AI Helps Predict Which Heart-Failure Patients Will Worsen Within a Year

Heart failure remains one of the leading causes of illness and death worldwide, with nearly half of patients dying within five years of diagnosis. Despite advances in treatment, predicting how a patient’s condition will evolve remains a major challenge. Researchers have now developed an artificial intelligence (AI) model that can forecast worsening heart function up to a year in advance using routine ECG data.

The deep learning model called PULSE-HF, developed by researchers from MIT (Cambridge, MA, USA), Mass General Brigham (Boston, MA, USA), and Harvard Medical School (Boston, MA, USA), is designed to predict changes in left ventricular function in patients with heart failure. PULSE-HF analyzes electrocardiogram data to predict whether a patient’s left ventricular ejection fraction will fall below 40 percent within the next year. Ejection fraction reflects how effectively the heart pumps blood, with normal values typically ranging from 50 to 70 percent. A decline below 40 percent indicates severe heart failure.

The model was trained and tested using data from three patient cohorts, including Massachusetts General Hospital, Brigham and Women’s Hospital, and the MIMIC-IV database. Across these datasets, the model demonstrated strong predictive performance, achieving AUROC scores between 0.87 and 0.91. Notably, a version of the model using single-lead ECG data performed as well as the standard 12-lead version, suggesting potential for use in simpler and more accessible clinical settings.

Unlike existing ECG-based tools that focus on detecting current disease, PULSE-HF is designed to forecast future deterioration. This predictive capability could help clinicians identify high-risk patients earlier and prioritize them for closer monitoring or intervention. Patients identified as lower risk may be able to reduce the frequency of hospital visits and testing, easing the burden on both patients and healthcare systems. The ability to use single-lead ECG data also opens the possibility of deploying the tool in resource-limited settings where advanced imaging, such as echocardiography, is not readily available.

The researchers plan to evaluate the model in prospective clinical studies involving real patients to confirm its predictive value in routine care. If validated, PULSE-HF could support more personalized management of heart failure by enabling earlier intervention and better allocation of healthcare resources, particularly in settings where access to specialized cardiac imaging is limited.

“The biggest thing that distinguishes [PULSE-HF] from other heart failure ECG methods is instead of detection, it does forecasting,” said Tiffany Yau, an MIT PhD student in Stultz’s lab who is also co-first author of the PULSE-HF paper.

Related Links:
MIT 
Mass General Brigham
Harvard Medical School