Evaluation of the Consensus Sleep Diary in a community sample: comparison with single-channel electroencephalography, actigraphy, and retrospective questionnaire.

STUDY OBJECTIVES

The Consensus Sleep Diary (CSD) was developed by experts to promote standardization of sleep diary data across the field, but studies comparing the CSD with other assessments of sleep parameters are scarce. This study compared the CSD with 3 other methods to assess sleep duration, efficiency, and timing.

METHODS

Participants (n = 80) were community adults (mean age = 32.65 years, 63% female) who completed the time-stamped CSD and used single-channel electroencephalography (EEG) and actigraphy for 7 days at home, then completed a retrospective sleep questionnaire. Total sleep time (TST), sleep efficiency (SE), and sleep midpoint were compared using correlations, Bland-Altman plots, and limits of agreement (adjusted for repeated measures).

RESULTS

Correlations between the CSD and all methods on TST were large (rs = .63-.75). Adjusted CSD average TST was 40 minutes greater than with EEG and 31 minutes greater than with actigraphy. Correlations between CSD, actigraphy, and EEG for SE were small (rs = .18), and there was a medium correlation with questionnaire (r = .42). Adjusted CSD average SE was 7% greater than EEG and 6% greater than actigraphy; both demonstrated heteroscedasticity. Sleep midpoint correlations between CSD and all methods were large (r = .92-.99). Adjusted CSD was, on average, 6 minutes later than EEG and 1 minute later than actigraphy. Questionnaire-derived sleep parameters demonstrated nonconstant bias; lesser values had positive bias and greater values had negative bias.

CONCLUSIONS

The time-stamped CSD led to meaningful overestimations of TST and SE as measured by objective/inferred methods. However, sleep timing was rather accurately assessed with the CSD in comparison to objective/inferred measures. Researchers should carefully consider which sleep assessment methods are best aligned with their research question and parameters of interest, as methods do not demonstrate complete agreement.