Kenneth Bilchick, MD, MS, is a professor of cardiology, clinical cardiac electrophysiologist, and director of electrophysiology research at UVA.
His research focuses on employing advanced statistical methods and AI to analyze data and improve the effectiveness of procedural interventions for patients with heart rhythm disorders and heart failure. His ultimate goal is to offer personalized therapies for this population that result in the best outcomes. Here, he discusses this research's potential impact on heart rhythm care.
What are you working on right now?
One of the exciting projects we have right now is an analysis of cardiac magnetic resonance (CMR) images of the heart and ECGs with deep learning. "Deep learning" refers to an artificial intelligence approach that uses multi-layer neural networks to solve problems.
A particular problem of interest for AI with deep learning involves identifying people who are at the highest risk of dying suddenly. An apparent paradox in cardiovascular medicine is that patients with known structural heart disease have a higher proportional risk of arrhythmic death, while patients without preserved or only mildly compromised heart function have a lower proportional risk of sudden death, but a far greater absolute risk of sudden death, because there are more patients in this latter group.
We offer implantable cardioverter defibrillators (ICDs) to patients with known severe structural heart disease, but many patients in the latter group with mild or no known compromise of their heart function are never offered ICDs and die suddenly. Therefore, there is a clinical need to identify patients in this at-risk group, particularly those with coronary artery disease and prior myocardial infarction, who could benefit from treatments to rescue them from sudden death.
In this regard, we have developed very innovative AI pipelines to use ECGs and CMR to identify these patients. We are presently applying these methods to thousands of patients with ECGs and CMR. We believe this will have a major impact on the prevention of sudden death, which takes many people away from their families prematurely and without warning.
What are the most intriguing potential clinical applications of your work?
A particularly intriguing clinical application of this work is the potential to develop an innovative and cost-effective strategy to identify patients with the greatest risk of sudden cardiac arrest. Specifically, we want to use deep learning to identify patients with no more than mildly compromised heart function (left ventricular ejection fraction [LVEF] > 35%) who presently are not offered treatments to protect them against sudden cardiac arrest, even though they are at high risk.
In turn, we are also using these methods to identify patients with more significant heart pump dysfunction who presently receive treatments, such as ICDs, but who do not need them because they are not forecasted to have sudden cardiac arrest despite their heart pump dysfunction.
These observations highlight a problem of poor resource utilization in clinical medicine based on the fact that the degree of heart pump dysfunction (from the LVEF) is only moderately correlated with the risk of sudden cardiac arrest. As a result, many patients with normal or mildly compromised heart pump function could benefit from ICDs but do not get them, and many patients with more severe heart pump dysfunction who are not expected to benefit from ICDs actually do get them, anyway.
We are working on a promising strategy to address this problem, which involves starting with the AI-based analyses of the tests that are most widely available and least expensive, followed by AI applications to more involved tests, such as CMR, as needed. In this regard, a logical progression of testing would start with a clinical risk score and ECG, followed by CMR as needed based on the initial testing results.
Building on our prior work on risk scores for patients with LVEF ≤ 35%, we have recently developed a risk score to identify patients with LVEF > 35% at risk of sudden cardiac arrest. Regarding ECGs, we have developed very innovative AI methods based on the standard 12-lead ECG to predict the risk of sudden cardiac arrest and ventricular arrhythmias. I find this part of the research program very intriguing because, for decades, people thought that we could only extract parameters based on criteria apparent to the human eye from 12-lead ECGs. AI has demonstrated a “tip of the iceberg” paradigm such that AI methods reveal incredible information that lies “under the surface.” Finally, CMR is considered the gold standard for imaging cardiac structure and function; however, as with ECG analysis, we have only recently figured out how to apply AI-based methods, such as radiomics and deep learning with neural networks, to extract this information.
What made you choose UVA Health as the place to do your research?
I have been a faculty member at the University of Virginia for 18 years. I'm now a professor of Medicine. After completing internal medicine, cardiovascular medicine, and clinical cardiac electrophysiology training at Johns Hopkins, I considered several great faculty opportunities at prominent universities. I chose UVA over these other options because I was convinced we have the optimal combination of clinical excellence, research innovation, mentorship networks, state-of-the-art facilities, and collaborative opportunities.
With respect to clinical excellence, my early clinical mentor, John DiMarco, MD, PhD, was one of a handful of pioneers who started the field of clinical cardiac electrophysiology (also called EP or heart rhythm cardiology), and he was director of our clinical EP section when I accepted the faculty position here in 2007. It is remarkable how the field has evolved since that time. In fact, our electrophysiology practice today treats many patients from all over the state of Virginia and beyond, and the growth has been amazing.
At UVA Health, our research programs and collaborations with industry facilitate the acquisition of the latest technology to treat patients with many arrhythmia disorders, including atrial fibrillation, ventricular tachycardia, heart failure, pacing needs, supraventricular tachycardia, genetic arrhythmias, heart failure diagnoses, and other conditions. We also continue to be part of national and international clinical studies that bring the technology our patients want to Charlottesville.
With respect to mentorship and collegiality, mentorship from DiMarco, Chris Kramer, MD (currently Cardiology division chief), and Frederick Epstein, PhD (MRI physicist and recent department chief for Biomedical Engineering) has been critical for the success of my research program. In addition, UVA physicians have the opportunity to collaborate across grounds with faculty who have expertise in other key areas, allowing our patients to benefit from cross-disciplinary research.
In my research, I have collaborated with computer scientists, data scientists, cardiac immunologists, statisticians, biomedical engineers, data scientists, and others. In addition, I have had the opportunity to serve on doctoral thesis committees for many trainees working with my collaborators.
What do you wish more people knew about your area of research?
I would like people to appreciate that there are exciting opportunities to use math, statistics, artificial intelligence, and biomedical engineering to solve important clinical problems for patients with cardiovascular disease. In addition, other areas of clinical medicine, such as advanced cardiac imaging and immunology, are also very relevant to my research related to heart failure and arrhythmias. For this reason, I maintain expertise in statistics, artificial intelligence, and cardiac magnetic resonance (CMR) in addition to clinical cardiac electrophysiology, which is my main focus for clinical practice. And I highly value collaborative work with my colleagues who specialize in biomedical engineering and computer science.
An example of this collaborative work with biomedical engineers is my research on improving clinical outcomes for patients receiving implantable cardioverter defibrillators and cardiac resynchronization therapy (CRT) devices. CRT refers to a pacing therapy that can improve cardiac function, symptoms, and survival in heart failure by correcting electrical abnormalities. An issue with CRT is that its beneficial effect depends on whether it is implemented in the best way for each individual. In very innovative collaborative work with biomedical engineer and MRI physicist Epstein, we collected information about each patient’s cardiac structure and function from a CMR examination and ECGs. We then used sophisticated mathematical methods to generate parameters to predict response from these studies, and developed a method to integrate the CMR images into the clinical procedure to help operators guide the critical CRT pacing lead to the left ventricular pacing location expected to yield the best response.
We then worked with cardiac immunologist Coleen McNamara, MD, to demonstrate how CRT modulates the inflammatory response in patients with heart failure, and used a sophisticated flow cytometry method to demonstrate the importance of a key protein on B cells that is modified by CRT.
Another recent exciting development in this field of pacing for heart failure is the growth of conduction system pacing, with pacing of the left bundle branch as an alternative approach to CRT implemented as biventricular pacing. We are part of groundbreaking multicenter studies in this area, and there are key opportunities to use AI-based analyses of CMR and ECGs to personalize the choice of pacing therapy and pacing approach to improve outcomes in patients with heart failure.
How did you become interested in your area of research?
My research is focused on the analysis of cardiac magnetic resonance (CMR) imaging studies, electrocardiographic data, and biological data related to inflammation using advanced statistical methods and AI to improve outcomes and the effectiveness of procedural interventions for patients with heart failure and arrhythmias. My interest in this area of research is closely linked to my clinical specialty of heart rhythm cardiology.
In my clinical specialty, much of my clinical effort is devoted to performing procedures for patients with heart rhythm disorders and patients with heart failure. These procedures include:
- Catheter ablation for atrial fibrillation, ventricular tachycardia, and supraventricular tachycardia.
- Pacemakers, with a focus on left bundle branch pacing and leadless pacing for patients with and without heart failure.
- Implantable cardioverter defibrillators implemented as transvenous and subcutaneous systems, usually for patients with heart failure and/or a prior myocardial infarction.
For me, research was never a secondary vocation relative to my clinical work, but rather an integral part of my clinical mission from the start. In other words, I have always wanted to deliver the highest quality of care to my patients, and innovating to develop new therapies and personalize existing therapies is clearly a part of this mission.
Along the way here at UVA, I have completed two masters degrees (in clinical investigation and, most recently, statistics), mentored over 40 trainees in research, served on key research committees at UVA (Institutional Review Board for Health Sciences Research, Cardiovascular Research Oversight Committee, and the iPrecision Immunomedicine Collaborative Research Program), served as chair of several national/international professional committees (including my current roles as chair of the Heart Rhythm Society Research Committee and Society for Cardiovascular Magnetic Resonance Clinical Trials Committee), and contributed to the mission of the American Heart Association and National Heart, Lung, and Blood Institute as faculty for national workshops, a recipient of research awards, and a (current) NIH study section member.
This year, I have an exciting leadership opportunity working to develop the Heart Rhythm Society Research Network. This network will facilitate networking among researcher leaders nationally and internationally, facilitate effective mentor-mentee relationships, establish an infrastructure for multicenter registries and studies, and generate a data warehouse for the analysis of ECGs and other data specimens for patients with conduction system pacing in the context of appropriate metadata.
The network will also establish data repositories for clinical areas such as left bundle branch area pacing, which should lead to major advances in deep learning and generative AI through the development of these large datasets.
