A bold new finding shows that physical function matters just as much as cardiac metrics for predicting mortality in elderly heart failure patients. Traditional risk models for heart failure days rely mainly on heart-focused clinical data, but they may underestimate danger for older East Asian individuals. In a study led by researchers at Juntendo University in Japan, scientists applied machine learning to a national registry of elderly heart failure patients and built a model that includes physical function measures, improving risk reclassification by about 20% over existing models and potentially guiding future treatment choices.
Managing and predicting outcomes in heart failure is complex at any age. Classic risk tools—like AHEAD (Atrial fibrillation, Hemoglobin, Elderly, Abnormal renal parameters, Diabetes mellitus) and the BIOSTAT compact model—assess survival based on cardiac-related factors such as arrhythmia, anemia, age, diabetes, and ejection fraction. Yet prior work consistently showed these tools underestimate risk in older East Asian populations. Could adding non-cardiac factors sharpen predictions of who is most at risk?
The Juntendo team, led by Professor Tetsuya Takahashi and Assistant Professor Kanji Yamada, with Associate Professor Nobuyuki Kagiyama, sought to answer that. They used data from the nationwide J-Proof HF registry, covering 9,700 elderly patients treated for heart failure and discharged between December 2020 and March 2022 at 96 hospitals across Japan. Using advanced machine learning (an eXtreme Gradient Boosting, or Full XGBoost, approach), they trained models to predict one-year mortality after treatment.
Their first model, Full XGBoost, identified the most important predictors for survival. A second model, Top-20 XGBoost, used the 20 top variables from the first model. Notably, seven of these top factors were related to physical function and other non-cardiac domains. “The prominence of the Barthel Index (BI) and the Short Physical Performance Battery (SPPB) is clinically coherent,” Dr. Yamada explained. “Unlike some scores that rely on subjective activities of daily living, these performance-based assessments offer better reproducibility and directly capture functional limitations.”
Both XGBoost models demonstrated similar accuracy in predicting one-year mortality. Importantly, the Top-20 XGBoost model outperformed established tools like AHEAD and BIOSTAT compact in classifying patients by risk. Because the model was built from a nationwide Japanese cohort, it may offer a more context-specific tool for risk assessment in older Japanese patients with heart failure.
Beyond precision, the Top-20 XGBoost framework can help clinicians allocate resources more efficiently. It can better identify patients who would benefit from closer follow-up or tailored post-discharge care, emphasizing the value of incorporating physical rehabilitation and strategies to preserve or improve physical function as part of long-term heart failure management. The study’s authors also highlighted that discharge physical function is a powerful predictor of survival, comparable to traditional cardiovascular risk factors.
As promising as these results are, the researchers stress that the model needs further validation in Japan and in other countries. In the meantime, they are developing a practical tool based on Top-20 XGBoost, enabling clinicians to input patient details and receive an estimated mortality risk within the next year.
Bottom line: integrating functional and non-cardiac assessments with standard cardiac data can improve mortality risk prediction for elderly heart failure patients. This approach supports more personalized care plans, underlines the importance of physical rehabilitation, and opens the door to a targeted, resource-efficient approach to post-discharge management. Do you think incorporating geriatric and functional evaluations should become a standard part of heart failure care everywhere, or are there meaningful downsides to broader adoption? Share your thoughts in the comments.
