Sleep
11(5):430-436, Raven Press, Ltd., New York
© 1988 Association of Professional Sleep Societies
Identifying Sleep Apnea from Self-Reports
*tLinda E. Kapuniai,*David J. Andrew, *:j:David H. Crowell, and
*§James W. Pearce
*Sleep Disorders Center of the Pacific,
Straub Clinic and Hospital, and the t School of Public
Health, the :f:Department of Psychology and Pacific Biomedical Research Center, and the
§lohn A. Burns School of Medicine, University of Hawaii, Honolulu, Hawaii, U.S.A.
.~
Summary: An apnea score (AS) was developed as a potential screening tool for
sleep apnea. This was based on self-report questionnaire responses of 76 sleep
disorder center patients and 20 sleep survey volunteers. Twenty volunteers
and 23 patients (group I) comprised the initial AS development group. Their
questionnaire responses were compared to polysomnographic apnea indexes
(AI) and apnea plus hypopnea indexes (AHI). Stepwise multivariate discriminant analysis was used to test whether or not selected group I questionnaire
responses could be used to correctly classify respondents into apnea (AI or
AHI >5) or nonapnea (AI, AHI ~5) groups. Self-reports of "stops breathing
during sleep," "loud snoring," and history of adenoidectomy best discriminated normal (AI ~5) from apnea (AI >5) cases. The AS derived from group
I responses to these three variables was then computed for group II (n = 53).
Mter examination of the AS results, the AS was modified to include just
"stops breathing" and "loud snoring" and the AI criterion was raised to 10 per
hour. This revised AS correctly identified 100% of the cases with moderatesevere sleep apnea (AI or AHI >40) and 70-76% of all sleep apnea cases with
AI or AHI >5. Predictive accuracy was 88% for AI > 10. The two questions
that comprise the AS should be incorporated into risk appraisal instruments or
interviews to screen for sleep apnea. Key Words: Sleep apnea-Self-report
score-Prevalence-Sensitivity-Specificity.
There is a growing body of literature suggesting that the prevalence of sleep apnea
may be greatly underestimated for adults in industrialized nations 0-10). In 1981,
Carskadon and Dement (7) found some degree of sleep apnea in 37.5% of 24 asymptomatic elderly subjects ". . . selected for good health and absence of sleep
complaints." Lavie (5) has reported sleep apnea in 0.89% of presumably normal men in
the Israeli work force and has suggested that this be considered only a minimum
estimate of probable cases. In an ambulatory screening study (10), sleep apnea was
detected in 20% of an unselected sample of male outpatients, and in 16% of male power
plant employees, a rate higher than previously expected.
Data from sleep disorder centers have demonstrated that the prevalence of sleep
Accepted for publication March 1988.
Address correspondence and reprint requests to Dr. Linda E. Kapuniai at Sleep Disorders Center of the
Pacific, Straub Clinic and Hospital, Inc., 888 S. King Street, Honolulu, HI 96813, U.S.A.
430
I
SELF-REPORTS AND SLEEP APNEA
431
apnea appears to increase with age, a fact with significant public health implications
given the rapidly increasing proportion of elderly in countries such as the United States.
Although it appears to occur most frequently in men after 40 years of age, the prevalence in women after menopause is thought to approximate that found in men. There is
no method, apart from polysomnography, however, for determining the prevalence of
sleep apnea in the general population.
Given the potentially adverse effects of sleep apnea on cardiopulmonary functioning
and daytime performance, possible negative drug interactions, and the high cost of
polysomnography, an inexpensive, reliable, and valid self-report sleep apnea instrument would be a valuable screening tool. Such an instrument could yield prevalence
estimates from various sectors of the adult population to aid in planning preventive
health care and in identifying cases for subsequent diagnostic studies.
The purpose of this preliminary study was to assess the predictive accuracy of
self-reported factors known to be present in adults with polysomnographically defined
sleep apnea. In this study, combinations of self-reported characteristics were evaluated
in terms of their association with the polysomnographic apnea index (AI) and apnea
plus hypopnea index (AHI). Those characteristics most significantly associated with AI
and AHI were formed into an apnea score (AS). The AS then was tested in a second
group of subjects for sensitivity, specificity, and predictive accuracy.
METHODS
Design
In this exploratory study, responses to selected items on the Hawaii Sleep Questionnaire (HSQ) (11) from 20 adult survey respondents (11) and 23 sleep disorder center
(SDC) patients (group I) were compared with Als and AHls derived from nocturnal
polysomnography (NPSG) conducted after HSQ completion. An AS was developed
from the results of these comparisons and applied to a second group (group II, n = 53)
of adult SDC patients who also had completed the HSQ and had AI and AHI results.
Group II, therefore, served as a test of the AS as a predictor of sleep apnea.
Subjects
Twenty survey cases plus 23 SDC patients comprised the original study sample,
group I (n = 43); 53 additional SDC patients formed the application sample, group II.
The SDC cases were consecutive patients who had completed a HSQ followed by an
all-night PSG. These patients were selected without regard to complaint or subsequent
diagnosis. Group I included 25 men and 18 women; group II had 42 men and 11 women.
Ages ranged from 36 to 69 years for group I (mean 50, SD 13.6) and 17-78 years (mean
46; SD 14.2) for group II. Educational level was 14 (±3.9) years of school completed in
group I and 14 (±2.8) years in group II. Ethnic group distributions for groups I and II,
respectively, were: white (44%, 38%), Oriental (28%, 30%), Hawaiian-part-Hawaiian
(21%, 19%), and other (7%, 13%). Groups I and II comparisons revealed no significant
differences in age, education, income, ethnicity, or body mass index (BMI) (12). Group
II had a significantly higher percentage of men (79%) than group I (58%), X2 (df = 1) =
4.06 (p = 0.04).
Data Acquisition
Self-report
The HSQ is a general sleep history questionnaire that includes questions on characteristics known to be present in sleep apnea patients. Apnea-associated questions inSleep, Vol. 11, No.5, 1988
432
L. E. KAPUNIAI ET AL.
clude frequency graded in terms of never, rarely, sometimes, often, or always of (a)
stopping breathing during sleep, (b) loud snoring, and (c) waking from sleep gasping or
short of breath. It also includes questions on sex, age, height, weight, and sleep history.
All respondents were asked to indicate if they had had a tonsillectomy or adenoidectomy.
Poiysomnography
A single NPSG was obtained for all group I and II cases. Apneas and hypopneas of
10 s or greater duration were identified by observation of air flow measured by right and
left nasal and oral thermocouples and by rib cage and abdominal inductive plethysmography. (Some of the earlier studies used a single abdominal respi-cuff or thoracic and
abdominal strain gauges.) The same technician scored all the polysomnograms.
Data analysis
In order to obtain a clinically useful AS, data analysis was carried out in four phases.
Phase I: Mter the NPSG, group I cases were divided into apnea and nonapnea subgroups; Phase II: group I AS development; Phase III: group II AS application; and
Phase IV: post hoc evaluation. In Phase II, step-wise discriminant analyses (13) were
conducted to ascertain the combination of selected HSQ responses that best discriminated group I apnea or nonapnea subgroups established in Phase I. The most discriminating subset of variables was examined to identify those variables best suited to
computing an AS. In Phase III, the resultant AS was computed for an independent
group II sample. The degree to which the AS correctly identified sleep apnea and
nonapnea cases was summarized in terms of specificity, sensitivity, and predictive
accuracy. Phase IV consisted of post hoc analyses of the group II AS and NPSG
indexes. Details of these data analysis phases are as follows.
Phase I
Prior to the analysis of the HSQ responses, AI (14) and AHI (14) were calculated for
each individual using a computer-generated NPSG report program. Apnea-nonapnea
comparison groups were formed for both the AI and AHI. An apnea case was defined
by an AI >5 or AHI >5. Although five apneas per hour may not be considered indicative of clinically significant sleep apnea, it is a common cutoff criterion reported in
apnea-nonapnea studies (6,15-17).
Phase II
As there were only 13 group I cases with AI >5, the number of self-report variables
that could be included as discriminating variables in the step-wise discriminant analysis
was limited. The HSQ items selected were chosen on the basis of clinical judgment.
These variables, along with the values entered in the discriminant analysis, were loud
snoring and stops breathing during sleep (never, rarely = 1, sometimes, often, always
= 2); wakes gasping or short of breath (never = 1, rarely = 2, sometimes = 3, often
= 4, always = 5), adenoidectomy (1 = yes, 2 = no), and BMI.
It is important to note that excessive daytime sleepiness and gender were not among
those selected. A preliminary correlational analysis of self-report responses with AI
and AHI values revealed that excessive daytime sleepiness was not significantly related
to AI and AHI; and therefore, it was not included in the set selected for this study. Even
though male gender was expected to be highly associated with sleep apnea, it was felt
Sleep, Vol. 11, No.5, 1988
SELF-REPORTS AND SLEEP APNEA
433
that inclusion of gender in a screening instrument would prejudice against detection of
sleep apnea in women.
Two separate step-wise discriminant analyses were conducted: first, comparing
cases with AI ~5 (nonapnea) and >5 (apnea); second, comparing cases with AHI ~5
(nonapnea plus hypopnea) and >5 (apnea plus hypopnea) cases. The combinations of
discriminating variables consistently yielding the highest associations with AI and AHI
would be used to formulate the AS. Mter the AS set of variables was selected, a
"positive for apnea" response (e.g., stops breathing sometimes, often, or always) had
to be present on each variable selected in order to designate a respondent as a probable
sleep apnea case.
Phase III
The AS developed from group I data was determined for each group II case. Then,
all group II cases were classified by AS into apnea or nonapnea subgroups. AS sensitivity and specificity were calculated. In addition, predictive accuracy was computed
for presence and for absence of sleep apnea. Predictive accuracy for sleep apnea was
calculated as the number of cases with AI (or AHI) >5 divided by the number of cases
with AS positive for sleep apnea. The accuracy of predicting nonapnea cases was
calculated as the number of cases with AI (or AHI) ~5 divided by the number of cases
with AS indicative of no sleep apnea.
RESULTS
AS development
The order of the step-wise selection of significantly discriminating variables was as
follows: (a) for AI, stops breathing, adenoidectomy, BMI, and wakes gasping or short
of breath, X2 (df = 4) = 13.90 (p = 0.008) and (b) for AHI, stops breathing, loud
snoring, and adenoidectomy, X2 (df = 4) = 12.20 (p = 0.007). These results defined the
criteria for selecting the variables for the AS: (a) Two of the three variables from the
AHI analysis-stops breathing and adenoidectomy-were chosen because they also
appeared in the AI subset. (b) Even though loud snoring only appeared in the AHI
subset, it was chosen because of its strorig clinical correlation with sleep apnea. (c) BMI
and wakes gasping or short of breath were discarded from the AS computation for
several reasons. First, both appeared only in the AI subset of discriminating variables,
and second, BMI requires additional computational steps that might deter use in a
clinical setting. Finally, (d) waking gasping or short of breath was eliminated because
it may be a symptom of other disorders, such as gastroesophageal reflux or asthma. In
summary, responses to these three variables, stops breathing, loud snoring, and adenoidectomy, constituted the AS.
With the variables identified, the actual AS computation was as follows. A value of
1 was assigned to each of the following: (a) a response of stops breathing sometimes,
often, or always, (b) a response of loud snoring sometimes, often, or always, and (c) a
report of no history of adenoidectomy. To be considered a probable sleep apnea case,
responses had to yield an AS of 3, that is, meet all three criteria. Any other combination
of responses (a score of less than 3) would place a respondent in the nonapnea group.
AS score application in group II
As a test of the AS utility as a clinical screening tool, the AS was determined for each
group II subject. AS specificity in relation to the AI was 67%, and for AHI, it was 69%;
Sleep, Vol, 11, No, 5,1988
434
L. E. KAPUNIAI ET AL.
sensitivity was 61% for AI and 59% for AHI (Table lA). Predictive accuracy of AS was
69% for the nonapnea cases (AI ~5 criteiion) and 58% for sleep apnea cases (.LA... I >5).
For AHI, 67% of the sleep apnea cases were identified correctly, as were 62% of the
nonapnea cases.
In the Phase IV post hoc evaluation, other combinations of group II AS and polysomnographic criteria were examined to determine whether or not improved screening
outcomes could be revealed. First, adenoidectomy was reported by only 26% of group
II cases. In light of this, the AS was recomputed without adenoidectomy, that is, only
stops breathing and loud snoring were included. In this revised AS, a score of 2 indicated a probable apnea case. When adenoidectomy was deleted, sensitivity increased
from 61% to 76% for AI >5, (Table lB), but specificity remained about the same: 65%
instead of 67% (a difference of one case). For AHI, specificity decreased from 69% to
65% and sensitivity increased to 70%.
Using this revised AS, the next modification was to raise the polysomnographic
criterion from 5 to 10/h, which is more in line with current clinical judgment for clinically significant amounts of sleep apnea. This change resulted in substantial improvements in sensitivity for both AI (83%) and AHI (78%) (Table lC). For those group II
cases with AI >20, 83% were identified by the AS as probable sleep apnea cases; for
AHI > 20, the figure was 81 %. When a criterion of 40 apneas or apneas plus hypopneas
per hour was imposed, 100% of these moderate-severe sleep apnea cases were in the
AS = 2 group. The criterion of lO/h appeared to yield the best all-around results.
DISCUSSION
The purpose of this study was to evaluate how selected self-report responses in the
form of an AS could be used to screen for sleep apnea. The AS, based on responses of
stops breathing and loud snoring sometimes, often, or always on the Hawaii Sleep
Questionnaire, was 100% effective in identifying sleep apnea cases with AI or AHI
>40. Adults with the revised AS of <2 could be expected to have little or no significant
sleep apnea (AI or AHI ~10) 80-88% of the time.
The accuracy of a screening instrument can be evaluated in two ways. First, negative
(nonapnea) cases should be excluded from further assessment, and second, positive
TABLE 1. Group II (n = 53) sensitivity, specificity, and predictive accuracy for (A) AI and
AHI ",;;;5 or >5 based on AS derived from self-reports of loud snoring, breathing cessation
during sleep and adenoidectomy, (B) loud snoring and breathing cessation only, (C) AI and
AHI > 10 for criteria in (B)
Apnea score
Predictive accuracy
Sensitivity
Specificity
Apnea
No apnea
AI
AHI
61%
59%
67%
69%
58%
67%
69%
62%
AI
AHI
76%
70%
65%
65%
61%
68%
76%
68%
AI
AHI
83%
78%
63%
67%
54%
64%
88%
80%
A
B
C
Sleep, Vol.ll, No.5, 1988
SELF-REPORTS AND SLEEP APNEA
435
cases should be flagged for further diagnostic evaluation. Based on the above results,
the AS was most effective when a sleep apnea criterion of at least 10/h was used. At this
level, 83% (AI) and 78% (AHI) of actual sleep apnea cases would have been flagged for
further evaluation. However, the AS also resulted in a substantial number of false
positives. Conversely, the AS yielded few false negatives; that is, the revised AS
accurately identified cases with AI ~ 10 88% of the time and AHI ~ 10 80% of the time.
Few cases with AI or AHI >20 and none with indexes >40 were placed in the nonapnea
category. These results were based on a simple yes-no additive formula from responses
to these two questions: When you are sleeping how often do you do the following or
someone has told you you do the following: (a) stop breathing, (b) snore loudly.
The AS was not as discriminating at lower levels of apnea frequency, but was very
accurate at higher levels. An AS of 2 could alert the clinician to. consider the potential
presence of sleep apnea and possible need for polysomnography. These cases also
could be followed for weight gain, excessive daytime sleepiness, or other indications
that sleep apnea might be present. Polysomnographic evidence of clinically significant
sleep apnea would be required to make a definite diagnosis.
With regard to the results obtained in this study, there were several factors that might
have contributed to the rather low AS specificity: (a) Levels of apnea signified by
indexes less than 10 may be too close to normal amounts of sleep apnea to be detected
in a cursory screen; (b) reliance on a single NPSG for the AI and AHI; and (c) lack of
direct observation.
1. AS specificity was not sufficient to warrant use of the AS in general population
surveys because of the large number of false-positives. These individuals might not be
available for further evaluation, and therefore, accurate estimates of sleep apnea prevalence could not be obtained. In a clinical setting, however, the AS could playa role in
preventive education. Cases with positive AS (including false-positives) could receive
information about snoring and sleep apnea as they relate to factors such as weight gain,
hypertension, or otolaryngologic abnormalities. At the same time, additional questioning could be pursued relative to presence of excessive sleepiness or sleeping partner
reports of apneic episodes. As apnea indexes of > 5 continue to serve as standards for
sleep apnea (6,15-17), it is important to persevere and develop a score that would be
reliable at various levels of sleep apnea.
2. Reliance on a single NPSG to define a sleep apnea case has been challenged
(15,16). In one study of NPSGs on 2 consecutive nights (15), AI varied over nights to
the extent that 15% of their sample would have been misdiagnosed using only first night
indexes >5. This degree of error could have inflated the false-positive rate in this study.
Subtracting 15% from the AI specificity of 37% and AHI rate of 33% (Table lC) would
reduce the error to 22% and 18%, respectively, which is still on the high side but
substantially better than chance.
3. The absence of direct reports from sleeping partners may have reduced the probability of accurate reporting of snoring and breathing cessation. Although it may be
preferable to include a sleeping partner's evaluation, this requirement would place
other restrictions on who could be screened. Taken together, however, these three
factors may account for some of the predictive error in this study population.
With the current availability of effective treatments for obstructive sleep apnea, such
as continuous positive airway pressure and uvulopalatopharyngoplasty, unnecessary
morbidity and mortality could be avoided through more education and earlier diagnoses. Currently, most health risk appraisals include one or two questions on sleep
Sleep, Vol. /1, No.5, 1988
436
L. E. KAPUNIAI ET AL.
difficulties, but these questions are not an adequate screen for a condition that seems
to be present in large numbers of adults. The resuits from this study iend quantitative
support to clinically observed signs associated with sleep apnea. Inclusion of the AS in
clinical histories and risk appraisals would be an important and simple step toward
preventive health education and early diagnosis of sleep apnea.
Acknowledgment: This research was supported in part by the Leahi Trust and the George F.
Straub Trust. Access to clinic visitors for the survey was made possible by Straub Clinic physicians, the Kaiser Permanente Medical Group, and Kaiser Foundation Hospital of Honolulu.
Computer and other facilities were made available by the School of Medicine and the International Center for Health Promotion and Disease Prevention, School of Public Health, University
of Hawaii.
REFERENCES
1. Block A, Boysen P, Wynne J, Hunt L. Sleep apnea, hypopnea and oxygen desaturation in normal
subjects. N Engl 1 Med 1979;300:513-7.
2. Coleman RM, Roffwarg H, Kennedy SJ, et al. Sleep-wake disorders based on a polysomnographic
diagnosis, a national cooperative study. lAMA 1982;247:997-1003.
3. Guilleminault C, Van de Hoed J, Mitler M. Clinical overview of the sleep apnea syndromes. In: Guilleminault C, Dement WC, eds. Sleep apnea syndromes. New York: Alan R. Liss, 1981:1-12.
4. Krieger J, Turlot J, Mangin P, Kurtz D. Breathing during sleep in normal young and elderly subjects:
hypopneas, apneas, and correlated factors. Sleep 1983;6:108-120.
5. Lavie P. Incidence of sleep apnea in a presumably healthy working population. A significant relationship
with excessive daytime sleepiness. Sleep 1983;6:312-318.
6. Ancoli-Israel S, Kripke DF, Mason WJ, Gabriel S, Kaplan O. Sleep apnea and PMS in a randomly
selected elderly population: final prevalence results. Sleep Res 1986;15:101.
7. Carskadon MA, Dement WC. Respiration during sleep in the aged human. 1 GerontoI1981;36:420-423.
8. Coleman RM, Miles LE, Guilleminault C, Zarcone VP, Van den Hoed J, Dement WC. Sleep-wake
disorders in the elderly: a polysomnographic analysis. lAm Geriatr Soc 1981 ;29:289-296.
9. Guilleminault C, Dement WC. Sleep apnea syndromes and related sleep disorders. In: Williams RL,
Karacan I, eds. Sleep disorders diagnosis and treatment. New York: Wiley, 1978:9-78.
10. Peter JH, Fuchs E, Kohler U, et al. Studies in the prevalence of sleep apnea activity (SAA): evaluation
of ambulatory screening results. Eur 1 Respir Dis 1986;146:451-458.
11. Kapuniai LE. Psychosocial factors associated with physician utilization for sleep problems. Unpublished
doctoral dissertation. Honolulu: University of Hawaii, 1984.
12. Keys A, FidanzaF, Karvonen MJ, Kimura N, Taylor HL. Indices of relative weight and obesity. 1 Chron
Dis 1972;25:322-343.
13. SPSS Inc. SPSSX user's guide (ed 2). Chicago, IL: McGraw-Hill, 1986.
14. Bornstein SK. Respiratory monitoring during sleep: polysomnography. In: Guilleminault C, ed. Sleeping
and waking disorders: indications and techniques. Menlo Park, CA: Addison-Wesley, 1982:183-212.
15. Kramer M, Silva C. Night to night variability of apnea. Sleep Res 1986;15:138.
16. Dickel MJ, Sassin J, Mosko S. Sleep disorders in an aged population: preliminary findings of a longitudinal study. Sleep Res 1986;15: 116.
17. Hoch CC, Reynolds CF, Stack JA, et al. Sleep apnea and psychopathology in geriatric patients. Sleep
Res 1986;15: 130.
Sleep, Vol. JI, No.5, 1988