Periodic Reporting for period 2 - COVID-RED (COVID-19 infections - Remote Early Detection)
Berichtszeitraum: 2022-01-01 bis 2022-12-31
On 31 December 2019, the local authorities of Wuhan, Hubei province, China, reported a cluster of pneumonia cases of unknown origin. On 9 January 2020, the China Centre for Disease Control reported a novel coronavirus - now referred as SARS-CoV-2 to be the causative agent. As of 24 February 2020, 79 360 cases of novel coronavirus infection COVID-19 (in accordance with the applied case definitions in the affected countries) have been reported, including 2 618 deaths. The disease has already spread to 31 countries outside China, with new cases continuing to emerge daily [1].The World Health Organization (WHO) declared the COVID-19 pandemic a Public Health Emergency of International Concern. As of December 2022, over 649 million confirmed cases and over 6.6 million deaths have been reported globally. During the WHO Committee in January 2020, a univocal belief was expressed that it is still possible to interrupt virus spread, provided that countries put in place strong measures to detect disease early, isolate and treat cases, trace contacts, and promote social distancing measures commensurate with the risk. In light of this, the European Commission called for projects that would help contain the spread of the volatile pandemic and ultimately minimize the burden on public health.
At the time when the call was published in March 2020, there were no specific vaccines or treatments proven effective against this viral disease. Therefore, early and accurate detection of COVID-19 has been critical to containing its spread, improving health outcomes and limits the negative impacts on society.
The COVID-RED consortium was brought together as a result of this call in order to address issues of high public health need that emerged from the pandemic. The COVID-RED project has been working towards developing a reliable tool for easy, at-home use to detect potential COVID-19 infections in symptomatic and asymptomatic individuals. The project employed wearable digital technology (smart armbands) and Artificial Intelligence (AI) algorithms to assess user health status in real-time following known or unknown exposure to the virus. Information is exchanged with the user via a mobile health application. The tool was designed to provide a daily health status check and make recommendations regarding referrals to medical care facilities for confirmatory diagnoses of COVID 19 infections if required.
The overall goal of the COVID-RED public and private partner consortium had been to conduct a study that evaluated the use and performance of a repurposed CE-marked wearable device, which measures breathing rate, pulse rate, skin temperature, and heart rate variability, to detect and monitor changes in the measured bio parameters that are indicative of a potential COVID-19 infection. Participants also reported clinical symptoms via the device’s mobile application. The project has aimed to improve early detection of COVID-19 cases, and thus limit the spread of the disease, optimize triage for medical care, and improve patient prognosis.
At the time when the call was published in March 2020, there were no specific vaccines or treatments proven effective against this viral disease. Therefore, early and accurate detection of COVID-19 has been critical to containing its spread, improving health outcomes and limits the negative impacts on society.
The COVID-RED consortium was brought together as a result of this call in order to address issues of high public health need that emerged from the pandemic. The COVID-RED project has been working towards developing a reliable tool for easy, at-home use to detect potential COVID-19 infections in symptomatic and asymptomatic individuals. The project employed wearable digital technology (smart armbands) and Artificial Intelligence (AI) algorithms to assess user health status in real-time following known or unknown exposure to the virus. Information is exchanged with the user via a mobile health application. The tool was designed to provide a daily health status check and make recommendations regarding referrals to medical care facilities for confirmatory diagnoses of COVID 19 infections if required.
The overall goal of the COVID-RED public and private partner consortium had been to conduct a study that evaluated the use and performance of a repurposed CE-marked wearable device, which measures breathing rate, pulse rate, skin temperature, and heart rate variability, to detect and monitor changes in the measured bio parameters that are indicative of a potential COVID-19 infection. Participants also reported clinical symptoms via the device’s mobile application. The project has aimed to improve early detection of COVID-19 cases, and thus limit the spread of the disease, optimize triage for medical care, and improve patient prognosis.
The COVID-RED study harnessed recent advances in artificial intelligence and wearable sensor technology to detect potential coronavirus disease 2019 (COVID-19) infections prior to symptom development using a custom mobile phone application (“Ava COVID-RED”) and complementary wearable device (the Ava bracelet). The trial’s primary objective was to determine if an algorithm ingesting wearable-measured biophysical parameters could outperform symptom-based standard of care.
Between 22 February 2021 and 07 February 2022, 17 825 subjects in the Netherlands participated in the study. In line with a randomised, single-blind, two-period, two-sequence crossover design, subjects were randomised 1:1 to either Sequence 1 (experimental condition followed by control condition; n=8915) or Sequence 2 (control condition followed by experimental condition; n=8910). The study started with an initial Learning Phase (maximum of 3 months), followed by a 3-month Period 1 and a 3-month Period 2. Based on demographics, medical history, and/or profession, each subject was stratified at baseline within each sequence based on their risk to contract SARS-CoV-2 as well as their risk of developing severe COVID-19 (i.e. a high-risk and a normal-risk group). This design aimed to allocate 10 000 subjects per sequence: approximately 6500 normal-risk individuals and 3500 high-risk individuals per sequences.
Taken together, the primary analyses demonstrated that the algorithm applied in the experimental condition was superior in the overall and early detection of SARS-CoV-2 infections than the control condition. Its higher sensitivity, however, came at the cost of inferior specificity, due to an increased amount of (false) positive signals to get tested received by subjects in the experimental condition.
Secondary analyses revealed the final version of the experimental condition algorithm relied heavily on positive variance (e.g. net increases compared to a baseline mean) in a subject’s breathing rate and heart rate in the 1 and 4 days preceding the alert, respectively. Additional analyses also revealed the pattern of change in physiological parameters across the SARS-CoV-2 disease trajectory and considered the impact of potential confounders on wearable measurements. Finally, we investigated procedural compliance rates across several stratifications, including risk class, study periods, and pre- versus post-COVID-19 vaccination.
Despite a rapidly evolving epidemiological landscape and the unforeseen roll out of vaccines during the course of the investigation, the COVID-RED study achieved its primary objectives. Namely, it developed and tested a novel algorithm ingesting data from a wearable device.
Between 22 February 2021 and 07 February 2022, 17 825 subjects in the Netherlands participated in the study. In line with a randomised, single-blind, two-period, two-sequence crossover design, subjects were randomised 1:1 to either Sequence 1 (experimental condition followed by control condition; n=8915) or Sequence 2 (control condition followed by experimental condition; n=8910). The study started with an initial Learning Phase (maximum of 3 months), followed by a 3-month Period 1 and a 3-month Period 2. Based on demographics, medical history, and/or profession, each subject was stratified at baseline within each sequence based on their risk to contract SARS-CoV-2 as well as their risk of developing severe COVID-19 (i.e. a high-risk and a normal-risk group). This design aimed to allocate 10 000 subjects per sequence: approximately 6500 normal-risk individuals and 3500 high-risk individuals per sequences.
Taken together, the primary analyses demonstrated that the algorithm applied in the experimental condition was superior in the overall and early detection of SARS-CoV-2 infections than the control condition. Its higher sensitivity, however, came at the cost of inferior specificity, due to an increased amount of (false) positive signals to get tested received by subjects in the experimental condition.
Secondary analyses revealed the final version of the experimental condition algorithm relied heavily on positive variance (e.g. net increases compared to a baseline mean) in a subject’s breathing rate and heart rate in the 1 and 4 days preceding the alert, respectively. Additional analyses also revealed the pattern of change in physiological parameters across the SARS-CoV-2 disease trajectory and considered the impact of potential confounders on wearable measurements. Finally, we investigated procedural compliance rates across several stratifications, including risk class, study periods, and pre- versus post-COVID-19 vaccination.
Despite a rapidly evolving epidemiological landscape and the unforeseen roll out of vaccines during the course of the investigation, the COVID-RED study achieved its primary objectives. Namely, it developed and tested a novel algorithm ingesting data from a wearable device.
Our findings highlight the potential role of wearable devices in early detection of pre-symptomatic respiratory illnesses. Although the final version of the algorithm tended to overestimate COVID-19’s prevalence, future iterations could be fine-tuned to increase its specificity. Additional research building on our novel findings may wish to consider a differentiation process for illness testing, thereby allowing the creation of labelled data for non-COVID-19 respiratory infections; with this information, the machine learning algorithm could be trained to detect COVID-19 versus influenza versus a common cold, for example. The largest test of a SARS-CoV-2 wearable-based infection detection algorithm to date, the COVID-RED study generated a unique dataset with high external validity and demonstrated the potential of personalized biofeedback during a global pandemic. With additional resources and refinement, the underlying COVID-19 health status algorithm and study-generated database could prove beneficial to scientific researchers for years to come.