Skip to main content
. 2020 Oct 7;44(2):zsaa170. doi: 10.1093/sleep/zsaa170

Table 3.

EBE accuracy metrics for sleep/wake and sleep stages detection

EBE metric Binary definition (sleep/wake) Categorical definition (sleep staging)
Sensitivity Proportion of “true” sleep epochs (i.e. based on the reference) that are correctly classified as sleep by the device. Proportion of epochs classified as a given sleep stage by the reference that are correctly classified as that stage by the device.
Specificity Proportion of “true” wake epochs (i.e. based on the reference) that are correctly classified as wake by the device. Proportion of epochs not classified as a given stage by the reference that are correctly not classified as that stage by the device.
Accuracy Proportion of correctly classified sleep and wake epochs over the total number of epochs. Proportion of correctly classified epochs for a given stage over the total number of epochs.
PPV Proportion of epochs classified as sleep by the device that are “true” sleep epochs (i.e. based on the reference). Proportion of epochs classified as a target stage by the device that are classified as that stage by the reference.
NPV Proportion of epochs classified as wake by the device that are “true” wake epochs (i.e. based on the reference). Proportion of epochs not classified as a target stage by the device that are not classified as that stage by the reference.
PI Proportion of “true” sleep epochs (i.e. based on the reference) over the total number of epochs. Proportion of epochs classified as a target stage by the reference over the total number of epochs.
BI Difference in sleep epoch proportion (i.e. sleep epochs over the total number of epochs) between device and reference. Difference between device and reference in the proportion of epochs classified as a target stage over the total number of epochs.

Metrics are defined based on proportions (generally ranging from 0 to 1), but it is common to report them as percentages, by simply multiplying the result by 100. PPV, positive predictive value; NPV, negative predictive value; PI, Prevalence Index; BI, Bias Index.