Obstructive sleep apnea (OSA) is a common medical disorder with general health and quality-of-life implications [1]. Associations with important medical conditions, including diabetes mellitus, coronary arterial disease, congestive heart failure, hypertension, and cerebrovascular accident, are well-documented, especially in moderate-to-severe OSA [1-6]. Untreated OSA may result in excessive daytime sleepiness, impaired decision-making and automobile accidents [7-9]. Continuous overnight polysomnography (PSG) performed in a sleep laboratory remains the current gold standard for diagnosis of OSA. Screening strategies for use in the primary care setting have been developed, with the goal of detecting patients at risk for OSA, and subsequent referral for PSG.

One such commonly used screening strategy is the Epworth Sleepiness Scale (ESS). This questionnaire relies on self-reported patient symptoms, asking “How likely are you to doze off or fall asleep?” in a set of 8 hypothetical situations, each scored 0-3, giving a total score 0-24. This test has provided mixed results in the detection of OSA. The scale effectively discriminated between primary snoring and OSA in early studies [10, 11]. An ESS score of 10 is most often considered to be the upper limit of normal, though more recent work has shown that a lower score (8) may be associated with abnormal daytime sleepiness [12, 13].

While primarily designed and developed as a measure of excessive daytime sleepiness, the ESS emerged as an important clinical tool in the workup and management of OSA. Some studies have relied on the use of the ESS to screen patients for sleep apnea and clinically this remains the most widespread tool for primary care physicians to triage patients for sleep evaluations [14]. Not only is it part of the screening process in determining referrals to sleep labs for PSG, but it remains part of the clinical decision-making process in determining who should get treatment, especially in cases of mild OSA [15]. The predictive properties of this instrument in the high-risk population of patients referred to the sleep laboratory for a sleep study for evaluation of clinically suspected OSA are not well established. Moreover, the impact of OSA on individual patients is partially affected by their demographic characteristics, i.e. gender and ethnicity [16-18]; body mass index may affect how OSA will impact a patient. The ESS is also subject to variations in self-reported symptoms by these same populations. It has been reported that African American subjects reported higher ESS scores than Caucasians, while gender and age did not influence the average score, in a study of insomnia [19]. Other studies have reported that being Maori in New Zealand is independently associated with elevated ESS score [20]. An analysis of normal patients in the Sleep Heart Health Study showed no association of age, sex, or BMI on the ESS [21], however ESS scores varied despite similar rates of subjective sleepiness [22]. Thus, it is likely that the ESS, which assesses only daytime sleepiness, may vary among different demographic groups; its relationship with OSA may also vary between these same groups.

This retrospective study was designed to examine the differences in ESS scores between demographic groups (and subgroups) of a cohort of patients referred for PSG. Because the ESS is commonly used to screen patients for OSA, we sought to determine if the predictive value of the ESS for OSA varies between these same groups and subgroups.

Data were collected from 2112 consecutive eligible patients referred to the Torr Sleep Center (Corpus Christi, TX) for PSG evaluation of suspected OSA between February 5, 2007 and June 26, 2009. All subjects were ≥12 years of age without a prior diagnosis of OSA. Subjects were excluded from the final analysis if the ESS was incomplete or if they failed to undergo PSG and recording of the respiratory disturbance index (RDI). The protocol was approved by the CHRISTUS Spohn Hospital Corpus Christi Institutional Review Board (#08 08013).

Prior to undergoing PSG, the subjects completed questionnaires. A general health questionnaire obtained information about demographics and general sleep health, including objective daytime sleepiness, as measured by the Epworth Sleepiness Scale [10]. Gender and ethnicity were self-reported by the participants. Physical examination was performed by trained technologists. The general examination included height and weight measurement.

Overnight comprehensive PSG was performed in the sleep laboratory, with multichannel recordings monitoring electroencephalogram, electrocardiogram, electrooculograms, submentalis electromyogram, airflow, respiratory effort, oxygen saturation, and anterior tibialis electromyogram. Data were scored by a technologist manually, according to the American Academy of Sleep Medicine Scoring Guidelines [23]. An apnea was scored if there was cessation of airflow for ≥10 seconds; a hypopnea was scored if there was ≥30% reduction in airflow for ≥10 seconds, associated with a drop in SaO₂ ≥4%; a respiratory effort related arousal (RERA) was scored if there was a sequence of breaths, not qualifying as apnea or hypopnea, lasting ≥10 seconds with increasing respiratory effort or flattening of the nasal pressure waveform leading to an arousal from sleep. The RDI was calculated by summing the number of obstructive apneas, hypopneas, and RERAs per hour. Technologists were chosen with minimum experience of scoring 500 PSG, and intra- and inter-scorer variability were standardized by means of a point system in place at the sleep center [24].

Comparisons between the means of 2 normally distributed groups were performed with the unpaired t-test, between 2 non-normally distributed groups with the Mann-Whitney U test, and between 3 or more non-normally distributing groups with the Kruskal-Wallis test. Sensitivity and specificity of the ESS in the detection of OSA were calculated based using ESS ≥10 as a positive test result and RDI ≥15 events/hour as the diagnostic standard for OSA. Correlation analysis was performed by calculating the Spearman rank correlation coefficient for nonparametric samples. A p-value of <0.05 was considered statistically significant.

A total of 2112 patient records were evaluated for inclusion into the study, of which 1900 had complete data collected and were included in the final analysis. Of the excluded patients, the vast majority were because of missing ESS questionnaires. Excluded patients did not differ significantly from those included in terms of age, height, body mass index (BMI), RDI, or lowest O2 saturation during the PSG; the excluded patients did have a higher mean weight than included patients (71 vs 68 kg, p=0.047) and were more likely to be male (65% vs 58%. P=0.037). The baseline clinical characteristics and demographics of the included participants, separated by gender are shown in Table 1. In general, men were larger by height and weight, though the average body mass index (BMI) was higher in females. Indices of OSA, such as RDI and lowest SaO₂ were more severe in males, and the mean ESS was higher in males (11 vs 10). The subjects predominantly identified themselves as Hispanic or Caucasian in both groups, and the mean BMI fell within the range of obesity; subject characteristics were otherwise unremarkable. Three subjects (all Caucasian) did not report their gender on the pre-PSG questionnaires. Table 2 shows the demographic and clinical characteristics separated by race.

The mean ± s.d. ESS score for all patients was 10.7 ± 5.6. The distributions of ESS within each of the major groups evaluated (separated by gender, ethnicity, and obesity status) and subgroups (combinations of demographic and clinical factors such that only one subgroup is appropriate for each patient) are shown in Fig. (1A, B), respectively. ESS scores varied significantly by group and subgroup (P<0.0001 by 1-way ANOVA for both sets). Within the major groups, the non-obese patients had significantly lower ESS scores than obese patients and the cohort as a whole. Males reported significantly higher ESS scores than females. The highest scores were noted in males (11.1 ± 5.7), Hispanics (11.1 ± 6.0), and obese patients (11.1 ± 5.6). The only group with a mean score <10 was non-obese patients (9.7 ± 5.5).

Fig. (1). Epworth Sleepiness Scale (ESS) scores of patients referred for polysomnography. (A) Box (25-75 percentile) and whisker (10- 90 percentile) plots are shown depicting the ESS scores of All (N = 1900), Male (N = 1092), Female (N = 805), Caucasian (N = 998), Hispanic (856), Other ethnicities (Other, N = 46), Obese (N = 1370), and Non-obese (N = 529) patients. (B) Box and whisker plots are shown for subgroups of patients: obese Caucasian males (OCM, N = 397), non-obese Caucasian males (NCM, N = 198), obese Hispanic males (OHM, N = 364), non-obese Hispanic males (NHM, N = 104), obese Caucasian females (OCF, N = 269), non-obese Caucasian females (NCF, N = 130), obese Hispanic females (OHF, N = 303), and non-obese Hispanic females (NHF, N = 82). # Non-obese patients differed significantly in ESS scores than all patients. * ESS scores differed significantly between males and females, and between obese and non-obese patients. ** ESS scores differed significantly between obese and non-obese Caucasian males.

Fig. (2). Test characteristics of the Epworth Sleepiness Scale (ESS) in the diagnosis of obstructive sleep apnea (OSA), related to gender, ethnicity, and body mass index. (A) The sensitivity and specificity of the ESS for the diagnosis of OSA are plotted for the overall study population (N = 1900), males (N = 1092), females (N = 805), Caucasians (N = 998), Hispanics (N = 856), other races (N = 46), obese (N = 1370), and non-obese (N = 529). (B) The sensitivity and specificity of the ESS for the diagnosis of OSA are plotted for obese Caucasian males (OCM, N = 397), non-obese Caucasian males (NCM, N = 198), obese Hispanic males (OHM, N = 364), non-obese Hispanic males (NHM, N = 104), obese Caucasian females (OCF, N = 269), non-obese Caucasian females (NCF, N = 130), obese Hispanic females (OHF, N = 303), and non-obese Hispanic females (NHF, N = 82).

Only subgroups involving either Caucasian or Hispanic patients were analyzed, because a) the “other ethnicities” group was too small to effectively subdivide and b) clinically important conclusions could not be drawn from such heterogeneous subgroups. Most of the subgroups showed similar ESS scores to other subgroups that differed by a single factor (i.e., obese Caucasian males were similar to obese Caucasian females); the exception was obese Caucasian males who reported higher ESS scores than non-obese Caucasian males. Otherwise, obese Hispanic males scored higher than non-obese Caucasian males, obese Caucasian females, non-obese Caucasian females, and non-obese Hispanic females. The highest ESS scores were from obese Hispanic males (12.0 ± 6.0) and obese Caucasian males (11.2 ± 5.4). Non-obese Caucasian males (9.5 ± 5.1), non-obese Caucasian females (9.5 ± 5.2), non-obese Hispanic females (9.5 ± 5.8), and obese Caucasian females (9.97 ± 5.08) scored the lowest.

In the overall cohort of patients, those with OSA (RDI ≥15) had significantly higher ESS scores than those without OSA (P<0.0001). The mean ± s.d. ESS scores are shown in Table 3 for patients with and without OSA in the groups and subgroups mentioned above. The major groups for which the ESS was significantly higher in those with OSA included males, Caucasians, Hispanics, obese, and non-obese patients. For subgroups, the same held true for obese Caucasian males, obese Hispanic males, and non-obese Hispanic males. In non-obese Caucasian females, the ESS was actually higher in patients without OSA, though not significantly. The 90^th percentile ESS scores for patients without OSA was calculated for each group, as this would serve as a potential upper limit of normal (ULN) when the ESS is applied as a screening tool for OSA. Each group produced a calculated ULN too high to be clinically useful for detection of OSA. The 10^th percentile of ESS in patients with OSA (i.e., lower limit of abnormal) was consistently in the 3-4 range for all groups and subgroups.

The correlations of between ESS and indices of OSA severity (RDI and lowest SaO₂) also varied between groups. The correlations for the overall study group, as well as each group and subgroup are listed in Table 4. In the overall study cohort, the ESS correlated significantly with RDI (r = 0.17, P<0.0001) and lowest SaO₂ (r = -0.20, P<0.0001). Of the groups that demonstrated a significant correlation between ESS and the markers of OSA, that association was almost universally stronger with the lowest SaO₂ than with RDI, though if there was a significant correlation with one of the markers, there was most likely also a significant correlation with the other within the same group. The correlation between ESS and RDI was most pronounced in males (r = 0.21), Hispanics (r = 0.19), and those that had OSA on the PSG (r = 0.20); significant correlations were seen with lowest SaO₂ were in the overall group (r = -0.20), males (r = -0.22), Caucasians (r = -0.22), Hispanics (r = -0.24), obese (r = -0.20), and those with OSA (r = -0.25). In the subgroups, consistently significant associations between the ESS and severity of OSA were seen in obese Caucasian males (r = 0.21 for RDI; r = -0.21 for lowest SaO₂), obese Hispanic males (r = 0.18 for RDI; r = -0.22 for lowest SaO₂), and non-obese Hispanic males (r = 0.29 for RDI; r = -0.27 for lowest SaO₂).

In all participants, the sensitivity of ESS for predicting OSA was 54% and specificity was 57%. The positive predictive value (PPV) was 64% and negative predictive value (NPV) was 47. In males, the sensitivity was 56%, specificity was 58%, PPV was 74%, and NPV was 38%; in females, sensitivity was 49%, specificity was 56%, PPV was 48%, and NPV was 57%. Among the self-reported ethnicity groups, the ESS displayed the best sensitivity in Hispanics (56%) and best specificity in Caucasians (57%). The PPV was substantially better in Hispanics (69%) than either Caucasian (59%) or other races (50%), while the NPV was <50% for all 3 ethnicity groups. In obese subjects, sensitivity was 55%, specificity was 53%, PPV was 68%, and NPV was 40%, compared with sensitivity of 47%, specificity of 63%, PPV of 49%, and NPV of 61% for non-obese subjects. Sensitivities and specificities for each group are shown in Fig. (2A).

The sensitivities and specificities in clinical and demographic subgroups are shown in Fig. (2B). The highest sensitivity was seen in obese Hispanic males (59%); the corresponding specificity was 54%. The highest specificity was seen in non-obese Hispanic males (76%), which also had a sensitivity of 55%, making this subgroup the one in which the ESS is most accurate in predicting OSA. As above, sensitivity and specificity analyses were not performed for males or females of other ethnicities due to small group sizes. Calculations of the test characteristics were repeated using other ESS ULN to determine if what ULN would result in a sensitivity >90% in each subgroup. These results were similar to those from the overall group: in each group, only an ESS ULN in the 3-4 range resulted in sufficient sensitivity to serve as a screening test, though also provided very poor specificity (7-22%).