Novel Atrial Fibrillation Phenotypes Defined by Functional-Anatomical, Machine-Learned Classifications

Stanford University, Stanford CA

Investigators

Linked publications & trials

Abstract

Abstract Atrial fibrillation (AF) is a pervasive disease which affects over 30 million individuals worldwide, in whom it is associated with morbidity and mortality, yet for which therapeutic outcomes are suboptimal. One major limitation to mechanistic and clinical advances in AF is its taxonomy, which is based on number of days of detected AF rather than increasingly reported functional and personalized mechanisms. I reasoned that a digital and scalable AF taxonomy, based on interactions of anatomic and functional factors and clinical features, may better guide existing therapy and catalyze future mechanistic and therapeutic advances. I set out to create a predictive tool to guide therapy in AF patients using machine learning of rich mechanistic data from a large multicenter registry of patients undergoing ablation. I hypothesized that clinically actionable AF phenotypes can be defined by statistical clustering between electrophysiologic features, anatomic regions and clinical indices, that can be uncovered by physiological and statistical quantification and machine learning. I have two Specific Aims: 1) To construct a multimodal digital atlas of atrial fibrillation which registers functional indices at absolute and relative spatial locations in both atria from a multicenter registry, and make this atlas available as an open-source software resource. This deliverable will uniquely map the probability that specific mechanisms will be relevant to AF in a specific patient of given clinical characteristics. Novel pathophysiological phenotypes will be defined via probabilistic interactions in these individual components. 2) To develop a predictive tool using machine learning to estimate the likelihood that ablation at any site(s) will contribute to success tailored to individual characteristics, by learning clusters of electrophysiologic features, clinical indices, and anatomic regions in a training population and applying it to a validation cohort from a large multicenter registry. This project uses state-of-the-art computational tools and statistical methods that may reconcile divergent AF mechanistic hypotheses to define novel functional AF phenotypes and guide therapy. In the process, I will be mentored by world leading mentors, in an extraordinary training environment to facilitate this development into an independent physician-scientist in bioengineering-heart rhythm medicine.