Detection of COVID-19 using multimodal data from a wearable device: results from the first TemPredict Study

Area of Science:

• Digital Health
• Infectious Disease Surveillance
• Machine Learning in Healthcare

Background:

• Early disease detection is crucial for limiting transmission and enabling prompt medical care.
• Consumer wearable devices offer continuous physiological monitoring for potential illness identification.
• The COVID-19 pandemic highlighted the need for innovative detection methods.

Purpose of the Study:

• To develop and validate a machine learning algorithm for early COVID-19 detection using wearable sensor data.
• To assess the algorithm's performance in identifying illness onset prior to diagnostic testing.
• To evaluate the impact of physiological features, particularly temperature, on detection accuracy.

Main Methods:

• Collected daily questionnaire and physiological data from over 63,000 participants using the Oura Ring wearable.
• Trained a machine learning classification algorithm on data from 73 PCR-confirmed COVID-19 cases.
• Validated the algorithm using SARS-CoV-2 antibody confirmed cases and analyzed performance variations by demographics.

Main Results:

• The algorithm detected COVID-19 an average of 2.75 days before participants sought testing, with 82% sensitivity and 63% specificity (ROC AUC 0.819).
• Inclusion of continuous temperature data improved the AUC by 4.9%.
• Validation in antibody-confirmed cases showed 90% sensitivity and 80% specificity.

Conclusions:

• Wearable devices and machine learning show significant promise for early COVID-19 detection.
• Continuous temperature monitoring and diverse population data are vital for optimizing algorithm accuracy.
• This approach can support public health by enabling earlier self-isolation and intervention.