Detection of COVID-19 using multimodal data from a wearable device: results from the first TemPredict Study
Level 3 - non-randomized controlled study
Prospective cohort study developing and validating a diagnostic machine learning algorithm
OpenAlex W4214931945 · doi:10.1038/s41598-022-07314-0
What was done
Researchers gathered daily self-reported questionnaire data and continuous physiological metrics from 63,153 participants wearing a consumer device (Oura Ring). From 704 individuals self-reporting possible COVID-19, 73 with PCR confirmation and high-quality physiological data were used to train a machine learning classification algorithm to identify COVID-19 onset. An additional validation was performed in a subset of 10 participants with antibody-confirmed COVID-19.
What was found
The algorithm detected COVID-19 an average of 2.75 days before participants sought diagnostic testing, with a sensitivity of 82%, a specificity of 63%, and a ROC AUC of 0.819 (95% CI [0.809, 0.830]). Incorporating continuous temperature increased the AUC by 4.9%. In the antibody-confirmed validation cohort (n = 10), the model achieved 90% sensitivity, 80% specificity, and an AUC of 0.819 (95% CI [0.809, 0.830]). Accuracy varied substantially according to age and biological sex.
Why it matters
This study shows that continuous physiological data from consumer wearables, particularly temperature metrics, can identify COVID-19 infection days before formal clinical testing, potentially supporting earlier self-isolation.
Limits
The algorithm development and validation relied on very small numbers of confirmed cases (73 PCR-confirmed and 10 antibody-confirmed) derived from self-reports within a massive cohort. The primary model specificity was modest (63%), which would yield substantial false positives in broad populations.
Cited by
- supports A study conducted by Oura and Stanford showed that biometric data from the Oura Ring could detect infection onset days before clinical symptoms appeared.