Concordance of polysomnographic and actigraphic measurement of sleep and wake in older women with insomnia

Abstract

Study objectives: The objective of this secondary analysis was to evaluate concurrent validity of actigraphy and polysomnography (PSG) in older women with insomnia.

Methods: Concurrent validity of actigraphy and PSG was examined through (1) comparison of sleep outcomes from each recording method; (2) calculation of sensitivity, specificity, accuracy, and predictive values from epoch-by-epoch data; and (3) statistical and graphical exploration of the relationship between sleep disturbance severity and concordance of actigraphy and PSG. Subjects were 16 community-dwelling older women (mean age 69.4 ± 8.1) with insomnia who underwent 8 nights of concurrent actigraphy and PSG.

Results: Sleep efficiency reflected much greater sleep disturbance on PSG (66.9%) than actigraphy (84.4%). Based on generalized linear models, the parameter estimates for agreement between actigraphy and PSG were statistically significant (p < 0.05) for total sleep time and sleep latency, verged on significance for WASO (p = 0.052), but was not significant for sleep efficiency (p = 0.20). Epoch-by-epoch analysis showed high sensitivity (96.1%), low specificity (36.4%), and modest values on agreement (75.4%) and predictive values of sleep (74.7%) and wake (80.2%). Generalized linear models showed that overall accuracy of actigraphy declined as sleep efficiency declined (unstandardized Beta = 0.741, p < 0.001). Based on this model, sleep efficiency of 73% was the point at which accuracy declined below an acceptable accuracy value of 80%.

Conclusions: Actigraphy offers a relatively inexpensive and unobtrusive method for measuring sleep, but it appears to underestimate sleep disturbance, particularly at sleep efficiency levels below 73%, in older women with insomnia.

Keywords: Actigraphy; aging; insomnia; sleep; sleep initiation and maintenance disorders.

Figure 1. Scatterplots showing the relationship between actigraphy and PSG sleep outcomes within subjects

Data points represent a subject on one night of measurement (4-8 data points per subject). Lines represent the within-subject relationship between PSG and actigraphy. The graphs show linear relationships between PSG and actigraphy on total sleep time (A) and WASO (C) within participants. The graph shows a positive linear relationship between PSG and actigraphy on sleep efficiency (B) for most participants, but of low magnitude, as reflected in the nonsignificant results on the generalized linear model analysis. Although the scatterplot and generalized linear model show overall agreement between PSG and actigraphy on sleep latency, the within-subject lines (D) show substantial variability within the sample.

Figure 2. Concordance of actigraphy and polysomnography

This figure illustrates concordance outcomes at the three wake-detection threshold settings. For all outcomes, zero represents no concordance, and one represents perfect concordance.

Figure 3. Within-subject relationships between PSG sleep efficiency and actigraph accuracy

Data points represent a subject on one night of measurement (4-8 data points per subject). In this graph, accuracy (i.e., percentage of epochs on which PSG and actigraphy are in agreement) declined as PSG sleep efficiency declined in all except one participant.

Figure 4. Bland-Altman plots of whole-night sleep outcomes

The graphs show the difference between actigraphy and PSG (actigraphy value – PSG value) plotted against the mean of actigraphy and PSG on each of the sleep outcome: total sleep time (A), sleep efficiency (B), wake after sleep onset (C), and sleep latency (D). The horizontal reference line represents zero difference between the two recording methods on each outcome. Values above the zero-difference line indicate higher values on actigraphy than PSG (i.e., overestimation by actigraphy); values below the zero-difference line indicate lower values on actigraphy than PSG (i.e., underestimation by actigraphy).