The Dynamics of Human Atrial Fibrillation

About This Grant

Project Summary Atrial fibrillation (AF) is a major cause of morbidity, stroke and mortality affecting 2-5 million Americans. Recent trials have revolutionized the field, by showing that treating patients classified early in their AF process with anti-arrhythmic strategies can reduce morbidity and increase survival versus treatment with rate control medications. Unfortunately, current definitions of `early AF' are highly inconsistent, so that many patients may be excluded from rapid therapy while others may be treated who are unlikely to benefit. This proposal focuses on developing novel computational phenotypes for patients with early stage atrial fibrillation. This proposal builds on preliminary data from our last funded cycle, novel tools we have developed and datasets we have curated in thousands of AF patients. Aims 1 and 2 will develop computational phenotypes for various forms of early stage AF, which address actionable clinical treatment with non-interventional or ablation therapy. For this, we will develop state-of-the-art machine learning models that we apply to very large datasets from our medical records. For Aim 2, we will integrate body surface mapping, imaging and rich clinical data. We will test models in independent cohorts to test the generalizability of our results across populations. Aim 3 will test models prospectively in a pilot study. Our studies address academic rigor, specifically including a broad range of patients, and we will study sex as a biological variable. This study will deliver immediate clinical and translational impact, accelerating personalized medicine for patients with AF. Results from the project will enable clinicians to classify patients with early-stage AF that are amenable to therapy, versus forms of AF which are less amenable. The project will provide translational insights which could inform future mechanistic studies. We will develop unique datasets and novel computational tools which we will share with the research community through our github and established Institutional portals. Our multi-disciplinary team comprises experts in electrophysiology, imaging, computer science, artificial intelligence and biostatistics. The proposal is thus highly feasible.