Comparison of rest-activity rhythm metrics from Apple Watch and ActiGraph devices.

STUDY

objectivesObjective rest-activity rhythm (RAR) disturbances are linked with major disease outcomes. If consumer wearables yield RAR measures comparable to traditional research devices, these popular devices could provide a scalable option for clinical applications of RAR monitoring, e.g. risk factor screening.

methodsWe asked a convenience sample of participants (analytic n = 23; mean age = 27 years; 80% female) to wear Apple Watch and ActiGraph devices on separate wrists for one week. We derived a time series of activity counts from the raw accelerometer data, then extracted nonparametric (interdaily stability [IS], relative amplitude [RA], intradaily variability [IV]), and extended-cosine (pseudo-F, amplitude, up-mesor, acrophase, down-mesor) RAR variables. Spearman correlation coefficients (R) assessed association. Intraclass correlation coefficients (ICCs) assessed both agreement (closeness of values) and consistency (similarity of rankings) of RAR measures from the two devices.

resultsThe devices' RAR measures were highly correlated (Spearman R range: 0.7-0.9, p < .001). Agreement ICCs indicated good reliability for most metrics (ICC = 0.74-0.85), except for amplitude (which is highly dependent on the activity count's scale; ICC = 0.01). Agreement ICCs had wide confidence intervals; most reached at least the moderate agreement range (e.g. IS agreement ICC = 0.77; 95% CI: 0.47 to 0.90). Consistency ICCs were higher and exhibited narrower 95% CIs, with estimates in the good-to-excellent range (e.g. IS consistency ICC = 0.84; 95% CI: 0.61 to 0.92).

conclusionsGood-to-excellent consistency ICCs indicate that these devices yield similar participant rank-orderings on these RAR measures. However, there was systematic disagreement, suggesting absolute values from different devices' measures should not be pooled. Statement of Significance Consumer wearable devices could potentially be used in clinical-translational applications of sleep/circadian science, e.g. using the Apple Watch to monitor rest-activity rhythm (RAR) factors that are associated with disease risk. We evaluated if RAR measures derived from the Apple Watch's raw accelerometer data had consistency and agreement with RAR measures from a research-grade device. We found that, although RAR measures from these devices were strongly correlated and highly consistent, there was moderate agreement indicative of systematic discrepancies between the two devices' RAR measures. These findings show this consumer-wearable approach can be used to detect between-subject differences in RAR risk factors, but absolute measures should not be pooled across devices until further study determines and resolves sources of measurement discrepancies.