An explainable artificial intelligence approach for predicting cardiovascular outcomes using electronic health records

Abstract

Understanding the conditionally-dependent clinical variables that drive cardiovascular health outcomes is a major challenge for precision medicine. Here, we deploy a recently developed massively scalable comorbidity discovery method called Poisson Binomial based Comorbidity discovery (PBC), to analyze Electronic Health Records (EHRs) from the University of Utah and Primary Children’s Hospital (over 1.6 million patients and 77 million visits) for comorbid diagnoses, procedures, and medications. Using explainable Artificial Intelligence (AI) methodologies, we then tease apart the intertwined, conditionally-dependent impacts of comorbid conditions and demography upon cardiovascular health, focusing on the key areas of heart transplant, sinoatrial node dysfunction and various forms of congenital heart disease. The resulting multimorbidity networks make possible wide-ranging explorations of the comorbid and demographic landscapes surrounding these cardiovascular outcomes, and can be distributed as web-based tools for further community-based outcomes research. The ability to transform enormous collections of EHRs into compact, portable tools devoid of Protected Health Information solves many of the legal, technological, and data-scientific challenges associated with large-scale EHR analyses.

American Heart Association's Children's Strategically Focused Research Network Experience

“…Finally, collaborative efforts within members of the Utah SFRN program were instrumental in achieving the goals of the Basic Science project, culminating in published work describing integrated data‐science approaches to predicting the risk of CHD and CHD outcomes. 16 , 28 The Utah Center developed a novel and rigorous approach to risk‐factor identification and prediction, leveraging state‐of‐the‐art computational workflows to compute on EHRs at scale. 16 These data serve as inputs to an explainable artificial intelligence based platform that captures and quantifies synergistic (nonadditive) effects between variables that affect the outcome under study.…”

“… 16 These data serve as inputs to an explainable artificial intelligence based platform that captures and quantifies synergistic (nonadditive) effects between variables that affect the outcome under study. 28 The ability to capture synergistic relationships between variables is a major advantage and innovation for risk prediction. Moreover, the ability to transform enormous collections of EHRs into compact, portable tools devoid of protected health information solves many of the legal, technological, and data‐scientific challenges associated with large‐scale EHR analyses.…”

“…Nearly a century after AHA began and half a century after commissioning of Cardiovascular Disease in the Young, the Children's Strategically Focused Research Network (SFRN) was established to apply state‐of‐the‐art basic, clinical, and population science strategies and train the next generation of pediatric cardiology scientists while still embodying Cardiovascular Disease in the Young's original focus on RHD, CHD, and preventive cardiology. The Children's SFRN directly and proudly builds on work done by the early pioneers in of pediatric cardiovascular science within AHA in the 1940s and 1950s.This article highlights the hypotheses, goals, accomplishments, challenges, and future directions of each center; provides an overview of collaboration efforts (Table 1 ) 1 , 2 , 3 , 4 and trainee accomplishments (Table 2 ) 1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 ; and summarizes how we can leverage the output of this SFRN to advance pediatric cardiovascular science.…”

An explainable AI approach for discovering social determinants of health and risk interactions for stroke in patients with atrial fibrillation

“…Leveraging modern, high-powered computational workflows, we selected features for analysis with a novel Poisson binomial-based approach to comorbidity discovery (PBC). 3, 4 We then explored the association between clinical and social determinants of health variables and stroke in patients with AF by implementing a Probabilistic Graphical Model (PGM) AI technique. We use the multimorbidity network derived from this technique to understand the landscape of stroke risk and social determinants of health in patients with AF.…”

Genetic, demographic and clinical variables act synergistically to impact neurodevelopmental outcomes in children with single ventricle heart disease

“…As a result, the relative contribution of genomic variation to neurodevelopment and growth outcomes remains poorly described, especially in the context of clinical variables and other comorbidities, which can combine with one another in a conditionally dependent manner to create a constellation of influence on any given outcome. 9 Two clinical variables are conditionally dependent if the probability of encountering both variables is not equal to the multiplication of their individual probabilities. For example, sex and bicuspid aortic valve (BAV) are conditionally-dependent, in that the probability of encountering an individual with bicuspid aortic valve is greater for males versus females.…”