Table 1.
Summary of key characteristics, metrics and COVID-19 test results among symptomatic participants.
Fig 1.
a-c, green: included data; red: excluded data; R: the day receiving test results: (a) include all data; (b) exclude data on and after the day receiving test results; (c) exclude data before the day receiving test results since symptom onset. Ninety-two subjects (87.6%) received their results within the symptomatic period (0 to 7 days after symptom onset).
Fig 2.
AUCs based on RHR, sleep, activity and all-sensor metric derived from wearable sensors to differentiate symptomatic subjects who were tested positive and negative corresponding to data schemes I-III.
(a-d): Scheme I—all data; (e-h) Scheme II—remove data on and after knowing the test results; (i-l): Scheme III—remove data since symptom onset and before the test results. For each data scheme in the row, the four panels are for RHR, sleep, activity and sensor metrics, respectively.
Fig 3.
One-sided conditional permutation test for assessing the null that, compared to random data reduction, no additional change in AUC caused by removing data.
(a-d) on or after receipt of test result and (e-h) in the symptomatic period and prior to receipt of test results. The random data removal and AUC calculation are done for RHR, sleep, step and the all sensor data, respectively (shown in four panels in each row). In each panel, the red line indicates the observed change of AUC; the blue line is at zero, indicating no change. The reference distributions are not centered at zero despite data removal being random; because on average there are more days after the receipt of test results than before, random data removal may still impact AUCs. For each metric, if a red line is at the left tail of the histogram, we conclude a statistically significant additional decrease in AUC.