Wearable AI Tool Predicts Hospitalization Risk in Heart Failure

Heart failure, a condition in which the heart cannot pump enough blood to meet the body’s needs, is a leading driver of unplanned hospital use. Clinicians often lack continuous insight into symptom changes that unfold between visits, limiting timely intervention. A new study shows that data from consumer smartwatches can flag early signs of worsening status days to weeks before urgent care is needed. The approach aims to support proactive management and reduce costly admissions.

Researchers at University Health Network (UHN; Toronto, Canada) and the University of Toronto developed and validated an artificial intelligence model that analyzes smartwatch data to estimate daily cardiopulmonary fitness. The system was designed to surface real-time changes that correlate with clinical deterioration in people living with heart failure. The work involved teams affiliated with the Ted Rogers Centre for Heart Research and Transform HF.

The method uses data streams commonly available on consumer wearables, including heart rate, physical activity, and oxygen saturation. Investigators reported that smartwatch-derived fitness estimates closely matched results from formal cardiopulmonary exercise testing performed in hospital at study entry and again at three months. Monitoring shifts in cardiopulmonary fitness over time provided a clinically relevant readout that relates to the likelihood of unplanned medical care.

In the TRUE-HF observational cohort, 217 people with heart failure were followed in free-living conditions for three months. Apple supplied 200 iPhones and Apple Watches, provided feedback on the manuscript, and collaborated with all authors to build the study-specific mobile application, while the research team independently led the study design, model development, analysis, and writing. Participants who experienced a 10% or greater decline in daily cardiopulmonary fitness had more than a three-fold increased risk of unplanned hospitalization or urgent treatment. The model’s predictions were assessed for concordance with cardiopulmonary exercise testing metrics and clinical events including declines in fitness, hospitalization, emergency visits, and intravenous furosemide administration. The findings were published in Nature Medicine.

The study underscores the need for widely available tools that can monitor heart failure outside the clinic and create an earlier window for medication adjustments or other interventions. Further research will evaluate how continuous wearable monitoring can be integrated into routine care pathways.

"The really novel thing about our study is that it captures unobtrusive, free-living data from patients in the real-world,” said Chris McIntosh, senior scientist at University Health Network and assistant professor at the University of Toronto. “We’re not only measuring how fast someone walks down a hallway in the hospital while their clinical team is standing behind them and encouraging them. We’re seeing what happens to their heart rate when they’re walking at the mall, on the street or at home."

“The findings of this study are a potential game-changer because they allow us to identify signals that would tell us a patient was in trouble before they ended up coming to the emergency room,” said Heather Ross, cardiologist at the Peter Munk Cardiac Centre and professor at the University of Toronto’s Temerty Faculty of Medicine.

Related Links
University Health Network
University of Toronto Temerty Faculty of Medicine