Research team led by Weil Institute Data Science Director awarded $500,000 to develop model capable of detecting underdiagnosed heart disease

ANN ARBOR, MI – Researchers at the University of Michigan have received a two-year, $500,000 grant from the American Heart Association (AHA) to develop and test a machine learning (ML) model capable of detecting hypertrophic cardiomyopathy (HCM), a dangerous yet often underdiagnosed cardiac condition.

Dr. Sardar Ansari, a Research Assistant Professor of Emergency Medicine and Director of Data Science at the university’s Max Harry Weil Institute for Critical Care Research and Innovation, will serve as the project lead alongside co-investigators from Precision Health at U-M and the Departments of Internal Medicine and Learning Health Sciences. The grant will be conducted as an ongoing collaboration between the Weil Institute and Precision Health. Training and validation of the prospective model will also involve partner sites at Johns Hopkins University, UT Southwestern and Stanford University.

Getting to the heart of the data security challenge

HCM is a rare condition that causes the walls of heart to thicken, restricting blood flow. The disease is most often caused by a genetic abnormality and can sometimes occur with few or no symptoms, resulting in many patients not being diagnosed early enough for treatment. ML models show great promise for improving the detection of underdiagnosed diseases such as HCM, yet there remains a barrier to utilization, given that the data used to train these models includes protected health information.

“Moving data from hospital to hospital is challenging,” said Dr. Ansari. “When we build these models, we typically do so with data from multiple sites, yet we conduct the training and validation in one place. This can lead to potential issues with data sharing and patient privacy as all of the unique data necessary for training the model will need to be moved into that central hub.”

Compared to more conventional approaches to machine learning, Ansari and the team, in collaboration with AHA, will instead train their model using a method called “federated learning”. This will provide greater data security as it will allow each site in the network to train and validate the model locally on their end without needing to share unique patient data with the central training hub. This method will also allow the team to train their model on more diverse patient populations than would otherwise be accessible if the training were conducted with data from a single site.

"Moving data from hospital to hospital is challenging. When we build these models, we typically do so with data from multiple sites, yet we conduct the training and validation in one place. This can lead to potential issues with data sharing and patient privacy as all of the unique data necessary for training the model will need to be moved into that central hub."

Sardar Ansari, PhD
Research Assistant Professor, Emergency Medicine; Director, Data Science Team, Weil Institute
University of Michigan

Next steps

The first phase of the project will see the U-M team gathering the data necessary for training the model, which will include electrocardiograms, echocardiograms, and data from the electronic health record. They will also be working with the partner institutions and with AHA to ensure accuracy and consistency in how HCM is labeled and defined across the network. They will also establish a plan for how the model should be validated once training is complete.

The project’s second phase will involve simulated rollouts of the newly developed model at each of the participating sites. Teams will evaluate the model based on how it impacts clinical practices and outcomes and how much additional workload could be incurred if the model were to be formally adopted. They will also analyze the potential feasibility of adopting the model at both local and national levels.

Once trusted models are derived, further studies will be needed before simulation becomes implementation. Dr. Ansari and team are hopeful that their work could serve to show how federated learning models could be applied with the support of organizations such as AHA for the detection of various other disease states across multiple institutions.

Project Team
Sardar Ansari, PhD (Emergency Medicine; Weil Institute); Brahmajee Nallamothu, MD (Internal Medicine, Precision Health at U-M); Adam Helms, MD (Internal Medicine); John Donnelly, MSPH, PhD (Learning Health Sciences) 

About the Weil Institute, formerly MCIRCC
The team at the Max Harry Weil Institute for Critical Care Research and Innovation (formerly the Michigan Center for Integrative Research in Critical Care) is dedicated to pushing the leading edge of research to develop new technologies and novel therapies for the most critically ill and injured patients. Through a unique formula of innovation, integration and entrepreneurship that was first imagined by Weil, their multi-disciplinary teams of health providers, basic scientists, engineers, data scientists, commercialization coaches, donors and industry partners are taking a boundless approach to re-imagining every aspect of critical care medicine. For more information, visit weilinstitute.med.umich.edu.

About Precision Health at U-M
Precision Health at the University of Michigan expands far beyond medical treatment to include discovery, implementation, and community health. Spanning across 19 colleges and schools, Precision Health brings the best and brightest together to 1) develop fundamental social, medical, computational, and engineering science; 2) translate these basic science discoveries into promising treatments that are evaluated in partnership with Michigan Medicine patients and regional health systems; and 3) evaluate and increase the public health impact of effective therapies, working with community health systems, policy makers, and payers to implement these therapies nationally. For more information, visit precisionhealth.umich.edu.