Heart Rate, Movement May Used to Predict VO2 max

A new artificial intelligence (AI) model may make it easier to assess patients’ fitness without the need for expensive in-clinic testing or at-home exercise tracking.

Investigators from the University of Cambridge in Great Britain

"Considering that the best nonexercise benchmark in clinical practice to understand your overall fitness and mobility is the 6-Minute Walk Test, there's plenty of additional information from the everyday lives of individuals our models can incorporate," says Dimitris Spathis, a researcher at the University of Cambridge.

say they have developed a model that can predict a patient’s maximal oxygen consumption (VO2max) based on heart rate and accelerometer data that can be passively collected by a device like a smart watch.

Visiting researcher Dimitris Spathis, Ph.D., explained that VO2max is seen as an important clinical vital sign and proxy for overall fitness. Spathis is also a researcher at Nokia Bell Lbas

“Low VO2max is a stronger predictor of mortality than traditional clinical risk factors such as diabetes, hypertension, or smoking,” he told Managed Healthcare Executive®.

Yet, calculating a patient’s VO2max is costly and requires a costly, rigorous in-clinic exercise test, which in practice means such testing is rarely done, he said.

“These tests require reaching one's maximum heart rate, something not possible for most people,” he said.

Some commercial smartwatches have VO2max tools, but those approaches require users to track their workouts, meaning such readings are not available for people who do not exercise or do not log their exercise. Moreover, the models commercial smartwatches use to predict

VO2max are generally not made public and thus not verifiable by independent researchers.

Spathis and colleagues wondered whether it might be possible to use AI and machine learning to develop a model that could use data from everyday activity (rather than workouts) to accurately predict VO2max.

“There's a well-known, almost-linear relationship between heart rate and VO2max when we exercise, however this association becomes non-linear in lower heart rates (where people spend most of their time),” he said. “As a result, estimating nonexercise VO2max is a much harder problem, but potentially more meaningful.”

The study outlining Spathis and colleague’s model was published in the open-access journal npj Digital Health.

The investigators used data from more than 11,000 people participating in the Fenland Study, a study designed to better understand how genetic and environmental factors affect the risk of common diseases. As part of the study, participants wore sensors that collected a variety of data points. A subset of patients (2,675) were tested again sevven years after the initial study. Pathos and colleagues then used an external validation cohort of 181 patients to ensure their model worked. In the end, the authors found their model had 82% agreement with the baseline VO2max predictions and 72% agreement in follow-up testing.

Spathis said he believes the model could one day be useful as a standalone assessment of patient health.

“Considering that the best nonexercise benchmark in clinical practice to understand your overall fitness and mobility is the 6-Minute Walk Test, there's plenty of additional information from the everyday lives of individuals our models can incorporate,” he said.

He noted that VO2max does not change abruptly, so it would be a meaningful way to see how patient fitness changes over time. However, he added that other biomarkers are also worthy of evaluation, such as heart-rate variability,resting heart rate, and deep sleep.

“By tracking a range of health metrics, individuals can get a more complete picture of their health and can make more informed decisions about their health and well-being,” he said.

Investigating those data will require access to the data generated by fitness trackers and smartwatches. Spathis said most trials have a standardized process of obtaining patient consent to use their data. That has led to the ability to publish studies of significant size, though Spathis said the largest datasets are largely walled off.

“However, the manufacturers of these wearables such as Fitbit/Samsung usually own the largest datasets of this type (hundreds of millions of users) which academics cannot easily access,” he said.