Study objectives: To determine the short-term variability of indices of disturbed respiration and sleep during 2 nights of unattended nonlaboratory polysomnography conducted several months apart.
Design: Participants were randomly selected using a block design with stratification on preliminary estimates of 2 criteria: respiratory disturbance index [RDI3% (apnea or hypopnea events associated with > or = 3% O2 desaturation): < 15/hour total sleep time, > or = 15/hour total sleep time] and sleep efficiency (SEff: < 85% and > or = 85%). The RDI and sleep data from initial and repeated polysomnography were compared.
Participants: A subset of 99 participants in the Sleep Heart Health Study who agreed to have a repeat polysomnogram within 4 months of their original study.
Measurements and results: Acceptable repeat polysomnograms were obtained in 91 subjects (mean study interval: 77 +/- 18 [sd] days; range: 31-112 days). There was no significant bias in RDI between study nights using several different RDI definitions including RDI3% and RDI4% (apnea or hypopnea events associated with > or = 4% O2 desaturation). Variability between studies estimated using intraclass correlations (ICC) ranged from 0.77 to 0.81. For subjects with a RDI3% < 15, variability increased as a function of increasing RDI, but for those with a RDI3% > or = 15, variability was constant. Body mass index, SEff, gender, or age did not directly predict RDI variability. Using RDI4% cutpoints of < or = 5, < or = 10 and < or = 15 events per hour of sleep demonstrated that 79.1%, 85.7%, and 87.9% of subjects, respectively, had the same classification of SDB status on both nights of study. There also was no significant bias in sleep staging, sleep efficiency, or arousal index between studies. However, variability was greater with ICC values ranging from 0.37 (% time in REM) to 0.76 (arousal index).
Conclusion: In the Sleep Heart Health Study, accurate estimates of the severity of sleep-disordered breathing and the quality of sleep were obtained from a single night of unattended nonlaboratory polysomnography. These findings may be applicable to other large epidemiologic studies provided that similar recording techniques and quality-assurance procedures are followed.