For decades, the NHLBI has aggressively studied heart failure – a chronic, debilitating condition that develops when the heart can’t pump enough blood to meet the body’s needs. Heart failure affects more than 6.5 million adults in the United States alone and continues to be a growing public health threat, largely because of an aging population and increasing rates of high blood pressure and diabetes, which are both risk factors for the condition.
Just as troubling, however: Those living with some types of heart failure have few effective treatment options. Over the years, researchers have developed new medicines and identified lifestyle practices that may help stave off the complications that can come with this progressive condition – shortness of breath, extreme tiredness, even sudden death. But the biological changes that ultimately lead to heart failure can occur without any obvious symptoms, making it difficult to diagnose – and even harder to understand.
Now, the study of heart failure may get a welcome boost thanks to a unique, nationwide research competition recently hosted by the NHLBI. Called the Big Data Analysis Challenge: Creating New Paradigms for Heart Failure Research, the contest opened in February 2020 and closed in August 2020.
The goal: to get entrants to use machine learning—a form of artificial intelligence—to gain a better understanding of heart failure, improve classification of different types, and do it with the help of large-scale NHLBI datasets from longstanding observational studies, clinical trials, and other research efforts. In machine learning, computers automatically learn to identify patterns and predict outcomes over time based on input from such datasets.
“We wanted to draw attention to this field and attract new minds and new approaches to the study of heart failure,” said Mollie Minear, Ph.D., a program officer in the NHLBI’s Division of Cardiovascular Sciences and part of the team that helped create the contest. The so-called Heart Failure Challenge is part of an ongoing federal government crowdsourcing effort to engage citizens in prize competitions to help solve scientific problems.
Clinicians currently diagnose heart failure using a variety of methods, including a physical exam, blood tests, and cardiac imaging. One of the most common imaging techniques, echocardiography, uses sound waves to create a graphic representation of a patient’s heart and measures how well it pumps blood when it beats.
Yet “no single test can diagnose heart failure,” noted Renee Wong, Ph.D., chief of the NHLBI’s Heart Failure and Arrhythmias Branch. “Heart failure is a complex, multi-organ syndrome that involves the heart, kidney, lungs, and muscles. We need more measurements and new approaches in order to get better, more targeted therapies.”
The Heart Failure Challenge addresses this need. In all, 12 teams entered the research competition, which offered up to $50,000 to the winners. The judging criteria emphasized novel and innovative ideas, the inclusion of datasets from diverse participants, and the development of an easy-to-use, free, and publicly accessible research tool that improves the classification of different types of adult heart failure.
After two months of rigorous review, a panel of experts at NHLBI named five winning teams.
One of the winning teams was led by Partho Sengupta, M.D., a researcher at the West Virginia University Heart and Vascular Institute and the chief of its Cardiology Division who also directs its cardiac imaging department. His team developed a machine learning tool for early diagnosis of heart failure with preserved ejection fraction, or HFpEF. Ejection fraction is a measure of how well your heart pumps blood and is measured in a percentage, with a healthy heart producing an ejection fraction of 50-55% or higher. As heart failure can occur even with a normal ejection fraction, HFpEF is one of the most difficult forms of heart failure to detect and treat.
The team used datasets from three different NHLBI-funded heart failure clinical trials and showed that their algorithm, which they call a “deep learning classifier,” could more easily and accurately grade the severity of heart dysfunction than current methods. The researchers also showed that their classifier can help predict which heart failure patients from this group are more likely to benefit from certain heart medications. As more patients are screened and their data is entered into the database, the tool becomes better able to predict patient outcomes, the researchers said. Earlier diagnosis means lives saved.
“It is a great honor to receive this recognition,” said Sengupta, who hopes to use the winnings to support future clinical trials on machine learning for heart failure, which he calls “the wave of the future.”
“Traditional statistical models are not good at predicting heart failure due to its complexity,” Sengupta said. “We need to rediscover the breadth of conditions that present under heart failure using the lens of machine learning and hope that it will lead to better understanding and management of heart failure.”
Contest organizers hope that the prizes they award will help advance heart failure research and lead to scientific publications in this field. The competition comes at a critical time: the rate of heart failure-related deaths has increased nationwide since 2013, according to a recent NHLBI-funded study published in the Journal of the American College of Cardiology. The study indicates that this increase is occurring particularly among adults ages 35 to 64 and is disproportionately higher in young black men and women.
“We need to do better at preventing and treating heart failure,” NHLBI’s Wong said. “We believe that this unique contest will help lay the groundwork for future heart failure research in this area.”
Laura Hsu, DrPH, NHLBI’s challenge manager in the Division of Extramural Research Activities, said that she hoped the current competition would have another benefit: “We hope that winners, and even those who don’t win, might come back for other funding or grants,” Hsu said.
People who do not already have heart failure can focus on reducing their disease risk by adopting better lifestyle habits, including exercising more, eating a heart-healthy diet, aiming for a healthy weight, and avoiding tobacco products, Wong noted.
- A team led by Ambarish Pandey, M.D., of the University of Texas Southwestern Medical Center, developed and tested a machine learning tool to better identify a subgroup of patients with acute decompensated heart failure (ADHF), who are unlikely to respond to diuretics, the standard treatment for severe congestion that accompanies this type of heart failure. ADHF involves a sudden worsening of heart failure symptoms and is a leading cause of hospitalization in older persons in the United States.
- A team led by Josephine Lamp, a Ph.D. candidate at the University of Virginia, developed a machine learning tool to characterize patient conditions and predict patient risk of outcomes such as death and rehospitalization for advanced heart failure. The tool uses measures of heart function (called hemodynamics) to make these predictions, and could allow patients to receive life-saving therapies in a timely manner.
- A team led by David Kao, M.D. of the University of Colorado Anschutz Medical School, is developing a tool to characterize clinical outcomes in patients with acute heart failure to identify management strategies to reduce post-discharge complications, including readmission and hospitalization.
- A team led by Rishi J Desai, Ph.D., of Harvard Medical School, is developing novel machine learning-based predictive modeling approaches for grouping people with HFpEF. The goal is to identify HFpEF subgroups that are most likely to derive benefit from certain treatments and develop more targeted treatments for the condition.