A Method for Intelligent Allocation of Diagnostic Testing by Leveraging Data from Commercial Wearable Devices: A Case...

Area of Science:

• Epidemiology and Public Health
• Biomedical Data Science
• Wearable Technology Applications

Background:

• Mass surveillance testing is crucial for controlling infectious diseases like COVID-19.
• Global diagnostic test shortages necessitate improved methods for efficient mass surveillance.
• Targeting tests to individuals most likely to be infected can increase testing positivity rates.

Approach:

• Developed an Intelligent Testing Allocation (ITA) method using data from the CovIdentify and MyPHD studies.
• Leveraged continuous digital biomarkers from smartwatch data (resting heart rate and steps) from 1,265 individuals.
• Optimized monitoring time periods and aggregate metrics to enhance COVID-19 diagnostic test positivity.

Key Points:

• Resting heart rate features detected COVID-19 infection earlier (up to 10 days prior) than step features (up to 5 days prior).
• Combining resting heart rate and step features improved model performance (AUC-ROC 7-11% increase, AUC-PR 38-50% increase).
• The best model achieved AUC-ROC of 0.77 using high-resolution Fitbit data, increasing positivity rates up to 6.5-fold.

Conclusions:

• The ITA method significantly increases COVID-19 test positivity rates, up to 3-fold in an independent test set.
• ITA effectively identifies both symptomatic and asymptomatic individuals (up to 27%), reducing the burden of test shortages.
• Large-scale deployment of ITA, without symptom reporting, can optimize diagnostic testing resource allocation.