Gregg EW, Cheng YJ, Cadwell BL, Imperatore G, Williams DE, Flegal KM, Narayan KMV, Williamson DF. Secular Trends in Cardiovascular Disease Risk Factors According to
Body Mass Index in US Adults. JAMA. 2005;293(15):1868-1874. doi:10.1001/jama.293.15.1868
Author Affiliations: Division of Diabetes Translation,
National Center for Chronic Disease Prevention and Health Promotion, Centers
for Disease Control and Prevention, Atlanta, Ga (Drs Gregg, Cheng, Imperatore,
Williams, Narayan, and Williamson and Ms Cadwell); National Center for Health
Statistics, Centers for Disease Control and Prevention, Hyattsville, Md (Dr
Context Prevalence of obesity in the United States has increased dramatically
in recent decades, but the magnitude of change in cardiovascular disease (CVD)
risk factors among the growing proportion of overweight and obese Americans
Objective To examine 40-year trends in CVD risk factors by body mass index (BMI)
groups among US adults aged 20 to 74 years.
Design, Setting, and Participants Analysis of 5 cross-sectional, nationally representative surveys: National
Health Examination Survey (1960-1962); National Health and Nutrition Examination
Survey (NHANES) I (1971-1975), II (1976-1980), and III (1988-1994); and NHANES
Main Outcome Measures Prevalence of high cholesterol level (≥240 mg/dL [≥6.2 mmol/L]
regardless of treatment), high blood pressure (≥140/90 mm Hg regardless
of treatment), current smoking, and total diabetes (diagnosed and undiagnosed
combined) according to BMI group (lean, <25; overweight, 25-29; and obese,
Results The prevalence of all risk factors except diabetes decreased over time
across all BMI groups, with the greatest reductions observed among overweight
and obese groups. Compared with obese persons in 1960-1962, obese persons
in 1999-2000 had a 21-percentage-point lower prevalence of high cholesterol
level (39% in 1960-1962 vs 18% in 1999-2000), an 18-percentage-point lower
prevalence of high blood pressure (from 42% to 24%), and a 12-percentage-point
lower smoking prevalence (from 32% to 20%). Survey × BMI group interaction
terms indicated that compared with the first survey, the prevalence of high
cholesterol in the fifth survey had fallen more in obese and overweight persons
than in lean persons (P<.05). Survey × BMI
changes in blood pressure and smoking were not statistically significant.
Changes in risk factors were accompanied by increases in lipid-lowering and
antihypertensive medication use, particularly among obese persons. Total diabetes
prevalence was stable within BMI groups over time, as nonsignificant 1- to
2-percentage-point increases occurred between 1976-1980 and 1999-2000.
Conclusions Except for diabetes, CVD risk factors have declined considerably over
the past 40 years in all BMI groups. Although obese persons still have higher
risk factor levels than lean persons, the levels of these risk factors are
much lower than in previous decades.
There has been a marked increase in overweight and obesity in the United
States over the past 25 years, with the prevalence of obesity among adults
aged 20 to 74 years rising from 13% to 31%.1- 3 This
increase in prevalence of obesity has occurred among both men and women and
across all racial/ethnic and age groups. Prospective cohort studies as well
as national surveys have shown that obese individuals have an increased risk
of several adverse health outcomes, notably hypertension, diabetes, cardiovascular
disease (CVD), arthritis, disability, and mortality.4- 8 Accordingly,
numerous clinical consensus panels and public health organizations have recommended
that persons with a body mass index (BMI) of 30 or higher, or with risk factors
and a BMI of 25 or higher, achieve and maintain a lower weight.9,10
Despite the scope and magnitude of the observed increases in obesity,
the health implications of these trends for the US population remain unclear.
Although increases in obesity have been accompanied by increases in diabetes,11- 13 the association between
obesity trends and other CVD risk factors remains less certain. In the overall
population, mortality rates from ischemic heart disease as well as levels
of key CVD risk factors have declined during the past 30 years,14- 19 but
whether comparable improvements have occurred across levels of BMI remains
Using data from the 5 consecutive nationally representative health surveys
conducted during the last 40 years, we examined whether long-term secular
changes in levels of key cardiovascular risk factors have been different in
overweight or obese persons compared with lean persons. We examined changes
in current smoking status, diagnosed and undiagnosed diabetes, and high blood
pressure and high total cholesterol levels. For blood pressure and cholesterol,
we analyzed actual values, regardless of treatment, so that levels could be
influenced by trends in medical care as well as by environmental and behavioral
changes that may have occurred.
The National Health Examination Survey (NHES) and the National Health
and Nutrition Examination Surveys (NHANES) are a series of cross-sectional
health examination surveys representative of the United States civilian noninstitutionalized
population. They were conducted in 1960-1962 (NHES), 1971-1975 (NHANES I),
1976-1980 (NHANES II), 1988-1994 (NHANES III), and 1999-2000 (NHANES 1999-2000).
Each of the surveys followed a stratified multistage probability design in
which a sample of the US population is selected. Detailed descriptions of
the plan and operation of each survey have been published.20- 24 To
maximize comparability across the 5 surveys, we restricted analyses to examined,
nonpregnant adults who were aged 20 to 74 years. This consisted of 6257, 12 911,
11 765, 14 319, and 3601 persons in the 5 surveys, respectively.
NHANES III and NHANES 1999-2002 have undergone institutional review
board approval and included written informed consent. Institutional review
board approval using current standards was not obtained for NHES, NHANES I,
or NHANES II, but internal human subjects review was conducted.
Each survey included a standardized examination to obtain information
on height, weight, and CVD risk factors. Weight and height were measured by
a standard protocol and used to calculate BMI; ie, weight in kilograms divided
by the square of height in meters. We refer to BMI groups of <25, 25 to
29.9, and ≥30 as “lean,” “overweight,” and “obese,”
respectively. Four CVD risk factors were analyzed: serum total cholesterol
concentration, blood pressure, smoking status, and diabetes. These risk factors
were chosen because they were assessed in at least 4 surveys, using similar
methods across all surveys.
Cholesterol analyses were conducted on venous blood serum samples and
standardized according to previously published methods.16,25,26 The
samples were frozen at −20°C and shipped weekly on dry ice to the
laboratory conducting the lipid analyses. The NHES and NHANES I measurements
were made in the Centers for Disease Control and Prevention Lipid Standardization
Laboratory. The NHANES II, III, and 1999-2000 measurements were made in lipid
research clinic laboratories that were standardized for cholesterol measurements
according to the criteria of the CDC–National Heart, Lung, and Blood
Institute Lipid Standardization Program.
Blood pressure was measured according to standard protocols that differed
somewhat across the 5 surveys.15 For consistency,
we used only blood pressure measurements conducted in the mobile examination
center with participants in a seated position. Because large adult and thigh
blood pressure cuffs were available only for the latter 2 surveys, blood pressure
could be overestimated among obese persons in early surveys. Thus, we computed
an adjusted blood pressure according to methods suggested by Maxwell et al27 to account for arm circumference and cuff size.28
Previously diagnosed diabetes was determined by asking participants
whether a physician or other health care professional had ever told the respondent
that he or she had diabetes. For the latter 3 surveys, undiagnosed diabetes
was assessed among subsamples (n = 3786, n = 5791, and
n = 1434 in NHANES II, III, and 1999-2000, respectively) of nondiabetic
persons randomly assigned to a morning fasting examination session.29- 32 Procedures
for blood collection and processing have been described.23,24,29,31,32 We
used American Diabetes Association diagnostic criteria for undiagnosed diabetes
(fasting glucose level ≥126 mg/dL [6.99 mmol/L]). Total diabetes prevalence
was calculated as the sum of the diagnosed diabetes from the interview sample
and undiagnosed diabetes from the fasting morning sample, adjusted using methods
To determine current cigarette smoking, respondents were asked, “Have
you smoked at least 100 cigarettes in your entire life?” and “Do
you smoke cigarettes now?” Current smoking was defined as a positive
answer to both questions. Smoking was not assessed in the NHES and, thus,
is reported only for the latter 4 surveys. For NHANES I, only a subsample
of persons aged 25 to 74 years were asked about smoking. Thus, for smoking,
we restricted our analyses to persons aged 25 to 74 years for the remaining
surveys (n = 3823, n = 10 389, n = 12641,
and n = 3271 among NHANES I, II, III, and 1999-2000, respectively).
We defined high blood pressure as systolic blood pressure of at least
140 mm Hg or diastolic blood pressure of at least 90 mm Hg28 using
the cuff-size–corrected measurements. We defined high total cholesterol
level as at least 240 mg/dL (6.20 mmol/L).33 High
blood pressure and high cholesterol level were defined according to levels
of control regardless of use of medications. These definitions better capture
the impact of secular clinical and public health factors and are less vulnerable
to recall bias than definitions based on treatment status. Although current
standards recommend lower levels of total cholesterol and blood pressure,
these cut points represent common definitions of risk factors across the period
of our study (1960-2000). Smoking was defined as smoking status at the time
of the survey. Total diabetes was defined as either previously diagnosed diabetes
(by self-report) or, for those without previously diagnosed diabetes, a fasting
glucose level of at least 126 mg/dL (6.99 mmol/L). Since fasting glucose measurements
were available only for the latter 3 surveys, we conducted additional analyses
of diagnosed diabetes across all 5 surveys.
Participants were asked about whether they had been told by a physician
to take prescribed medicine to lower blood cholesterol level and whether they
were currently taking medication for high blood pressure. These were used
to define the prevalence of medication use, with the entire population used
as the denominator. We also used this information to perform additional analyses
in which we defined the outcome as having a high level or taking medication
treatment for the condition. For the latter analysis, prevalence of lipid-lowering
medication use during the first 3 surveys (1960-1980) was assumed to be zero.
For all surveys, sampling weights had been calculated that take into
account unequal probabilities of selection resulting from the sample design,
nonresponse, and planned oversampling of certain subgroups. Data management
was conducted using SAS software, version 9.1.34 Standard
errors were calculated with SUDAAN software, version 9.0, to account for the
complex sample and to apply survey weights to make results representative
of the civilian, noninstitutionalized US population.35 For
estimates of total diabetes, we computed standard errors using a jackknife
variance estimator because the summation of diagnosed and undiagnosed diabetes
violated assumptions of statistical independence.36 For
high cholesterol, high blood pressure, smoking, and diagnosed diabetes, we
used multiple logistic regression and computed predictive marginals to estimate
the prevalence of each CVD risk factor by survey and BMI group when controlling
for age and sex.36 Predictive marginals are
a type of direct standardization in which the predicted values from the logistic
regression models are averaged over the covariate distribution of the population.
Using the predictive marginals and standard errors, we present differences
and 95% confidence intervals (CIs) in prevalence of risk factors over time
according to BMI group. Significance of interaction terms between BMI groups
and survey years from the logistic models were assessed to determine if changes
throughout the 40-year period and between the first and last surveys differed
by BMI group.
Table 1 lists the characteristics
of the participants across the 5 NHANES surveys, weighted to the US population.
In addition to the previously reported increases in body weight and BMI,1,2 substantial increases in educational
level occurred across the period covered by the surveys. Within BMI strata,
there were generally only minor increases in mean BMI across surveys. Within
the obese category, however, mean BMI increased from 33.6 in 1960-1962 to
35.8 in 1999-2000.
As has been previously documented,15,16,19 there
were large reductions (33% to 52% relative to baseline) in the prevalence
of high cholesterol level, high blood pressure, and smoking in the overall
population during the past 30 to 40 years (Table
2). However, the prevalence of obesity and diagnosed diabetes doubled
during this period, and between 1976-1980 and 1999-2000, the prevalence of
total (diagnosed and undiagnosed) diabetes increased by 55%.1,2,29,32
Between 1960-1962 and 1999-2000, the age- and sex-adjusted prevalence
of high cholesterol, high blood pressure, and smoking levels decreased among
all BMI groups (Table 2). The prevalence
of total diabetes remained stable within BMI groups across decades, while
the prevalence of diagnosed diabetes increased among overweight and obese
Absolute reductions in the prevalence of high cholesterol between 1960
and 2000 were greater for overweight (21-percentage-point decline; 95% CI,
17%-25%) and obese persons (21-percentage-point decline; 95% CI, 15%-27%)
than for those of lean persons (12-percentage-point decline; 95% CI, 8%-16%).
Obese persons had a 9- to 12-percentage-point higher prevalence of high cholesterol
level than lean persons across all surveys, but by 1999-2000, this difference
was only 2.7 percentage points. The overall test for the 8 survey × BMI
interaction terms was not statistically significant. However, the interaction
terms comparing change in prevalence of high cholesterol level between the
first and fifth surveys showed a statistically significantly greater reduction
in prevalence of high cholesterol level among obese (P = .02)
and overweight (P<.05) persons compared with lean
Prevalence of high blood pressure also tended to decline more over time
among obese (18 percentage points; 95% CI, 13%-23%) and overweight (17 percentage
points; 95% CI, 13%-21%) persons than among lean persons (14 percentage points;
95% CI, 11%-18%). However, the prevalence of high blood pressure among obese
persons remained at least twice that of lean persons (24% vs 11%) in 1999-2000.
Moreover, survey × BMI group interaction terms comparing the
first and fifth surveys were not statistically significant for obese (P = .25) and overweight (P = .76) persons compared with lean persons, indicating that
secular declines in high blood pressure did not differ by BMI.
The prevalence of smoking was lowest among obese persons and highest
among lean persons across all 5 surveys (Table
2). This prevalence declined among all BMI groups by a similar magnitude
(12- to 13-percentage-point change across groups) such that about one third
of lean persons and one fifth of obese persons reported smoking in 1999-2000.
Survey × BMI interaction terms were not significant for smoking.
Prevalence of total diabetes (diagnosed and undiagnosed combined) was
stable over time within BMI groups, increasing by only 1 to 2 percentage points
between 1976-1980 and 1999-2000. In contrast, prevalence of diagnosed diabetes
was about 2.5 to 3.5 times as high in 2000 as in 1960 among overweight (from
1.6% to 4.2%) and obese (from 2.9% to 10.1%) persons (Table 2). The survey × BMI group interaction comparing
diagnosed diabetes prevalence in the first vs fifth surveys was statistically
significant for obese persons (P = .03)
but not overweight persons (P = .28), indicating
that secular increases in prevalence of diagnosed diabetes were greater among
obese than lean persons.
The proportion of the overall population receiving medication treatment
for high cholesterol level and high blood pressure increased among all BMI
groups (Table 3). The prevalence of
treatment for high cholesterol increased more among obese persons (from 3.5%
to 9.2%) than among lean persons (from 2.2% to 4%), although the survey × BMI
group interaction terms were not significant. The prevalence of treatment
for high blood pressure increased significantly more among obese persons (from
11% to 28%) and overweight persons (from 6% to 17%) than among lean persons
(from 5% to 8%) (P<.01 for interaction terms).
When we examined prevalence estimates using a definition of risk as
having either high levels or receiving treatment, there were still reductions
in prevalence over time (Table 4). However,
the magnitude of reduction in prevalence over time was less than observed
in our primary analyses when risk levels were defined irrespective of treatment
(Table 2). For example, the magnitude
of reductions in prevalence of having high cholesterol levels or taking medication
ranged from 9 to 13 percentage points across the 3 BMI groups. The magnitude
of reductions in prevalence of having high blood pressure levels or receiving
medication ranged from 4 to 10 percentage points across the 3 BMI groups,
with the smallest reduction noted in the obese group (P<.01 for interaction term).
In this unique series of nationally representative surveys of the US
adult population, we documented a substantial decline in the prevalence of
key CVD risk factors over the last 3 to 4 decades, affecting obese, overweight,
and lean segments of the population. Among obese persons today, prevalence
of high cholesterol, high blood pressure, and smoking are now 21, 18, and
12 percentage points lower, respectively, than among obese persons 30 to 40
years ago. The corresponding reductions among lean persons have been somewhat
less, with average declines of 12 to 14 percentage points. Although obesity
remains associated with a higher prevalence of important CVD risk factors,
differences in total cholesterol levels across BMI groups may be narrowing,
and for blood pressure and smoking improvements have been similar across BMI
groups. Thus, obese and overweight persons may be at lower risk of CVD now
than in previous eras.
Diabetes is a notable exception to the observed reduction in risk factors,
as prevalence of total diabetes (ie, diagnosed and undiagnosed combined) did
not decrease within BMI groups. This was accompanied by a 55% increase in
total diabetes among the overall population (ie, all BMI groups combined),
presumably due to an increasing proportion of the population moving into the
obese categories. Diagnosed diabetes alone, on the other hand, increased more
among obese persons than other groups, but presumably due to greater increases
in detection among obese than lean persons. In a detailed examination of diabetes
trends,13 we recently found that undiagnosed
diabetes decreased considerably among the most obese persons (particularly
those with BMI >35), and the proportion of total cases that were diagnosed
increased dramatically. These observations, combined with our findings of
increased cholesterol-lowering and antihypertensive medication use, also suggest
that the health care system is placing greater emphasis on preventive care
among obese persons.
The observation that obese persons now have better CVD risk factor profiles
than their leaner counterparts did 20 to 30 years ago, combined with trend
data in the general population, suggests that other factors may be compensating
for the increases in obesity.15,17,18,37- 43 Studies
have documented a mixture of both deleterious and beneficial trends in the
US population. Trends promoting obesity include increases in total calories,
portion sizes, refined carbohydrates, and fast food intake.18,39- 41,44,45 Healthy
trends include decreases in the proportion of dietary saturated fat; increases
in fruits, vegetables, and the proportion of polyunsaturated and monounsaturated
fats in the diet17,18,39- 43;
and decreased smoking, which, ironically, may have contributed somewhat to
increasing obesity.43 Although trends in overall
physical activity levels remain unclear, recent estimates suggest walking
prevalence may have increased, particularly among obese persons.46 All
of these changes appear to be overlaid on a background of increased awareness,
aggressive identification, and pharmacological treatment of high cholesterol
and blood pressure levels.15,37,38 Increased
awareness about obesity and its association with CVD may have further fueled
the changes observed among overweight and obese persons. The net result of
these phenomena may be a population that is, paradoxically, more obese, diabetic,
arthritic, disabled, and medicated, but with lower overall CVD risk.2,4- 7,47
Although our study does not directly address the causes of changes in risk factors, our results are consistent with
both pharmacological and nonpharmacological factors playing a causal role.
Two observations support a role for pharmacological factors: First, we found
that the prevalence of lipid-lowering and antihypertensive medication increased
in all groups, particularly among obese persons. Second, when we defined our
outcomes as either high levels or receipt of treatment, we found that the
magnitude of reduction was less than in our primary analyses, where risk was
defined irrespective of treatment. However, we interpret these latter findings
with caution because changes in clinical definitions and treatment guidelines
over time make people more likely to be treated at lower levels of risk in
later surveys. This could have the effect of overestimating prevalence in
later surveys when we define risk as either high levels or receipt of treatment.
Furthermore, our observation that risk factors were lower over time regardless
of whether treatment was part of the definition suggests that nonpharmacological
factors influenced these trends as well.
Our findings are supported by other ecologic observations, including
reductions in CVD and ischemic heart disease mortality in the general population,14,19 as well as simultaneous BMI increases
and CVD risk factor reductions in the Minnesota Heart Survey.18 A
report from the WHO-MONICA study48 that examined
the correlation between changes in risk factors and coronary heart disease
(CHD) incidence rates from 38 separate international populations found that
increasing BMI trends were actually associated with declining CHD rates among
men; among women, there was no association between changes in BMI and CHD.
These findings from the WHO-MONICA study suggest that changes in obesity are
a relatively minor determinant of CHD trends and remind us that BMI is only
one among many determinants of CVD risk.
Our analyses have several limitations. First, we did not examine BMI-specific
trends in dietary intake, physical activity levels, lipid subfractions, body
fat distribution, inflammation markers, musculoskeletal disorders, or health-related
quality of life. Second, the lack of extra-large blood pressure cuffs in early
surveys could exaggerate earlier blood pressure levels as well as declines
in blood pressure among obese persons over time. However, we corrected all
blood pressure values for cuff size and arm circumference. Third, the sample
size for NHANES 1999-2000 was considerably smaller than the earlier surveys.
Inferences about changes between the latter 2 surveys should thus be drawn
with caution. Fourth, our analyses did not adjust for shifts in race/ethnicity
over survey years because of changes in the way race/ethnicity was assessed.
However, when we adjusted for nonwhite race, we found essentially no difference
in our findings. Finally, we examined some risk factors irrespective of treatment.
This assumes, not quite correctly, that a person is considered not at risk
despite receiving treatment as long as their risk factor levels are controlled.
We selected this approach because we were interested in capturing the full
range of secular trends that may have occurred, whether from medical care,
behavioral, or environmental sources. Note, however, that for diabetes we
had no comparable measure for risk subsequent to treatment because levels
of glycemia were not assessed among persons with diabetes. Thus, there may
have been some level of CVD risk reduction associated with improving glycemic
control and better treatment that we could not measure.
In summary, our study found that with the exception of diabetes, CVD
risk factor levels have declined over recent decades among all BMI groups.
While obesity remains associated with elevated levels of several CVD risk
factors compared with lean persons, their levels of risk factors have now
diminished such that they are lower than those of lean individuals 20 to 30
years ago. In addition to determining the key reasons for the favorable trends
we observed, future studies should examine whether these changes extend to
incidence of morbidity, including cardiovascular disease incidence and disability.
Despite our encouraging findings, a considerable proportion of lean as well
as obese persons still have elevated levels of modifiable risk factors, particularly
when one considers that the current definitions of risk factor control are
more aggressive than the definitions used in this trend analyses. Clinical
and public health efforts should continue to emphasize maintenance of healthy
lifestyle behaviors for lean as well as overweight and obese persons.
Corresponding Author: Edward W. Gregg, PhD,
Division of Diabetes Translation, Centers for Disease Control and Prevention,
4770 Buford Hwy NE, Mailstop K-10, Atlanta, GA 30341 (email@example.com).
Author Contributions: Dr Gregg had full access
to all of the data in the study and takes responsibility for the integrity
of the data and the accuracy of the data analysis.
Study concept and design: Gregg, Cheng, Imperatore,
Williams, Narayan, Williamson.
Acquisition of data: Imperatore.
Analysis and interpretation of data: Gregg,
Cheng, Cadwell, Imperatore, Williams, Flegal, Williamson.
Drafting of the manuscript: Gregg, Cheng, Flegal.
Critical revision of the manuscript for important
intellectual content: Gregg, Cheng, Cadwell, Imperatore, Williams,
Flegal, Narayan, Williamson.
Statistical analysis: Gregg, Cheng, Cadwell,
Imperatore, Williams, Flegal, Williamson.
Administrative, technical, or material support:
Study supervision: Narayan.
Financial Disclosures: None reported.