Detecting early signs of SARS-CoV-2 infection using heart rate data

COVID-19 has caused death, disability and inconvenience to millions of Australians since early 2020. Many people might benefit from a convenient, automated way to detect the early signs of SARS-CoV-2 infection, especially when there are no obvious symptoms.

Heart rate is a convenient measure of health status. Unlike state-dependent measures, such as steps taken and sleep quality, passive continuous heart rate measures can be obtained from commercial wearable devices, such as an Apple Watch or Fitbit. For example, resting heart rate measured while people are not active might be useful for population level surveillance of influenza.

Researchers from Stanford University’s Genome Laboratory showed that resting heart rate could be used to detect the early signs of SARS-CoV-2 infection. Of 4642 volunteers in the study who reported wearing a smartwatch, 114 (2.5%) eventually self-reported having COVID-19. Using the smartwatch data for a small subset of 25 participants diagnosed with COVID-19, the early signs of COVID-19 were detected, on average, 4 days before symptoms appeared and 7 days before a formal diagnosis. This technique was 63% successful, with four patients showing delayed recovery, possibly an indication of long COVID — a problem that affects a difficult-to-determine number of patients diagnosed with COVID-19, due to difficulties in data acquisition and diagnosis definitions. In a follow-up investigation with new participants and a different data analysis technique, detection of infection was possible for 80% of 84 people who had COVID-19 at or prior to the onset of symptoms.

Application of a new model to detect the first sign of SARS-CoV-2 infection

Technology that had successfully detected the early signs of a manic relapse in a person with bipolar disorder using wearable heart rate data (published on psyRxiv; not yet peer reviewed) was applied to heart rate difference time series data obtained from the Stanford study. Using this proxy to heart rate variability, a daily health index was estimated with a model that considers heart rate difference fluctuations to be fractal on all time scales, a multifractal property demonstrated for heart rate by Ivanov and colleagues.

In our article published on preprint server medRxiv, which is yet to be peer reviewed, heart rate difference data were obtained from a Fitbit smartwatch worn by patients eventually diagnosed with COVID-19. For 90% of 31 cases, the health index tracked COVID-19 infection with the virus and subsequent recovery. The first sign of COVID-19 was detected on average 9 days before symptoms were reported.

The median presymptomatic infection prediction time was 9 days, sufficient for potential COVID-19 patients to self-isolate while they are still asymptomatic.

Prospects for the future of COVID-19 monitoring

As SARS-CoV-2 infection can occur well before any symptoms are experienced, the use of smartwatch heart rate data to warn of infection before symptoms appear may have benefits in reducing the spread and community impact of COVID-19, including a recommendation that the person self-isolate. The technology described briefly here and in more detail in the associated preprint article might be implemented in a mobile application (app) that anyone can use. However, the warning offered by the app to the user is not specific to COVID-19 infection; it could result from a decline in health from any source, including mental health issues. When SARS-CoV-2 infection has occurred, there are clear declines in the health measure that coincide with early precursors of infection; the maximum decline occurs when symptoms are evident.

Although no comprehensive review studies exist for SARS-CoV-2 infection detection using smartwatches, application of these devices to detect atrial fibrillation indicates specificity and sensitivity exceeding 90%. A recent article analysing the same data presented here has shown a low specificity of around 40% but a reasonable sensitivity of around 85%.

Clearly, more research needs to done before this technology can be used more generally in clinical practice as a source of specific information for the treating team. Despite this reservation, smartwatch data may suggest a decline in health as COVID-19 progresses from an initial diagnosis to the possibility of longer term complications, loosely defined as long COVID. The time course of such complications can be traced using commercial smartwatches that many people already own.

Honorary Associate Professor Rachel Heath is from the School of Psychological Science at the University of Newcastle.

Acknowledgments: I thank the Stanford University Genomics Laboratory for permission to use their public datasets. We adjusted the year in these data files to be 2020.

The statements or opinions expressed in this article reflect the views of the authors and do not necessarily represent the official policy of the AMA, the MJA or InSight+ unless so stated.

Subscribe to the free InSight+ weekly newsletter here. It is available to all readers, not just registered medical practitioners.

If you would like to submit an article for consideration, send a Word version to mjainsight-editor@ampco.com.au.