Current models for predicting mortality risk in heart failure (HF) patients primarily use cardiac-specific clinical variables. These models may underestimate risk in elderly East Asian patients.
Researchers in Japan have utilized machine learning to analyze data from a nationwide registry of elderly HF patients, developing a new model that incorporates metrics of physical function.
This new model improved risk reclassification by approximately 20% compared to existing models and holds potential for enhancing future treatment options for patients.
Background on Existing Models
Monitoring and treating heart failure is a complex condition. Models such as Atrial fibrillation, Hemoglobin, Elderly, Abnormal renal parameters, Diabetes mellitus (AHEAD) and BIOlogy Study to TAilored Treatment in Chronic Heart Failure (BIOSTAT) compact predict patient survival likelihood based on clinical factors.
However, previous studies indicate these tools, developed for European and North American populations, consistently underestimate risk in older East Asian patients. These models predominantly rely on cardiac-specific and biomedical variables.
This reliance potentially underemphasizes non-cardiac factors like physical function, frailty, and nutritional status. These non-cardiac factors are important for prognosis in older adults and may represent modifiable targets through rehabilitation and supportive care.
Development of the New Model
Researchers from Juntendo University, including Professor Tetsuya Takahashi, Assistant Professor Kanji Yamada, and Associate Professor Nobuyuki Kagiyama, developed an improved model to predict long-term survival after HF. Their findings were published on February 3, 2026, in Volume 67 of The Lancet Regional Health – Western Pacific.
The team utilized machine learning algorithms and data from the J-Proof HF registry, which tracks elderly HF patients across 96 institutions in Japan. Data from 9,700 patients treated between December 2020 and March 2022 and discharged from the hospital were used to train an eXtreme Gradient Boosting (Full XGBoost) algorithm. This algorithm was designed to predict the risk of mortality within one year of treatment.
Model Components and Findings
In addition to the Full XGBoost model, the team developed a Top-20 XGBoost model using the 20 most significant variables identified by the first model. Seven of these 20 variables were related to physical function and other non-cardiac factors, such as the Barthel Index (BI) and Short Physical Performance Battery (SPPB). Performance-based assessments like BI and SPPB offer greater reproducibility and capture functional limitations more directly compared to subjective assessments.
Both XGBoost models demonstrated similar accuracy in predicting the one-year risk of death. The Top-20 XGBoost model classified patients according to their risk of death more effectively than the AHEAD and BIOSTAT compact models. As the model was developed using a nationwide Japanese patient cohort, it may provide a more context-specific tool for risk assessment in older patients with HF in Japan.
Implications and Future Outlook
This model could enable healthcare professionals to identify patients who may benefit from closer monitoring or more tailored post-discharge care. It represents a move beyond a "one-size-fits-all" treatment approach for elderly HF patients and could contribute to more efficient use of medical resources.
The prominence of physical function metrics in this model highlights the role of physical rehabilitation in long-term heart failure management. It also underscores the potential value of maintaining physical function before and after hospitalization.
Findings suggest that physical function at discharge is a significant determinant of survival, comparable to traditional cardiovascular risk factors. The study emphasizes integrating comprehensive geriatric and functional assessments into the routine management and risk stratification of older patients with HF. The team acknowledges that further refinement and testing are required, both in Japan and other countries. A tool based on Top-20 XGBoost is currently under development to assist physicians and healthcare professionals in estimating patient mortality risk.