Monitoring and treating heart failure (HF) remains difficult at any age, especially among older adults. Doctors often rely on prediction models to estimate a patient’s chances of survival. Well-known tools such as the AHEAD and BIOSTAT compact models use clinical factors like irregular heart rhythm, anaemia, age, diabetes, kidney function, and ejection fraction. However, research has shown that these models, which were mainly developed using European and North American populations, tend to underestimate risk in older East Asian patients. This raises an important question about whether including additional factors could improve survival predictions.
A research team from Juntendo University in Japan has worked to address this issue by developing a more accurate model for predicting long-term survival after heart failure. Professor Tetsuya Takahashi, Assistant Professor Kanji Yamada, and Associate Professor Nobuyuki Kagiyama led the study. They used machine learning methods to identify which patient characteristics are most important for predicting outcomes. Their findings were published in February 2026 in The Lancet Regional Health – Western Pacific.
One of the key limitations of existing models is that they mainly focus on heart-related and biological measurements. According to Dr Yamada, this approach often overlooks other important factors such as physical function, frailty, and nutritional status. These factors are especially important in older adults because they strongly influence recovery and survival. Unlike age or other fixed characteristics, physical function and nutrition can potentially be improved through rehabilitation and supportive care, making them valuable targets for treatment.
To build a better model, the researchers used data from the J-Proof HF registry, which tracks elderly heart failure patients across 96 hospitals in Japan. They analysed data from 9,700 patients who were treated between December 2020 and March 2022 and later discharged from the hospital. Using this data, they trained a machine learning algorithm called XGBoost to predict the risk of death within one year after treatment. They then created a simpler version of the model using the 20 most important variables identified by the algorithm.
Interestingly, seven of these top 20 variables were related to physical function and other non-cardiac factors. Measures such as the Barthel Index (BI) and Short Physical Performance Battery (SPPB) were especially important. These are performance-based tests that assess a patient’s ability to carry out daily activities and physical tasks. Compared to more subjective assessments, these tests provide more consistent and reliable results and better reflect a patient’s true functional ability.
Both versions of the model performed well in predicting one-year mortality, but the simplified Top-20 model was particularly effective at classifying patients into different risk levels. It also outperformed traditional models like AHEAD and BIOSTAT compact. Because it was developed using data from Japanese patients, it may be better suited for assessing risk in older individuals with heart failure in Japan and similar populations.
This new model could help healthcare professionals move away from a one-size-fits-all approach to treating heart failure. Instead, doctors could use it to identify patients who need closer monitoring or more personalised care after leaving the hospital. It could also help make better use of medical resources. Importantly, the strong role of physical function in the model highlights the value of rehabilitation and maintaining physical ability both before and after hospitalisation.
The researchers emphasise that physical function at discharge is just as important as traditional heart-related risk factors in predicting survival. Their findings support the idea that routine care for older heart failure patients should include broader assessments that consider overall physical and functional health. While the team is optimistic about the model’s potential, they note that further testing in different populations is needed. They are also working on developing a practical tool that clinicians can use to estimate a patient’s risk more easily.
More information: Kanji Yamada et al, Machine learning prediction of 1-year mortality in older patients with heart failure: a nationwide, multicenter, prospective cohort study, The Lancet Regional Health – Western Pacific. DOI: 10.1016/j.lanwpc.2026.101808
Journal information: The Lancet Regional Health – Western Pacific Provided by Juntendo University Research Promotion Center
