Machine learning can identify older heart failure patients at risk of losing independence

For older adults with heart failure (HF), leaving the hospital alive does not always mean they can return to living independently. Losing independence can lower quality of life, place more demands on caregivers, and increase the need for care. However, most methods for predicting patient outcomes focus on death or returning to the hospital. This makes it harder to identify patients who may survive but lose their ability to live independently.

To address this problem, a research team led by Assistant Professor Kanji Yamada from the Department of Physical Therapy, Faculty of Health Science, Juntendo University, Japan, studied 6,382 patients aged 65 or older who were hospitalized for HF, received physical therapy, and could carry out daily activities independently before admission. The team also included Associate Professor Nobuyuki Kagiyama and Specially-Appointed Professor Tetsuya Takahashi, both from Juntendo University. Using data from the prospective J-Proof HF Registry, the researchers developed an eXtreme Gradient Boosting (XGBoost) machine learning model to predict whether patients would maintain their physical abilities, experience a decline in their abilities, or die within 1 year after leaving the hospital. The study was published online in the Journal of Cardiac Failure on August 20, 2026.

The team considered 77 possible predictors. These included tests carried out by physical therapists at discharge, such as maximum walking speed, the Short Physical Performance Battery (SPPB), handgrip strength, and the Barthel Index (BI). They also included frailty before hospital admission, measured using the Kihon Checklist (KCL). The researchers tested a simpler model using the 10 most important predictors. They assessed its performance using a method called nested leave-one-site-out validation and compared it with scores based on the KCL.

One year after discharge, 2,076 patients had lost some functional ability, and 928 had died. The other 3,378 remained independent. The 10-predictor XGBoost model performed significantly better than both the modified KCL score and the total KCL score. Walking speed, the KCL score before admission, the discharge BI score, and the SPPB score were among the most important predictors. These measures were more useful for prediction than heart-related measures such as ejection fraction.

Being able to live independently is just as important to many patients as how long they live. Our findings show that physical function at discharge can reveal a risk that models focused on death may miss. These models can help identify patients who are likely to survive but lose their independence.

Dr. Kanji Yamada, Assistant Professor, Department of Physical Therapy, Faculty of Health Science, Juntendo University

The model identified four groups of patients based on their predicted risks of losing functional ability and dying. One important group included 1,732 patients who had a low risk of death but a high risk of functional decline. In this group, 57% experienced a decline in functional ability, even though the predicted chance of surviving for 1 year was 89%. Identifying these patients could help healthcare professionals plan support after discharge, including more intensive rehabilitation at outpatient clinics or in patients' homes.

Instead of asking only whether a patient will survive, we need to ask whether they will be able to maintain the life they value after discharge, Dr. Yamada says. A practical risk estimate could help doctors and patients discuss rehabilitation needs earlier and use limited resources more effectively.

To help doctors use the model in practice, the researchers developed a prototype web application. It estimates each patient's risk at discharge and shows how changes in physical ability, such as walking faster, might affect the predicted risk. The application is still a prototype. Further studies are needed to find out whether decisions guided by the model lead to better patient outcomes.

Overall, the study shows that machine learning may help doctors identify the risk of losing physical ability in older adults with HF, alongside the risk of death. By combining clinical information with direct tests of physical performance, this approach could help doctors plan rehabilitation for patients who need it most and make discharge plans that reflect patients' needs and goals. The aim is to help patients maintain their independence and quality of life.