Artificial Intelligence Model Accurately Predicts Heart Failure Deterioration Within One Year

Heart failure is a chronic disease that affects more than 6 million people in the United States alone, and its prevalence is rising worldwide. The condition occurs when the heart muscle becomes too weak or stiff to pump blood efficiently, leading to fluid accumulation in the lungs, legs, and abdomen. Despite advances in medication, device therapy, and lifestyle management, the prognosis for many patients remains grim. Roughly 50 % of people diagnosed with heart failure do not survive beyond five years, and readmission rates are alarmingly high. These statistics highlight the urgent need for tools that can predict which patients are most likely to experience a worsening of their condition, enabling clinicians to intervene early and allocate resources more effectively.

Introducing PULSE‑HF: A Deep Learning Tool for Predicting Heart Failure Progression

In a recent breakthrough, researchers from MIT, Mass General Brigham, and Harvard Medical School have developed a deep‑learning model called PULSE‑HF (Predict changes in left ventricULar Systolic function from ECGs of patients who have Heart Failure). The model uses electrocardiograms (ECGs)—a routine, non‑invasive test—to forecast whether a patient’s heart failure will deteriorate within the next year after hospitalization. The study, led by Professor Collin Stultz and PhD student Teya Bergamaschi, demonstrates that PULSE‑HF can predict heart failure worsening with unprecedented accuracy, outperforming traditional clinical risk scores.

How PULSE‑HF Works and Why It Matters

At its core, PULSE‑HF is a convolutional neural network trained on thousands of ECG recordings from heart‑failure patients. The algorithm learns subtle patterns in the electrical activity of the heart that are invisible to the human eye but correlate strongly with future changes in left ventricular function. By feeding the model with a patient’s ECG and a handful of clinical variables—such as age, sex, and medication use—the system outputs a probability that the patient’s heart failure will worsen within 12 months.

Traditional risk scores rely on a limited set of clinical parameters and often fail to capture the complex, dynamic nature of heart failure. PULSE‑HF, on the other hand, integrates the full waveform of the ECG, allowing it to detect micro‑abnormalities that precede clinical deterioration. In validation studies, the model achieved an area under the receiver operating characteristic curve (AUC) of 0.87, compared with 0.68 for the most widely used clinical score. This level of predictive performance could transform how clinicians triage patients, prioritize follow‑up visits, and decide when to intensify therapy.

Clinical Implications and Potential Benefits

By accurately identifying patients at high risk of rapid decline, PULSE‑HF offers several tangible benefits:

• Personalized Care Plans—High‑risk patients can receive more aggressive medication adjustments, earlier device implantation, or closer monitoring.
• Reduced Hospital Readmissions—Targeted interventions may prevent the costly and harmful readmissions that plague heart‑failure management.
• Efficient Resource Allocation—Healthcare systems can focus limited resources on those most likely to benefit, improving overall outcomes.
• Patient Empowerment—Patients gain a clearer understanding of their prognosis, enabling them to make informed lifestyle choices.

Moreover, because ECGs are inexpensive and widely available, the model can be integrated into routine clinical workflows without significant additional cost. The researchers are currently working on a user‑friendly interface that will allow cardiologists to input an ECG and receive an instant risk score, facilitating real‑time decision making.

Future Directions and Ongoing Research

While the initial results are promising, the team acknowledges that further work is needed before PULSE‑HF can be deployed in everyday practice. Key next steps include:

1. External validation in diverse populations across multiple countries to ensure generalizability.
2. Integration with electronic health records (EHRs) to automatically pull ECG data and clinical variables.
3. Prospective clinical trials to assess whether using the model actually improves patient outcomes