Inter-observer variability in the assessment of EEG arousal

Chronic sleep deprivation leads to excessive daytime sleepiness and impaired performance. Recently, it has become apparent that recurrent sleep disruption can have equally serious consequences. However, traditional sleep assessment criteria have inadequate resolution to document reliably these transient disturbances or arousals. Accordingly, the American Sleep Disorders Association (ASDA) recently proposed detailed criteria to specifically and reliably identify the occurrence of transient arousals. The aim of this study was to investigate interobserver variability in the assessment of EEG arousal. Polysomnographic data were recorded from ten subjects being investigated for suspected obstructive sleep apnoea. From each, nine 40 s epochs were identified. Epochs were equally distributed by base sleep stage i. light (I/II), ii. deep (III/IV) and iii. REM, and as those which (by our judgement) a. contained EEG arousal, b. did not contain EEG arousal, and c. could not be readily categorised. The order of the 90 epochs was randomised. Copies were distributed to 20 expert observers who assessed each epoch for arousal according to the ASDA guidelines, indicating the arousal duration where appropriate. Agreement was quantified using the Kappa (κ) statistic. To date, results have been collated from 11 centres. For 40 epochs, the consensus (6+ observers) rated the epoch 'arousal', for 50 epochs, 'no arousal'. Good agreement (9+ observers concur) was reached for 58 epochs, with moderate overall agreement (κ = 0.47). Degree of agreement was not related to arousal duration, but was dependent on base sleep stage (p < 0.05), being best for deep sleep (κ = 0.59), moderate for REM (κ = 0.50) and poor for light sleep (κ = 0.30). We conclude that variability exists in the interpretation of the ASDA criteria, particularly for arousals occurring during light sleep.