The impact of obstructive sleep apnea variability measured in-lab versus in-home on sample size calculations
© Levendowski et al. 2009
Received: 10 September 2008
Accepted: 02 January 2009
Published: 02 January 2009
When conducting a treatment intervention, it is assumed that variability associated with measurement of the disease can be controlled sufficiently to reasonably assess the outcome. In this study we investigate the variability of Apnea-Hypopnea Index obtained by polysomnography and by in-home portable recording in untreated mild to moderate obstructive sleep apnea (OSA) patients at a four- to six-month interval.
Thirty-seven adult patients serving as placebo controls underwent a baseline polysomnography and in-home sleep study followed by a second set of studies under the same conditions. The polysomnography studies were acquired and scored at three independent American Academy of Sleep Medicine accredited sleep laboratories. The in-home studies were acquired by the patient and scored using validated auto-scoring algorithms. The initial in-home study was conducted on average two months prior to the first polysomnography, the follow-up polysomnography and in-home studies were conducted approximately five to six months after the initial polysomnography.
When comparing the test-retest Apnea-hypopnea Index (AHI) and apnea index (AI), the in-home results were more highly correlated (r = 0.65 and 0.68) than the comparable PSG results (r = 0.56 and 0.58). The in-home results provided approximately 50% less test-retest variability than the comparable polysomnography AHI and AI values. Both the overall polysomnography AHI and AI showed a substantial bias toward increased severity upon retest (8 and 6 events/hr respectively) while the in-home bias was essentially zero. The in-home percentage of time supine showed a better correlation compared to polysomnography (r = 0.72 vs. 0.43). Patients biased toward more time supine during the initial polysomnography; no trends in time supine for in-home studies were noted.
Night-to-night variability in sleep-disordered breathing can be a confounding factor in assessing treatment outcomes. The sample size of this study was small given the night-to-night variability in OSA and limited understanding of polysomnography reliability. We found that in-home studies provided a repeated measure of sleep disordered breathing less variable then polysomnography. Investigators using polysomnography to assess treatment outcomes should factor in the increased variability and bias toward increased AHI values upon retest to ensure the study is adequately powered.
Obstructive Sleep Apnea (OSA) has recently gained recognition as one of the most common, under-diagnosed chronic diseases [1, 2]. It is characterized by frequent loud snoring and recurrent failures to breathe adequately during sleep (termed apneas or hypopneas), as a result of full or partial collapse of the upper airway. OSA can be a confounding factor in clinical trials because it causes daytime drowsiness, and has been associated with hypertension, increased risk of congestive heart failure, coronary artery disease, myocardial infarction, cardiac arrhythmias, diabetes and stroke [3–8].
The most commonly referred prevalence statistic on obstructive sleep apnea syndrome is that 4% of men and 2% of women have the disease . What doesn't get the same recognition is that 24% of the men and 9% of the women in the cohort were found to have sleep disordered breathing/obstructive sleep apnea (OSA)(i.e., AHI ≥ 5 without daytime somnolence). Because daytime somnolence and OSA severity are not well correlated [9, 10] and there are individual differences in susceptibility to sleep deprivation [11, 12], reliance on the former prevalence statistic could substantially underestimates the degree that this disease might affect subjects in clinical studies. More recent reports suggest the prevalence of OSA is increasing in part due to the rise in obesity, a major risk factor for OSA. In St. Louis MO, where the obesity rate is 28%, 19% of an adult surgical population was estimated to have moderate, severe or very severe OSA . In a recent study of 331 dental patients, 59% of men and 22% of women were estimated to have an AHI > 10, and 43% and 12% were estimated to have an AHI > 20 .
In the United States laboratory polysomnography serves as the gold-standard for evaluating OSA. The PSG is used to obtain an overall AHI, and this is compared to some threshold value both for diagnosis and for treatment recommendations. In treatment outcome studies, however, the percentage change in AHI is the primary determinant of success. Thus, the inherent variability associated with the subject's, conditions, and the measurement tool become important factors, and it is presumed that the variability can be accurately estimated for purposes of sample size calculations.
The literature on repeated-measures PSG on symptomatic patients is somewhat limited. Chediak et al.  reported that 12 of 37 (32%) of their cases exhibited a difference in AHI ≥ 10 in two sequential nights of PSG. Le Bon et al.  studied 243 subjects during sequential nights of PSG and evaluated the benefit of improved sensitivity and specificity as a result of having multiple nights of data. Dean and Chaudhary  investigated PSG variability in nine patients who appeared negative during the first PSG study and positive in the second PSG. Carlile and Carlile  found that 48% of patients with an AHI < 5 had AHI values ≥ 5 on a repeat study.
Most recently Levendowski et al.  reported a fairly weak correlation (r = 0.44) between overall Apnea/Hypopnea indices from the two PSG studies conducted approximately 40 days apart with a 7 event/hour bias toward increasing AHI values upon retest.
The goal of this study was to assess the test – retest variability in untreated mild to moderate OSA patients studied with both a laboratory PSG and an in-home recording with a substantial time between the test and retest.
The data were obtained from a multi-site, double blind placebo controlled study whose aim was to assess the efficacy of palatal implants as a treatment for mild to moderate OSA. Exclusion criteria were an AHI < 10 or > 40 based on the in-home baseline study, body mass index > 32 kg/m2, evidence of airway obstruction other than retro-palatal, prior airway surgery other than nasal, adenoid or tonsil, and presence of another sleep disorder (restless legs, insomnia, narcolepsy, etc.)
Thirty-seven patients from the placebo group underwent a baseline PSG and in-home sleep study followed by a second set of studies under the same conditions. The in-home studies were conducted approximately 60 days prior to the initial PSG studies (mean 59 ± 34 days). The time between the test and retest for PSG was almost five months (mean 140 + 35 days) and for the in-home studies was over six months (mean 188 + 47 days). The follow-up PSG and in-home studies were typically conducted within three weeks (mean 18 ± 21 days). Each subject had his or her test and retest PSG done at the same accredited sleep disorders center, with 12 subjects studied at the University of Cincinnati, 13 studied at the Medical College of Wisconsin, and 12 studied at Total Sleep Diagnostics of Indianapolis. The PSG data were acquired and scored by registered polysomnographic technicians applying the pre-2007 American Academy of Sleep Medicine guideline to classify apneas (i.e., 10-second cessation in airflow) and hypopneas (i.e., > 30% decrease in airflow and > 3% desaturation), and sleep staging (i.e., Rechtschaffen and Kales criteria) across sites. Total sleep times were 308 ± 61 SD minutes during the PSG test and 323 ± 50 SD minutes during the PSG retest.
The in-home studies were performed with Apnea Risk Evaluation System (ARES™) Unicorder (Advanced Brain Monitoring, Carlsbad, CA, USA). The ARES Unicorder measures oxygen saturation, pulse rate, airflow, respiratory effort, snoring levels, head movement, and head position from a wireless recorder self applied with a single strap to the forehead. Reflectance oximetry is used to obtain the SpO2 and pulse rate signals. Respiratory effort is derived from the measurement of changes in forehead venous pressure acquired using a combination of photoplethysmography and changes in surface pressure of the reflectance oximetry sensor, and head movement. Airflow is obtained via a nasal cannula and a pressure transducer. A calibrated acoustic microphone is used to acquire quantified snoring levels (dB). Accelerometers are used to measure head movement and derive head position. The recorder was designed to be easily affixed by the patient, and provide alerts during the study if poor quality airflow or SpO2 is detected so the device could be adjusted.
Automated scoring algorithms were applied off-line to detect sleep disordered breathing. The AHI was computed using a time-in-bed measure based on recording time with acceptable signal quality minus periods when the patient was upright or presumed to be awake based on actigraphy. Apneas, based on a 10-s cessation of airflow detected by the automated algorithms and hypopnea events that met the Medicare criteria required a 50% reduction and recovery in airflow, and a minimum 3.5% reduction in SpO2 and at least a 1.0% recovery were included in the apnea-hypopnea index (AHI-4%). Fifteen of the patients recorded two nights of in-home data during the test, two subjects completed a multi-night study during the retest. The total and valid recording times for the initial in-home test were 500 ± 211 and 451 ± 160 minutes respectively, and 406 ± 106 and 365 ± 104 minutes for the in-home retest.
Pearson correlations and Bland-Altman plots were used to assess the association in sleep disordered breathing between the test vs. retest data. Differences between the test and retest in demographic data and sleep parameters were tested with Student paired t-test.
Sample sizes required to provide nominal statistical power for 50% treatment effect within subject.
1-β = .80
1-β = .90
1-β = .80
1-β = .90
No bias, no reliability
No bias, observed reliability
Observed bias, no reliability
Observed bias, observed reliability
This is the first study which compares AHI variability in untreated mild to moderate OSA patients studied with PSG and in-home with a substantial time between the test and retest (i.e., five to six months). When comparing the overall AHI and apnea index values, the in-home results provided approximately 50% less variability than the PSG data (Figures 2 and 4) and showed virtually no bias. Most of the PSG bias and variability came from the six subjects whose PSG-AHI had increased by more than 30 between the test and retest. By way of comparison, the ARES AHI increased by more than 20 in two of these six subjects, in one case it decreased by 21, and the change was less than 5 events/hr in the remaining three. If the six subjects (17%) are excluded from the analyses, the results from the two methods become quite similar (ARES AHI: bias = 0.6, SD = 8.5; PSG AHI: bias = -0.7, SD = 8.6). Neither the demographic data nor the sleep parameters (TST, % time supine) supported these six subjects being dropped as their results did not differ significantly with the rest of the cohort (all p > 0.1). At best, the variability might be explained by a first-night effect for PSG that is known to under-report the severity of the disease in OSA subjects [17–19].
The substantially lower apnea and hypopnea variability obtained with the in-home system was likely influenced by the use of the same equipment and auto-scoring algorithms. The in-home device and auto-scoring algorithms used in this study was described and validated in two previous studies. The first had 284 valid comparisons of the in-laboratory simultaneous PSG and ARES and 187 valid comparisons of the in-laboratory PSG with a separate two nights unattended self-applied ARES Unicorder . The second study with 102 participants had 92 simultaneous in-laboratory comparisons and 86 in-home to in lab comparisons . Both studies showed that the ARES had high sensitivity and specificity. It is uncertain whether the findings in this study apply to other in-home Level III sleep study devices.
A limitation of this study is that the conditions used to acquire the PSG data reflect the conventional practice at sleep centers, as opposed to employing extra-ordinary means to control variability. For example, it was previously reported that the depth of desaturation varies by equipment manufacturer  and the AHI used in this study was based on a 4% desaturation. However we could not ensure the same type of pulse-oximeter was used in all rooms and at all sites. Although the procedures used to score the sleep and respiratory events within and across the sleep centers were standardized, intra- and inter-rater reliability assessments were not made. We found no statistically significant differences in test-retest reliability among the three sleep centers. There was, however, a trend suggesting one facility provided more reliable results even though all three centers followed identical guidelines for data acquisition and scoring. This finding suggested that future multi-site PSG-based outcome studies should consider utilizing a single facility for the scoring of the data to improve test-retest reliability. Of interest, the overall AHI scored by multiple technicians in this study showed improved correlations as compared to a previous report in which a single technician scored all the records (r = 0.56 vs. 0.44) .
Another factor which may have impacted reliability for both the in-home and PSG studies was the selection of patients with AHI values between 10 and 40 based on the initial in-home study. Previous reports suggest that patients with mild to moderate OSA tend to have positional dependent severity that is more readily influenced by the time spent in the supine position.
Differences in the supine AHI variability were less severe with the ARES studies as compared to PSG may be explained by the percentage of time supine being more consistent in-home vs. in-lab. The correlations in the percent time supine for the in-home study in this group (r = 0.72) studied approximately five months apart was very similar to the results from a group studied under similar circumstances 40 days apart (r = 0.70) . Patients spent less time supine in the lab during their retest, thus the expected impact of gravity on AHI does explain the bias toward increased PSG-AHI values upon retest. The in-home study recorded head position while the PSG studies used chest position, but it's uncertain what role this played in the supine positional results.
Several of the factors that might have impacted the test vs. retest AHI values were controlled by comparing laboratory and in-home results obtained under relatively similar circumstances. The subjects represented a homogenous group of subjects with baseline AHI values in the mild to moderate range, not excessively obese and without other sleep disorders. Although the initial in-home study was conducted two months earlier than the initial PSG study, the weight, BMI and neck circumference of the subjects did not significantly change between any of the in-home and PSG studies (all t-test p > 0.1). If a patient changed medication between the test and retest, it should have impacted both the PSG and in-home data similarly. In a previous study, Levendowski et al  was unable to relate differences in sleep architecture with differences in repeated measure PSG-AHI values. This study points toward a need for a well controlled prospective study to further clarify the issues which impact PSG variability.
The results from this study suggest major ramifications for the planning of future treatment studies. If one assumes no bias, and incorporates the observed reliability for the respective measures, the sample size required for PSG would be approximately three times that required for an ARES in-home study. Assuming the PSG is biased toward increasing values upon retest, the sample size required is more than 2 orders of magnitude larger than for the ARES and effectively impossible to implement. The bias toward increasing AHI values upon retest by PSG in this study (7.7 events/hour, p < .01) is very similar to the results reported by Levendowski et al. (i.e., 7 event/hour increase in PSG-AHI values at a 40 day retest)  and it is unlikely a coincidence. Other studies have also reported a "first-night" effect whereby PSG AHI values increase at retest [17, 18].
Test-retest variability in the sleep disordered breathing severity can be a confounding factor in assessing treatment outcomes. These data suggest that in-home studies provide a repeated measure of sleep disordered breathing less variable that that obtained by PSG at multiple sites. Guidelines developed to standardize the scoring of sleep and detection of related events (i.e., accreditation by the American Academy of Sleep Medicine) appears ineffective in controlling the inherent variability of OSA when measured by PSG. The acquisition of multi-night PSG studies is both difficult for subjects and very expensive. Investigators using PSG for assessing treatment outcomes should factor in the increased variability and tendency toward increased AHI values upon retest when selecting a sample size to ensure the study is adequately powered.
obstructive sleep apnea
Apnea Risk Evaluation System
The authors wish to thank Timothy Zavora and Daniela Scarfeo for providing assistance in acquiring the in-home data.
- Dement WC, Vaughan CC: The promise of sleep: A pioneer in sleep medicine explores the vital connection between health, happiness, and a good night's sleep. New York: Delacorte Press 1999.Google Scholar
- Kripke DF, Ancoli-Israel S, Klauber MR, Wingard DL, Mason WJ, Mullaney DJ: Prevalence of sleep-disordered breathing in ages 40–64 years: a population-based survey. Sleep 1997,20(1):65–76.PubMedGoogle Scholar
- Bradley TD: Right and left ventricular functional impairment and sleep apnea. Clin Chest Med 1992,13(3):459–479.PubMedGoogle Scholar
- Duchna HW, Guilleminault C, Stoohs RA, Faul JL, Moreno H, Hoffman BB, Blaschke TF: Vascular reactivity in obstructive sleep apnea syndrome. Am J Respir Crit Care Med 2000,161(1):187–191.PubMedGoogle Scholar
- Guilleminault C, Connolly SJ, Winkle RA: Cardiac arrhythmia and conduction disturbances during sleep in 400 patients with sleep apnea syndrome. Am J Cardiol 1983,52(5):490–494.View ArticlePubMedGoogle Scholar
- Guilleminault C, Suzuki M: Sleep-related hemodynamics and hypertension with partial or complete upper airway obstruction during sleep. Sleep 1992,15(6 Suppl):S20–24.PubMedGoogle Scholar
- Palomaki H: Snoring and the risk of ischemic brain infarction. Stroke 1991,22(8):1021–1025.PubMedGoogle Scholar
- Partinen M, Guilleminault C: Daytime sleepiness and vascular morbidity at seven-year follow-up in obstructive sleep apnea patients. Chest 1990,97(1):27–32.View ArticlePubMedGoogle Scholar
- Young T, Palta M, Dempsey JA, Skatrud J, Weber SM, Badr MS: The occurrence of sleep-disordered breathing among middle-aged adults. N Engl J Med 1993, 328:1230–1235.View ArticlePubMedGoogle Scholar
- Bixler EO, Vgontzas AN, Lin HM, Calhoun SL, Vela-Bueno A, Kales A: Excessive daytime sleepiness in a general population sample: the role of sleep apnea, age, obesity, diabetes, and depression. J Clin Endocrinol Metab 2005,90(8):4510–4515.View ArticlePubMedGoogle Scholar
- Berka C, Levendowski D, Westbrook P, Davis G, Lumicao M, Olmstead R, Popovic MV, Zivkovic V, Ramsey C: EEG Quantification of Alertness: Methods for Early Identification of Individuals Most Susceptible to Sleep Deprivation. Proceedings of SPIE Defense and Security Symposium, Biomonitoring for Physiological and Cognitive Performance during Military Operations (Edited by: Caldwell JA, Wesensten NJ). Orlando, FL: SPIE: The International Society for Optical Engineering 2005, 5797:78–89.Google Scholar
- Doran SM, Van Dongen HP, Dinges DF: Sustained attention performance during sleep deprivation: evidence of state instability. Arch Ital Biol 2001,139(3):253–267.PubMedGoogle Scholar
- Finkel K, Leif S, Safarzadeh E, Bottros M, Avidan M: Obstructive Sleep Apnea: The Silent Pandemic [abstract]. Anesthesiology 2006., 105:
- Levendowski DJ, Morgan T, Montague J, Melzer V, Berka C, Westbrook PR: Prevalence of probable obstructive sleep apnea risk and severity in a population of dental patients. Sleep Breath 2008,12(4):303–9.View ArticlePubMedGoogle Scholar
- Chediak AD, Acevedo-Crespo JC, Seiden DJ, Kim HH, Kiel MH: Nightly variability in the indices of sleep-disordered breathing in men being evaluated for impotence with consecutive night polysomnograms. Sleep 1996,19(7):589–592.PubMedGoogle Scholar
- Le Bon O, Hoffmann G, Tecco J, Staner L, Noseda A, Pelc I, Linkowski P: Mild to moderate sleep respiratory events: one negative night may not be enough. Chest 2000,118(2):353–359.View ArticlePubMedGoogle Scholar
- Dean RJ, Chaudhary BA: Negative polysomnogram in patients with obstructive sleep apnea syndrome. Chest 1992,101(1):105–108.View ArticlePubMedGoogle Scholar
- Carlile J, Carlile N: Repeat study of 149 patients suspected of having sleep apnea but with an AHI < 5 [abstract]. Sleep 2008.,31(A153):
- Levendowski D, Zack N, Rao SM: Assessment of the test-retest reliability of laboratory polysomnography. Sleep Breath 2008, in press.
- Westbrook P, Levendowski D, Cvetinovic M, Zavora T, Velimirovic V, Henninger D, Nicholson D: Description and Validation of the Apnea Risk Evaluation System: A Novel Method to Diagnose Sleep Apnea-Hypopnea in the Home. Chest 2005,128(4):2166–2175.View ArticlePubMedGoogle Scholar
- Ayappa I, Norman RG, Seelall V, Rapoport DM: Validation of a self-applied unattended monitor for sleep disordered breathing. J Clin Sleep Med 2008,4(1):26–37.PubMedGoogle Scholar
- Zafar S, Ayappa I, Norman R: Choice of oximeter affects apnea-hyponea index sleep staging reliability. Sleep 2004, 27:1171–1179.Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.