Fig. 5 | Scientific Reports

Fig. 5

From: An infection prediction model developed from inpatient data can predict out-of-hospital COVID-19 infections from wearable data when controlled for dataset shift

Fig. 5

Hospital model performance and risk scores in detecting COVID-19 infection from wearable dataset. (A) Receiver Operating Characteristic (ROC) curves. Experiment I (blue dash-dot line): hospital model directly applied to wearable daily features. Experiment II (green dashed line): hospital model applied to wearable sleep-only features. Experiment III (red solid line): hospital model applied to wearable sleep-only features after feature transformation. Experiment IV (orange dotted line): hospital model applied to wearable daily features after feature transformation, without using sleep data exclusively. Area under the ROC curve (AUROC) for each experiment is included in the figure legend. (B) Precision-recall curve. Colors are the same as described in subplot A. Average Precision (AP) score for each experiment is included in the figure legend. (C) Mean infection risk score based on the output of the best generalized hospital model (Experiment III: sleep-only features + feature transformation) in 301 COVID-19 positive subjects (black) and 2,111 COVID-19 negative subjects (blue) as a function of number of days relative to the COVID-19 test time (red). Grey and light-blue shaded area depicts 95% confidence interval calculated from the standard error.

Back to article page