AI Tool Predicts Mortality in Elderly Heart Failure Patients with Unprecedented Accuracy

Researchers at Juntendo University have built a new machine learning tool to predict one-year mortality in elderly heart failure patients. The model uses advanced algorithms to assess risk more accurately than existing methods. It could help doctors identify high-risk patients and adjust treatment plans sooner.

The team developed two versions of the model: a full XGBoost model and a simplified Top-20 XGBoost model. The streamlined version, based on the 20 most influential variables, performed just as well as the full model. Both outperformed established prognostic tools like AHEAD and BIOSTAT compact.

The data came from the J-Proof HF registry, which includes records from 9,700 patients treated at 96 institutions across Japan between December 2020 and March 2022. The model's algorithm, eXtreme Gradient Boosting (XGBoost), excels at analysing complex interactions between multiple factors. Rigorous testing within this large, prospective cohort confirmed its reliability.

Unlike traditional models that focus mainly on cardiac factors, this one also weighs physical function heavily. Seven of the top 20 predictors were linked to non-cardiac elements, such as mobility and daily activity levels. Measures like the Barthel Index and Short Physical Performance Battery were included alongside standard heart-related metrics. Physical function at hospital discharge proved just as critical for survival prediction as classic cardiovascular risks—and, unlike some other factors, it can be improved with targeted interventions.

The research team is now refining the model into a user-friendly clinical tool. This would allow real-time risk assessments, giving healthcare providers immediate insights to guide patient care.

The new model offers a more precise way to forecast mortality in elderly heart failure patients. By combining cardiac and physical function data, it highlights areas where early intervention could make a difference. Doctors may soon have a practical tool to personalise treatment and improve outcomes for this vulnerable group.