Assessment of the test-retest reliability of laboratory polysomnography.

Statement Of The Problem: When conducting a treatment intervention study, it is assumed that a level of reliability can be obtained from the measurement tool such that the outcome can be reasonably assessed.

Purpose Of Study: Investigate the reliability of laboratory polysomnography, the gold standard for assessment of treatment outcomes for obstructive sleep apnea, at a 1-month interval.

Materials And Methods: In a clinical trial of 118 patients recruited to assess the effects of a pharmaceutical treatment intervention, a subset of 20 patients designated as placebo controls completed two polysomnography studies, one at baseline and one at least one month later.

Results: The correlation between the overall Apnea/Hypopnea indices from the two polysomnography (PSG) studies was poor (r = 0.44) and the results were biased, with a mean increase of seven events per hour on night 2. Twenty-five percent of the subjects had an increase greater than 20 events/hour on night 2 and only 45% of participants had a night-to-night difference of < or =5 events/hour. The correlation between overall apnea indexes for nights 1 and 2 (r = 0.61) was improved, compared to the overall apnea/hypopnea indexes. The correlation in sleep efficiency across the two nights was relatively week (r = 0.52) but significant. The correlations between nights 1 and 2 for the percentage of time supine (r = 0.70) and the supine apnea-hypopnea index (AHI) (r = 0.69) were similar and highly significant. The correlation for the non-supine AHI was only 0.25

Conclusions: In this study, the reliability of a single-night PSG in measuring treatment outcome was compromised as a result of the large night-to-night variability of subjects' obstructive sleep apnea (OSA). Studies employing the AHI as an outcome need to be adequately powered with respect to the inherent night-to-night variability in the measurement. When assessing treatment intervention outcomes, there may be benefit from the acquisition and averaging of multiple nights of data in order to mitigate the inherent night-to-night variability of OSA and improve the accuracy of the outcome assessment.