Physiology-informed machine learning for precision phenotyping of heart failure with preserved ejection fraction

JI Program: Cardiovascular Medicine

Heart failure (HF) with preserved ejection fraction (HFpEF) is common, increasing in prevalence, and associated with poor quality of life and outcomes. There are currently few effective therapies for HFpEF due to phenotypic heterogeneity. This proposed collaborative research project aims to develop, extend, and apply state-of-the-art computational approaches for precision phenotyping of HFpEF. The overarching goal is to use patient-specific computer modeling and simulation, combined with data-driving machine learning approaches, to classify patients into phenogroups that represent distinct categories of underlying cardiovascular dysfunction. In a retrospective study we will extend and refine analytic technology developed in preliminary studies in an application to several hundred patients drawn from clinical records at UMich and PKU. In addition, in a proposed prospective study we will further extend and refine our technology to assess cardiovascular function data obtained during cardiopulmonary exercise testing, to deepen our understanding of the mechanisms governing exercise intolerance in HFpEF patients, and to determine if and how these mechanisms can be identified from analysis of baseline-state data alone. Finally, using the results from these retrospective and prospective studies, we will design and test an optimal patient classifier that uses solely noninvasively obtained data. With a multi-disciplinary collaborative team that brings a unique combination of clinical expertise and data, experimental tools, computational technology, and novel technology to this leading edge problem, we are uniquely positioned to fundamentally improve the ways in which HFpEF is diagnosed and ultimately treated.