Heart failure weakens the heart muscle, causing fluid to build up in the lungs and other body parts. This condition is chronic and has no cure. It often leads to serious issues like irregular heartbeats or sudden cardiac arrest.
Historically, treatments were quite primitive, involving bloodletting and leeches. Today, managing heart failure includes lifestyle changes, medications, and sometimes pacemakers. Despite these advances, heart failure remains a major cause of illness and death, putting a huge strain on health care systems worldwide.
Predicting Worsening Heart Failure
About half of people diagnosed with heart failure die within five years. Predicting how a patient will do after being hospitalized is crucial for managing limited resources.
We're a new kind of news feed.
Regular news is designed to drain you. We're a non-profit built to restore you. Every story we publish is scored for impact, progress, and hope.
Start Your News DetoxResearchers at MIT, Mass General Brigham, and Harvard Medical School developed a deep learning model called PULSE-HF. This model predicts changes in a patient's left ventricular ejection fraction (LVEF). LVEF measures the percentage of blood pumped out of the heart's left ventricle with each beat. A healthy heart pumps out 50 to 70 percent. Anything less can signal a problem.
PULSE-HF takes an electrocardiogram (ECG) and predicts if a patient's ejection fraction will drop below 40 percent within the next year. This level indicates the most severe form of heart failure.
If the model predicts a patient's condition will worsen, doctors can prioritize them for follow-up care. This also means lower-risk patients might need fewer hospital visits and less time hooked up to ECG machines. The model could also be useful in rural clinics that don't always have cardiac sonographers available for ultrasounds.
Tiffany Yau, an MIT PhD student and co-first author of the PULSE-HF paper, noted that the model's main difference from other methods is its ability to forecast. No other existing methods predict future LVEF decline in heart failure patients.
How PULSE-HF Works
During testing, PULSE-HF achieved high accuracy, with AUROC scores ranging from 0.87 to 0.91 across different patient groups. AUROC measures a model's ability to distinguish between different outcomes, with 0.5 being random and 1 being perfect.
The researchers also created a version of PULSE-HF for single-lead ECGs, which only requires one electrode. This version performed just as well as the 12-lead version, which is typically considered more comprehensive.
Developing PULSE-HF was a multi-year effort. One of the biggest challenges was collecting and cleaning the ECG and echocardiogram data. Training machine learning models requires clear, labeled data. However, echocardiogram files often come as PDFs, which become difficult for the model to read when converted to text. Real-life issues like restless patients or loose leads also created messy data that needed extensive cleaning.
Teya Bergamaschi, another co-first author, described the data cleaning as a "never-ending rabbit hole." While more complex methods could filter the data, the team had to consider the practical use case. They aimed for a model that works well even with slightly messy data, as this is often the reality in clinical settings.
The next step for PULSE-HF is to test it in a prospective study with real patients whose future ejection fraction is unknown.
Despite the challenges, the researchers found the work rewarding. Yau, who joined the lab after a personal health event, believes that anything that eases suffering is a valuable use of her time.
Deep Dive & References
Predicting future decline in left ventricular ejection fraction among patients with heart failure using deep learning on the electrocardiogram - Lancet eClinical Medicine, 2026
