Professional Documents
Culture Documents
Population-Based Sleep-Disordered Breathing: Study
Population-Based Sleep-Disordered Breathing: Study
Background: Clinical observations have linked sleep- ing apnea-hypopnea index (P=.003 for systolic, P=.01
disordered breathing, a condition of repeated apneas and for diastolic, adjusted for confounding factors). The mag-
hypopneas during sleep, with hypertension but evi- nitude of the linear association increased with decreas-
dence for an independent association has been lacking. ing obesity. At a body mass index (weight in kilograms
Understanding this relationship is important because the divided by the square of the height in meters) of 30
prevalence of sleep-disordered breathing is high in adults. kg/m2, an apnea-hypopnea index of 15 (vs 0) was asso-
ciated with blood pressure increases of 3.6 mm Hg for
Objective: To test the hypothesis that sleep-dis- systolic (95% confidence interval, 1.3-6.0) and 1.8
ordered breathing is related to elevated blood pressure mm Hg for diastolic (95% confidence interval, 0.3-3.3).
independent of confounding factors. The odds ratio for hypertension associated with an apnea-
hypopnea index of 15 (vs 0) was 1.8 (95% confidence
Methods: The sample included 1060 employed women interval, 1.3-2.4).
and men aged 30 through 60 years who had completed
an overnight protocol as part of the Wisconsin Sleep Co- Conclusions: There is a dose-response relationship be-
hort Study. In-laboratory polysomnography was used to tween sleep-disordered breathing and blood pressure, in-
determine sleep-disordered breathing status, quantified dependent of known confounding factors. If causal, the
as the number of apneas and hypopneas per hour of sleep high prevalence of sleep-disordered breathing could ac-
(apnea-hypopnea index). Blood pressure was measured count for hypertension in a substantial number of adults
on the night polysomnography was performed. in the United States.
Results: Blood pressure increased linearly with increas- Arch Intern Med. 1997;157:1746-1752
_I_
THE
repeatedepisodes of ap- including hypoxemia, hypercapnia, arous-
and hypopnea in sleep-
nea als from sleep, and large negative intratho-
disordered breathing are racic pressures, may affect BP regulation
known to cause transient through both neural and humoral mecha¬
elevations in blood pres¬ nisms. Although a causal mechanism has
sure (BP) during sleep, and it has been hy¬ not been established, there is some evi¬
pothesized that these episodes result in el¬ dence that patients with sleep-disordered
evated daytime BP as well.1 Because the breathing have increased sympathetic nerve
prevalence of untreated sleep-disordered activity,3 decreased baroreceptor sensitiv¬
breathing, ranging from mild to severe, is ity,4 accentuated vascular responsive¬
high in both women and men (9% and ness,5 and abnormal salt and water metabo¬
24%, respectively), quantifying the role of lism,6 all of which could contribute to
this condition in the development of hy¬ hypertension. A few studies710 have shown
pertension is particularly important.2 Even BP to decrease in patients with sleep-
a modest role for sleep-disordered breath¬ disordered breathing after successful treat¬
ing in BP elevation would place a large ment by tracheostomy or by the most com¬
From the Departments number of people at increased risk for car¬
of Preventive Medicine
monly used therapy of nasal continuous
diovascular morbidity and mortality. positive air pressure.
(Drs Young and Palta and
Mr Peppard and Ms Finn),
A causal role of sleep-disordered Widely cited clinical observa¬
Medicine (Drs Hla and breathing in hypertension is supported by tions1113 that 50% to 90% of patients with
Skatrud), and Kinesiology strong biological plausibility and some ex¬ sleep-disordered breathing have hyper¬
(Dr Morgan), University perimental evidence. The acute physiologi¬ tension add support to the hypothesis.
of Wisconsin, Madison. cal consequences of apnea and hypopnea, However, inconsistent findings have re-
suited from recent clinic-based studies14'21designed to sin Sleep Cohort Study, a longitudinal study of the
specifically test thehypothesis thatsleep-disordered natural history of sleep-disordered breathing. The study
breathing is related to increased BP afteradjustment for is unique in having the combined strengtbs of a large
obesity, age, and other potential confounding factors. Fur¬ population-based sample and of assessment of sleep-
thermore, with the exception of our study22 using am¬ disordered breathing by in-laboratory polysomnogra¬
bulatory BP monitoring in a subset of a population sample, phy, the standard of clinical practice for diagnosis.28-29
no population-based study to date has found a statisti¬
tingency tables, multiple linear regression, and logistic re¬ consumed 24 hours before the sleep study), education, and
gression. To account for the stratified sampling of the sleep physical activity were investigated as confounding fac¬
cohort, all analyses were weighted to give unbiased esti¬ tors. Confounding was assessed by the standard method
mates. The SUDAAN software was used to compute ap¬ of comparing the association of interest before and after ad¬
propriate SEs for the weighted analyses. dition of each potential confounding factor.36 When ad¬
The association between sleep-disordered breathing justment changed the regression coefficient for AHI by 15%
and BP adjusted for confounding factors was quantified by or more, the covariate was retained in the final model. In
multiple regression techniques. Apnea-hypopnea index, the addition, interactions between the covariates and AHI with
primary independent variable, was used in all models as a respect to BP were tested for statistical significance. The
continuous variable. statistical significance of linear regression coefficients was
Multiple linear regression was used to estimate the assessed by í tests and that of logistic regression coeffi¬
change in BP associated with increases in the AHI. Al¬ cients was assessed by Wald x2 tests. Two-tailed P values
though linear regression has the advantage of modeling BP of less than .05 were considered to indicate statistical sig¬
as a continuous variable, the ability to estimate the true re¬ nificance. Standard regression diagnostics were per¬
lationship is hampered by the influence of antihyperten¬ formed to assess model fit and adequacy of compliance with
sive medication on BP. To account for this, we fit models the modeling assumptions.
lighting the need to account for these factors that are also mate the association of AHI and BP, it is necessary to
correlates of hypertension. specify a BMI level. Final models of systolic and dias¬
Multiple linear regression modeling, adjusting for tolic BPs, centered at a BMI of 30 kg/m2 for individuals
age, sex, and several indicators of body habitus (BMI, not using antihypertensive medication, are given in
waist-hip ratio, neck girth, and skinfolds thickness), Table 2. Under the conditions of the model, the ß co¬
showed that sleep-disordered breathing, measured by AHI, efficient for AHI indicates an increase of 0.24 and 0.12
was significantly related to systolic and diastolic BPs. The mm Hg in systolic and diastolic BPs, respectively, for each
body habitus variables were all significant in the regres¬ additional apnea or hypopnea per hour of sleep. The model
sion models and reduced the coefficient for AHI to the predicts, for example, that BPs will be 3.6 mm Hg (sys¬
same degree. Use of combinations of the body habitus tolic) and 1.2 mm Hg (diastolic) higher for mild sleep-
variables produced the same results. Body mass index has disordered breathing (AHI, 15) vs no sleep-disordered
been most widely used, so this measure of body habitus breathing (AHI, 0). To illustrate the decrease in this
was adopted. There was no interaction of AHI and age effect with increasing obesity, predicted BP increases
or sex, indicating that the relationship between AHI and associated with mild to moderate sleep-disordered breath¬
BP did not vary by age or sex. There was, however, a sig¬ ing over 3 BMI levels are given in Figure I.
nificant negative interaction of AHI and BMI, indicating The final multiple logistic regression model of sleep-
the association of AHI and BP decreases with increasing disordered breathing and hypertension, with terms for
BMI. Although the decrease is small, to correctly esti- AHI, sex, age, BMI, and an interaction term for BMI and
*
N= 1069. AHI Indicates apnea-hypopnea index; body mass index, a measure of weight in kilograms divided by the square of the height in meters; and BP, blood
pressure.
tRange for AHI, 30.5 to 97.5.
Table 2. Multiple Linear Regression Models for Systolic and Diastolic BPs: Sleep Cohort Study*
Systolic BP Model Diastolic BP Model
r 1 r 1
Independent Variable SE(ß) SE(ß)
AHI, events per h .24 0.08 .003 .12 0.05 .01
Age, 1 y .34 0.07 <.001 .13 0.05 .004
Sex, Male 5.50 0.95 <.001 3.77 0.66 <.001
BMIf .71 0.09 <,001 .39 0.08 <.001
AHIxBMIf -.014 0.006 .03 -.010 0.004 .01
*
N=936. The formulas to calculate the millimeters of mercury increase in systolic and diastolic blood pressure (BP) associated with other apnea-hypopnea
Index (AHI)-body mass index (BMI, a measure of weight in kilograms divided by the square of the height in meters) combinations can be calculated as millimeters
of mercury=[0.24+(BMI- 30)x (-0.014)]xAHI. The increase in millimeters of mercury in diastolic BP associated with any AHI-BMI combination can be calculated
as millimeters of mercury=[0.12+(BMI-30)x (-0.01)]xAHI.
\Body mass Index is centered at 30 kg/m2.
AHI, indicated that each additional apnea or hypopnea thickness). We consistently saw a dose-response trend:
per hour of sleep increased the risk of having hyperten¬ BP increased with increasing sleep-disordered breathing
sion by approximately 4% (ß coefficient, .037). The odds severity. Of particular public health relevance, even
ratios and 95% confidence intervals for hypertension as¬ mild sleep-disordered breathing (eg, AHI, 5-15) appears
sociated with AHI levels of 5, 15, and 30 predicted by to carry a slightly increased risk for elevated BP.
the models (centered at a BMI of 30 mm Hg) are given The use of probability sampling from a well-
in Table 3; the slight decrease in effect with increasing defined base of employed men and women increases con¬
BMI is illustrated in Figure 2. fidence in the validity of the study findings, as well as
the generalizability of our results to middle-aged adults.
COMMENT Furthermore, the large sample size allowed precise esti¬
mates of associations and enhanced our ability to con¬
This investigation, based the largest community
on trol simultaneously for numerous potential confound¬
sample that has been studied with in-laboratory poly¬ ing factors.
somnography, has shown explicitly that occult, An additional important strength of our study lies
untreated sleep-disordered breathing is significantly in the quality of the measurements of sleep-disordered
associated with elevated systolic and diastolic BPs and breathing and BP. Use of attended, laboratory-based poly¬
with hypertension, independent of age, sex, and body somnography, the diagnostic standard for assessment of
habitus (BMÎ, waist-hip ratio, neck girth, and skinfold sleep-disordered breathing,28-29 ensured high-quality data
Table 3. Odds Ratios for Sleep-Disordered Breathing and Hypertension: Sleep Cohort Study*
Systolic BP ==140 mm Hg
Systolic BP a140 mm Hg Diastolic BP -90 mm Hg orDiastolic BP 90 mm Hg
or the Use of or the Use of or the Use of
Antihypertensive Medication Antihypertensive Medication Antihypertensive Medication
r ~l 1 1
Independent Variable OR 95% CI OR 95% CI OR 95% CI
AHI
5vs0 1.21 1.10-1.34 1.18 1.07-1.30 1.21 1.09-1.34
15 vsO 1.78 1.32-2.38 1.64 1.22-2.21 1.75 1.28-2.40
30vs0 3.15 1.75-5.67 2.68 1.48-4.86 3.07 1.65-1.74
*N=1069. The formulas to calculate odds ratios (ORs) for each definition of hypertension associated with other apnea-hypopnea index (AHI)-body mass index
(BMI, a measure of weight in kilograms divided by the>square of the height In meters) combinations are as follows: for the logistic regression models, the OR for
hypertension defined as systolic blood pressure (BP) 140 mm Hg or the use of antihypertensive medication associated with any AHI-BMI combination can be
calculated as ORHm=el003S*<mi-3O,x<-0JX"98>lxAHI. The OR for hypertension defined as diastolic BP >90 mm Hg or the use of antihypertensive medication associated
with any AHI-BMI combination can be calculated as ORHm=eiao33,tBMI~m'<<~omi>,xA"1. The OR for hypertension defined as systolic BP > 140 mm Hg or diastolic BP
>90 mm Hg or the use of antihypertensive medication associated with any AHI-BMI combination can be calculated as ORHm=eIO037*<BMI-3O>xl-OO022>,xAHI. HTN
indicates hypertension. Body mass index centered at 30 kg/nf.
studies.16"18 Nearly all the other studies141519"21-23"25 show by our findings that demonstrate a statistically signifi¬
a strong association between sleep-disordered breath¬ cant relationship, independent of known confounding fac¬
ing and elevated BP that becomes weaker and statisti¬ tors. Our findings are of both clinical and public health
cally insignificant when BMI, age, and sex are taken into significance.
consideration. Although these findings are generally taken A patient with even mild unrecognized sleep-
as evidence against an association of sleep-disordered disordered breathing is more likely to have an elevated
breathing with BP, méthodologie limitations may have BP than an otherwise similar patient without sleep ap-
made the studies incapable of detecting a true but small nea. Recognition of sleep-disordered breathing in such
association. patients may be helpful in designing treatment strate¬
Our analysis indicates that the association of sleep- gies for both the elevated BP and the sleep-disordered
disordered breathing with BP is not large. To detect the breathing. Of particular clinical importance, sleep-
effect with statistical significance requires a sample size disordered breathing is a stronger risk factor for el¬
that provides adequate study power, accurate measure¬ evated BP in the nonobese compared with the obese
ments, and proper multivariable modeling. Many of the patient. Although obesity is a strong risk factor for sleep-
clinic studies1419"21 with negative findings had small sample disordered breathing, there is a significant prevalence of
sizes or used control groups of symptomatic patients who unrecognized sleep-disordered breathing in the non-
could be considered to have mild or preclinical sleep- obese patient. Currently, suspicion for sleep-disordered
disordered breathing. In view of our finding that even breathing is focused on obese snorers; nonobese pa¬
mild sleep-disordered breathing is related to elevated BP, tients who do not present with sleep complaints are un¬
use of these controls would underestimate the associa¬ likely to be questioned on their snoring or sleeping his¬
tion. Another concern is random error in the measure¬ tory. Our findings suggest that during the routine medical
ment of sleep-disordered breathing, which would bias the examination, questions relevant to sleep-disordered
effect estimate toward the null. This problem may ac¬ breathing should not be restricted to obese patients.
count for insignificant findings of the previous population- A prevalent risk factor associated with even a small
based studies, none of which used polysomnography to increase in BP is of public health significance. Although
measure sleep-disordered breathing. Surrogate mea¬ the causal direction of the sleep-disordered breathing and
sures with poor or unknown validity were used in all these elevated BP association has not been established, it is pos¬
studies.23"25 Finally, body habitus was controlled for in sible using our estimates to appreciate what the popula¬
all studies, but an interaction term has never been con¬ tion impact would be if sleep-disordered breathing does
sidered previously. Our finding that the association be¬ directly contribute to hypertension, defined as systolic
tween sleep-disordered breathing and BP diminishes with BP of 140 mm Hg or more, diastolic BP of 90 mm Hg or
increasing BMI levels may explain negative findings from more, or using antihypertensive medication. For this, we
studies with a high proportion of extremely obese pa¬ used standard formula for attributable risk,41 in conjunc¬
tients. In support of this explanation, positive findings tion with (1) 1990 US Census population estimates, (2)
were reported from investigations in Sweden16 and Aus¬ National Health and Nutrition Examination Survey hy¬
tralia17 in which the average BMI of the patients studied pertension prevalence estimates,37 (3) the prevalence of
was less than 30 kg/m2. sleep-disordered breathing in middle-aged women (5%
In summary, the current controversy of whether an for an AHI of 5-15 and 4% for an AHI of > 15) and men
association between sleep-disordered breathing and el¬ (15% for an AHI of 5-15 and 9% for an AHI of >15),2
evated BP truly exists should be lessened considerably and (4) the relative risk estimates for hypertension as-