the gender and are potentially preventable by optimizing procedure techniques. And that's the end of my presentation. Thank you for listening. Dr. Nam, that was an excellent presentation. Your two judges are Dr. Vesaghi and Dr. Sundaram. Dr. Vesaghi would be your primary judge. She will be asking most of the questions, and Dr. Sundaram will do some follow-up questions. Dr. Vesaghi, the microphone is yours. Thank you. Very nice presentation, Linh. I had a few questions for you. One is, you may be aware that there have been a number of other national and international registries, including sub-studies of clinical trials, including the FIRE and ICE trial that analyzed over 30,000 patients and showed that women indeed have a higher complication rate, largely driven, as you have found, by vascular complications and bleeding. Was there something of weakness in the current data that drove you to try to analyze this in your cohort? Thank you for your question. Yes, I am aware about the FIRE and ICE trial that provides results regarding the gender differences in procedure safety of AF ablations. But I think that the FIRE and ICE trial were conducted in selected cohorts and usually high-experience centers, and usually it was operators with high experience that performed these procedures. So I was wondering, will we still find the gender differences with a nationally representative cohort that included every ablation, regardless of patient's age or payer or experience of the ablation center? Okay, but there have also been national registries, including registries from the U.S., Germany, China, some of which have included more than 80,000 patients that have shown similar findings and probably limited by the same limitations that you have, which is diagnostic criteria. Is there something that you think you found in your study that's different than what's already been shown? So based on the literature review that I did, I found that the largest, the studies that, I found a study from the U.S. that included more than 80,000 patients, but that study reported in-hospital complications only. So our study, on the other hand, included both in-hospital and post-discharge complications, and we also showed that the gender gap occurred exclusively during the hospital stay, but not at the post-discharge phase. Okay, I have a couple of other questions. What was the proportion of persistent versus proxismal AFib patients in women and men in your cohort? So in our data set, unfortunately, that we could not distinguish the types of AF, so we could not tell the proportion of persistent or paroxysmal AF in our study. Is that because that's not in your ICD diagnosis, in your current diagnostic criteria when you code for AFib? Do you think that the fact that women have been shown to generally present later for ablation and therefore have persistent AF may have been a driver for some of your results? Because generally for persistent AF, we tend to be a little bit more aggressive, and we tend to do more lines, and we tend to do posterior wall isolation. Yes, exactly. So studies have shown that females usually present later at ablation have more persistent, more females have persistent AF at the time of ablations, and therefore more ablations were done in females, and therefore they are more vulnerable to procedure complications. And yes, but because of the ICD-10 code, there's no specific code for persistent or paroxysmal AF, so we couldn't identify that in our data set. Okay, I have two more quick questions. One is, you looked at the study between 2010 to 2015. As you know, actually methods of ablation over the past, certainly the decade, but five years have significantly changed with advent of contact force catheters, as well as the second generation of cryo-balloon catheters. Why, if you were only looking at 30-day discharge data, why only look between 2010 and 2015? Why not do a study that looks like 2015 to 2020, when your follow-up is 30 days? So, thank you. That's a great question. But currently, we only have data that's available from 2010, 2015 with updated data coming in, so that would be a great area that we would love to look into in the near future when we have updated data. Is that largely because of access to the codes, or what was the driving factor? So, we ask for the access of data from every state and territories, and from New Zealand as well, and we could achieve data from 2010 and 2015. We are waiting for the updated data from the data custodian units in each state and territory. Then the last question I have, and I'll promise I'll hand it over to Sri, is why not look at things like pulmonary vein stenosis and phrenic nerve injury, which are more likely to occur in women because of smaller atrial sizes or smaller vein sizes? In the large population that you had available to you, I would have thought those were two things you would look for specifically. Instead, that's my question. Yes, thank you. I could still hear your questions. So, of course, we want to look at these complications, but unfortunately, with ICD-10, we don't have specific codes for these complications, and they also usually present beyond the 30-day follow-up period of our study, and their rates are usually quite low. Phrenic nerve injury is pretty immediate, right? It actually presents. You know about it if you tested it right after the ablation procedure. But I think if there was a way to capture that information, it would significantly strengthen your study. Yes, thank you. Thank you. I would love to, but yeah, unfortunately, that we don't have specific code for that complication. I'm going to hand it over. Thank you. Hi. Excellent presentation. I really enjoyed it. A few quick questions here. First, you mentioned in your paper that with the use of ultrasound for groin access and intracardiac echo, the complication rates decreased. Do you do that at the University of Adelaide now? Do you use ultrasound in ice? I'm actually a PhD student, so I'm not a practicing electrophysiologist. I'm sorry. I cannot tell that. Okay. All right. Then I guess my next question too is going to be on that lines. Do you guys discuss these complications specifically with the female patients? Sorry, can you repeat the questions? I couldn't hear you. Do you discuss with female patients and tell them that they are going to have a higher rate of complications? Well, so actually, we'll just use the gender specific rates of complications, not like we don't tell them they are going to have a high rate of complications, but we use the specific risk because we think that the overall rate of complications may not truly reflect the rate of complications in females specifically. You do tell females that there's a higher complication rate than with men? What I want to say is that we use the specific rate in females, not the overall complications rate that may be lower than the rate of complications. And one last more small question. You listed high rates of AV block in atrial fibrillation ablation. I've been over 2,500 atrial fibrillation ablations that have never had a high-grade AV block. What is the mechanism of the high-grade AV block? Yes. Thank you for your questions. We were also wondering why we have captured complete heart block after AF ablation, and we did a literature review, and we did find some studies that reported complete heart block after AF ablations, but it's not caused by the pulmonary vein isolation alone, but it was caused by additional ablations. For example, the ablation of complex fractionated electrograms near the coronary sinus ostium. That study reported a complications rate of 0.16% among more than 7,000 patients, and another study reported a complete heart block after cavo-tricuspid isthmus ablations with a rate of 0.4%. And our rate of complications for a complete heart block complication is 0.21%. And there's another population study in Germany where they also reported a 2% to 0.3% rate of complete heart block, and specifically, that study just reported in-hospital complications. Thank you. Thank you. Thank you for your excellent presentation and discussion, Dr. Nong. Our next presenter will be Dr. Chen. Thank you very much, everybody, for having me and the privilege of presenting here today. I'll be talking about optimization of current paths in the measurement of impedance change during radiofrequency ablation, providing more accurate characterization of lesion formation. These are my disclosures. Radiofrequency catheter ablation has become a cornerstone in the treatment of cardiac arrhythmias. However, there remains a lack of direct methods to measure thermal ablation lesion dimensions in the conventional EP lab, which is critical for further improving procedural safety and efficacy. While MRI can provide real-time thermal lesion imaging, it requires significant redesign of catheter lab equipment and consumables, and furthermore, is precluded in some patients. Ablation parameters are used to dose RF energy, and calculated indices based on power, duration, and contact force have improved the consistency of lesion quality. However, real-time measures that cooperate, tissue heating, remain important for titrating energy delivery and avoiding overheating. Among available measurements, circuit impedance drop is readily accessible in real time, and can reflect temperatures. This is because tissue heating causes a significant reduction in conductance of about 2% per degree Celsius, as well as reductions in tissue capacitance from disruption of cell membranes. With tissue overheating, air bubble formation can increase impedance and herald steam pops. However, circuit impedance drop has only a modest correlation with lesion dimensions. This is due to many components that contribute to circuit impedance, other than myocardial impedance alone, such as the blood pool, body tissues, contact impedance at the patient return electrode. A better way of obtaining a measure of myocardial impedance change would be to place a second electrode to direct the measurement path through the region of heating, independent of the ablation circuit. Even better would be to have a set of current injection electrodes separate from the measurement electrodes in order to eliminate electrode contact impedance at the measurements, in the measurement. Based on this concept, we created an impedance-based imaging system that uses electrodes on the ablation catheter and a set of skin electrodes to optimize the path for impedance measurement through the ablation lesion, and showed in a gel phantom model that this system could visualize heating patterns. We hypothesize in the study that a multi-channel impedance thermal imaging system can be designed and built to interface with standard EP lab equipment, and that it is feasible to obtain four terminal impedance measurements using skin and intracardiac electrodes during cardiac radiofrequency ablation. And that by optimizing the path of impedance measurement and using a four terminal method, the correlation of impedance fall with lesion dimension will significantly improve. We designed and built a prototype impedance thermal imaging system and validated the system in seven sheet. Irrigated RF ablation was performed using a thermal cool catheter at 40 Watts for 60 seconds on the RV septum and LV at 10 predetermined sites guided by fluoroscopy. The operator was blinded to both circuit impedance and ITIS impedance. Lesions were sectioned along the long axis depth, width and surface area were measured. If the lesion was not found, it was designated with dimensions of zero. The total change in ITIS impedance and ablation circuit impedance was correlated with lesion dimensions. The measurement system used a 64 electrode patient belt and electrodes on the ablation catheter. An algorithm selected three surface electrodes with the lowest impedance to the ablation electrode. Using a current, current was injected between the ring electrode on the ablation catheter and a skin electrode. And voltage was measured between the flanking skin electrodes and the ablation electrode. We gated ITIS impedance measurements to systole on the first QRS at end expiration. While only two milliseconds was required for an impedance measurement because environmental noise was unknown, 20 millisecond measurement intervals were used and eight sets of measurements were performed for averaging. The system interface with the respirometer, ECG, electrode belts, RF return patch, ablation catheter and RF generator. During ablation RF energy passes through the impedance measuring system, which switches between ablation and measurement modes. 68 ablation lesions were successfully delivered. 59 of these were identified at their expected location and nine were not found. Median lesion dimensions were as follows. Depth 3.5 millimeters, width 8.3, surface area 23.8. Median impedance drop was 23.4 ohms for the ablation circuit impedance and 5.3 ohms only for the ITIS impedance. However, because electrical noise was low at only 0.6 ohms, this was an excellent signal to noise ratio. ITIS measurement were made every 4.4 seconds taking up of 4.8% of the duty cycle. Ablation system processing and image reconstruction times were rapid taking 0.6 and 0.2 seconds respectively. We found that correlation between total impedance drop and lesion depth, width and surface areas when analyzed with after exclusion of four transmural lesions and one steam pump in these ablation lesions, dimensions could not be accurately determined, showed that circuit impedance here in blue was inferior to ITIS impedance shown in red in correlation, correlating with lesion dimensions. And in a linear mixed effects model, ITIS impedance significantly outperformed circuit impedance as a predictor of lesions with p-values as shown here. In these examples, these example ablations, ITIS impedance drop is shown on the left panel for the three lesions of varying sizes here on the right. ITIS impedance predicted lesion images are superimposed on top of these ablation lesions to show a correlation with actual ablation lesion size. While our data is insufficient to analyze detection of tissue overheating, given that only one steam pump occurred, it is nonetheless interesting to note that ITIS impedance tracing during this ablation showed an impedance rise approximately, so impedance fall of approximately 15 ohms before a transient impedance rise, which was associated with the steam pump. This drop in impedance was very large given that the upper quarter for ITIS impedance was only 8.3 ohms. Interestingly, while the impedance rise was seen in the ITIS measurement, it was not evidence in the circuit impedance measurement. This finding suggests that by taking a more direct impedance measurement path through the ablation region, gas formation in tissue may be more readily detected, though further study is needed to confirm this. In this study, we've demonstrated that it is possible to obtain cardiorespiratory-gated four terminal impedance measurements during radiofrequency ablation using a combination of skin and intracardiac electrodes to optimize the current path for measurement. The impedance change that's obtained is better correlated with lesion dimensions compared with conventional ablation circuit impedance. The low noise floor and rapid data acquisition and processing times means that there is scope to greatly improve the spatial temporal resolution of the system by taking multiple measurements using various combinations of intracardiac and surface electrodes, and we're using shorter measurement intervals to get greater temporal fidelity. As such, an impedance imaging system may provide a real-time visualization of ablation lesions created in the EP lab. Future directions will include developing this early prototype system and optimizing impedance measurement algorithms, using this system to validate the system in situations where there's abnormal myocardium, such as myocardial scar, inner myocardial tissue, as we only did ventricular ablations in this study, and using various ablation settings. And the other utility of such a system might be in terms of mapping the baseline impedance of myocardial tissue to determine where there is any fibrotic or fatty infiltration. That might be a method of detecting arythmogenic substrate for VT ablation. I'd like to thank our collaborative project team at the Cardiology Department at Westmead Hospital and the Faculty of Engineering at the University of Sydney. Thank you very much for your attention. Dr. Chen, excellent presentation. Your two judges today are Dr. Sundaram and Dr. Vesaghi. Dr. Sundaram is your primary judge. He will ask most of the questions, and Dr. Vesaghi will have some follow-up questions. Dr. Sundaram, the microphone is yours. Hi, excellent presentation, Pierre. Absolutely enjoyed it. A couple of questions. First, how did you pick 40 watts, 60 seconds? Yeah, so we wanted to make sure that we had a decent-sized lesion to do the analysis. So we decided that 40 watts was a reasonable, more of the middle-of-the-path type of ventricular ablation type of setting. We sort of secretly wanted to have a few more steam pops so that we could look for signs of overheating on our system measurement. So that's basically why we chose that setting. Second, I noticed you did not use a contact force catheter, is that correct? Yes, that's correct. Okay, so explain the, you said you had nine lesions you could not find. Yes, so we did 10 lesions per animal. And of course, it's challenging fluoroscopically to deliver 10 ablations into the right and left ventricles and have them all at separate anatomical locations. The other challenge is that when you try to perform an ablation, particularly with non-contact force sensing catheters in the ventricle, you may end up making a very small lesion, which we were okay with doing because the point of the study was to have a spread of different lesion sizes. So it's not surprising to me really that we have about nine out of 70 or 68 ablations, which we could not find. We designated them with dimensions of zero because the usual reason for not being able to find them is that simply they're small and they're not obvious from the endocardial surface. So when we look at our correlation graph, we find that in fact, whenever the lesion could not be found, the ITIS impedance was very minimal and change. So again, corroborating that these lesions were small and that was the reason we couldn't find them. Interestingly, when you look at the ablation circumference, there was still a quite a widespread, some of them had five or 10 drops. That is difficult to explain. But it's again, an important to validate the zero coordinate on the correlation graph. So how do you think your data is gonna change in the setting of contact force catheters? Well, as I alluded to in the beginning of the presentation, I think of ablation index or LSI as a way of prescribing the energy. You intend to create a lesion with a certain dose and whether or not you create that lesion depends on factors that are independent of power duration and contact force. It is largely determined by those factors, but perhaps you're making it in myocardial tissue of different properties, perhaps the stability of the catheter is different or the contact angle, all of these factors. So I think of those indices as outputs, output functions of the system and of the operator, but the tissue heating aspect of the equation is an independent measure that I think is a verification. It's like a procedural endpoint, if you like, of each single ablation. And I think it'll always be useful whether or not you have contact force or without contact force. I think that it would be interesting to look at the data in terms of controlling ablations with LSI or with ablation index and seeing whether or not this, in that setting, using an impedance imaging system can provide extra information on ablation region dimensions. You mentioned this as one of your last slides. What is your next steps? What are you taking this to eventually get to the point that we do this in humans? Yeah, this is a very early study. We wanted to know a whole lot of things, mostly to take the next steps on system design. We didn't know whether it was feasible at all to do this. We didn't know whether the noise level in the EP lab would preclude very noise sensitive impedance measuring systems. And so we wanted to know what the quality of the raw signal was. And we found that the quality of raw signal was very good. And the measurement intervals were very short. That means that we don't need a lot of averaging. And it means that we can take measurements very rapidly. So our next steps will be to use a multitude of electrodes. We have 64 electrodes there. We may not need all of them, but instead of using just three, we'll use a lot more. And we want to take multiple sections through the lesion. Because if there's sliding movement of the catheter, and for instance, the lesion dimensions will be different. They may be shallower and broader versus deeper and more focal. And also we want to try different powers, different contact force to create lesions with different thermal distributions. It would be interesting to see whether we can tell in terms of lesion dimension geometry using multiple measurements, or whether we can assay the hottest point in the lesion using the rate of impedance drop, for instance, or another signal that we can detect. And that I think will give us a lot more information on lesion formation and in terms of efficacy and safety. Thank you, Pierre. I'll turn this over to Marmar. Thank you. Great, thank you very much. So I had a couple of questions. Obviously the presentation was excellent. This is very exciting. But I wasn't sure, can you talk a little bit about what was, I noticed there are two co-first authors on this paper. What was specifically kind of your contribution to the study as opposed to Dr. Wenz? Yes. So this type of study is not possible without strong collaboration and equal collaboration between cardiology and engineering. As a cardiologist, we simply don't, we have the problem, we have the problem and the engineers have the tools, and we sort of need to combine forces to apply the right tool to the right problem. And this is an excellent example of equal contribution to a project to resolve a problem that neither one of us could have tackled on their own. So my contribution to this presentation, to this project, to this paper, and to this program is to be the clinical lead, identify the need for the problem, design the, contributing to the design of the impedance imaging system. I'm an inventor of the imaging system. And I also performed the animal experiments and designed the animal experiments, performed the animal experiments, performed part of the data analysis draft of the manuscript. And lead the project. But this would not be possible without my engineering counterpart. And that is why he has a co-authorship with this author. Okay. I mean, I was wondering because there are several references, including ones you've listed on your paper, where he's already described the system in several biomedical engineering journals. So was the unique aspect of the study, the feasibility in testing it on a large animal? He, so those publications you referenced, I'm also an author on those publications, but the reason why he is first author on those publications is because they're in engineering journals. And he is describing things like signal noise and specifics about current injection and algorithms. Now, not all of those, out of those three or four you mentioned, only one is based on this system. The others are based on impedance imaging or like impedance tomography methods that we have not yet applied clinically that we were exploring. Also to which I devoted creative energies to. But this particular system has been validated only, tested only in a gel tank model, which we developed. And this is the first animal validation. And that leads me actually right into my second question. This is thermal image. There's a whole title of imaging. Yes. You didn't show us any of the images. So where is the imaging aspect of this play into whether this system performs better or not than our current techniques and measures or parameters for ablation lesions? Yeah, so that is a good question. It is a thermal imaging system in the sense that it can take multiple impedance measurements through the lesion and reconstruct lesion dimensions. In this study, because we did not know what the quality of the signal would be if we did it in an animal rather than in a very controlled in vitro environment, we took only one measurement through the lesion. And that measurement, we want it to be the best measurement. So we took the optimal impedance path through the measurement and showed that that was a good quality signal. That is the first step to creating a lesion image that is visible on real time. What we did instead in this paper was to use that signal and the standard curves that were generated from the depth and width to create a pseudo image. And that pseudo image is a post hoc image. So it's not an image that you can see in real time. But because the processing speed is less than about 0.4 of a second in generating that image and the data acquisition is so rapid, like within 200 milliseconds. So it's reasonable to think that this type of system can create a high fidelity image in real time. Okay, we're looking forward to seeing those images. The other questions that I had was with regards to, since you were the designer of the study, why not look at EGM measurements, capture and use a contact force catheter where most of us are now using to look at correlations with FTI and ablation index? I think that this is exactly what we wanna do in the next study. And the reason why we didn't do it in this study is because it's a proof of concept study and we wanted to demonstrate that the quality of the data is better than the current method of measuring impedance. I think that measure of measuring impedance currently is now already outdated is sort of like why I think that additional measures would have been helpful in showing that you're really on top with the IDA system. My last question was in terms of your graphs. Now, as you know, many of us don't look at impedance continuously. We look at an impedance drop of greater than 10 ohms, for example, as a sign that we have created a lesion and then go from there. Obviously, if it's greater than 10, 15 or 20, even better, that would suggest a larger lesion. Why would you use a continuous impedance as opposed to using anything above 10 ohms as a comparison to your IDA system? Well, when you take a cutoff, you're sort of throwing out some data in a way when you're doing a- Right, but we already know that, let's say something less than 10 ohms may not be a very good lesion or may not have had contact. So I'm saying that if you're using it as a comparator, why wouldn't you use the way that people use our current impedance systems? You're suggesting to dose each ablation at a 10 ohm sort of cutoff. I'm suggesting that your cutoff curve and your values for looking at your correlation for impedance perhaps should start at 10 ohms or more, basically for actually making a lesion. Oh, I see. Yeah, I think that's a very good point. And I think this is something that we wanna do in a separate study where we wanna use this imaging tool or this tool to produce more consistent lesions. So for instance, a transmural linear lesion, for instance, or consistent lesions, focal lesions. In this study, we wanted to see what the quality of the signal was in terms of correlating with lesion dimensions. So we purposely wanna have a good spread of different lesion sizes. So that's why we didn't design it that way. As a follow-up to that, for the nine lesions that Dr. Sunderbhoy and kindly for that you couldn't find, did IDA show a significant impedance drop? No. It did not show a significant. So did you actually analyze that data? Yes, it's included in the graph. It's included in the correlation graph. So the nine lesions you couldn't find are still in the correlation graph? Yes. They're at zero? Yes. Okay, just to clarify. And then on a personal note, you have a very impressive CV with patents in renal denervation, and now you're working on ITIS. Can you tell us a little bit about, is it because you just decided to leave the renal denervation world or what happened to all that creative activity that you had done? We're doing renal denervation. I'm actually at, I was at Westmead Hospital. Yes. And I went on a two-year fellowship and I'm now at Brigham and Women's Hospital. Yes, yes. And so I'm going back and forth to do my renal deprivation stuff. That's ongoing. And we actually have an industrial design catheter that we just finished and got a batch of this month. So we're going to do more testing on that. So that is going. But because it's industrial design phase, it took so long, that's why we haven't had any output from that. But we did publish a couple of papers on that last year. This ITIS impedance thing, I think it seems a bit different to what I'm doing. But really, I'm interested in new ablation technology. And ablation monitoring is very important when considering if you're going to make deeper lesions with, say, microtechnology. It's important to know that you're safe. So this is entirely consistent, I think, with the direction of our group. Pia, I have a clarification question. You used the word thermal impedance early on in your presentation, thermal imaging. When you look at ablation physiology and the ones that you described was based on electrical impedance, there's a whole field of embedding thermistors in a catheter and tissue. I was misled when you started talking about thermal imaging, thermal impedance. My understanding was we were going to show thermistor data from the tissue and the catheter. Did I get this wrong, or was the word used loosely at the beginning? We validated this. Well, we tested the system. We developed the system in a gel model where the thermal change in the gel creates only a conductive decrease, which is detected as an impedance change. So just a purely thermal conductivity change is visible on the system. That is why we use thermal imaging. But of course, when you ablate tissue, you have both thermal-mediated conductive changes, and as you're alluding to, you have capacitative changes. So this impedance is a complex number. So it's true. The impedance change of signal that we're detecting is not purely thermal. But because of the way we've developed it and validated in a thermal model, we have called it a thermal imaging. So you used the word thermal only because you developed it in your gel model, not because these impedance run, like for instance, thermal runoffs drop in temperature throughout the tissue. That's what most people use the word, thermal impedance and runoffs. Impedance you measured was totally electrical measurements. Yes, yes, yes. I may have misspoke. I meant, I didn't mean to say thermal impedance. More like, yes, impedance imaging system, yeah. Because there are, as you know, big companies who are now working on thermal-guided imaging on thermistors and tissue temperatures. Anyways, lovely presentation, really enjoyed. Our next presenter will be Dr. Howell. Good afternoon, I'm Stacey Howell. I'm a second year general cardiology fellow at Oregon Health and Science University in Portland, Oregon. And I'm applying into EP fellowships this upcoming cycle. It's a pleasure to present my work titled, Sex-Specific Prediction of CRT Response Using Machine Learning, Insights from the SMART-AB Study. By way of background, we know that patients with advanced heart failure and evidence of left ventricular dyssynchrony benefit from CRT implantation. That is demonstrated here in these echocardiograms for one patient, in which after a BIV pacing, one can appreciate in the right-sided echocardiograms that the patient has evidence of reverse remodeling as indicated by an improvement in the LV ejection fraction, a decrease in the LVN diastolic diameter, and an improvement in mitral regurgitation to trace. And by way of background, the clinical trials of the early 2000s demonstrated that CRT therapy improves clinical outcomes in patients with CRT. And this included measures of reverse remodeling, like LV ejection fraction or LVN systolic volume, also quality of life measures, as well as reduced heart failure hospitalizations and mortality. This informed our current ACC AHA class one guidelines, which are listed below into the next slide as well. However, the issue remains that roughly a third of patients who receive CRT therapy are considered non-responders, meaning that despite CRT implantation, these patients have a clinical decline as though they did not receive CRT at all. Additionally, this data from the RAFT trial demonstrated that women with CRT therapy had significantly better outcomes as compared to men. And lastly, we've seen in the more recent years, the emergence of machine learning in cardiology populations, particularly the heart failure population. Machine learning algorithms are able to identify complex interactions between large amounts of data and to help to improve prediction of a specified outcome as compared to standard statistical methods. Next slide. With this in mind, this informed our current objectives, which were one, to create sex-specific prediction models for CRT response at six months using machine learning, and to compare machine learning model performance to current ACC AHA class one guidelines. Our second objective was to identify sex-specific predictors of CRT response. For our methods, we used the SMART-AV trial population, which was previously described. This was a randomized control trial in which patients who received CRT were randomized to one of three AV delay strategies. This included a fixed AV delay, an echo-optimized AV delay, and lastly, a SMART-AV algorithm delay. Overall, this was a negative study. In total, we had 741 patients, of whom roughly a third were women. The inclusion criteria are listed there, and we defined CRT response as greater than or equal to 10% reduction in the LVN systolic volume index, and or freedom from death or heart failure hospitalization at six months. We divided our total population into a machine learning training and testing cohort, which was 80% of the population, and a machine learning validation cohort, which was 20% of the population. This is customary for machine learning studies. Within the training and testing cohort, we trained our machine learning algorithms, which are listed there, on 56 different candidate predictors, and within the training and testing cohort, we then tested the performance of the machine learning models within that population. We selected the best machine learning model based on the area under the curve, and we also applied calibration using deviance and deviance ratio. We then validated our selected machine learning model in the validation cohort, which was out of sample. And lastly, we compared the machine learning model performance to that of ACC AHA Class 1 guidelines using the area under the curve. We had a number of different candidate predictors, which included clinical variables, device-related characteristics, electrocardiographic and echocardiographic data, and lastly, biomarker data. And this included markers of neurohormonal activation, systemic inflammation, as well as markers of extracellular matrix. Getting into our results, here are the baseline characteristics. Mean age was 66. It was a predominantly white population. Next, we found that the type of cardiomyopathy did vary between women and men, with more women having a non-ischemic cardiomyopathy as compared to men. We also found that both men and women had mostly left bundle branch block, although there was a smaller percentage that had different conduction disease. And now looking at the receiver operating curves for the machine learning models as compared to guidelines, starting first with all patients, our retained model had 28 predictors. And for all patients, the machine learning model outperformed guidelines to the area under the curve of 0.76 as compared to 0.62, and that was statistically significant. Similarly, looking just at women, the model retained eight predictors, and similarly, the machine learning model outperformed that of guidelines. And lastly, for men, the machine learning model retained 15 predictors and also outperformed guidelines. And now looking at the specific predictors that were retained in all three subgroups, this is depicted as a Venn diagram to point out those that are similar and different. First, we'll look at those that are similar, which are highlighted in red. QRS morphology with left bundle branch block was more predictive of CRT response, as well as C-reactive protein. Looking at those predictors that were shared between women and all patients and men in all patients, I've highlighted those in red that had the highest beta coefficient, indicating that they had more weight or were more predictive of CRT response. For men and all patients, that was the type of cardiomyopathy, with non-ischemic cardiomyopathy being more predictive, a wider QRS duration, presence of AV block, and a primary rather than secondary prevention device. And then for women and all patients, the presence of revascularization was associated with worse CRT outcome, as well as lead location. And then lastly, there were a few predictors that were unique to each of the different subgroups. That which had the highest beta coefficient was the LV ejection fraction, with a lower EF being less predictive of CRT response. Depicted here is the retained machine learning predictor importance. Plotted on the x-axis is the beta coefficient, and on the y-axis is the different predictors that are retained in order of their importance. As I had mentioned previously, for all patients, the highest predictors were LV ejection fraction and non-ischemic cardiomyopathy. For women, the presence of valvular disease or non-left bundle branch block. And for the machine learning model for men, the type of cardiomyopathy, as well as presence of AV block and a non-left bundle branch block were the highest predictors. And so the limitations in future directions for our study is we would like to validate our findings of our machine learning models in a larger study, given that our sample size was around 700 patients. We also would like to aim to create a risk score calculator, which would be available online. And clinicians could use that as part of their decision-making process for CRT implantation. And lastly, we hope that our findings can help to inform future studies on sex-specific management of patients with CRT. In conclusion, we found that machine learning can clinically improve patient selection for CRT. And two, that predictors of CRT response vary considerably by sex. And this has the potential to perhaps guide sex-specific management for CRT outcomes. Thank you, and thank you to my study co-investigators. Dr. Howell, that was an excellent presentation. Now we are up to the question session. Your primary judge will be Dr. Vasaghi. She'll ask most of the questions, followed by Dr. Dostler. Dr. Vasaghi, the microphone is yours. Stacey, very nice presentation. I enjoyed it. The first question is, can you tell us a little bit about your, why you chose the machine learning algorithms that you chose, the logistic, the four I think that you listed, and give us a little bit of insight as to which ones perhaps, or why you didn't choose others. Sure, and thank you for that question. So we were intentional in the algorithms which we used, in part because of our second aim, which we had for the study, which was to identify specific predictors of CRT response. With that in mind, we favored using the LASSO machine learning algorithm, which actually is the algorithm which we retained as the best fit model for all subgroups. The reason for that being, is that LASSO tries to identify the least number of predictors in the model. This is compared to some other machine learning algorithms, such as neural networks, which if you contrast that, actually use all of the candidate predictors. And so since we wanted to identify specific predictors, we favored those that were more selective. Secondly, it also serves to be more applicable in actual clinical practice, that you can imagine some of these clinical predictors might not be as readily available. And so as we can envision this in the future, in future directions, we'd like this to be available for clinicians like myself to use. And so if you were to retain all of the variables, that would make it more cumbersome and a little bit less able to do so. And that also somewhat brings up the point of some of the biomarkers, which I'll point out was a unique aspect of our study compared to other machine learning studies in the context of CRT. And that was something we had discussed about the importance of including those, because it could be a little bit less applicable to clinical practice. For example, the tissue inhibitors of matrix metalloproteinases, those we're not checking in a usual clinical setting, it would be a little bit more difficult to use. So with all of those things in mind, we favored the machine learning algorithms, which would narrow in on specific predictors rather than being a wider catchment. So as a follow-up to that question, yes, the unique thing about your study is that you used biomarkers and the biomarkers that I don't routinely order on my patients ended up being useful. So do you think that those biomarkers are those biomarkers that we should be ordering on these patients? And are those readily available in a hospital laboratory? So some are, namely those like NT-proBNP, which are neurohormonal markers, or those of systemic inflammation, like CRRP. And as I had mentioned, that was shared among all three subgroups. Though a majority of the others, like those that are reflective of extracellular matrix and some of the other markers of systemic inflammation are not. So the other thing I would say to take into consideration is the relative contribution of the biomarkers. And all of them, despite being retained, had relatively small beta coefficients. So I think one possibility is another way to look at the data or to consider moving forward is acknowledging what you brought up is that a lot of these are not readily available. Could we omit those as candidate predictors, rerun the models and see how our model performance do without those? And so I think that's absolutely reasonable and potentially a helpful thing to do as you think about translating this to the clinical setting. Okay, and I had two quick questions about the study in general. How many of the patients had just the death and hospital follow-up available? Because you said not everyone, I think, had echocardiographic data, right? Right, and that's in part why we made our definition of CRT response a composite outcome. It was a small percentage who did not have echo data. The specific number I don't have off the top of my head, but the majority of patients did have echo data. A small percentage did not. And so that is a limitation of the study. And also something to kind of compare and contrast with existing machine learning CRT papers that the definition of CRT response has varied considerably from study to study, majority of whom are looking at markers of reverse remodeling, like improvement in LVEF or change in LV ancestral volume index or clinical outcomes as two separate things. So we had taken that into consideration and thought that overall, since in the overwhelming majority of patients, we did have echo data, that it was a reasonable outcome. But I think you bring up an excellent point. Okay, and then one last question before I pass it on. One of the factors you analyzed was LV location, which I thought actually was interesting and a relevant parameter to analyze. But you ended up looking at base, mid, versus apical locations, as opposed to what I commonly find or have been published as sort of maybe a better predictor, which is a lateral versus a septal location. What was the rationale for a mid, base, and apical versus lateral and septal location of leads? I think it was in part that as part of the SMART-AB trial did not include septal versus lateral. And, you know, I acknowledge that that would probably carry more clinical significance. You may remember in the slide when I had pointed out the different predictors that there was kind of two ways in which we coded the LV lead location. One was looking if it was basal, mid, or apical, and then one was looking if it was apical or not. And so we, you know, tried to look at that somewhat in two different perspectives, but, you know, I realized we weren't able to look at septal versus lateral, and that was a function of the data in the trial. Great, I will pass it on. Thank you. Thank you. That was a wonderful presentation. Few questions you kind of touched on already a bit, but one of the distinct advantages of the ACCAHA guidelines are that they're simple, right? Anybody can follow these guidelines without a calculator based on data that's readily available. Could you just comment on clinical use of a predictive algorithm that has 28 variables? Are people gonna gather all that information, plug it into a calculator and use it? What are your thoughts on that? I think that's an excellent question. You know, I think there's a balance between making it simple as well as also involving the complexity of the patients, and we know that heart failure, the heart failure population is very heterogeneous, that these patients have, you know, a background of different underlying cardiomyopathy, different types of conduction delay on different combinations of medications, and so, you know, unfortunately, the heart failure population is not simple, and so I think that that does need to be accounted for in our prediction models. So, you know, I would argue that I think that incorporating some degree of the complexity is important to better capture the patients, but then the challenges you bring up, becomes how do we as clinicians actually use this? And so that's why I hope that if we were able to make a prediction risk calculator online, that that's a relatively, you know, easier thing to do, even if there are a number of variables which are retained, and I, and from the kind of the course of our conversation, I think it's really great feedback, and it makes me think maybe excluding the biomarkers could simplify things while not shortcoming the overall prediction of the model, so that may be something that we will consider. Yeah, along those same lines, I was just wondering, yeah, if you could boil this down to half a dozen variables, how much power would you lose in, if you eliminated the less, the ones with the smaller beta, you know, impact on the model how much better would it be than the guidelines? By incorporating a more limited number of inputs. Sure. Yeah, so I think you kind of talked about that. That was my main question. Thank you. Thank you. Thank you very much, Dr. Howell, for a nice presentation and a great discussion. I would like to now close and have some remarks. I wanna thank all three finalists. I wanna remind you that the Heart Rhythm Society will publish the outcome of this competition, but irrespective of that, all three of you are winners in our eyes, primarily because you have gone through a rigorous process of being selected up to this point, and I would like to congratulate you on being a finalist. I would also like to take this opportunity to thank our judges, Dr. Wasegi, Dr. Dostal, and Dr. Shri Sundaram. Thank you very much for your contribution. Thank you.

HRS 2020 Science Young Investigator Competition

cardiac electrophysiology

gender differences

complications

catheter ablation

atrial fibrillation

impedance change

radiofrequency ablation

cardiac resynchronization therapy

machine learning