AI Predicts Heart Failure Progression: A Year in Advance

Researchers at MIT, Mass General Brigham, and Harvard Medical School have developed a deep-learning model that can predict which heart failure patients are at risk of their condition worsening within a year. The model, called PULSE-HF (Predict changes in left ventricULar Systolic function from ECGs of patients who have Heart Failure), represents a significant advance in cardiac care by enabling clinicians to forecast disease progression rather than simply detecting current problems.

The development of PULSE-HF addresses a critical need in heart failure management. Heart failure affects millions worldwide and remains one of the leading causes of morbidity and mortality, with about half of diagnosed patients dying within five years. The condition, characterized by weakened or damaged heart musculature, leads to fluid buildup throughout the body and often results in arrhythmias or sudden cardiac arrest.

Traditionally, heart failure management has relied on a combination of lifestyle changes, medications, and sometimes pacemakers. However, predicting which patients will experience worsening symptoms has remained challenging. This is where PULSE-HF offers a transformative approach.

How PULSE-HF Works

The model analyzes electrocardiogram (ECG) data to predict changes in left ventricular ejection fraction (LVEF), which measures the percentage of blood pumped out of the heart's left ventricle with each beat. A healthy heart pumps out 50-70% of blood per beat, while anything less indicates potential problems.

"The model takes an ECG and outputs a prediction of whether or not there will be an ejection fraction within the next year that falls below 40 percent," explains Tiffany Yau, an MIT PhD student and co-first author of the PULSE-HF paper. "That is the most severe subgroup of heart failure."

If PULSE-HF predicts that a patient's ejection fraction is likely to worsen within a year, clinicians can prioritize that patient for follow-up care. Conversely, lower-risk patients might reduce hospital visits and the time spent on ECG testing.

Technical Performance and Advantages

During testing across three patient cohorts from Massachusetts General Hospital, Brigham and Women's Hospital, and the MIMIC-IV public dataset, PULSE-HF achieved area under the receiver operating characteristic curve (AUROC) scores between 0.87 and 0.91. This performance level indicates strong predictive capability, as AUROC scores range from 0 to 1, with 0.5 representing random chance and 1 representing perfect prediction.

Notably, the researchers also developed a single-lead ECG version of PULSE-HF. While 12-lead ECGs are typically considered more comprehensive and accurate, the single-lead version performed just as strongly as the 12-lead version. This finding is particularly significant because single-lead ECGs require only one electrode placement, making the technology more accessible for use in rural areas and other low-resource clinical settings that may not have cardiac sonographers available.

The Development Journey

Creating PULSE-HF required years of work and multiple iterations. One of the biggest challenges was collecting, processing, and cleaning the ECG and echocardiogram datasets. The researchers faced several obstacles:

• Echocardiogram files typically come in PDF format, which becomes difficult for machine learning models to read when converted to text files due to formatting issues
• Real-world data collection is complicated by factors like restless patients or loose leads, creating signal artifacts that need cleaning
• The unpredictable nature of clinical scenarios means data is often messy and inconsistent

The team had to make practical decisions about data quality. "At what point do you stop?" Yau asks. "You have to think about the use case — is it easiest to have this model that works on data that is slightly messy? Because it probably will be."

Clinical Impact and Future Directions

PULSE-HF represents a shift from detection to forecasting in cardiac care. "The biggest thing that distinguishes [PULSE-HF] from other heart failure ECG methods is instead of detection, it does forecasting," Yau notes. The paper emphasizes that no other methods currently exist for predicting future LVEF decline among heart failure patients.

The model's ability to predict deterioration up to a year in advance could transform how clinicians allocate resources and prioritize patient care. By identifying high-risk patients early, healthcare providers can intervene sooner, potentially preventing hospitalizations and improving outcomes.

Looking ahead, the researchers anticipate that the next step will be testing PULSE-HF in prospective studies with real patients whose future ejection fraction is unknown. This would validate the model's effectiveness in actual clinical settings.

The Human Element

For the researchers, the work carries personal significance. Yau joined Stultz's lab after a health event made her realize the importance of machine learning in healthcare. "There's too much suffering in the world," she says. "Anything that tries to ease suffering is something that I would consider a valuable use of my time."

Bergamaschi reflects on the rewarding nature of challenging research: "I think things are rewarding partially because they're challenging. A friend said to me, 'If you think you will find your calling after graduation, if your calling is truly calling, it will be there in the one additional year it takes you to graduate.'"

Broader Context

The development of PULSE-HF comes amid growing interest in AI applications for healthcare. While doctors, computer scientists, and policymakers remain cautiously optimistic about AI's potential in medicine, tools like PULSE-HF demonstrate how machine learning can address specific clinical needs. The model's success in predicting heart failure progression could pave the way for similar forecasting tools in other chronic conditions.

As healthcare systems worldwide grapple with resource constraints and aging populations, AI tools that can predict disease progression and help prioritize care may become increasingly valuable. PULSE-HF represents an important step toward more proactive, personalized cardiac care that could ultimately reduce suffering and improve outcomes for heart failure patients.