ORIGINAL ARTICLE Polysomnography vs Self-reported Measures in Patients With Sleep Apnea Edward M. Weaver, MD, MPH; Vishesh Kapur, MD, MPH; Bevan Yueh, MD, MPH Background: While obstructive sleep apnea syndrome is tween PSG indices and self-reported measures were fur- defined by both polysomnographic (PSG) abnormalities and ther assessed with multivariable regression techniques, symptoms, severity is quantified primarily by the apnea- adjusting for age, sex, body mass index, comorbidity, and hypopnea index (AHI) alone. PSG type. Objective: To determine the correlation between stan- Results: The PSG parameters correlated poorly with self- dard PSG indices (AHI and others) and self-reported reported measures (15 correlations; range of magni- sleepiness, mental health status, and general health in pa- tude, 0.004-0.24; mean, 0.09). AHI was not associated tients with sleep apnea. with self-reported sleepiness or general health, and it was associated with the SF-36 Health Status mental health do- Design: Cross-sectional study. main only on multiple linear regression (P=.04) but not on multiple logistic regression (adjusted odds ratio, 1.02; Setting: University-affiliated outpatient sleep laboratory. 95% confidence interval, 1.00-1.04; P=.09). Patients: Ninety-six consecutive patients with PSG- Conclusions: In general, PSG measures, and AHI in par- confirmed sleep apnea (AHI Ն5). ticular, correlated poorly with self-reported measures in a clinical sleep laboratory sample. After adjustment for Measurements: Patients completed a questionnaire that potentially confounding variables, weak associations were included the Epworth Sleepiness Scale, Medical Out- found between some PSG indices and selected self- comes Study 36-Item Short-Form Health Survey (SF- reported measures. These findings suggest that sleep ap- 36) mental health domain, and self-rated health on the nea disease burden should be quantified with both physi- evening of diagnostic PSG. Spearman correlation coef- ologic and subjective measures. ficients were computed. This sample had 85% power to detect a correlation of 0.3 or greater. The associations be- Arch Otolaryngol Head Neck Surg. 2004;130:453-458 OLYSOMNOGRAPHY (PSG) IS eral health status in a cross-sectional analy- considered the gold standard sis of a clinical OSAS population. From the Department of for the diagnosis of obstruc- Otolaryngology–Head and tive sleep apnea syndrome METHODS Neck Surgery (Drs Weaver and (OSAS), the estimation of its Yueh), Sleep Disorders Center severity, and measurement of treatment re- PATIENTS P1 (Drs Weaver and Kapur), sponse. While OSAS involves both a PSG Center for Cost and Outcomes abnormality and symptoms, its severity is We retrospectively reviewed the records of all Research (Drs Weaver and patients (N=96) who had diagnostic PSG often defined by the apnea-hypopnea in- Ն Yueh); Pulmonary and Critical 2 findings consistent with OSAS (AHI 5) in Care Medicine Division, dex (AHI) alone. Improvement of symp- the University of Washington Sleep Labora- Department of Medicine toms is an important outcome, especially for tory between July 1, 1998, and January 31, (Dr Kapur), and Department of patients with mild OSAS, with which seri- 1999. Age, sex, height, weight, and self-re- Health Services (Dr Yueh), ous medical complications are less likely to ported comorbid conditions (specifically, an- University of Washington, occur.3-6 gina, myocardial infarction, congestive heart Seattle; and Surgery and Clinically, we have noted discor- failure, stroke, and chronic obstructive pul- Perioperative Care Service and dance between the severity indicated by monary disease) were recorded for each pa- Health Services Research and PSG results and the degree of symptoms tient on the evening of PSG. The comorbidity Development Service reported by some patients with OSAS. score (range, 0-5) was the number of comor- (Drs Weaver and Yueh), bid conditions listed above. Comorbidity Veterans Affairs Puget Sound Thus, we sought to determine the corre- score was considered missing if the patient Health Care System, lation between the severity of abnormali- answered “don’t know” or did not respond to Seattle, Wash. The authors ties as measured by standard PSG indices the comorbidity questionnaire (n=24). Body have no relevant financial and the severity of self-reported sleepi- mass index was calculated as weight in kilo- interest in this article. ness, symptoms of depression, and gen- grams divided by the square of height in me- (REPRINTED) ARCH OTOLARYNGOL HEAD NECK SURG/ VOL 130, APR 2004 WWW.ARCHOTO.COM 453 ©2004 American Medical Association. All rights reserved. Downloaded From: http://archotol.jamanetwork.com/ by a University of California - Los Angeles User on 03/23/2014 ters. This study was approved by the University of Washing- reported measures and PSG parameters. Spearman correla- ton Human Subjects Review Committee. tions are presented because the assumptions of normal distributions for Pearson correlations were not met. With 96 POLYSOMNOGRAPHY subjects, this study had 85% power to detect an important cor- relation coefficient (Ն0.3) at the 2-tailed significance level of All patients underwent overnight, monitored, in-laboratory di- .05.15 agnostic PSG. When split-night PSG studies were performed (com- Multiple linear regression was used to further assess for bined diagnostic and continuous positive airway pressure titra- significant associations between each self-reported measure (ESS tion; n=46), only data from the diagnostic portion (mean±SD or SF-36 mental health domain) as the dependent (continu- sleep time, 118±72 minutes) was analyzed for this study. The PSG ous) variable and each PSG index as the independent (con- included recordings of sleep state parameters (4-lead electroen- tinuous) variable, adjusting for age, sex, body mass index, co- cephalogram, bilateral electro-oculogram, and submental and bi- morbidity score, PSG type (split-night vs full-night), and lateral leg electromyogram), breathing (oronasal airflow by therm- presence of periodic limb movement disorder. The AHI was also istors and thoracic and abdominal excursion by strain gauge), analyzed separately as an ordinal variable of clinical categories oximetry, electrocardiogram, and infrared video. All studies were and as an ordinal variable of 33rd percentiles. Residual diag- manually scored by trained technicians and confirmed by a single nostics were analyzed visually to confirm assumptions of lin- board-certified sleep physician (V.K.). 7 8 earity and homogeneity of variance (no heteroscedasticity). Each Sleep stages and arousals were scored in standard fashion. adjustment variable is a potential confounder because each may Apnea was defined as 75% or greater reduction of airflow for 10 be associated with PSG indices and self-reported measures. Ad- seconds or longer. Apnea index was defined as the number of ap- justment for the presence of periodic limb movement disorder neas per hour of sleep. Hypopnea was defined as 25% reduction did not affect the results, so only analyses without this adjust- or more of airflow for 10 seconds or longer associated with either ment are shown. Adjustment for the presence of individual co- cortical arousal or 3% oxyhemoglobin desaturation. Apnea- morbid conditions (dummy variable model) was not signifi- hypopnea index was defined as the number of apneas and hy- cantly different from adjustment for comorbidity score; the latter popneas per hour of sleep and was clinically categorized as nor- is reported. Ͼ 2 mal (0-4.9), mild (5-15), moderate (15.1-30), or severe ( 30). Multivariable logistic regression was used to further as- Arousal index was defined as the number of arousals per hour of sess for significant associations between each dichotomized self- sleep. The lowest oxyhemoglobin saturation was the lowest re- reported measure (ESS, SF-36 mental health domain, or self- corded during sleep, and hypoxemic burden was defined as the rated health) as the dependent variable and each PSG index as percentage of sleep time with an oxyhemoglobin saturation of less the independent variable, adjusting for age, sex, body mass in- than 90%. Each of these PSG indices was analyzed as a continu- dex, comorbidity score, PSG type (split-night vs full-night), and ous variable, and AHI was also analyzed as an ordinal variable presence of periodic limb movement disorder. Multivariable or- categorized clinically and as 33rd percentiles. dinal logistic regression was then used to further evaluate the association between each ordinal self-reported measure (ESS SELF-REPORTED SYMPTOM AND categories or self-rated health) as the dependent variable and HEALTH STATUS MEASURES each PSG index as the independent variable, adjusting for the same potential confounders as above. Ordinal logistic regres- During the evening of PSG, each patient had completed a self- sion produces a common odds ratio that estimates the odds of administeredclinicalquestionnairethatincludedtheEpworthSleepi- the dependent ordinal variable increasing by 1 higher cat- ness Scale (ESS),9 the mental health domain of the Medical Out- egory for each unit increase in the independent variable. The comes Study 36-Item Short-Form Health Survey questionnaire common odds ratio is the same between any 2 consecutive cat- (SF-36),10 and self-rated health.11 The clinical questionnaire did egories of the dependent variable, and it is not dependent on not include other parts of the SF-36. Patients were blind to PSG the category cut points. The assumption of