Content uploaded by Enrico Marcelli
Author content
All content in this area was uploaded by Enrico Marcelli on Jan 09, 2024
Content may be subject to copyright.
Short and long sleep are positively associated with obesity, diabetes,
hypertension, and cardiovascular disease among adults in the United States
Orfeu M. Buxton
a
,
1
,
*
, Enrico Marcelli
b
,
1
a
Brigham and Women’s Hospital, Department of Medicine, Harvard Medical School, Division of Sleep Medicine, BLI-438, 221 Longwood Avenue, Boston, MA 02115, United States
b
Department of Sociology and Center for Behavioral and Community Health Studies (BACH), San Diego State University, San Diego, CA, United States
article info
Article history:
Available online 16 June 2010
Keywords:
USA
Body weight
Cardiovascular diseases
Diabetes
Hypertension
Metabolism
Obesity
Sleep disorders
Sleep duration
Sociogeographic
Social ecological model
abstract
Research associates short (and to a lesser extent long) sleep duration with obesity, diabetes, and
cardiovascular disease; and although 7e8 h of sleep seems to confer the least health risk, these findings
are often based on non-representative data. We hypothesize that short sleep (<7 h) and long sleep (>8h)
are positively associated with the risk of obesity, diabetes, hypertension, and cardiovascular disease; and
analyze 2004e2005 US National Health Interview Survey data (n¼56,507 observations, adults 18e85)
to test this. We employ multilevel logistic regression, simultaneously controlling for individual charac-
teristics (e.g., ethnoracial group, gender, age, education), other health behaviors (e.g., exercise, smoking),
family environment (e.g., income, size, education) and geographic context (e.g., census region). Our
model correctly classified at least 76% of adults on each of the outcomes studied, and sleep duration was
frequently more strongly associated with these health risks than other covariates. These findings suggest
a7e8 h sleep duration directly and indirectly reduces chronic disease risk.
Ó2010 Elsevier Ltd. All rights reserved.
The proportionof overweightand obese adults in the US has been
increasing for decades, with nearly two-thirds now overweight or
obese. Obesity predicts a broad range of health risks, including
diabetes, hypertension, cardiovasculardisease (CVD), early mortality,
and lower quality of life (Ogden et al., 2006; Ogden, Yanovski, Carroll,
& Flegal, 2007). As ‘actual causes’of death, obesity, diabetes, hyper-
tension, and CVD (McGinnis & Foege, 1993) are critical federal
research and public health priorities (U.S. Department of Health and
Human Services, 2000a: pp. B18-1eB18-17; U.S. Department of
Health and Human Services, 2000b: pp. B19-1eB19-42).
In addition to individual characteristics (e.g., age), a variety of
risk factors affecting obesity, diabetes, hypertension, and CVD have
been identified, including individual behaviors (e.g., diet, exercise,
smoking); access to insurance and care; and physical, economic,
and social environment (e.g., physical context, demographic
composition). For example, lower socioeconomic status has been
estimated to be associated with lower levels of treatment and
a higher prevalence of chronic disease morbidity/mortality
(Uchino, Cacioppo, & Kiecolt-Glaser, 1996). Far more limited
information supports the influence of sociodemographic factors on
sleep duration (Hale, Peppard, & Young, 2007) and sleep disorders,
or on how these in turn are associated with or influence health
outcomes (Bliwise, King, Harris, & Haskell, 1992; Patel, 2007).
Sleep duration and quality may contribute to the increasing inci-
dence of chronic disease. The average sleep duration of US adults
appears to have fallen to nearly 7 h per night (Hale, 2005; Hale et al.,
2007; National Sleep Foundation, 2003; National Sleep Foundation,
2005; National Sleep Foundation, 2006; Roffwarg, Muzio, &
Dement, 1966). Insufficient sleep duration has been linked to
elevated body mass index (Hasler et al., 2004; Kohatsu et al., 2006),
weight gain (Patel et al., 2004; Patel & Hu, 2008), obesity (Cizza,
Skarulis, & Mignot, 2005; Gangwisch, Malaspina, Boden-Albala, &
Heymsfield, 2005; Taheri, Lin, Austin, Young, & Mignot, 2004;
Vioque, Torres, & Quiles, 2000), metabolic dysfunction (Karlsson,
Knutsson, Lindahl, & Alfredsson, 2003), and diabetes mellitus (Ayas
et al., 2003; Gottlieb et al., 2005; Hayashino et al., 2007; Knutson,
Ryden, Mander, & Van Cauter, 2006; Mallon, Broman, & Hetta,
2005; Yaggi, Araujo, & McKinlay, 2006). Modulations of cardiovas-
cular function by sleep duration have been demonstrated for eleva-
tions of blood pressure (Lusardi et al., 1999); (Meier-Ewert et al.,
2004), consistent with observed associations of sleep duration with
blood pressure (Gottlieb et al., 2006; Meier-Ewert et al., 2004; Sakata
Abbreviation: Body mass index, BMI; Cardiovascular disease, CVD; Coronary
heart disease, CHD; C-reactive protein, CRP; High blood pressure, HBP; National
Health Interview Survey, NHIS
*Corresponding author. Tel.: þ1 6175079177.
E-mail address: orfeu_buxton@hms.harvard.edu (O.M. Buxton).
1
Both authors contributed equally to this paper.
Contents lists available at ScienceDirect
Social Science & Medicine
journal homepage: www.elsevier.com/locate/socscimed
0277-9536/$ esee front matter Ó2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.socscimed.2010.05.041
Social Science & Medicine 71 (2010) 1027e1036
et al., 2003)andCVD(Wingard & Berkman, 1983). And short sleep
duration is associated with premature mortality (Dew et al., 2003;
Kripke, Garfinkel, Wingard, Klauber, & Marler, 2002; Mallon,
Broman, & Hetta, 2002; Patel et al., 2004; Wingard & Berkman, 1983).
Although seven-to-eight-hour sleep duration is generally
associated with the least health risk, results typically are based on
non-representative samples. Epidemiologic studies associate short
sleep with obesity and diabetes [reviewed in (Knutson & Van
Cauter, 2008; Spiegel, Knutson, Leproult, Tasali, & Van Cauter,
2005)], consistent with the results of laboratory studies of sleep
restriction or disruption (Spiegel, Leproult, L’Hermite-Balériaux,
et al., 2004; Spiegel, Leproult, & Van Cauter, 1999; Tasali, Leproult,
Ehrmann, & Van Cauter, 2008). Datasets with a sufficient number
of observations of long-sleeping individuals generally find an
elevated risk of disease or mortality, with socioeconomic status as
a strong correlate, leading to the notion that ‘too much’sleep is also
a health risk (Patel, 2007). The US National Health and Nutrition
Examination Survey (NHANES) I national cohort follow-up data
from the early 1990s revealed an association of long sleep duration
with stroke and coronary heart disease (Qureshi, Giles, Croft, &
Bliwise, 1997) and a cross-sectional association of short sleep
duration with increased obesity (Gangwisch et al., 2005). There is
a compelling need to identify the long-term risks to human health
involving sleep or sleep disorders (Institute of Medicine, 2006).
We test the hypothesis that both short (<¼6 h) and long (>¼9h)
self-reported sleep duration are positively associated with risk of
chronic diseases, obesity, diabetes, hypertension, and CVD using
a multilevel logistic regression approach influenced by developments
in social epidemiology (Berkman, 2000;Committee on Assessing
Interactions among Social, Behavioral, and Genetic Factors in Health,
2006;Evans & Stoddart, 1990; Marcelli & Heer, 1997; Studenmund,
2006:pp.449e451) that controls for other individual characteris-
tics, health behaviors other than sleep, health insurance coverage,
family environment, and geographic context.
Methods
Source of data
We merged the Person, Adult Sample, Household, and Family
files of the 2004e2005 US National Health Interview Survey (NHIS:
http://www.cdc.gov/nchs/nhis.htm) at the individual level to
estimate the association between short or long sleep and obesity,
diabetes (type 2), high blood pressure (HBP), or CVD among US
residents aged 18e85. Data were collected by approximately 400
interviewers using standard computer-assisted personal inter-
viewing (CAPI) procedures. Although the NHIS attempts to collect
data from all adult members of each randomly selected household
as part of the Family Core component, only one adult per family is
randomly selected for the Sample Adult questionnaire. This adult
responds directly unless he or she is physically or mentally
incapable, in which case a knowledgeable proxy is permitted to
answer (National Center for Health Statistics, 2006).
Of the 71,287 adults in the Person File in 2004, 31,263 remained
after merging with Adult Sample files. Of the 68,299 adults
included in the 2005 Person File, 31,383 remained, providing a total
initial sample size of 62,646. After dropping all observations for
those who did not provide a valid response for any variables in our
analysis (Table 1), our final sample (56,507) is weighted using
NHIS-provided person-level sample weights.
Statistical methods
Consistent with a multifactorial, multilevel model of health
(Evans & Stoddart, 1990), and building on recent work on the
demography of sleep by Hale and colleagues (Hale, 2005; Hale
et al., 2007), we control for factors grouped into four analytical
categories e(1) individual sociodemographic and economic
characteristics (i.e., age, gender, ethnoracial group, nativity, marital
status, educational attainment, labor force participation, health
insurance coverage as a measure of medical care access); (2)
individual behaviors (i.e., smoking, alcohol consumption, vigorous
exercise); (3) individual health conditions (i.e., psychological
distress); and (4) extra-individual family and regional context (i.e.,
college-educated family member; census region). Inclusion of
a variable for whether a family member has been graduated from
college is included because research has shown that family envi-
ronment can profoundly influence health and socioeconomic
behavior, and consequently long-term health. While such an
approach does not investigate the within- and between-group
variable interactions likely to influence our four health outcomes, it
does permit analysis of a random sample of the entire non-insti-
tutionalized U.S. adult civilian population to estimate the marginal
association of sleep on important chronic diseases. Although we
estimated how current smoking and moderate exercise were
associated with our four health outcomes, these are not included in
our final models because they were not statistically significant at
the 90 percent confidence level, and alternative measures for these
health behaviors were significant and included (e.g., vigorous
leisure-time physical activity, being a former smoker). Table 1 lists
the definitions, means, and standard deviations of all variables
included in the models. Importantly, although climate and other
place-based factors vary widely by geography and influence the
four outcomes we study in this article, the NHIS data only provide
regional-level identifiers. We control for these rather large aggre-
gated areas in our models, but recognize that it would be better to
have lower-level geographic identifiers.
We employ logistic regression to estimate whether respondents’
answers to the following question e“On average, how many hours
of sleep do you get in a 24-h period?”ewere associated with having
been obese, diabetic, or having had HBP or CVD. We created three
categorical variables from the sleep variable, enabling us to
examine whether those who sleep less or more than is conven-
tionally considered healthy were likely to have one of the four
health outcomes. Obesity is defined as a body mass index (BMI)
30 kg/m
2
. Our remaining three outcome variables are also
dichotomous and defined by respondents’answers to the question
“Have you EVER been told by a doctor or other health professional
that you had diabetes; HBP; or CVD (including coronary heart
disease (CHD), angina, heart attack, stroke or “other”heart condi-
tion)?”The list of variables for each of our models included
comorbidity factors incrementally, such that the diabetes model
included obesity, the HBP model included obesity and diabetes, and
the CVD model included obesity, diabetes, and HBP.
We estimated how short sleep (<7 h/night) and long sleep (>8
h/night) were associated with each health outcome, employing
a sociogeographic model to incorporate potentially important
social and behavioral components often overlooked in more
traditional clinically-focused notions of cardiometabolic disorders.
Since responses were in integers only, comparisons of short and
long sleep are made to self-reported seven or 8 h of sleep per night.
Regression results are displayed in two ways. First, they are
reported together in four separate columns and in conventional
summary format as estimated parameter coefficients (Table 2).
Second, we transformed each regression coefficient into a change in
the probability of an outcome equaling one as the result of a one-
unit change in an explanatory variable for each of our four
outcomes separately (Figs. 1e4). Each filled bar in these figures
represents an estimated relationship between an explanatory and
outcome variable that is statistically significant at the noted level
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e10361028
(see Table 2), and each empty bar represents an association that is
not statistically significant at this level. We report parameter
coefficients rather than odds ratios in Table 2, and probabilistic
changes in Figs. 1e4, for two straightforward reasons. First,
reporting parameter coefficients permits one to show the sign of
the estimated association between each explanatory variable and
the dummy dependent variable explicitly rather than inferring it
from whether an odds ratio is greater or less than one. Second,
regardless of whether mathematically identical odds ratios or
parameter coefficients are reported, researchers eventually employ
the language of probability, and there are three well-known
methods for converting coefficients into probabilities (Liao, 1994;
Pampel, 2000; Petersen, 1985, Studenmund, 2006: pp. 449e451).
Detailed significance levels for each explanatory variable, however,
are provided in Table 2, and were estimated using robust standard
errors (Huber, 1967: pp. 221e233).
Results
Approximately one-fourth (24.3%) of all adults who were 18-to-
85 years old in the United States in 2004 or 2005 according to NHIS
data are estimated to have been obese, and mean body mass index
was 27.2 kg/m
2
(Table 1). A similar proportion (25.2%) are estimated
to have been told by a doctor that they had high blood pressure,
about 13% had been told they had a cardiovascular disease, and
seven percent had been told they had diabetes. The mean age of all
adults was 45 years, almost half were male (49%), and fully 71%
were non-Latino white. Latinos represented 12.5% of the sample,
non-Latino blacks represented 11%, and relatively small propor-
tions were either non-Latino Asian (3.6%) or other (1.6%). Similar
proportions of U.S. adults were married or had been graduated
from high school (58%), and 26% had earned at least an under-
graduate college degree. Most (68.4%) were either employed or
actively seeking work.
Results concerning health behaviors and characteristics show
that most (69.2%) adults were covered by a private health insurance
plan, and only 16% did not have insurance. Most importantly for
purposes of the current study, more than a fourth (28.6%) reported
sleeping less than 7 h nightly, and slightly fewer than one in ten
(8.6%) reported sleeping more than 8 h nightly. Only 3.8% are
estimated to have been experiencing serious psychological distress
as measured by the well-known K-6 scale.
Two family-level constructs ehighest level of education
attained by a family member and homeownership ehelp us assess
Table 1
Descriptive Statistics, 2004e2005 National Health Interview Survey, USA
Variable definition
m
SD Min Max
Health outcome
OBESE Dummy ¼1ifBMI)>¼30 24.3% e01
BMI Height-adjusted body mass index ¼kg/m
2
27.2 5.9 9.9 88.6
DIABETES Dummy ¼1 if told by a doctor that (s)he has diabetes 7.3% e01
HBP Dummy ¼1 if told by a doctor that (s)he has hypertension 25.2% e01
CVD Dummy ¼1 if told by a doctor that (s)he has cardiovascular disease 12.8% e01
Demographic/risk factor
AGE Subject’s age in years since birth 45 17 18 85
AGESQ Subject’s age in years since birth, sqaured 2357 1736 324 7225
MALE Dummy ¼1 if subject is male 48.8% e01
LATINO Dummy ¼1 if subject's ethno-racial group ¼Latino (of any race) 12.5% e01
ASIAN Dummy ¼1 if subject’s ethno-racial group ¼Asian 3.6% e01
BLACK Dummy ¼1 if subject’s ethno-racial group ¼African American or black 10.9% e01
WHITE Dummy ¼1 if subject’s ethno-racial group ¼white 71.4% e01
OTHERG Dummy ¼1 if subject’s ethno-racial group ¼another or multiple race 1.6% e01
FOREIGNBORN Dummy ¼1 if subject was born outside of the USA 14.6% e01
MARRIED Dummy ¼1 if subject is married 57.7% e01
NOHIGHSCH Dummy ¼1 if subject did not graduate from high school 16.3% e01
HIGHSCH Dummy ¼1 if subject graduated from high school or earned GED 57.9% e01
COLLEGE Dummy ¼1 if subject graduated from college 25.9% e01
INLABFORCE Dummy ¼1 if subject is employed or actively seeking work 68.4% e01
Health behavior/characteristic
PRIVINSR Dummy ¼1 if subject is covered by a private health insurance plan 69.2% e01
PUBINSR Dummy ¼1 if subject is covered by a public health insurance plan 14.5% e01
NOINSR Dummy ¼1 if subject is not covered by a health insurance plan 16.3% e01
PASTSMKR Dummy ¼1 if subject ever smoked >¼100 cigarettes & currently does not smoke 21.6% e01
DRNKSPWK Average number of alcoholic drinks consumed per week during previous year 2.7 9.2 0 297
EXERVGPWK Number of times subject engaged in vigorous leisure-time physical acitivty each week 1.4 2.6 0 28
SLEEPLT7 Dummy ¼1 if average amount of sleep in a 24-our period <7 h 28.6% e01
SLEEP7_8 Dummy ¼1 if average amount of sleep in a 24-our period ¼7 or 8 h 62.8% e01
SLEEPGT8 Dummy ¼1 if average amount of sleep in a 24-our period >8 h 8.6% e01
DISTRESS Dummy ¼1 if Kessler 6-item, 24-point index >¼13 3.8% e01
Family environment
F_NOHIGHSCH Dummy ¼1 if adult family member with the most education was not a high school graduate 9.3% e01
F_HIGHSCH Dummy ¼1 if adult family member with the most education was a high school graduate 55.0% e01
F_COLLEGE Dummy ¼1 if adult family member with the most education was a college graduate 35.7% e01
F_HOMEOWN Dummy ¼1 if family member owns home in which subject resides 71.3% e01
Region/period
WEST Dummy ¼1 if subject resided in a western state 21.1% e01
MIDWEST Dummy ¼1 if subject resided in a mid-western state 24.6% e01
NORTHEAST Dummy ¼1 if subject resided in a northeastern state 18.3% e01
SOUTH Dummy ¼1 if subject resided in a southern state 36.0% e01
OBS2004 Dummy ¼1 if observation In 2004 National Health Interview Survey (NHIS) data 49.6% e01
OBS2005 Dummy ¼1 if observation in 2005 National Health Interview Survey (NHIS) data 50.4% e01
Note: Merged 2004 and 2005 Person, Adult Household and Family National Health Interview Survey (NHIS) data (N¼56,507).
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e1036 1029
whether family environment may have influenced the probability
of an individual having been obese, or having had diabetes, high
blood pressure or cardiovascular disease. Table 1 shows that
slightly more than one-third (35.7%) of all adults had a family
member who had earned a college degree, and more than half
(55%) had a family member who had graduated from high school.
Only about nine (9.3)% did not have a family member who had
either graduated from high school or college. A majority (71.3%),
however, resided in a home which someone in the family owned.
While more than one in five of all adult residents lived in
a Western (21.1%) or mid-Western (24.6%) state, a plurality resided
in the South (36%) and fewer than one in five resided in the
Northeast. Lastly, our sample is split evenly between 2004 (49.6%)
and 2005 (50.4%).
Measuring overall fit of a logit model is not as straightforward as
is the case when performing conventional ordinary least squares
(OLS) regression analysis. Percent concordant pairs ethat is, the
proportion of observations in a sample that a logit regression
equation explains correctly eis a standard method for character-
izing how robustly a logit model explains variation in a dichoto-
mous (“dummy”) dependent variable such as those we investigate
in this paper (Marcelli & Heer, 1997; Studenmund, 2006: pp.
449e451). Our models correctly classified respondents on each of
the four outcomes as follows: the percent concordant pairs for
obesity was 75.7%, for diabetes it was 92.7%, for HBP it was 79.5%,
and for CVD it was 87.6%. These overall model fit statistics, as well
as those for individual explanatory variables, are reported in
Table 2, but below we will focus our discussion on their probabi-
listic interpretation (Figs. 1e4).
Obesity
The relative (normalized) probability of having been obese
associated with a one-unit change in each explanatory variable is
depicted in Fig. 1. As can be seen, all but one variable (having had
private insurance) were statistically significant, and most were at
the 99 percent confidence level (p<0.01). Table 2 reports signifi-
cance levels for all explanatory variables. Respondents surveyed in
2005 had a slightly higher probability of having been obese than
those in 2004, reflecting the secular trend of increasing obesity rates.
Both short (6%) and long sleep (3%) duration were significantly
associated with obesity, relative to sleeping seven to 8 h. Residing in
the northeast or west was associated with a one and two percent
lower probability of having been obese, and residing in the Midwest
Table 2
Logistic regression results for obesity, diabetes, high blood pressure, and cardiovascular disease (using Robust Standard Errors)
OBESE DIABETES HBP CVD
ß RSE ß RSE ß RSE ß RSE
Demographic/risk factor
AGE 0.107 (0.008)
a
0.184 (0.014)
a
0.114 (0.003)
a
0.035 (0.008)
a
AGESQ 0.001 (0.000)
a
0.001 (0.000)
a
0.001 (0.000)
a
0.000 (0.000)
MALE 0.092 (0.038)
b
0.356 (0.033)
a
0.034 (0.044) 0.187 (0.055)
a
LATINO 0.381 (0.076)
a
0.421 (0.096)
a
0.112 (0.025)
a
0.435 (0.034)
a
ASIAN 0.780 (0.233)
a
0.688 (0.090)
a
0.364 (0.079)
a
0.398 (0.199)
b
BLACK 0.428 (0.038)
a
0.407 (0.074)
a
0.409 (0.022)
a
0.416 (0.022)
a
OTHERG 0.492 (0.063)
a
0.423 (0.057)
a
0.185 (0.064)
a
0.042 (0.177)
FOREIGNBORN 0.595 (0.066)
a
0.044 (0.107) 0.241 (0.085)
a
0.349 (0.051)
a
MARRIED 0.065 (0.024)
a
0.053 (0.054) 0.053 (0.026)
b
0.066 (0.024)
a
HIGHSCH 0.081 (0.024)
a
0.081 (0.027)
a
0.095 (0.033)
a
0.110 (0.079)
COLLEGE 0.348 (0.068)
a
0.094 (0.125) 0.142 (0.049)
a
0.241 (0.099)
b
INLABFORCE 0.081 (0.040)
b
0.542 (0.059)
a
0.227 (0.035)
a
0.472 (0.053)
a
Health behavior/characteristic
PRIVINSR 0.020 (0.034) 0.171 (0.109) 0.222 (0.059)
a
0.073 (0.048)
PUBINSR 0.148 (0.063)
b
0.307 (0.110)
a
0.320 (0.046)
a
0.287 (0.041)
a
PASTSMKR 0.224 (0.019)
a
0.126 (0.045)
a
0.088 (0.033)
a
0.299 (0.033)
a
DRNKSPWK 0.009 (0.001)
a
0.040 (0.008)
a
0.007 (0.001)
a
0.004 (0.003)
EXERVGPWK 0.058 (0.008)
a
0.032 (0.006)
a
0.024 (0.003)
a
0.021 (0.011)
b
3LEEPLT7 0.322 (0.056)
a
0.103 (0.042)
b
0.216 (0.039)
a
0.229 (0.035)
a
SLEEPGT8 0.136 (0.036)
a
0.314 (0.059)
a
0.098 (0.037)
a
0.255 (0.034)
a
DISTRESS 0.291 (0.060)
a
0.411 (0.142)
a
0.396 (0.059)
a
0.644 (0.089)
a
BMI ee0.099 (0.002)
a
0.088 (0.003)
a
0.004 (0.002)
c
DIEABETES eeee0.960 (0.020)
a
0.531 (0.021)
a
HBP eeeeee0.805 (0.039)
a
Family environment
F_COLLEGE 0.249 (0.046)
a
0.152 (0.109) 0.134 (0.029)
a
0.047 (0.050)
F_HOMEOWN 0.054 (0.012)
a
0.116 (0.025)
a
0.027 (0.030) 0.233 (0.034)
a
Region/period
WEST 0.094 (0.012)
a
0.278 (0.022)
a
0.137 (0.014)
a
0.053 (0.019)
a
MIDWEST 0.045 (0.010)
a
0.028 (0.014)
b
0.078 (0.004)
a
0.014 (0.005)
a
NORTHEAST 0.071 (0.008)
a
0.211 (0.008)
a
0.168 (0.011)
a
0.054 (0.007)
a
OBS2005 0.069 (0.035)
c
0.072 (0.027)
a
0.005 (0.018) 0.014 (0.040)
INTERCEPT 3.372 (0.205)
a
11.132 (0.376)
a
7.705 (0.084)
a
3.940 (0.279)
a
N (Weighted) 195,579,249 195,579,249 195,579,249 195,579,249
Percent Concordant Pairs 75.7% 92.7% 79.5% 87.6%
Note: Merged 2004 and 2005 Person, Adult, Household and Family National Health Interview Survey (NHIS) data (N¼56,507). Logistic regression results with robust standard
errors (RSE) statistically significant at the two-tailed p<0.01 (a), p<0.05 (b) or p<0.10 (c) test level.
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e10361030
was associated with a 2% higher probability of having been obese
than residing in the southern census region. Among family-level
factors, having a college-educated family member was associated
with a 5% lower probability of obesity. Being married was positively
associated with obesity (augmenting the probability of obesity by
nearly 2% relative to single status). Relatively exogenous individual-
level characteristics were stronglyassociated with the probability of
obesity; Asian heritage (relative to non-Latino white) and foreign-
born status (relative to US-born) conferred a 14% and 11% relative
reduction in the probability of obesity, respectively. Age was asso-
ciated with a one percent increase per year in the probability of
obesity, on average, but it is important to note that age is top-coded
at 85 years. Compared with females, males had a 1% greater proba-
bility of obesity. Exhibiting psychological distress was associated
with a 5% greater probability of being obese. Of other individual-
level factors, having a college or high school education was associ-
ated with a nearly 5% or 2% lower probability of being obese
(respectively) compared with not having graduated from high
school. Public health insurance coverage was associated with a 3%
increased probability of being obese compared with having no
insurance, and as noted above access to private insurance was not
significant.
Of individual health behaviors, regular vigorous exercise was
associated with a 2% reduction in the probability of obesity. Asso-
ciations with alcohol consumption were statistically significant but
nearly zero. Being a former smoker was associated with a 4%
increase in the probability of obesity, and being a current smoker
was an insignificant factor, relative to nonsmoker, in all of our
models. We therefore excluded the latter smoking metric from our
final models.
Diabetes
Fig. 2 depicts estimated associations between each explanatory
variable in our model and the probability of a diagnosis of type 2
diabetes. The model is identical to the obesity model, except that
BMI is included. And once again we show all estimated probabilistic
associations: empty bars indicate those that are estimated to be
statistically insignificant at the p<0.10 level, and all filled bars
except that concerning having resided in a mid-Western state
(MIDWEST, p<0.05) are significant at the 99 percent confidence
level (p<0.01). It is important to emphasize that both short and
long sleep were adversely and strongly associated with obesity, and
that this ein addition to the relatively large direct independent
effect sleep is estimated to have on diabetes (reported below) e
represents an potential indirect effect of sleep on being diagnosed
with diabetes. Respondents surveyed in 2005 were more likely to
have had diabetes than respondents in 2004, reflecting the secular
trend of increasing rates of diabetes diagnosis (0.5%), independent
of all other variables. Relative to the obesity model, several signif-
icant factors became insignificant in the diabetes model (having e
or having a family member with ea college education, being
currently married, and being foreign-born). In addition, the influ-
ence of individual factors is generally reduced, and two factors
change direction (living in the Midwest from positive to negative
association, and Asian heritage from strong negative to positive
association). Both short and long sleeping are positively associated
with the probability of a diabetes diagnosis.
In general, it also appears that residing in a state that is not in
the South is associated with a reduced probability of having had
diabetes (2%, 1%, and <1%, respectively). Of the family environment
-15% -5% 5% 15% 25% 35%
live in the west
live in the northeast
live in the midwest
family member w/college education
family own home
married
2005 survey (vs 2004)
asian
foreign-born
age
male
psychological distress
latino
black
other ethnoracial group
college education
labor force participant
high school education (no college)
private health insurance
public health insurance
regular vigorous exercise
number of drinks/week
long sleep duration >8 hours/night
former smoker
short sleep duration <7 hours/night
Probability of obesity
less likely <-- --> more likely
Census Region
Family
Environment
Time Control
Individual-level
Characteristics
Individual
Health
Behaviors
Individual-level
factors (other)
Fig. 1. Logistic Regression of Having Been Obese among Adults on Geographic and Family Environment factors, Individual Characteristics and Behaviors, and Individual Health
Behaviors, United States, 2004e2005 (NHIS). Obesity is defined by a Body Mass Index >30 kg/m
2
. Probability estimates are normalized to the same scale by converting beta
coefficients to probabilities (see Methods). All statistically significant factors (p<0.05) are shown, and depicted by filled bars. The Control of 2005 Survey (vs 2004) had p<0.10. The
variable AGESQ was associated with a 34.2% probability, or a lesser probability of obesity.
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e1036 1031
factors, only someone in the respondent’s family owning a home
was associated with a reduced probability of having had diabetes.
Individual characteristics were most strongly associated with
increased probability of having diabetes: Latino and non-Latino
Asian, black, and other (Native American) ethnoracial groups were
more likely to have had diabetes (5%, 3%, 3%, and 3%, respectively)
than non-Latino whites.
Age was associated with a slightly higher probability of having
diabetes (<1%). Males were 2% more likely than females to have
diabetes. Of the other individual-level factors, labor force partici-
pation and high school education were negatively associated with
the probability of having diabetes (4% and 0.5%, respectively),
whereas access to public insurance was positively associated with
the probability of diabetes (2%) and private insurance was insig-
nificant). Of the individual health behaviors, the amount of exercise
per week and number of alcohol drinks per week were positively
associated with a reduced likelihood of diabetes, though magnitude
was small (0.2%, 0.1%). Former smoker status, long sleep duration,
and short sleep duration were significantly associated with an
increased probability of diabetes (2%, 1% and 1%, respectively).
High blood pressure
Fig. 3 depicts all estimated associations between all explanatory
variables and the probability of a having received an HBP diagnosis.
All filled bars are significant at the p<0.01 level except that
reported for having been married (p<0.05). In this model, there
was no appreciable change from 2004 to 2005 surveys in the
probability of HBP, independent of other factors. The highest
likelihood of HBP (11%) was conferred by a diabetes diagnosis. As in
the obesity model, the effects on risk of HBP of both short and long
sleep were statistically significant relative to sleeping 7e8 h and
had the largest positive associations of Individual Health Behaviors
(2.4%, 1.1%, respectively).
As was the case with diabetes, residing outside of the South, was
associated with a reduced probability of having HBP (1.9%, 1.5%,
0.9%, respectively). Of the family environment-level factors,
residing in a family-owned home and having a family member with
a college education were associated with a reduced probability of
HBP. Among individual characteristics, being foreign-born was
associated with a 2.7% lower probability of having HBP than being
native-born. Ethnoracial groups exhibited some changes in relative
risks compared with the obesity and diabetes model. Latinos were
less likely to have HBP (1.3%) than black, Asian, and Other (e.g.,
Native American) ethnoracial groups, who had a greater likelihood
of HBP (4.6%, 4.1%, 2.1%, respectively) than non-Latino whites.
Psychological distress was also associated with an increased
likelihood of HBP (4.4%), as was age (0.4%/year). Of the other
individual-level factors, labor force participation and having
attended college or at least high school reduced the likelihood of
HBP. Obesity derived from BMI was slightly positively associated
with HBP. Access to private or public health insurance conferred an
elevated risk of HBP. Diabetes diagnosis was strongly associated
with an HBP diagnosis (10.7%).
Among individual health behaviors, greater regular vigorous
exercise was associated with only a slightly reduced risk of HBP;
and former smoker status was associated with an increased
probability.
-15% -5% 5% 15% 25% 35%
live in the west
live in the northeast
live in the midwest
family member w/college education
family own home
married
2005 survey (vs 2004)
BMI
asian
foreign-born
age
male
psychological distress
latino
black
other ethnoracial group
college education
laborforce participant
high school education (not college)
public insurance
regular vigorous exercise
number of drinks/week
long sleep duration >8 hours/night
former smoker
short sleep duration <7 hours/night
Probability of diabetes
Census Region
Family
Environment
Time Control
Individual
Characteristics
Individual
Health
Behaviors
Individual-level
factors (other)
less likely <-- --> more likely
Fig. 2. Logistic Regression of Having Diabetes among Adults on Geographic and Family Environment factors, Individual Characteristics and Behaviors, and Individual Health
Behaviors, United States, 2004e2005 (NHIS). Diabetes defined from respondent self-report of a physician diagnosis. Probability estimates are normalized to the same scale by
converting beta coefficients to probabilities (see Methods). All statistically significant factors (p<0.05) are shown, and depicted by filled bars; open symbols represent non-
significant variables retained from earlier models for comparison. The variable AGESQ was associated with a 14.7% probability, or a lesser probability of obesity.
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e10361032
Cardiovascular disease
Fig. 4 depicts estimated associations with all explanatory
variables and the probability of having been diagnosed with a CVD.
The model is identical to the HBP model, except that HBP is now
included. And all filled bars are significant at the 99% confidence
level except the following four: ASIAN, COLLEGE, and EXERVGPWK
(p<0.05), and BMI (p<0.10). The highest likelihood of CVD was
associated with having diabetes (10%) or HBP (9%). In general, the
CVD model results were highly similar to the results for the HBP
model, except for access to private health insurance, which was not
significant for CVD. Both short and long sleeping were indepen-
dently associated with an increased probability of having a CVD
diagnosis (2.5% and 1.1%, respectively).
Discussion
In all models, compared with sleeping 7e8 h/night, both short
and long sleep were significantly associated with the probability of
obesity, diabetes, HBP, and CVD at the 99% confidence level. Our
logistic regression estimates simultaneously control for all variables
shown in Table 1 and for sample clustering at the regional level to
provide relatively conservative estimates (using robust standard
errors). However, NHIS data are cross-sectional and self-reported;
thus we are unable to estimate the extent to which sleep is influ-
enced by (rather than influencing) our four health outcomes. That
said, although many individual-level characteristics and behaviors
were significantly associated with at least some of the four
outcomes, some were not significant in all models (open bars Figs.
1e4); short or long sleep duration were both significant in all
models. In most cases, the relationships between short or long
sleep duration and these four outcomes were even stronger than
better-established predictors of these chronic metabolic/CVD
outcomes. Sleep duration measures were also strongly associated
with various health risks relative to family and geographic
variables. Importantly, we are not suggesting that family and spatial
variables viewed collectively are more important than all behav-
iors. Our estimates simply suggest that sleep has a relatively larger
association with our four health outcomes than particular variables
within other analytical categories. Finally, we observed not only
direct and independent relationships between sleep duration and
chronic disease; we also observed an indirect relationship,
for example, on CVD, in that obesity, diabetes, and HBP were
significantly associated with CVD, and sleep was independently
associated with each. Increasing the proportion of the population
achieving 7e8 h of daily sleep may reduce chronic disease risk in US
adults, but only to the extent that reverse causation is absent.
As described in time-use studies (1965, 1975, 1985, 1999), mean
sleep duration was 7.7 h, somewhat longer than the current esti-
mate of 7.2 h (Hale, 2005; Hale et al., 2007). Here, the prevalence of
short and long sleeping (29% and 9%, respectively; see Table 1)was
highly concordant with estimates from the above time-use diary
-5% 0% 5% 10% 15% 20%
live in the northeast
live in the west
live in the midwest
family member w/college education
married
family own home
2005 survey (vs 2004)
foreign-born
latino
male
age
other ethnoracial group
asian
psychological distress
black
labor force participant
college education
high school education (no college)
Body Mass Index (BMI)
private health insurance
public health insurance
Diabetes
regular vigorous exercise
number of drinks/week
former smoker
long sleep duration >8 hours/night
short sleep duration <7 hours/night
Probability of high blood pressure
less likely <-- --> more likely
Census Region
Family
Environment
Time Control
Individual
Characteristics
Individual
Health
Behaviors
Individual-level
factors (other)
Fig. 3. Logistic Regression of Having High Blood HBP among Adults on Geographic and Family Environment factors, Individual Characteristics and Behaviors, and Individual Health
Behaviors, United States, 2004e2005 (NHIS). High blood HBP defined from respondent self-report of a physician statement. Probability estimates are normalized to the same scale
by converting beta coefficients to probabilities (see Methods). All statistically significant factors (p<0.05) are shown, and depicted by filled bars; open symbols represent non-
significant variables retained from earlier models for comparison). The variable AGESQ was associated with a 10.7% probability, or a lesser probability of obesity.
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e1036 1033
study [24.6% and 9.5%, respectively; (Hale, 2005; Hale et al., 2007)]
but discordant with estimates from another sample [14.4% and
23.9%, respectively; (Basner et al., 2007; Hale & Do, 2007)]. Rather
than considering short or long sleeping separately, ’extremes of
sleep duration’may be a more useful designation of sleep-related
predictors of chronic disease. In the current analyses, short and
long sleeping are consistently associated with increased risk of
chronic disease independent of individual characteristics and
socioeconomic factors.
Multiple known obesity risk factors were identified in our study,
but how might sleep duration independently influence body
weight? Insufficient sleep duration has been linked to higher BMI
(Hasler et al., 2004; Kohatsu et al., 2006), weight gain (Patel et al.,
2004; Patel & Hu, 2008), and obesity (Cizza et al., 2005;
Gangwisch et al., 2005; Taheri et al., 2004; Vioque et al., 2000). A
laboratory-based study in young men controlling for food intake by
means of a constant, eucaloric intravenous infusion revealed that
sleep-restricted subjects reported greater hunger and physiologic
drive for eating in the form of lower blood levels of leptin and higher
ghrelin (Spiegel, Leproult, et al., 2004). In a cross-sectional sample,
short sleep duration was similarly associated with lower leptin and
higher ghrelin levels, as well as higher BMI (Taheri et al., 2004). These
data support the hypothesis that short sleep duration may influence
weight by increasing the drive to eat. This has been extended in
a recent clinical study in middle-aged subjects imposing 5.5 h per
night compared to 8.5 h per night of time in bed in a crossover design.
With ad libitum palatable meals, ghrelin and leptin measures of
hunger and satiety were the same in both conditions because the
subjects chose to eat on average about 200 kcal more of food in the
form of snacks, supporting a role for adequate sleep duration in the
control of eating behavior (Nedeltcheva et al., 2009).
Furthermore, physiologic changes in metabolism due to sleep
restriction in young men include alterations of glucose metabolism
and hormonal changes (Spiegel et al., 2000; Spiegel, Leproult,
L’Hermite-Balériaux, et al., 2004; Spiegel, Tasali, Penev, & Van
Cauter, 2004; Spiegel et al., 1999; Spiegel, Tasali, et al., 2004) that
would tend to increase the likelihood of obesity, visceral adiposity,
and diabetes in the long term. Far less is understoodabout the role of
long sleeping in the development of obesity, above and beyond the
impact of sleep-disordered breathing, depression, or side effects of
medication that cause both an increase in weight and sleep
duration.
Epidemiologic studies support associations between short sleep
duration (Ayas et al., 2003; Mallon et al., 2005; Yaggi et al., 2006),
long sleep duration (Yaggi et al., 2006), insomnia (Nilsson, Roost,
Engstrom, Hedblad, & Berglund, 2004), and sleep-disordered
breathing (Punjabi et al., 2004) and type 2 diabetes. Interestingly, the
causal pathway between short sleep and diabetes may involve
weight gain, as concluded by Ayas et al.: the “association between
a reduced self-reported sleep duration and diabetes diagnosis could
be due to confounding by BMI, or sleep restriction may mediate its
effects on diabetes through weight gain. Sleep restriction may be an
-5% 0% 5% 10% 15% 20%
live in the northeast
live in the west
live in the midwest
family member w/college education
married
family own home
2005 survey (vs 2004)
foreign-born
latino
male
age
other ethnoracial group
asian
psychological distress
black
labor force participant
college education
high school education (no college)
Body Mass Index (BMI)
private health insurance
public health insurance
High blood pressure
Diabetes
regular vigorous exercise
number of drinks/week
former smoker
long sleep duration >8 hours/night
short sleep duration <7 hours/night
Probability of cardiovascular disease (CVD)
less likely <-- --> more likely
Census Region
Family
Environment
Time Control
Individual
Characteristics
Individual
Health
Behaviors
Individual-level
factors (other)
`
Fig. 4. Logistic Regression of Having Cardiovascular Disease among Adults on Geographic and Family Environment factors, Individual Characteristics and Behaviors, and Individual
Health Behaviors, United States, 2004e2005 (NHIS). CVD defined from respondent self-report of a physician diagnosis of cardiovascular and circulatory diseases including stroke.
Probability estimates are normalized to the same scale by converting beta coefficients to probabilities (see Methods). All statistically significant factors (p<0.05) are shown (family
own a home F_HOMEOWN p<0.10), and depicted by filled bars; open symbols represent non-significant variables retained from earlier models for comparison. The variable AGESQ
was associated with a 10.5% probability, or a lesser probability of obesity.
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e10361034
independent risk factor for developing diabetes.”(Ayas et al., 2003)
This mediation via BMI has not been detected in all cross-sectional
studies, (e.g., Yaggi et al., 2006), the relative risk of diabetes in short
sleepers (6 h/night) was not substantially modified by BMI. Labo-
ratory studies of sleep restriction described above include changes
in glucose metabolism that increase diabetes risk. It has been
hypothesized that the relationship between sleep duration and
metabolic disorders may be related to chronic inflammation
markers, as has been identified for the inflammatory marker
C-reactive protein (CRP) in a laboratory study (Gottlieb et al., 2006;
Meier-Ewert et al., 2004; Sakata et al., 2003) and a cross-sectional
study (Williams, Hu, Patel, & Mantzoros, 2007). Furthermore, in
a cross-sectional study of diabetics excluding those with frequent
pain, both self-reported sleep duration and sleep quality were
related to severity of diabetes, with glycemic control indexed by
HbA1c levels (Ayas et al., 2003; Gottlieb et al., 2005; Hayashino et al.,
2007; Knutson et al., 2006; Mallon et al., 2005; Yaggi et al., 2006).
We found consistent associations between extremes of sleep
duration and both HBP and CVD (controlling for HBP). Laboratory
studies have revealed that restricting sleep duration semichroni-
cally increases resting blood pressure, CRP levels (Meier-Ewert
et al., 2004), cortisol levels (Spiegel, Leproult, L’Hermite-
Balériaux, et al., 2004; Spiegel, Tasali, et al., 2004;Spiegel et al.,
1999;Tasali et al., 2008) and sympathetic nervous system activity
(Spiegel, Leproult, L’Hermite-Balériaux, et al., 2004;Spiegel et al.,
1999; Spiegel, Tasali, et al., 2004;Tasali et al., 2008). Less infor-
mation is available to address the direction of causality for longer
sleep duration, independent of sleep disorders such as sleep apnea
or psychiatric disorders such as depression that have been shown
to lead to chronic disease. Consistent with findings that familial
sources of social support appear to be related to better blood
pressure regulation (Uchino et al., 1996), we observed that several
family-level factors (e.g., college-education family member, being
married) are positively associated with the absence of HBP. Affec-
tive and anxiety disorders and chronic stress are related to CVD risk
(Rozanski, Blumenthal, & Kaplan, 1999). Lifestyle factors are
routinely addressed clinically, but psychosocial factors are less
commonly appreciated (Rozanski, Blumenthal, Davidson, Saab, &
Kubzansky, 2005). Area-level factors, socioeconomic status, and
education were associated with CVD markers (Metcalf et al., 2008),
but we noted an independent association of sleep duration and CVD
risk (Williams et al., 2007).
Sleep disorders that disrupt sleep continuity (e.g., sleep-disor-
dered breathing) have been linked increased risk of metabolic
(Punjabi et al., 2004) and cardiovascular dysfunction (Mehra et al.,
2006) and also predict early mortality (Dew et al., 2003; Kripke
et al., 2002; Mallon et al., 2002; Patel et al., 2004; Wingard &
Berkman, 1983). Insomnia, a sleep disorder that reduces sleep
quality and duration, has been linked to insulin resistance and
glucose intolerance (Punjabi et al., 2004), diabetes (Gottlieb et al.,
2005; Hasler et al., 2004), and early mortality (Dew et al., 2003;
Sakata et al., 2003).
Because these data are self-reported and not directly measured,
it is possible that systematic bias has influenced data on sleep
duration (Lauderdale, Knutson, Yan, Liu, & Rathouz, 2008), many of
the predictor variables, or even the presence or absence of the
chronic disease outcome. Diabetes, hypertension, and CVD may go
undiagnosed for years before frank symptoms elicit a diagnosis.
Although the percentage of concordant pairs reflected the robust
predictive capacity of the variables used, uncaptured variance
remains, and more proximate measures might better explain the
probability of the outcome, particularly if those measures were
objectively assessed. For example, sleep-disordered breathing
might explain HBP or CVD better than sleep duration alone in the
long-sleeping group. Self-reported sleep duration is a limitation
compared to objective measurements, recall bias is a possibility,
and responses were limited (fractions of an hourwere not allowed),
which likely reduced sensitivity. No questions enabled estimation
of sleep disorders, especially sleep-disordered breathing and
insomnia, prevalent disorders that might influence sleep duration.
The associations we observed may be evidence of reverse causality;
respondents with diabetes, cardiovascular, hypertension and
obesity may suffer from disturbed sleep that leads to either shorter
duration of sleep or longer sleep duration as a result of fragmen-
tation of sleep (Al Delaimy, Manson, Willett, Stampfer, & Hu, 2002;
Zee & Turek, 2006). NHIS does not include dietary behaviors that
play a role in chronic disease risk.
A clear need exists for consistent and accurate sleep duration
data collection on future NHIS and other representative samplings
of population health, including adults as well as adolescents, and
sleep disorder information. Longitudinal assessment of sleep
duration and quality, in association with objectively-measured
health status, are urgent priorities for future research (Institute of
Medicine, 2006).
Acknowledgements
This work was supported by a pilot grant to both authors from
the RWJ Health & Society Program, Harvard School of Public Health.
For their assistance with the data analysis and manuscript prepa-
ration, we thank Louisa Holmes, Keith Malarick, Vanessa Castro,
and Andrea Muirhead. Relationships with corporate entities: These
relationships have no presumed overlap with this manuscript, did
not fund this work, and are not benefited by the manuscript, but are
listed in the interest of complete disclosure. These relationships
include Investigator-initiated research support from Cephalon Inc.
(>2 years ago) and Sepracor Inc. (current and within the last 24
months); an unrestricted educational grant from Takeda Pharma-
ceuticals (>3 years ago); Consulting fees and/or honoraria from
Dinsmore LLC (2010), Sepracor Inc.(2008 and earlier), and Takeda
Pharmaceuticals (2008 and earlier).
References
Al Delaimy, W. K., Manson, J. E., Willett, W. C., Stampfer, M. J., & Hu, F. B. (2002).
Snoring as a risk factor for type II diabetes mellitus: a prospective study.
American Journal of Epidemiology, 155(5), 387e393.
Ayas, N. T., White, D. P., Al-Delaimy, W. K., Manson, J. E., Stampfer, M. J., Speizer, F. E.,
et al. (2003). A prospective study of self-reported sleep duration and incident
diabetes in women. Diabetes Care, 26(2), 380e384.
Basner, M., Fomberstein, K. M., Razavi, F. M., Banks, S., William, J. H., Rosa, R. R., et al.
(2007). American time use survey: sleep time and its relationship to waking
activities. Sleep, 30(9), 1085e1095.
Berkman, L. (2000). Social integration, social networks, social support, and health.
In L. Berkman (Ed.), Social epidemiology. New York, NY: Oxford University Press.
Bliwise, D. L., King, A. C., Harris, R. B., & Haskell, W. L. (1992). Prevalence of self-
reportedpoor sleep in a healthy population aged50e65. Social Science & Medicine,
34(1), 49e55.
Cizza, G., Skarulis, M., & Mignot, E. (2005). A link between short sleep and obesity:
building the evidence for causation. Sleep, 28(10), 1217e1220.
Committee on Assessing Interactions among Social, Behavioral, and Genetic Factors
in Health. (2006). In Lyla M. Hernandez, & Dan G. Blazer (Eds.), Genes, behavior,
and the social environment: Moving beyond the nature/nurture debate. Wash-
ington, D.C: The National Academy Press.
Dew, M. A., Hoch, C. C., Buysse, D. J., Monk, T. H., Begley, A. E., Houck, P. R., et al.
(2003). Healthy older adults’sleep predicts all-cause mortality at 4e19 years of
follow-up. Psychosomatic Medicine, 65(1), 63e73.
Evans, R. G., & Stoddart, G. L. (1990). Producing health, consuming health care. Social
Science & Medicine, 31(12), 1347e1363.
Gangwisch, J. E., Malaspina, D., Boden-Albala, B.,& Heymsfield, S. B. (2005). Inadequate
sleepas a risk factor forobesity: analysesof the NHANESI. Sl eep, 28(10),1289e1296.
Gottlieb, D. J., Punjabi, N. M., Newman, A. B., Resnick, H. E., Redline, S.,
Baldwin, C. M., et al. (2005). Association of sleep time with diabetes mellitus
and impaired glucose tolerance. Archives of Internal Medicine, 165(8), 863e867.
Gottlieb, D. J., Redline, S., Nieto, F. J., Baldwin, C. M., Newman, A. B., Resnick, H. E.,
et al. (2006). Association of usual sleep duration with hypertension: the sleep
heart health study. Sleep, 29(8), 1009e1014.
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e1036 1035
Hale, L. (2005). Who has time to sleep? Journal of Public Health (Oxford, England), 27
(2), 205e211.
Hale, L., & Do, D. P. (2007). Racial differences in self-reports of sleep duration in
a population-based study. Sleep, 30(9), 1096e1103.
Hale, L., Peppard, P. E., & Young, T. (2007). Does the demography of sleep contribute
to health disparities? In D. Leger, & S. R. Pandi-Perumal (Eds.), Sleep disorders,
their impact on public health (pp. 1e17) London: Informa Healthcare.
Hasler, G., Buysse, D. J., Klaghofer, R., Gamma, A., Ajdacic, V., Eich, D., et al. (2004).
The association between short sleep duration and obesity in young adults:
a 13-year prospective study. Sleep, 27(4), 661e666.
Hayashino, Y., Fukuhara, S., Suzukamo, Y., Okamura, T., Tanaka, T., & Ueshima, H.
(2007). Relation between sleep quality and quantity, quality of life, and risk of
developing diabetes in healthy workers in Japan: the high-risk and population
strategy for occupational health promotion (HIPOP-OHP) study. BMC Public
Health, 7, 129.
Huber, P. J. (1967). The behavior of maximum likelihood estimates under nonstandard
conditions. Fifth Berkeley symposium of mathematical statistics and probability.
Berkeley, CA: University of California Press.
Institute of Medicine. (2006). In H. R. Colten, & B. M. Alteveogt (Eds.), Sleep disorders
and sleep deprivation: An unmet public health problem (pp. 1e500). Washington,
D.C: National Academies Press.
Karlsson, B. H., Knutsson, A. K., Lindahl, B. O., & Alfredsson, L. S. (2003). Metabolic
disturbances in male workers with rotating three-shift work. Results of the
WOLF study. International Archives of Occupational and Environmental Health, 76
(6), 424e430.
Knutson, K. L., Ryden, A. M., Mander, B. A., & Van Cauter, E. (2006). Role of sleep
duration and quality in the risk and severity of type 2 diabetes mellitus. Archives
of Internal Medicine, 166(16), 1768e1774 .
Knutson, K. L., & Van Cauter, E. (2008). Associations between sleep loss and
increased risk of obesity and diabetes. Annals of the New York Academy of
Sciences, 1129, 287e304.
Kohatsu, N. D., Tsai, R., Young, T., Vangilder, R., Burmeister, L. F., Stromquist, A. M.,
et al. (2006). Sleep duration and body mass index in a rural population. Archives
of Internal Medicine, 166(16), 1701e1705.
Kripke, D. F., Garfinkel, L., Wingard, D. L., Klauber, M. R., & Marler, M. (2002).
Mortality associated with sleep duration and insomnia. (pp. 131e136).
Lauderdale, D. S., Knutson, K. L., Yan, L. L., Liu, K., & Rathouz, P. J. (2008). Self-reported
and measuredsleep duration:how similar are they?Epidemiology,19(6),838e845.
Liao, T. F. (1994). Interpreting probability models: Logit, probit, and other generalized
linear models. Thousand Oaks, CA: Sage Publications.
Lusardi, P., Zoppi, A., Preti, P., Pesce, R. M., Piazza, E., & Fogari, R. (1999). Effects of
insufficient sleep on blood pressure in hypertensive patients: a 24-h study.
American Journal of Hypertension, 12(1 pt.1), 63e68.
Mallon, L., Broman, J. E., & Hetta, J. (2002). Sleep complaints predict coronary artery
disease mortality in males: a 12-year follow-up study of a middle-aged Swedish
population. Journal of Internal Medicine, 251(3), 207e216.
Mallon, L., Broman, J. E., & Hetta, J. (2005). High incidence of diabetes in men with
sleep complaints or short sleep duration: a 12-year follow-up study of
a middle-aged population. Diabetes Care, 28(11), 2762e2767.
Marcelli, E. A., & Heer, D. M. (1997). Unauthorized Mexican workers in the 1990 Los
Angeles County labour force. International Migration (Geneva, Switzerland), 35
(1), 59e83.
McGinnis, J. M., & Foege, W. H. (1993). Actual causes of death in the United States.
Journal of the American Medical Association, 270(18), 2207e2212.
Mehra, R., Benjamin, E. J., Shahar, E., Gottlieb, D. J., Nawabit, R., Kirchner, H. L., et al.
(2006). Association of nocturnal arrhythmias with sleep-disordered breathing:
the sleep heart health study. American Journal of Respiratory and Critical Care
Medicine, 173(8), 910e916.
Meier-Ewert, H. K., Ridker, P. M., Rifai, N., Regan, M. M., Price, N. J., Dinges, D. F., et al.
(2004). Effect of sleep loss on C-reactive protein, an inflammatory marker of
cardiovascular risk. Journal of the American College of Cardiology, 43(4), 678e683.
Metcalf, P. A., Scragg, R. R., Schaaf, D., Dyall, L., Black, P. N., & Jackson, R. T. (2008).
Comparison of different markers of socioeconomic status with cardiovascular
disease and diabetes risk factors in the diabetes, heart and health survey. The
New Zealand Medical Journal, 121(1269), 45e56.
National Center for Health Statistics. (2006). In U.S.D.o.H.a.H.S., & Centers for
Disease Control and Prevention. (Eds.), 2005 National Health Interview Survey
(NHIS) public use data release. Hyattsville, MD.
National Sleep Foundation. (2003). Executive summary of the 2003 “Sleep in
America”poll.
National Sleep Foundation. (2005). Executive summary of the 2005 “Sleep in
America”poll.
National Sleep Foundation. (2006). In WB&A Market Research. (Ed.), Summary of
findings. 2006 sleep in America Poll (pp. 1e76), Washington, DC.
Nedeltcheva, A. V., Kilkus, J. M., Imperial, J., Kasza, K., Schoeller, D. A., & Penev, P. D.
(2009). Sleep curtailment is accompanied by increased intake of calories from
snacks. The American Journal of Clinical Nutrition, 89(1), 126e133.
Nilsson, P. M., Roost, M., Engstrom, G., Hedblad, B., & Berglund, G. (2004). Incidence
of diabetes in middle-aged men is related to sleep disturbances. Diabetes Care,
27(10), 2464e2469.
Ogden, C. L., Carroll, M. D., Curtin, L. R., McDowell, M. A., Tabak, C. J., & Flegal, K. M.
(2006). Prevalence of overweight and obesity in the United States, 1999e2004.
Journal of the American Medical Association, 295(13), 1549e1555.
Ogden, C. L., Yanovski, S. Z., Carroll, M. D., & Flegal, K. M. (2007). The epidemiology
of obesity. Gastroenterology, 132(6), 2087e2102.
Pampel, F. C. (2000). Logistic regression: A primer. Thousand Oaks, CA: Sage Publi-
cations, Inc.
Patel, S. R. (2007). Social and demographic factors related to sleep duration. Sleep,
30(9), 1077e1078.
Patel, S. R., Ayas, N. T., Malhotra, M. R., White, D. P., Schernhammer, E. S.,
Speizer, F. E., et al. (2004). A prospective study of sleep duration and mortality
risk in women. Sleep, 27(3), 440e444.
Patel, S. R., & Hu, F. B. (2008). Short sleep duration and weight gain: a systematic
review. Obesity (Silver Spring, Md.), 16(3), 643e653.
Petersen, T. (1985). A comment on presenting results from logit and probit models.
American Sociological Review, 50(1), 130e131.
Punjabi, N. M., Shahar, E., Redline, S., Gottlieb, D. J., Givelber, R., & Resnick, H. E.
(2004). Sleep-disordered breathing, glucose intolerance, and insulin resis-
tance: the sleep heart health study. American Journal of Epidemiology, 160(6),
521e530.
Qureshi, A. I., Giles, W. H., Croft, J. B., & Bliwise, D. L. (1997). Habitual sleep patterns
and risk for stroke and coronary heart disease: a 10-year follow-up from
NHANES I. Neurology, 48, 904e911.
Roffwarg, H. P., Muzio, J. N., & Dement, W. C. (1966). Ontogenetic development of
the human sleep-dream cycle. Science, 152, 604e619.
Rozanski, A., Blumenthal, J. A., Davidson, K. W., Saab, P. G., & Kubzansky, L. (2005).
The epidemiology, pathophysiology, and management of psychosocial risk
factors in cardiac practice: the emerging field of behavioral cardiology. Journal
of the American College of Cardiology, 45(5), 637e651.
Rozanski, A., Blumenthal, J. A., & Kaplan, J. (1999). Impact of psychological factors on
the pathogenesis of cardiovascular disease and implications for therapy.
Circulation, 99(16), 2192e2217.
Sakata, K., Suwazono, Y., Harada, H., Okubo, Y., Kobayashi, E., & Nogawa, K. (2003).
The relationship between shift work and the onset of hypertension in male
Japanese workers. Journal of Occupational and Environmental Medicine/American
College of Occupational and Environmental Medicine, 45(9), 1002e1006.
Spiegel, K., Knutson, K., Leproult, R., Tasali, E., & Van Cauter, E. (2005). Sleep loss:
a novel risk factor for insulin resistance and type 2 diabetes. Journal of Applied
Physiology, 99(5), 2008e2019.
Spiegel, K., Leproult, R., Colecchia, E. F., L’Hermite-Balériaux, M., Nie, Z.,
Copinschi, G., et al. (2000). Adaptation of the 24-h growth hormone profile to
a state of sleep debt. American Journal of Physiology, 279, R874eR883.
Spiegel, K., Leproult, R., L’Hermite-Balériaux, M., Copinschi, G., Penev, P. D., & Van
Cauter, E. (2004). Leptin levels are dependent on sleep duration: relationships
with sympathovagal balance, carbohydrate regulation, cortisol, and thyrotropin.
The Journal of Clinical Endocrinology and Metabolism, 89(11), 5762e5771.
Spiegel, K., Leproult, R., & Van Cauter, E. (1999). Impact of sleep debt on metabolic
and endocrine function. Lancet, 354, 1435e1439.
Spiegel, K., Tasali, E., Penev, P., & Van Cauter, E. (2004). Brief communication: sleep
curtailment in healthy young men is associated with decreased leptin levels,
elevated ghrelin levels, and increased hunger and appetite. Annals of Internal
Medicine, 141(11), 846e850.
Studenmund, A. H. (2006). Using econometris: A practical guide. Boston, MA:
Pearson.
Taheri, S., Lin, L., Austin, D., Young, T., & Mignot, E. (2004). Short sleep duration is
associated with reduced leptin, elevated ghrelin, and increased body mass
index. PLoS Medicine, 1(3), e62.
Tasali, E., Leproult, R., Ehrmann, D. A., & Van Cauter, E. (2008). Slow-wave sleep and
the risk of type 2 diabetes in humans. Proceedings of the National Academy of
Sciences of the United States of America, 105(3), 1044e1049 .
Uchino, B. N., Cacioppo, J. T., & Kiecolt-Glaser, J. K. (1996). The relationship between
social support andphysiological processes:a review with emphasis on underlying
mechanisms and implications for health. Psychological Bulletin, 119(3), 488e531.
U.S. Department of Health and Human Services. (2000a). Mental health and mental
disorders. Tracking healthy people 2010. Washington, D.C.: U.S. Government
Printing Office.
U.S. Department of Health and Human Services. (2000b). Nutrition and overweight.
Tracking healthy people 2010. Washington, D.C.: U.S. Government Printing Office.
Vioque, J., Torres, A., & Quiles, J. (2000). Time spent watching television, sleep
duration and obesity in adults living in Valencia, Spain. Internal Journal of
Obesity, 24(12), 1683e1688.
Williams, C. J., Hu, F. B., Patel, S. R., & Mantzoros, C. S. (2007). Sleep duration and
snoring in relation to biomarkers of cardiovascular disease risk among women
with type 2 diabetes. Diabetes Care, 30(5), 1233e1240.
Wingard, D. L., & Berkman, L. F. (1983). Mortality risk associated with sleeping
patterns among adults. Sleep, 6(2), 102e107.
Yaggi, H. K., Araujo, A. B., & McKinlay, J. B. (2006). Sleep duration as a risk factor for
the development of type 2 diabetes. Diabetes Care, 29(3), 657e661.
Zee, P. C., & Turek, F. W. (2006). Sleep and health: everywhere and in both
directions. Archives of Internal Medicine, 166(16), 1686e1688.
O.M. Buxton, E. Marcelli / Social Science & Medicine 71 (2010) 1027e10361036
