Wearable devices could transform the monitoring of heart failure and help to identify patients at greater risk of deterioration, before unscheduled healthcare assistance is required.
This is what was suggested in a study conducted by a group of researchers at the Ted Rogers Centre for Heart Research and by the Peter Munk Cardiac Centre at the University of Toronto, and published in the journal Nature Medicine (https://www.nature.com/articles/s41591-026-04247-3).
In people with heart failure, the heart is unable to pump enough blood to meet the body's needs. The disease progresses sporadically, with periods of relative stability alternating with sudden relapses that require urgent care or hospital admission.
Early detection of the signs of possible deterioration could facilitate intervention before the situation becomes more serious. One of the main parameters used to assess cardiorespiratory capacity and the prognosis for patients with heart failure is peak oxygen uptake (pVO₂), normally measured using the cardiopulmonary exercise test (CPET). However, this provides an instant snapshot at a specific moment rather than enabling the monitoring of the disease's development during everyday life.
The researchers investigated whether the data collected each day by a smartwatch, on the patient's various activities and physiological parameters, could be used to estimate pVO₂ over time and so identify any signs of deterioration.
This involved the development of an AI model known as TRUE-HF (Ted Rogers Understanding Exacerbations of Heart Failure). Based on deep learning algorithms, the model enables an estimation of pVO₂ levels, starting with the data collected by the smartwatch.
During the first stage of the study, the researchers monitored 154 patients with heart failure over about three months, using an Apple Watch to collect various data on activities and physiological parameters.
The pVO2 estimates obtained were consistent with the values measured using CPETs.
The most significant result was the change in pVO2 over time and its link with the risk of deterioration: a 10% drop was associated with an approximately 3.6-times greater risk of an unexpected healthcare event. It was also calculated that this early warning sign occurred about 7 days before the subsequent need for healthcare assistance.
To understand if the method would also work on other demographics and with a different wearable device, the researchers tested the model on a second group, consisting of 63 patients, using data from the All of Us Research Program collected using Fitbit devices, fitted with fewer sensors.
In this group too, a reduction in pVO₂ identified by the smartwatch was associated with the greater probability of a subsequent need for unscheduled healthcare assistance. The early warning came on average about 21 days before the event.
The study underlined that the ability to make predictions thus depends on the time and duration of the data collection, rather than on the type of device used.
According to the authors of the study, the results suggest that continuous monitoring via commonly used wearable devices, such as smartwatches, could provide a new way of tracking, over time, the cardiorespiratory capacity of people with heart failure and of identifying the changes associated with any subsequent deterioration. The same principle has already been successfully applied in other clinical settings, such as foetal monitoring during pregnancy.
It should in any case be clarified that the study's observational design means it's not possible to draw definitive conclusions; further research is required to establish whether the measurements of physiological parameters made using smartwatches can be used in clinical decision-making.