Making big data work for better heart health

May 31, 2018

Making big data work for better heart health

Institute for Healthcare Policy & Innovation

Two researchers are developing data-driven tools to improve the detection of serious heart conditions

 

Heart failure and atrial fibrillation (AF) are two common chronic conditions that can be managed if recognized early, but can have deadly consequences if they go undetected.

Unfortunately, many thousands of cases of heart failure and AF (a type of irregular heartbeat) are discovered too late because there are few simple and reliable measures to diagnose them. These are complex conditions whose clinical characteristics vary from day to day, and can show up very differently from person to person – or sometimes not at all. The constellation of symptoms associated with each condition are generally evaluated during a single point in time, such as a clinic visit, which can overlook indicators of the daily functioning of an individual’s heart at the moments when irregularities might make themselves apparent.

What if the most useful information about heart function could be seized from the largely untapped volumes of data already generated through a range of healthcare encounters, mobile devices, and what patients themselves report, and synthesized together to provide vastly improved new diagnostic tools?

Two U-M projects, supported by NIH career development awards, will use data science and analytic methods to attempt just that, with the goal of improving the ability to detect these conditions earlier, more reliably, and less expensively.

Michael Mathis, M.D., is a cardiothoracic anesthesiologist at Michigan Medicine who studies how to improve how people fare during the time surrounding a major surgery. He does this in part by integrating large databases across different medical specialties, including anesthesiology, cardiology, general surgery, and cardiac surgery. Mathis recognized that the period around a surgical procedure represents an opportunity to harness various existing but disconnected sources of electronic health data to better characterize a person’s cardiovascular function.

Mathis

Before surgery, patients undergo extensive workups to ensure their bodies can withstand the stresses imposed by the procedure. During surgery, patients’ responses to such stressors – including incisions, and the delivery of anesthesia – are carefully and continuously monitored, yielding highly detailed information comparable in quality to a routine treadmill stress test. Put together, these data could offer physicians an improved means to diagnose heart failure.

“This project will rely on robust data science methods to intelligently analyze these underutilized data, with the aim of improving our ability to efficiently and cost-effectively detect heart failure during early stages – a time at which intervention can dramatically change the long-term outcomes for these patients and avoid costly complications,” Mathis says.

These data science techniques have previously been successful in extracting features from the medical record to predict other diseases, such as coronary artery disease, but this will be among the first attempts to utilize these kinds of data to diagnose heart failure, Mathis says.

Hamid Ghanbari, M.D., M.P.H., is a cardiovascular electrophysiologist at Michigan Medicine who treats patients with arrhythmias such as AF. He also studies the symptoms of AF, as assessed in the clinical environment, and their relationship between information about daily functioning as described by individuals with AF and measured by wearable technology.

Ghanbari

Ghanbari notes that the short window that a physician has with a patient offers just a momentary snapshot of an individual’s symptoms, heart rhythm, and how those factors may affect daily quality of life in someone with AF – and, remarkably, some people with AF do not experience symptoms at all.

Using continuous heart rhythm recording, accelerometer data using wearable sensors, and a mobile application that allows individuals to report symptoms and their affect and well-being, Ghanbari’s project will allow for real-time assessment of symptoms and discover novel physiological and psychological correlates of disease severity.

Ghanbari’s project will integrate computational bioinformatics expertise, signal processing, machine learning, and a method known as “ecological momentary assessment” commonly used in social sciences to develop a more richly characterized picture of heart rhythm, affect, and functional status in people with AF. This would be the first application of these techniques focused on AF, Ghanbari says.

“One of the aims of this work is to create digital therapeutics for patients with atrial fibrillation, which can bring about meaningful behavior change through monitoring and feedback when used in conjunction with currently available medical and interventional therapies,” Ghanbari says. Ghanbari’s ultimate goal is to develop personalized therapies for AF that are more specifically matched to clinical symptoms and their impact on an individual’s quality of life.

Both Mathis and Ghanbari will utilize the collaborative expertise of the Institute for Healthcare Policy & Innovation and the Michigan Integrated Center for Health Analytics and Medical Prediction (MiCHAMP). Ghanbari is also a member of the Michigan Center for Integrated Analytics in Critical Care (MCIRCC), while Mathis is a member of the Multicenter Perioperative Outcomes Group (MPOG). Both are also members of U-M’s Frankel Cardiovascular Center.