Sep 9, 2019
Dr. Carolyn Lam: Welcome to Circulation on the Run, your weekly podcast summary and backstage pass to the journal and its editors. We're your cohosts. I'm Dr. Carolyn Lam, Associate Editor from the National Heart Center, and Duke National University of Singapore.
Dr. Greg Hundley: And I'm Greg Hundley, associate editor from the Poly Heart Center at VCU health in Richmond, Virginia.
Dr. Carolyn Lam: Greg, I'm so excited about the feature paper this week. You know it deals with machine learning. It's such a hot topic now, and this one particularly deals with machine learning and the prediction of the likelihood of an acute myocardial infarction. So everyone's going to want to listen to it. Let's discuss a couple of papers and get to it, shall we?
Dr. Greg Hundley: Absolutely Carolyn, would you like to go first?
Dr. Carolyn Lam: I sure would. So my first pick is the first study to investigate the overall importance of translational regulatory networks in myocardial fibrosis. This is the study from doctors Rackham and Cook from Duke NUS Medical School here in Singapore.
Dr. Carolyn Lam: What they did is they generated nucleotide resolution translatome data during transforming growth factor beta one, or TGF beta one-driven cellular transition of human cardiac fibroblasts to myofibroblasts. So this technique identified the dynamic changes of RNA transcription and translation at several time points during the fibrotic response, revealing transient and early responder genes.
Dr. Carolyn Lam: Now, very remarkably about one third of all the changes in gene expression in activated fibroblasts was subject to translational regulation and dynamic variation in the ribosome occupancy, affected protein abundance independent of RNA levels. Ribosome occupancy in the hearts of patients with dilated cardiomyopathy suggest that the same post-transcriptional regulatory network, which was underlying cardiac fibrosis. Now key network hubs included RNA binding proteins such as PUM2 and QKI that worked in concert to regulate the translation of target transcripts in the human disease hearts.
Dr. Carolyn Lam: Furthermore, the authors showed that silencing of both PUM2 and QKI inhibited the transition of fibroblasts towards profibrotic myofibroblast in response to TGF beta one.
Dr. Greg Hundley: You know, Carolyn, this whole aspect of fibroblasts and how they turn on and turn off, become myofibroblasts, such a hot topic in heart failure. What are the clinical implications of this work?
Dr. Carolyn Lam: Yes, I agree. Well, threefold. First, these authors identified previously unappreciated genes under translational control, which could be novel candidates for disease biology and therapeutic targets.
Dr. Carolyn Lam: Number two, they found that critical fibrosis factors impacted cellular phenotypes at a protein level only, and hence these cannot be appreciated using single cell, or bulk RNA sequencing approaches. So that was significant. Finally, RNA binding proteins was shown to be central to the fibrotic response and represent unexplored gene expression regulators, and of course potential diagnostic or therapeutic targets.
Dr. Greg Hundley: Very nice Carolyn. Well, my next paper is also from the world of basic science, and it comes from Dr. Joseph Hill. Have we ever heard of him? Well of course, he's our Editor in Chief. He's going to discuss, he and his team investigated Polycycstin-1. Well, what is Polycycstin-1? It's a trans membrane protein, originally identified in autosomal dominant polycystic kidney disease, where it regulates the calcium permeate cation channel polycystin-2. So autosomal dominant, polycystic kidney disease patients develop renal failure, hypertension, left ventricular hypertrophy, atrial fibrillation and other cardiovascular disorders. These individuals harbor PC1 loss of function mutations in their cardiomyocytes, but the functional consequences of this are relatively unknown.
Dr. Greg Hundley: Now PC1 is ubiquitously expressed in its experimental ablation in cardiomyocyte specific knockout mice reduces contractile function, and in this paper the authors set out to determine the pathophysiologic role of PC1 in these cardiomyocytes.
Dr. Carolyn Lam: Huh--very interesting. I liked the way you laid that out. So what did they find?
Dr. Greg Hundley: What the investigators identified is that PC1 ablation reduced action potential duration in cardiomyocytes. They decreased calcium transients and therefore myocyte contractility. PC1 deficient cardiomyocytes manifested a reduction in sarcoplasmic reticulum calcium stores due to reduced action potential duration and circa activity, an increase in outward potassium currents decreased action potential durations in cardiomyocytes lacking PC1. PC1 coimmunoprecipitated with a potassium 4.3 channel and modeled PC1 C terminal structure suggested the existence of two docking sites for PC1 within the end terminus of K4.3. Supporting a physical interaction between the cells. Finally, a naturally occurring human mutant PC1 manifested no suppressive effects on this potassium channel activity. Thus, Carolyn, Dr Hill and colleagues' results help uncover a role for PC1 in regulating multiple potassium channels, governing membrane repolarization and alterations in circa that reduce cardiomyocyte contractility.
Dr. Carolyn Lam: Oh wow. What a bonanza of really interesting papers in this week. Now my next pick is a secondary analysis of the reveal trial. It hinges on the hypothesis that was generated from prior trials that the clinical response to cholesterol ester transfer protein or CETP inhibitor therapy may differ by ADCY9 genotype. So in the current study, authors Dr. Hopewell and colleagues from Nuffield Department of Population Health, University of Oxford examine the impact of ADCY9 genotype on the response to the CETP inhibitor Anacetrapib within the reveal trial.
Dr. Greg Hundley: Tell me, I've forgotten a little bit, but can you remind me a little about what was the reveal trial?
Dr. Carolyn Lam: Yes, of course. So the randomized placebo controlled reveal trial actually demonstrated the clinical efficacy of the CETP inhibitor Anacetrapib among more than 30,000 patients with preexisting atherosclerotic vascular disease. Now, in the current study, among more than 19,000 genotyped individuals with European ancestry, 13% had a first major vascular event during four years median follow up. The proportional reductions in the risk of major vascular events did not differ significantly by ADCY9 genotype. Furthermore, the authors showed that there were no associations between the ADCY9 genotype and the proportional reductions in the separate components of major vascular events, or any meaningful differences in lipid response to Anacetrapib.
Dr. Carolyn Lam: So in conclusion, the reveal trial being the single largest study to date to evaluate the ADCY9 pharmacogenetic interaction provided no support for the hypothesis that ADCY9 genotype is materially relevant to the clinical effects of the CETP inhibitor Anacetrapib. The ongoing dal-GenE study, however, will provide direct evidence as to whether there's any specific pharmacogenetic interaction with dalcetrapib.
Dr. Greg Hundley: Oh, very good. So we've got some results coming from dal-GenE.
Dr. Carolyn Lam: Mm.
Dr. Greg Hundley: Well, Carolyn, my last selection relates to a paper regarding the incidence of atrial fibrillation among those that exercise, and I mean really exercise.
Dr. Carolyn Lam: Ooh.
Dr. Greg Hundley: So the paper comes from Dr Nicholas Svedberg from Uppsala University, and studies have revealed a higher incidence of atrial fibrillation among well trained athletes. The authors in this study aim to investigate associations of endurance training with the incidents of atrial fibrillation and stroke, and to establish potential sex differences of such associations in this cohort of endurance trained athletes. They studied all Swedish skiers, so 208,654 that completed one or more races of the 30 to 90 kilometer cross country skiing event called the Vasaloppet from 1989 through 2011, and they had a matched sample of 527,448 non-skiers, and all of the individuals were followed until their first event of either atrial fibrillation or stroke.
Dr. Carolyn Lam: Wow. What an interesting and what a big study. So tell us, what are the results and especially were there any sex differences?
Dr. Greg Hundley: Well, interesting that you ask about those sex and gender differences. So female skiers had a lower incidence of atrial fibrillation than female non-skiers, independent of their finishing time and the number of races, whereas male skiers had a similar incidence to that of non-skiers. Second, skiers with the highest number of races or fastest finishing times had the highest incidents of the AFib, but skiers of either sex had a lower incidence of stroke than non-skiers independent of the number of races and finishing time. Third, skiers with atrial fibrillation had a higher incidence of stroke than skiers and non-skiers without atrial fibrillation. That's true for both men and women. We would think that. Finally after one had been diagnosed with atrial fibrillation, skiers with atrial fibrillation had a lower incidence of stroke and a lower mortality compared to non-skiers with atrial fibrillation.
Dr. Carolyn Lam: Very interesting. Could you sum it up for us? What's the take home?
Dr. Greg Hundley: Couple things. One, female endurance athletes appear to be less susceptible to atrial fibrillation than male endurance athletes. Second, both male and female endurance athletes have a lower risk of stroke independent of their fitness level. Third, after the diagnosis of atrial fibrillation, participants in a long distance skiing event with atrial fibrillation had a 27% lower risk of stroke and a 43% lower risk of dying compared to individuals from the general population with the diagnosis of atrial fibrillation.
Dr. Greg Hundley: So there's some clinical implications. Although very well trained men have a higher incidence of atrial fibrillation than less trained men, the incidence is on par with that of the general population and not related to a higher incidence of stroke at that group level. This indicates that exercise has very beneficial effects on other risk factors for stroke. Then lastly, atrial fibrillation in well trained individuals should be treated according to our other usual guidelines for the population at whole.
Dr. Carolyn Lam: Wow. What a fantastic study to end our little coffee chat on, but it's time to move on to our feature discussion.
Dr. Carolyn Lam: Today's feature discussion touches on super-hot topics. First of all, the perennially interesting and hot topic of the prediction of acute myocardial infarction, or should I say the more precise predictions that we can do these days. The second part of the hot topic is machine learning. Oh my goodness. This is creeping into cardiovascular medicine like never before. So I'm so glad to welcome to this discussion corresponding author of the featured paper Professor Nicholas Mills from the University of Edinburgh, as well as our Associate Editor Doctor Deborah Diercks from UT Southwestern. So welcome both, and Nick, if I could start with you, tell us about MI Cubed.
Prof Nicholas Mills: First thing to say, it was a major international collaboration, involved researchers from over nine different countries and we got together to develop and test an innovative algorithm that estimates for individual patients the probability when they attend the emergency department with acute chest pain that they may or may not have had a myocardial infarction.
Prof Nicholas Mills: Machine learning is a really new area in cardiovascular medicine as you say. Our algorithm called MI Cubed uses a fairly simple algorithm which is a decision tree. It takes into consideration really important patient factors such as age, sex, troponin concentration at presentation, and troponin concentration on subsequent testing, and the change in troponin in between those two tests in order to estimate or calculate the probability of the diagnosis. One of the really interesting aspects of this is it's not just an algorithm for research, it's a clinical decision support tool as well. So what we've done is taken the output from that algorithm and translated it into something that is meaningful for clinicians. We've kept it quite simple. It gives an output between zero and a hundred, which is directly proportional to the likelihood of the patient having a myocardial infarct. We also provide estimated diagnostic metrics. So sensitivities and specificities that relate to that individual patient. It's really going to change the way we think about the interpretation of cardiac troponin in clinical practice.
Dr. Carolyn Lam: Indeed, and first audience please, please look up the beautiful figures of this paper. I think it summarizes it all. The algorithm shows you what MI Cubed is and then compares it to the ESC three hour algorithm, one hour algorithm. Then I love the last figure, where you actually show us that very important component that you just said. As a clinical support tool, how it's going to work. So we actually have pictures of your cell phone and showing you the pictures that you're going to get from it. So super cool. Beautiful paper.
Dr. Carolyn Lam: Now I just have so much to talk about, first the machine learning bit, always sexy sounding, but a bit scary for clinicians. So I really like the fact that you broke it down to actually say what components go in so that people aren't afraid of this black box. We don't know what's going on. Is there like a set time between samples, or how does this work? Do you need to have it within a certain timing? How does that fall in? Is it a particular type of troponin, what are some of the specs of the model that a practicing clinician needs to know?
Prof Nicholas Mills: Well, in order to answer that question, I might explain to you the rationale for developing it. So when you're assessing a patient in the emergency department, we all recognize in our daily practice that patients differ. So interpreting troponin has been challenging. One threshold for all may not be the right way to approach this really important clinical diagnosis. Troponin concentrations differ in men and women. They differ by age, and as a surrogate of the presence of comorbidities. They differ depending on the timing of when you take that sample and when you repeat that measurement, and that has introduced some complexity. So many interesting pathways have been developed for guidelines which try and apply fixed thresholds and fixed time points, and it's pretty tough to deliver in the real world setting of a super busy emergency department. So the premise for developing this algorithm was we wanted something that was really flexible, that recognized that patients are different, they're not all the same.
Prof Nicholas Mills: That's why we went for a machine learned approach rather than a more conventional statistical model. So you asked about the specification. You can do your two troponin tests whenever you like. So I had across the 11,000 patients huge variation in the timing of samples, but that is okay for MI Cubed. If you repeat the test within an hour, two hours, three hours, six hours, it still provides the same diagnostic performance. I think that's really important.
Prof Nicholas Mills: You also mentioned specification about the assay. This algorithm has been developed using a particular high sensitivity cardiac troponin assay developed by Abbott Diagnostics. It will be effective for other high sensitive troponin assays, but it's unlikely to be as effective using a contemporary assay. So if your hospital uses a contemporary or conventional cardiac troponin assay, this might not be the right algorithm for you.
Dr. Carolyn Lam: Great. Thank you for breaking down the issue so beautifully and practically. It really makes me think, oh my goodness, this paper's just far more than about MI. Because you know, natriuretic peptides, you could say the same thing. A prediction of heart failure is the same thing, you know? So the whole approach is novel. Deb, could you please share your thoughts and perspectives on where this is going perhaps?
Dr. Deborah Diercks: I think this study is terrific because I think it does, as Dr. Mills stated, reflect reality. We don't draw measures at zero, exactly at zero, and exactly at one and exactly at three, especially in a busy emergency department. So I think it provides flexibility to the physician and provider in using it to be able to interpret values in a world that doesn't fit complete structure like the guidelines are written out. What I find really interesting about this study, and I'd love to hear more about, is how you decided the thresholds of where low risk and high risk were cut at. It mentions by consensus, and I guess I would have loved to have been a fly on the wall to hear how those discussions went, and would love to hear more from you Dr. Mills about that.
Prof Nicholas Mills: Fascinating discussions amongst all the investigators on this project as to how we would define that. The first point I would make though is we designed the algorithm to provide a continuous output, a continuous measure of risk. So your MI Cubed score is between zero and a hundred. You don't have to apply a threshold, but we are used to in clinical practice having processes that support our triage of patients, and identifying people as low risk and high risk. Therefore we felt upfront that we should evaluate specific low risk and high risk thresholds.
Prof Nicholas Mills: So low-risk, we were completely unanimous on how to define that, and it was based on some really nice work done by emergency physicians in New Zealand. Martin Fan, who's the first author on this paper, surveyed many emergency physicians and asked about their acceptance of risk. They came up with the concept that an algorithm to be considered safe in emergency medicine would be acceptable if the sensitivity was greater than 99% or the negative predictive value was greater than 99.5%.
Prof Nicholas Mills: So we agreed up front that we would hold our low risk thresholds to those bars. Those metrics. Where there was less agreement was how you defined high risk. That didn't surprise me hugely. The positive predictive value of troponin is one of the most controversial topics around. Most cardiologists [crosstalk 00:20:52] of troponin has been difficult for them in clinical practice because with the improvements in sensitivity we are seeing lower specificity and lower causative link to value. If I put it into context, just measuring troponin and using the 99 percentile in consecutive patients gives you a positive predictive value of around about 45 to 50% in most healthcare systems for the diagnosis of type one myocardial infarction. Therein lies the problem. So one in every two patients has an abnormal troponin result but doesn't have the condition that we have evidence based treatments for, and whom cardiologists who are often quite simplistic in their approach to the assessment of these patients know how to manage.
Prof Nicholas Mills: Every second patient we don't know how to manage, and therefore we wanted an algorithm that would help us identify those patients who can go through our often guideline-based pathways and treatment pathways for acute coronary syndromes more effectively. We eventually agreed that a positive predictive value of 75% would be ideal. So three out of every four patients would have the diagnosis that we knew how to manage and treat. That was our target. We got pretty close to it in our test set. I think the actual positive predictive value at the threshold of around an MI Cubed value of 50 was 72%, so pretty effective. Certainly a lot better than relying on a kind of binary threshold such as the 99 percentile to identify high risk patients.
Dr. Deborah Diercks.: Thanks for that great answer. My next question is how do you think MI Cubed is going to integrate, or will it even replace the need for other risk stratification tools that we often use the emergency departments such as TIMI or the heart score?
Prof Nicholas Mills: Fabulous question. In this analysis, we haven't specifically compared the performance of MI Cubed with TIMI or heart, so my answer is going to be a little speculative. You can forgive me hopefully. Both those scores were developed prior to the widespread use of high sensitive cardiac troponin tests. I think what we've learned since the introduction of high sensitive cardiac troponin is that we're using this test as a risk stratification tool, and a lot of the power of the MI Cubed algorithm comes from the way that it identifies extremely low risk patients with very low and unchanging cardiac troponin concentrations way below the diagnostic threshold.
Prof Nicholas Mills: TIMI and heart simply consider troponin as a binary test, a positive or negative test, and do not take advantage of the real power of the test to restratify patients. All the evidence to date that has compared TIMI and heart with pathways that use high sensitive troponin in this way, both to restratify and diagnose patients show that these risk tools add very little in terms of safety, but do make pathways more conservative. So they identify fewer patients that are lower risk and permit discharge of those patients.
Prof Nicholas Mills: So my concern about using an algorithm like MI Cubed with an existing tool like heart is that it will undermine much of the effectiveness of this tool which identifies around about two thirds of patients as low risk. If you were to combine that with a heart score, you would reduce the effectiveness. I don't think you get a gain in performance, but further research is required to do a head to head comparison with these sorts of traditional restratification tools.
Dr. Carolyn Lam: I'm so grateful for this discussion, both Nick and Deb. In fact, I was about to ask what are the next steps and I think Nick you just articulated it. Deb, I want to leave the final words to you. Do you have anything else to add?
Dr. Deborah Diercks: I think this study represents a real change in how we can practice medicine, where we can actually take our biomarkers that actually have really strong value and utilize them in a manner that is pragmatic. It can actually introduce and take full advantage of them, and so I think this is a great opportunity for us to rethink our usual approach, which frankly, especially for troponin has really been very binary and very static. Thank you so much Dr Mills for the innovation and the willingness to look into this area.
Dr. Carolyn Lam: Thank you so much. This paper is like a sneak peak into the future of what we'll be practicing medicine like. Well, audience, you heard it right here on Circulation on the Run. Don't forget to tune in again next week.
This program is copyright American Heart Association 2019.