Health & Wellbeing

Wearable system detects COVID two days before symptoms appear

Wearable system detects COVID two days before symptoms appear
Using data from a wearable, an algorithm could predict 68% of COVID cases two days before symptoms appeared
Using data from a wearable, an algorithm could predict 68% of COVID cases two days before symptoms appeared
View 1 Image
Using data from a wearable, an algorithm could predict 68% of COVID cases two days before symptoms appeared
Using data from a wearable, an algorithm could predict 68% of COVID cases two days before symptoms appeared

What if a fitness tracker sitting on your wrist could detect COVID-19 before you even developed symptoms? An impressive new study claims this is not only possible, but preliminary investigations found infections can be detected nearly 48 hours before symptoms appear.

The new research began in early 2020, soon after the pandemic kicked off. A team of researchers wondered whether data from a wrist-worm health tracker could be leveraged to pick up small changes in a person’s vital signs that precede the onset of COVID-19.

Around 1,000 young participants were recruited from an ongoing observational health study and supplied with a commercially available wrist-worn device known as the Ava bracelet. The device is worn at night and every 10 seconds it measured heart rate, breathing rate, skin temperature, heart rate variability and blood flow. It is generally used as a fertility monitor due to its ability for tracking real-time changes to these five health measures.

Over the course of a year-long study 11% of the cohort came down with a lab-confirmed case of COVID-19. Around half of those COVID-positive subjects had a month of good wearable data preceding their infection, helping the researchers develop an algorithm that can detect small changes across the earliest stages of illness.

Noticeable changes in the days before COVID symptoms appeared were detected across all five measures recorded by the wearable. In particular, changes to heart rate, heart rate variability and wrist skin temperature were the most significant early features of COVID-19, preceding noticeable symptoms.

A novel machine-learning algorithm was trained on 70% of the COVID-positive cohort and then tested on the remaining 30%. Remarkably, the algorithm accurately caught 68% of the positive COVID cases two whole days before any symptoms appeared.

“Wearable sensor technology is an easy-to-use, low-cost method for enabling individuals to track their health and well-being during a pandemic,” the researchers concluded in the new study. “Our research shows how these devices, partnered with artificial intelligence, can push the boundaries of personalized medicine and detect illnesses prior to SO [symptom onset], potentially reducing virus transmission in communities.”

The idea that health wearables could detect infectious diseases before any tangible symptoms appear is not new. A fascinating study published last year tested out the idea on influenza and the common cold.

That research actually infected several dozen young volunteers with either rhinovirus or H1N1 and then tracked several health measures using a fitness tracker over the following days. Not only did the study establish that infections could be predicted using wearable health data around 24 hours before symptoms appeared but the severity of the subsequent infections could also be predicted with around 90% accuracy.

David Conen, an author on this new COVID-detection study, said the potential for detecting infections by combining advanced algorithms with real-time health data from wearables is promising. His team is now conducting a larger study testing the COVID-detection system in 20,000 subjects. Results from that investigation are expected later this year.

"That an existing medical device is able to be used in a different meaning [shows] that wearables have a promising future," said Conen. "This is not related only to COVID, in future diseases, it could also lead to preventative treatments and prevent significant complications."

The new study was published in the journal BMJ Open.

Source: McMaster University

If you do the arithmetic on this study, it was apparently conducted with about 50-60 subjects, and no discussion of false positives (which would be people told they probably have Covid when they don't). Not so much dinging the study as pointing out how hard it is to do research with enough subjects to get strong statistical results.
Treon Verdery
This is really encouraging as 48 hours gives time to do some sort of treatment before symptoms, similar to this would be a fitness tracker able to predict a cardiac event 68 of the time, with the 48 hours being used to take drugs like the dadle peptide that reduce harm from cardiovascular events. Another thing to predict is pneumonia
I got my first smartwatch in mid 2018. Wasn't going to keep it, play with it and send it back, but not having to grab my phone every couple minutes for a text, email was worth it. Then playing around with it, noticed it tracked sleep. So I started tracking sleep & HR (at the time that was all it did).
I started noticing that a couple time during the night, WHILE SLEEPING my heart rate would pop up into the 90 beats per minute range. Didn't think too much of it, but on my next yearly medical checkup, showed my doctor and he said do you snore. I said I live alone, but my relatives said I did. I had a mild case of sleep apnea. I've been on a CPAP ever since, and now my HR stays in the low-mid 50 range at night.
Amazing technology. Now it tracks HR, Blood O2, BP & ECG. Yeah, I know it's not as accurate as what the doctors use, but, if something looks fishy, good idea to have your doctor check up on it sooner, than later.
This would be good to give to health workers so that you could get them on the appropriate drugs early and hopefully back to work soonest with minimal side effects.
The basic device (Ava Fertility Tracker 2.0) costs about US$300 but may not be approved or available in your country, and the Covid-19 upgrade probably isn't approved yet.
Thanks for the synopsis Rich.
I don't know if I'll start analyzing patient data today given the data in the article. Yes, there are clear correlations, but by the time I have the data to analyze, I'll have a patient presenting me with direct presentation of illness. For my own purposes, if I were to wear a smart watch, it might give me an indication of why "I feel off" if indeed I would feel an illness coming on. I'm not betting on AI analysis for anything other than a secondary set of data points to confirm a diagnosis. I like the formulas developed so far for A-fib (and am aware of how APPLE blocked the APP after it duplicated it), and would bet my children and younger patients would like this data collection in the future. Personally, knowing two days early for SARS-CoV-2 would give a big headstart on antiviral usage, same during the influenza season. I can see benefit,