AI and Wearable Devices Offer Promise as an Early COVID-19 ID Method

While use of consumer wearable devices like smart watches and step counters for medical applications had been growing before COVID-19, the pandemic fueled the trend. With public health restrictions requiring people to stay home, healthcare providers suddenly needed ways to monitor their patients’ health remotely. In addition to driving increased utilization, the pandemic has also driven innovation and improvement in wearables technology. A notable example is recent research demonstrating that those technologies, when combined with artificial intelligence (AI), may provide a promising non-invasive method of identifying COVID-19 and other infectious diseases early on.

The Diagnostic Challenge

Though PCR testing remains the gold standard for diagnosing COVID-19 cases, its use is now largely being restricted to only those most at risk, such as healthcare workers. Rapid antigen tests are another option for the general public, but most are invasive, requiring a nasal swab, and don’t always provide accurate results. But what if there were a way to easily monitor large numbers of people and detect COVID-19 before symptoms appear? With COVID-19 cases rising and new variants continually emerging, such a tool would be especially valuable in keeping spread under control.

Since the pandemic’s start in 2020, several studies have explored using wearables to see if the physiological metrics they measure, such as heart rate and temperature, change before someone tests positive for COVID-19, and if that data could be used to train AI algorithms for use as COVID-19 detection tools.

Recent Studies on the AI-Wearables Combo as a COVID-19 Detection Tool

Most recently, a study published in the July 2022 issue of JAMIA Open explored how well heart rate variability measurements along with resting heart rate data collected from Apple Watches worn by 407 healthcare workers could predict COVID-19 infection before diagnosis with PCR testing. The workers were tested for COVID-19 via PCR testing, with 49 people testing positive for the SARS-CoV-2 virus that causes COVID-19. The team, which mostly included researchers from New York City’s Icahn School of Medicine at Mount Sinai, found that “parameters describing the circadian heart rate variability mean (MESOR) and peak-timing (acrophase), and age” best predicted COVID-19-positive cases.

Using this data, they then developed a machine learning algorithm to identify and predict SARS-CoV-2 infection, which boasted an average sensitivity of 82 percent and an average specificity of 77 percent. “While further validation is necessary, this non-invasive and passive modality may be helpful to monitor large numbers of people for possible infection with SARS-CoV-2 and help direct testing toward high-risk individuals,” the researchers state.

In previous research published in the Journal of Medical Internet Research in February 2021, the same group evaluated whether heart rate variability data collected by an Apple Watch and custom app could predict COVID-19 as well as symptoms of the illness. They found that heart rate variability was different in those who tested positive for COVID-19 in both the week before and the week after COVID-19 diagnosis, compared with heart rate variability during uninfected periods, showing that the metric could potentially be used to identify COVID-19 before an official diagnosis.

Similar to the JAMIA Open study, research published in June 2022 in BMJ Open used the Ava bracelet, originally designed to monitor fertility in women, as well as a modified app to collect data from 1,163 participants in Liechtenstein. Of that group, 127 tested positive for COVID-19 via either molecular and/or serological assays. Researchers determined that changes in the five physiological metrics collected—skin perfusion, wrist-skin temperature, heart rate variability, heart rate, and respiratory rate—could predict cases of COVID-19.

Like the Icahn School of Medicine researchers, the international research team then used their data to train an AI algorithm to detect COVID-19, successfully identifying 68 percent of COVID-19-positive cases in their testing set two days before symptom onset.

“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 team concluded. “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. Future research should focus on how medical-grade wearable sensor technology can aid in comba[t]ing the current pandemic by monitoring sensor data.”

Study Limitations and Barriers to Implementation

While there have been several other studies involving the effectiveness of various wearables in detecting signs of COVID-19 that have also shown promise, that research does have limitations, and there are other roadblocks that could limit wearables’ effectiveness in screening for COVID-19 and other infectious diseases. Some of the key limitations include:

• Sporadic data collection by wearables
• Reliability and variability among wearables
• Variability among individuals
• Limited data sets
• Inability to capture and control other environmental factors, such as stress, that may impact metrics collected
• Limitations of currently available wearable technologies, such as sensor performance, etc.
• Inability to distinguish COVID-19 from other flu-like illnesses
• The AI algorithms have mostly been used to detect already-confirmed COVID-19 cases

Other barriers to using wearables and AI to detect infectious illnesses include concerns over data sharing and privacy, as well as underreporting of cases. In addition, while some research has pointed out that wearable devices could offer a low-cost way to monitor health in low- and middle-income countries, there does not appear to be many studies on how affordable and accessible this technology actually is to low-income individuals. What does exist shows that such tech mainly benefits the wealthy, meaning that COVID-19 monitoring via AI and wearables, if proven effective, may not benefit everyone equally—a concerning problem considering that the illness has hit low-income groups particularly hard. However, researchers are working to find solutions to many of these key problems and note that as wearable technologies continue to improve, they will collect even more high-quality data and additional metrics, making them more effective tools for healthcare applications.

So far, the effectiveness of AI tools trained from wearables data has also varied greatly. According to a review of 12 articles and 12 study protocols related to identifying SARS-CoV-2 with wearables published May 2022 in The Lancet Digital Health, an AI tool’s success in detecting presymptomatic cases ranged from as low as 20 percent to as high as 88 percent. The authors of the review concluded that while such tools show promise, the use of AI and wearables for early detection of COVID-19 is still very much in its infancy.

“Large prospective, and preferably controlled, studies recruiting and retaining larger and more diverse populations are needed to provide further evidence” of the technique’s effectiveness, the authors add.

Outlook

However, the articles and study protocols that review looked at were collected in July 2021, and it seems progress has been made on these challenges since then, with researchers working hard on refining their AI tools.

Several studies relating to wearables and COVID-19 are still ongoing, such as the DETECT study from Scripps Research and the Infectious Disease and Wearables Study from the Stanford Healthcare Innovation Lab. Both DETECT and the Stanford study address participants’ privacy concerns with data privacy and security protocols listed clearly on their websites and other studies have made data privacy a priority as well. Research teams behind other recently-published studies also plan to further test and improve their algorithms, with the team behind the June 2022 BMJ Open paper planning to further train and validate their algorithm in an upcoming trial. As more data are collected as part of these projects, this will allow researchers to improve their AI algorithms to make them even more effective at detecting COVID-19. Research teams from the various studies have highlighted the potential to adapt their methods and algorithms to detect other infectious illnesses, and some have already done so, making the combination of AI and wearable tech potentially even more valuable to public health monitoring.