Content uploaded by Mark Pereira
Author content
All content in this area was uploaded by Mark Pereira on Feb 13, 2018
Content may be subject to copyright.
For personal use. Only reproduce with permission from Elsevier Ltd
Articles
36 www.thelancet.com Vol 365 January 1, 2005
Introduction
The frequency of obesity has risen at an alarming rate in
all age and ethnic groups in the USA.1,2 The age-
adjusted prevalence of obesity, defined as a body-mass
index (BMI) of 30 kg/m2 or greater, was 30·5% in
1999–2000 compared with 22·9% in 1988–1994, with
even higher rates in ethnic minority groups.1About two
of every three US adults and four of five African-
American women were overweight or obese in
1999–2000.1In children and adolescents, the prevalence
of being overweight rose by 50% in the past decade to
about 15%.2
The medical and economic outcomes of excessive
bodyweight are great, including an estimated
300 000 excess deaths and at least US$100 billion per
year in medical expenditures.3–6 One particularly
ominous public-health issue is the occurrence of glucose
intolerance7and type 2 diabetes in obese adolescents and
young adults.8
Because of its rapid development in genetically stable
populations, the obesity epidemic can be attributed to
environmental factors affecting diet, or physical activity
level. One potentially important dietary factor is
consumption of fast food, which can be defined as
convenience food purchased in self-service or carry-out
eating places.9,10 From its origins in the 1950s, fast food
has grown into a dominant dietary pattern, with a
current estimate of about 247 115 restaurants in the
USA.11 Consumption of fast food by children has risen
from 2% of total energy in the late 1970s to 10% of
energy in the 1990s.12
Several factors inherent to fast food as it now exists
could promote a positive energy balance11,13 and thereby
increase risk for obesity and diabetes, including:
excessive portion size, with single large meals often
approaching or exceeding individual daily energy
requirement; palatability, emphasising primordial taste
preferences for sugar, salt, and fat; high energy density;14
and high glycaemic load. Several dietary factors such as
trans-fatty acids15 and high glycaemic load16 might also
enhance risk for diabetes through energy-independent
mechanisms.
Surprisingly few studies have investigated the effects
of fast-food consumption on energy balance or body-
weight,17–20 and most of these are of cross-sectional
design. To our knowledge, no data for fast-food
consumption and diabetes-related endpoints are
available. For these reasons, we aimed to investigate the
association between reported fast-food habits and
changes in bodyweight and insulin resistance over a
15-year period in young black and white adults in
the USA.
Lancet 2005; 365: 36–42
See Comment page 4
Division of Epidemiology and
Community Health, School of
Public Health, University of
Minnesota, Minneapolis, MN,
USA (M A Pereira PhD,
Prof D R Jacobs Jr PhD); Clinical
Research Program, Children’s
Hospital, Boston, MA, USA
(A I Kartashov PhD);
Department of Medicine,
Children’s Hospital,
300 Longwood Ave, Boston,
MA 02115, USA
(C B Ebbeling PhD,
D S Ludwig MD); Department of
Preventive Medicine,
Northwestern University
Medical School, Chicago, IL,
USA (Prof L Van Horn PhD);
University of Utah Medical
School, Salt Lake City, UT, USA
(Prof M L Slattery PhD); and
Department of Nutrition,
University of Oslo, Oslo,
Norway (Prof D R Jacobs Jr)
Correspondence to:
Dr David S Ludwig
david.ludwig@childrens.
harvard.edu
Fast-food habits, weight gain, and insulin resistance
(the CARDIA study): 15-year prospective analysis
Mark A Pereira, Alex I Kartashov, Cara B Ebbeling, Linda Van Horn, Martha L Slattery, David R Jacobs Jr, David S Ludwig
Summary
Background Fast-food consumption has increased greatly in the USA during the past three decades. However, the
effect of fast food on risk of obesity and type 2 diabetes has received little attention. We aimed to investigate the
association between reported fast-food habits and changes in bodyweight and insulin resistance over a 15-year period
in the USA.
Methods Participants for the CARDIA study included 3031 young (age 18–30 years in 1985–86) black and white
adults who were followed up with repeated dietary assessment. We used multiple linear regression models to
investigate the association of frequency of fast-food restaurant visits (fast-food frequency) at baseline and follow-up
with 15-year changes in bodyweight and the homoeostasis model (HOMA) for insulin resistance.
Findings Fast-food frequency was lowest for white women (about 1·3 times per week) compared with the other
ethnic-sex groups (about twice a week). After adjustment for lifestyle factors, baseline fast-food frequency was
directly associated with changes in bodyweight in both black (p=0·0050) and white people (p=0·0013). Change in
fast-food frequency over 15 years was directly associated with changes in bodyweight in white individuals
(p<0·0001), with a weaker association recorded in black people (p=0·1004). Changes were also directly associated
with insulin resistance in both ethnic groups (p=0·0015 in black people, p<0·0001 in white people). By comparison
with the average 15-year weight gain in participants with infrequent (less than once a week) fast-food restaurant use
at baseline and follow-up (n=203), those with frequent (more than twice a week) visits to fast-food restaurants at
baseline and follow-up (n=87) gained an extra 4·5 kg of bodyweight (p=0·0054) and had a two-fold greater increase
in insulin resistance (p=0·0083).
Interpretation Fast-food consumption has strong positive associations with weight gain and insulin resistance,
suggesting that fast food increases the risk of obesity and type 2 diabetes.
For personal use. Only reproduce with permission from Elsevier Ltd
Articles
www.thelancet.com Vol 365 January 1, 2005 37
Methods
The Coronary Artery Risk Development in Young Adults
(CARDIA) study is a multicentre, population-based
prospective study of cardiovascular disease risk factor
evolution in a US cohort of African-American and white
young adults. The four study centres are Birmingham,
AL, Chicago, IL, Minneapolis, MN, and Oakland, CA.
Participants
We used recruitment stratification to obtain nearly equal
numbers of participants who were black and white,
young (18–24 years) and old (25–30 years), and with
more (high school or more) and less (less than high
school) education. Participants were followed up for
15 years and had six clinical examinations: in 1985–86
(baseline or year 0), 1987–88 (year 2), 1990–91 (year 5),
1992–93 (year 7), 1995–96 (year 10), and 2000–01 (year
15). More details of the CARDIA study design and
participants have been previously reported.21
We excluded participants from our analysis for the
following reasons: did not come to the year 15
examination; missing data for fast food, bodyweight, or
important covariates at baseline or follow-up; female
participants who were pregnant at baseline or within
180 days of year 15, or were breastfeeding; suspected
type 1 diabetes based on insulin treatment; and for the
insulin resistance analysis only, participants fasting for
fewer than 8 h at year 0 or year 15. Some participants
belonged to more than one of the above categories.
Procedures
We used standard questionnaires to maintain
consistency in the assessment of demographics (age,
sex, ethnic origin, and education) and behavioural
information across CARDIA examination visits. The
CARDIA physical activity history questionnaire22 queries
the amount of time per week spent in leisure,
occupational, and household physical activities over the
past 12 months. We estimated total physical activity—
expressed in exercise units as a product of
intensityfrequency—at every clinical examination.
Television watching was calculated as hours per week
and was reported at the examinations at years 5, 10, and
15. We quantified education as the number of years of
school completed at every examination and cigarette
smoking status as current smoker, former, or never-
smoker at every examination.
For dietary assessment, we did a structured interview
at every CARDIA examination and included a series of
questions on dietary practices, including food
preparation, and where meals were typically eaten. Only
the year 2 examination did not include fast-food habits.
We quantified fast-food habits based on responses to the
question: “How often do you eat breakfast, lunch, or
dinner at places such as McDonald’s, Burger King,
Wendy’s, Arby’s, Pizza Hut, or Kentucky Fried
Chicken?” Responses were recorded to the nearest
frequency per week (fast-food frequency) on a
semicontinuous scale and classified as less than 1, 1–2,
or greater than 2.
To assess possible confounding or mediation, we also
used dietary data obtained at year 0 and 7 from the
CARDIA diet history,23 which queried usual dietary
practices and obtained a quantitative food frequency of
the past month. Liu and colleagues24 reported on the
reliability and validity of the CARDIA diet history in
128 young adults. As an internal validation of the fast-
food frequency question, we identified all foods that
might have been obtained at fast-food restaurants—eg,
“double cheeseburger”, “chicken nuggets”, and “french
fries”. Frequency of consumption for every food item
(times per week) was used to estimate relative intake per
week for every food. We also considered intake of other
food groups and nutrients that could confound
associations or serve as mechanisms linking fast-food
intake with weight gain and insulin resistance, including
intake (times per week) of fruit, non-starchy vegetables,
dairy, soda and sugar-sweetened beverages, whole and
refined grains, total energy (kcal/day), alcohol (mL/day),
fibre (g/1000 kcal per day), trans-fatty acids (g/day), and
percentage of energy from animal and vegetable protein
and total, saturated, unsaturated, and trans-fatty acids.
As an additional approach to adjusting for the
confounding effects of correlated lifestyle habits, we
calculated a healthy lifestyle score, modified from our
previously used measure.25 We created binary variables
based on smoking status (current smoker [scored 0], non-
smoker [scored 1]) and median cutpoints for physical
activity (low [0], high [1]), television viewing (low [1], high
[0]), saturated fat intake (low [1], high [0]), wholegrain
intake (low [0], high [1]), fruit and non-starchy vegetable
intake (low [0], high [1]), low-fat dairy intake (low [0], high
[1]), and soft drink intake (low [1], high [0]). We summed
these binary scores into the healthy lifestyle score with a
range from 0, indicating a very unhealthful lifestyle, to 8,
indicating a very healthful lifestyle.
We undertook all clinical procedures in accordance
with the CARDIA study manual of operations. For
anthropometric measures, participants were standing
and dressed in light clothing without shoes. We
measured bodyweight to the nearest 0·2 kg with a
calibrated balance beam scale, height with a vertical
ruler to the nearest 0·5 cm, and waist size with a tape in
duplicate to the nearest 0·5 cm around the minimum
abdominal girth. BMI was calculated as weight in kg
divided by height in m2.
Before every CARDIA examination, we asked
participants to fast and to avoid smoking and heavy
physical activity for the final 2 h. For measurement of
insulin and glucose concentrations, we drew blood into
vacuum tubes containing no preservative. We separated
serum by centrifugation at 4ºC within 60 min, stored it
in cryovials, and froze it at –70ºC within 90 min until
laboratory analysis. The radioimmunoassay for insulin
For personal use. Only reproduce with permission from Elsevier Ltd
Articles
38 www.thelancet.com Vol 365 January 1, 2005
required an overnight, equilibrium incubation and used
a unique antibody that has less than 0·2% crossreactivity
to human proinsulin and its primary circulating split
form Des 31,32 proinsulin (Linco Research, St Louis,
MO, USA). Masked analysis of split serum samples
resulted in a technical error of 16·6% of the mean
(r=0·98). We measured fasting glucose by the
hexokinase method at every examination. The
homoeostasis model (HOMA) for insulin resistance was
calculated as glucose (mmol/L)insulin (mU/L)/22·5.26
Statistical analysis
We did all analyses with SAS statistical software version 9
(SAS, Cary, NC, USA). To assess changes in fast-food
intake over time, and the effects of age, time, and secular
trends on fast-food intake, we used repeated measures
regression analysis (PROC MIXED). We used general
linear models (PROC GLM) to calculate adjusted means
of demographic and lifestyle factors according to category
of fast-food intake and to analyse the relation between the
independent variables (baseline fast-food frequency and
15-year change in fast-food frequency) with changes in
the dependent variables (bodyweight and HOMA insulin
resistance) over the 15-year follow-up period. For these
analyses, 15-year change in fast food was calculated as the
baseline fast-food frequency subtracted from the value of
fast-food frequency at the final follow-up examination at
year 15. We modelled many potential confounding or
mediating factors as their baseline value and their change
over time, when available.
Ethnic origin-specific multivariable linear regression
models were constructed as follows. Model 1 included
demographic covariates (sex, age [continuous years],
centre, and education [continuous years]) and the
respective baseline value [continuous] of the dependent
variable (baseline bodyweight and height when modelling
weight change, and baseline HOMA insulin resistance
when modelling HOMA). Model 2 further included non-
dietary lifestyle covariates (alcohol consumption [mL/day
at baseline and year 15], smoking status [never, former,
current at baseline and year 15], physical activity [units per
day at baseline and 15-year change], and television
viewing [h/day at year 10 and change between year 10 and
year 15]). In model 3, we added baseline values and
changes in dietary factors, including total caloric intake,
dietary fibre (g/1000 kcal at year 7 only), percentage of
calories from saturated fatty acids, unsaturated fatty acids,
trans-fatty acids (g/day), and daily intake of soft drinks,
refined grains, wholegrains, low-fat and high-fat dairy
products, fruits, non-starchy vegetables, meat, and fish.
We also examined interactions between fast-food
frequency and ethnic origin, sex, and baseline overweight
status (BMI 25 kg/m2).
Role of the funding source
The sponsors of the study had no role in study design,
data collection, data analysis, data interpretation, or
writing of the report. The corresponding author had full
access to all the data in the study and had final
responsibility for the decision to submit for publication.
Results
5115 people attended the baseline examination, with 74%
retention of the surviving cohort in year 15. Those who
were not in the 15-year cohort were younger (mean
24·2 years [SD 3·7] vs 25·1 years [3·6], p<0·0001) and
more likely to be black (62·9% vs 47·1%, p<0·0001), men
(49·1% vs 44·1%, p=0·0014), and have less education
(61·0% vs 80·9%, p<0·0001) at baseline, but they did not
differ by BMI (mean 24·4 kg/m2[SD 5·2] vs 24·5 kg/m2
[5·0], p=0·2287) or frequency of fast-food visits
(2·1 times a week [2·3] vs 1·9 times a week [2·2],
p=0·1450) at baseline. After exclusion of 1443 people
who did not come to the year 15 examination, 492 who
had missing data, 223 who were pregnant or
breastfeeding, 17 with suspected type 1 diabetes, and 419
who fasted fewer than 8 h, 3031 participants could be
included in the bodyweight analysis and 2767 in the
insulin resistance analysis.
Age-adjusted fast-food frequency was relatively stable
over time in black people but fell in those who were white
(p<0·0001 for ethnic origin-time interaction; table 1).
Fast-food frequency was higher in black than in white
people (p<0·0001) and in men than in women
(p<0·0001) for every examination year. The reported fast-
food frequency in white women was notably low (table 1).
In both ethnic groups, fast-food frequency rose over
time in those with a low baseline frequency (<1 time per
week), whereas in individuals with high baseline
frequency it fell over time (table 2). The strong positive
association between fast-food frequency and consumption
of individual fast-food items, as reported during the diet
history interview (“Foods typical of fast food”), lends
support to the internal validity of the fast-food frequency
data. Individuals with high fast-food frequency were
younger than those with low frequency (table 2). No
association was seen between year 0 fast-food frequency
and year 0 bodyweight or HOMA insulin resistance in
either black or white people. Year 0 associations noted
between fast-food habits and lifestyle factors seemed to be
stronger for white people than for black individuals, with
the exception of energy intake, which was directly
associated with fast-food frequency in both populations.
1985–86 2000–01
Black men (n=647) 2·4 (0·09) 2·3 (0·09)
Black women (n=797) 1·8 (0·08) 2·0 (0·08)
White men (n=804) 2·4 (0·07) 1·9 (0·08)
White women (n=783) 1·6 (0·08) 1·3 (0·08)
Data are mean (SE) and are adjusted for age and study centre. Time, race, sex, and
timeracesex differences are significant (p<0·0001).
Table 1: Reported frequency of fast food restaurant visits (times per week)
by race, sex, and time in the CARDIA study
For personal use. Only reproduce with permission from Elsevier Ltd
Articles
www.thelancet.com Vol 365 January 1, 2005 39
The healthy lifestyle score was strongly inversely
associated with fast-food intake in white but not black
people (table 2). By comparison with individuals in the
low fast-food category, those in the high category had
higher year 0 intakes of total energy, total fat, saturated
fatty acids, soft drinks, refined grains, and meat, and
lower intakes of dietary fibre. An inverse association
between fast-food frequency and cigarette smoking was
noted in black people, whereas a direct association was
recorded in white individuals (table 2). For black, but not
white, people, high fast-food frequency was associated
with high intake of trans-fatty acids. For white but not
black individuals, by comparison with those in the low
fast-food category, those in the high category had fewer
years of education, had lower physical activity levels,
watched more hours of television, consumed more
alcohol, and had a lower intake of wholegrains, fruit and
non-starchy vegetables, and reduced fat dairy products.
Table 3 includes the adjusted mean 15-year changes in
bodyweight associated with differences in year 0 fast-
food frequency and 15-year change in fast-food
frequency of three times per week. Year 0 and change in
fast-food frequency were modeled simultaneously. A
difference of three times per week was chosen because,
as table 2 shows, this difference was close to the mean
between individuals with low versus high fast-food
frequency at baseline and also approximated the mean
range in change over time in these two baseline
categories. Year 0 fast-food frequency was associated
with a rise in bodyweight in both ethnic groups,
independent of many other lifestyle factors. Even after
thorough adjustment for many possibly confounding or
mediating dietary factors in model 3, a difference in
year 0 fast-food frequency of three times per week was
associated with mean weight gains of 2·2 kg (SE 0·72) in
black people (p=0·0014) and 1·6 kg (0·55) in white
people (p=0·0064). Change in fast-food frequency over
Black people p* White people p*
<1 (n=450) 1–2 (n=508) >2 (n=486) <1 (n=625) 1–2 (n=517) >2 (n=445)
Year 0 fast-food frequency (times per week) 0·4 1·4 4·6 <0·0001 0·4 1·4 4·8 <0·0001
15-year change in fast food (times per week) 1·4 0·7 –1·8 <0·0001 0·7 0·1 –2·5 <0·0001
Foods typical of fast food (times per week)† 1·5 2·4 4·3 <0·0001 0·8 1·7 3·6 <0·0001
Age (years) 25·0 24·4 24·0 <0·0001 26·1 25·5 25·1 <0·0001
Women (%) 61·7 56·6 47·7 <0·0001 59·5 46·8 37·9 <0·0001
Bodyweight (kg)‡ 72·4 73·5 72·9 0·6506 69·8 71·5 70·8 0·1598
HOMA insulin resistance score 2·4 2·4 2·6 0·1387 1·8 2·0 1·9 0·3277
Healthy lifestyle score 3·6 3·7 3·5 0·2257 4·9 4·6 4·1 <0·0001
Education (years) 13·8 13·9 14·0 0·1723 16·0 15·5 15·5 0·0011
Physical activity (exercise units) 360 389 394 0·0713 480 454 440 0·0298
Television (h/day)§ 3·2 3·1 3·1 0·5334 1·6 1·7 1·9 0·0003
Current smokers (%) 36·2 27·7 29·5 0·0256 20·1 23·6 30·7 0·0002
Alcohol intake (mL/day) 12·3 9·9 11·7 0·7304 12·1 11·9 17·6 <0·0001
Total energy intake (kcal/day) 2829 3169 3485 <0·0001 2451 2694 2978 <0·0001
Total fat (g/day) 110·1 123·2 138·7 <0·0001 91·5 103·9 117·7 <0·0001
Total fat (% energy) 34·3 34·7 35·6 0·0006 33·2 34·5 35·4 <0·0001
Saturated fat (% energy) 13·8 14·1 14·4 0·0011 13·6 14·1 14·5 <0·0001
Trans fat (g/day) 4·2 4·3 4·8 0·0192 3·3 3·5 3·5 0·1316
Soft drink intake (times per week) 6·2 6·9 8·6 <0·0001 2·5 4·3 6·4 <0·0001
Meat intake (times per week) 10·6 12·3 13·8 <0·0001 6·8 8·9 10·8 <0·0001
Refined grain intake (times per week) 23·4 26·9 26·5 0·0114 18·9 20·0 22·1 <0·0001
Wholegrain intake (times per week) 7·7 8·3 8·7 0·1242 11·9 10·5 8·6 <0·0001
Fruit and non-starchy vegetable intake (times per week) 14·2 15·1 15·2 0·9386 22·5 19·1 16·3 <0·0001
Fibre intake (g/1000 kcal per day)¶ 7·6 7·5 6·8 <0·0001 10·2 8·7 8·2 <0·0001
Reduced-fat dairy intake (times per week) 3·6 3·9 3·5 0·8452 11·2 10·5 9·6 0·0286
Data are mean values, and are adjusted for age, sex, and study centre. Data are from year 0 except for those noted. *Highest compared with lowest fast food category. †Sum of french
fries, hamburgers, breakfast items, and chicken items reported during the CARDIA diet history interview, which may have been obtained at fast-food restaurants. ‡Bodyweight was also
adjusted for height. §Measured at year 5. ¶Measured at year 7.
Table 2: Adjusted demographic and dietary factors by frequency of fast-food restaurant visits at year 0 (1985–86)
Fast-food variable Black people White people
b(SE) p b(SE) p
Model 1 Baseline 1·44 (0·58) 0·0126 2·68 (0·47) <0·0001
Change 0·70 (0·40) 0·0774 2·67 (0·41) <0·0001
Model 2 Baseline 1·72 (0·61) 0·0050 1·84 (0·50) 0·0013
Change 0·63 (0·42) 0·1004 1·97 (0·42) <0·0001
Model 3 Baseline 2·22 (0·72) 0·0014 1·56 (0·55) 0·0064
Change 0·74 (0·45) 0·1053 1·84 (0·44) <0·0001
Model 1 includes the following covariates: age (continuous years), study centre, sex, education (highest attained, continuous
years), baseline bodyweight (continuous), and height (continuous). Model 2 contains the extra covariates: cigarette smoking
status (never, former, current, at baseline and year change), alcohol intake (mL/day, at baseline and 15-year change), physical
activity (units per day, at baseline and 15-year change), television viewing (h/day, at year 10 and change between years 10 and
15). Model 3 contains the covariates in model 2 and dietary intake (at baseline and change between baseline and year 7) of total
energy (kcal/day), saturated fatty acids (% kcal), unsaturated fatty acids (% kcal), and trans-fatty acids (g/day), fibre
(g/1000 kcal at year 7 only), and daily servings of fruits, non-starchy vegetables, wholegrains, refined grains, soft drinks, meat,
dairy, and fish.
Table 3: Mean adjusted 15-year changes in bodyweight by three times per week year 0 differences in
fast food frequency between participants, and by three times per week 15-year change in fast food
frequency within participants
For personal use. Only reproduce with permission from Elsevier Ltd
Articles
40 www.thelancet.com Vol 365 January 1, 2005
15 years was also independently associated with changes
in bodyweight in white people (1·8 kg [SE 0·44],
p<0·0001), with a weaker association recorded in black
individuals (0·7 kg [0·45], p=0·1053). The association
between fast-food change and weight change was
stronger for white than black people (p=0·0075 for
interaction). Interaction terms between fast-food
frequency and year 0 BMI or sex were not significant.
As table 4 shows, the association between year 0 fast-
food intake and change in insulin resistance in black and
white individuals that was noted in model 2 was no
longer apparent in model 3 after adjustment for many
dietary factors that may be either confounders or on the
causal pathway. However, change in fast food was
directly associated with changes in insulin resistance in
both black and white people even after thorough
adjustment in model 3. Interaction terms between fast-
food frequency and ethnic origin or sex were not
significant. In white people only, we noted somewhat
stronger associations between fast-food habits and
changes in insulin resistance among those who were
overweight at baseline compared with those who were
normal weight at baseline (p=0·0121 for interaction
between baseline fast food and baseline overweight
status, and p=0·0042 for interaction between change in
fast food and baseline overweight status).
Adjustment for total, rather than saturated, fat intake
did not materially change these findings for either
endpoint. Percentage of energy from total fat and
saturated fat was not associated with weight gain or
insulin resistance in model 3 (p value range 0·14–0·88).
Further adjustment for the healthy lifestyle index did not
materially alter the findings (data not shown).
Figures 1 and 2 depict the joint associations of year 0
fast-food frequency (p=0·0389 for weight change and
p=0·0031 for insulin resistance) and 15-year changes in
fast-food frequency (p=0·0030 for weight change and
p=0·0001 for insulin resistance) with weight gain and
changes in insulin resistance. We recorded direct and
independent monotonic associations between both
year 0 and 15-year change in fast-food frequency and
15-year weight gain and changes in insulin resistance.
The interaction between year 0 fast food and change in
fast food was not significant for bodyweight (p=0·9340)
or insulin resistance (p=0·1461). By comparison with
the average 15-year weight gain in participants with
infrequent (<1 time per week) fast-food restaurant use at
baseline and follow-up (n=203), those with frequent
(>2 times per week) fast-food restaurant use at both
year 0 and follow-up (n=87) gained an extra 4·5 kg of
bodyweight (p=0·0054; figure 1) and had a 104% greater
increase in insulin resistance (p=0·0083, figure 2).
To further assess possible residual confounding by
physical activity and television viewing, we stratified our
models by five levels of a composite index of these two
behaviours, ranging from low activity and high
television viewing to high activity and low television
Fast-food variable Black people White people
b(SE) p b(SE) p
Model 1 Baseline 0·09 (0·14) 0·5011 0·32 (0·10) 0·0008
Change 0·26 (0·10) 0·0061 0·37 (0·08) <0·0001
Model 2 Baseline 0·22 (0·14) 0·0688 0·17 (0·10) 0·0056
Change 0·29 (0·09) 0·0015 0·28 (0·09) <0·0001
Model 3 Baseline 0·13 (0·16) 0·4112 0·04 (0·11) 0·7596
Change 0·32 (0·10) 0·0021 0·23 (0·09) 0·0127
Model 1 includes the following covariates: age (continuous years), study centre, sex, education (highest attained,
continuous years), baseline HOMA insulin resistance (continuous). Model 2 contains the extra covariates: cigarette smoking
status (never, former, current, at baseline and year 15), alcohol intake (mL/day, at baseline and 15-year change), physical
activity (units per day, at baseline and 15-year change), television viewing (h/day, at year 10 and change between years 10
and 15). Model 3 contains the covariates in model 2 and dietary intake (at baseline and change between baseline and year 7)
of total energy (kcal/day), saturated fatty acids (% kcal), unsaturated fatty acids (% kcal), and trans-fatty acids (g/day), fibre
(g/1000 kcal at year 7 only), and daily servings of fruits, non-starchy vegetables, wholegrains, refined grains, soft drinks,
meat, dairy, and fish.
Table 4: Mean adjusted 15-year changes in HOMA insulin resistance by three times per week year 0
differences in fast food frequency between participants, and by three times per week 15-year change in
fast food frequency within participants
< 1 1–2 > 2 < 0
0–1
> 1
10
12
14
16
Weight gain
(kg)
15-year change
in fast-food
frequency
(times per week)
Baseline fast-food frequency
(times per week)
Figure 1: Joint association of year 0 fast-food frequency and 15-year changes
in fast-food frequency with 15-year changes in bodyweight
The model is adjusted for the same covariates as in model 2 of table 3, with
ethnic origin as an additional covariate. Cell-specific sample sizes range from 87
(>2 times per week at baseline and >1 times per week change) to 730 (>2 times
per week at baseline and <0 times per week change).
< 1 1–2 > 2
< 0
0–1
> 1
0
0.5
1
1.5
15-year change
in fast-food
frequency
(times per week)
Baseline fast-food frequency
(times per week)
HOMA change
(units)
Figure 2: Joint association of year 0 fast-food frequency and 15-year changes
in fast-food frequency with 15-year changes in HOMA insulin resistance
The model is adjusted for the same covariates as in model 2 of table 4, with
ethnic origin as an additional covariate. Cell-specific sample sizes range from 79
(>2 times per week at baseline and >1 times per week change) to 672 (>2 times
per week at baseline and <0 times per week change).
For personal use. Only reproduce with permission from Elsevier Ltd
Articles
www.thelancet.com Vol 365 January 1, 2005 41
viewing. These stratified models were tested in the full
cohort with adjustment as in model 3 of tables 3 and 4
(evaluated for both year 0 fast food and change in fast
food). The interaction terms between fast-food
frequency (year 0 and 15-year change) and the activity
index were not significant for weight change or for
insulin resistance (p values range 0·26–0·51). Thus,
there was a direct, albeit underpowered, association
between fast-food frequency and 15-year changes in
bodyweight and insulin resistance among all activity
categories (data not shown). In fact, there seemed to be
an especially strong association between fast-food
frequency and changes in bodyweight and insulin
resistance among the least physically active
participants—eg, a change in fast-food frequency of
three times per week was associated with a mean
increase in bodyweight of 3·9 kg (SE 0·9, n=386;
p<0·0001).
Discussion
In young adult black and white men and women living
in the USA between 1985–86 and 2000–01, we recorded
strong positive associations between frequency of visits
to fast-food restaurants and increases in bodyweight and
insulin resistance, the two major risk factors for type 2
diabetes.27 By comparison with the average 15-year
weight gain in participants with infrequent fast-food
restaurant use at year 0 and follow-up, those with
frequent fast-food restaurant use at both baseline and
follow-up gained an extra 4·5 kg bodyweight and had a
two-fold greater increase in insulin resistance. The
associations seemed to be largely independent of other
potentially confounding lifestyle factors, such as
physical activity and television viewing.
Our findings accord with those of cross-sectional and
shorter prospective studies of fast food and bodyweight.
Binkley and colleagues19 reported that fast-food
consumption was independently associated with
bodyweight in a cross-sectional analysis of 16 103 adults.
French and co-workers17 reported that intake of fried
potatoes predicted 2-year weight gain in women but not
men. In a 3-year prospective observational analysis of
891 women participating in a weight loss study, fast-food
restaurant use was directly associated with bodyweight.18
The associations between fast food and insulin
resistance were not fully accounted for by adjustment for
many other lifestyle factors, energy intake, nutrients,
and food groups, raising the possibility that some
unmeasured factor inherent to fast food might be
mediating this effect. The one most obvious aspect of
fast food that might lead to weight gain is the large
portion size, with certain single-meal calorie levels
approaching total daily energy intake requirements.11,13
We attempted to control for portion size through our
adjustment for energy intake; however, energy intake
was assessed only at years 0 and 7 and is measured with
error. Therefore, residual confounding in energy intake,
and in the other nutrients and foods that we examined,
is quite likely. Independent of energy intake, many other
aspects of fast food, including direct effects of energy
density on passive overconsumption,14 might also
predispose to insulin resistance. Certain fast foods
contain large amounts of partially hydrogenated oils,
and this class of fatty acids can cause insulin resistance
and increase risk of type 2 diabetes.15 Fast food also
contains large amounts of highly refined starchy food
and added sugar, carbohydrates that have been
characterised as high in glycaemic index.28 Consumption
of a high glycaemic index or high glycaemic load diet has
been linked to risk for diabetes, independent of
bodyweight changes, in mechanistic studies,16 short-
term human trials,29,30 and observational studies,31,32
although this point remains controversial.33
The strengths of our study include: long-term
prospective study design with high rates of follow-up
over 15 years; standardised, repeated, and detailed
measurements of dietary practices; direct assessment of
specific fast-food habits, rather than proxy measures
such as meals away from home or fried foods; use of the
Nutrition Coordinating Center (University of
Minnesota) archival system, which attempts to keep pace
with the food supply, including changes in use of fats in
fast food; extensive data for covariates with which to
explore confounders and mediators of the associations
under investigation; and the demographics of the
cohort—young adult black and white men and women
from four US metropolitan areas who have been
examined during a period of life when substantial
weight gain occurs and chronic diseases arise.
The limitations of this study relate mainly to its
observational nature, with the possibility of residual
confounding precluding definitive conclusions about
causality, and to reliance on self-reported diet and other
lifestyle factors. Our conclusions are restricted to
frequency of fast-food restaurant use because we were
unable to analyse sufficiently the considerable range of
available fast-food items and their portion sizes.
Nonetheless, our results might underestimate the true
magnitude of the effect because of measurement error
and thorough adjustment for many covariates that might
be on the causal pathway—eg, food groups and
nutrients. Furthermore, several analytical issues might
have affected our results. Measurement error, if non-
differential, would tend to result in attenuated estimates
of the strength of association between fast-food intake
and the outcome variables. Similarly, analyses of
changes are limited by floor and ceiling effects in that
some people never eat fast food (so cannot decrease) or
eat most meals at fast-food restaurants (so cannot
increase). These difficulties would also tend to attenuate
noted relations compared with the true relation (such as
would be observed in a clinical trial of investigator-
controlled increase or decrease in fast-food intake). Only
data obtained in the first and last CARDIA examination
For personal use. Only reproduce with permission from Elsevier Ltd
Articles
42 www.thelancet.com Vol 365 January 1, 2005
are included in the reported analyses. Omission of
missing data or data from intermediate examination
could bias the results. To address these issues data were
also analysed by repeated measures regression analysis
with SAS PROC MIXED (data not shown), which
includes all examinations for which data are available for
every individual and properly handles correlated
observations within individuals. Because the repeated
measures regression analysis did not yield additional
insights into the data beyond that presented in the
tables, we did not present specific findings using that
more complex methodology.
In conclusion, fast-food habits have strong, positive,
and independent associations with weight gain and
insulin resistance in young black and white adults. Fast-
food consumption can be linked to adverse health
outcomes through plausible mechanisms, and results
from other studies lend support to our findings. In view
of the high and increasing rates of fast-food
consumption, further research into the effects of this
dietary pattern on public health should be given
priority.
Contributors
M Pereira designed the study, directed the statistical analyses, and wrote
the first draft of the report. A Kartashov did the statistical analyses.
C Ebbeling designed the study and revised the report. L Van Horn and
M Slattery supervised collection of dietary data and revised the report.
D Jacobs Jr designed the study, provided statistical expertise, and
revised the report. D Ludwig designed the study, provided supervision,
and wrote the report.
Conflict of interest statement
We declare that we have no conflict of interest.
Acknowledgments
This work was supported by the Charles H Hood Foundation, NIDDK
grant 1R01DK59240, NCRR grant M01 RR02172, and NHLBI contracts
N01-HC-48047, N01-HC-48048, N01-HC-48049, N01-HC-48050, and
N01-HC-95095.
References
1 Flegal KM, Carroll MD, Ogden CL, Clifford L. Prevalence and
trends in obesity among US adults, 1999–2000. JAMA 2002; 288:
1723–27.
2 Ogden CL, Flegal KM, Carroll MD, Johnson CL. Prevalence and
trends in overweight among US children and adolescents,
1999–2000. JAMA 2002; 288: 1728–32.
3 Thompson D, Edelsberg H, Colditz GA, Bird AP, Oster G. Lifetime
health and economic consequences of obesity. Arch Intern Med
1999; 159: 2177–83.
4 Allison D, Fontaine K, Manson J, Stevens J, Vanltallie T. Annual
deaths attributable to obesity in the United States. JAMA 1999; 282:
1530–38.
5 Colditz G. Economic costs of obesity and inactivity.
Med Sci Sports Exerc 1999; 31 (suppl 11):S663–67.
6 Sturm R. The effects of obesity, smoking, and drinking on medical
problems and costs. Health Affairs 2002; 21: 245–53.
7 Sinha R, Fisch G, Teague B, et al. Prevalence of impaired glucose
tolerance among children and adolescents with marked obesity.
N Engl J Med 2002; 346: 802–10.
8 Ludwig DS, Ebbeling CB. Type 2 diabetes mellitus in children:
public health and primary care considerations. JAMA 2001; 286:
1427–30.
9 Biing-Hwan L, Frazao E. Nutritional quality of foods at and away
from home. In: Food review. Washington, DC: US Department of
Agriculture, 1997: 33–40. http://www.ers.usda.gov/publications/
foodreview/aug1997/may97h.pdf (accessed Oct 8, 2004).
10 USDA. Continuing survey of food intakes by individuals,
1994–1996. Washington, DC: US Department of Agriculture, 1997.
11 Tecnomic Information Services. US foodservice industry forecast.
Chicago, IL: Technomic, 2002.
12 Guthrie JF, Lin B-H, Frazao E. Role of food prepared away from
home in the American diet, 1977–78 versus 1994–96: changes and
consequences. J Nutr Educ Behav 2002; 34: 140–50.
13 Ebbeling CB, Pawlak DB, Ludwig DS. Childhood obesity: public-
health crisis, common sense cure. Lancet 2002; 360: 473–82.
14 Prentice AM, Jebb SA. Fast foods, energy density and obesity: a
possible mechanistic link. Obesity Rev 2003; 4: 187–94.
15 Hu FB, Stampfer MJ, Manson JE, et al. Dietary fat intake and the
risk of coronary heart disease in women. N Engl J Med 1997; 337:
1491–99.
16 Ludwig DS. The glycemic index: physiological mechanisms relating
to obesity, diabetes, and cardiovascular disease. JAMA 2002; 287:
2414–23.
17 French SA, Jeffery RW, Forster JL, McGovern PG, Kelder SH,
Baxter JE. Predictors of weight change over two years among a
population of working adults: the healthy worker project. Int J Obes
1994; 18: 145–54.
18 French SA, Harnack L, Jeffery RW. Fast food restaurant use among
women in the Pound of Prevention study: dietary, behavioral and
demographic correlates. Int J Obes Relat Metab Disord 2000; 24:
1353–59.
19 Binkley JK, Eales J, Jekanowski M. The relation between dietary
change and rising US obesity. Int J Obes Relat Metab Disord 2000;
24: 1032–39.
20 Bowman SA, Gortmaker SL, Ebbeling CB, Pereira MA, Ludwig DS.
Effects of fast food consumption on energy intake and diet quality
among children in a national household survey. Pediatrics 2004;
113: 112–18.
21 Friedman GD, Cutter GR, Donahue RP, et al. CARDIA: study
design, recruitment, and some characteristics of the examined
subjects. J Clin Epidemiol 1988; 41: 1105–16.
22 Jacobs DR Jr, Hahn LP, Haskell WL, Pirie P, Sidney S. Reliability
and validity of a short physical activity history. J Cardiopulm Rehabil
1989; 9: 448–59.
23 McDonald A, Van Horn L, Slattery M, et al. The CARDIA dietary
history: development, implementation, and evaluation.
J Am Diet Assoc 1991; 91: 1104–12.
24 Liu K, Slattery M, Jacobs Jr D, et al. A study of the reliability and
comparative validity of the CARDIA dietary history. Ethn Dis 1994;
4: 15–27.
25 Pereira MA, Jacobs DR Jr, Van Horn L, Slattery ML, Ludwig DS.
Dairy consumption, obesity, and the insulin resistance syndrome in
young adults: the CARDIA study. JAMA 2002; 287: 2081–89.
26 Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF,
Turner RC. Homeostasis model assessment: insulin resistance and
beta-cell function from fasting plasma glucose and insulin
concentrations in man. Diabetologia 1985; 28: 412–19.
27 Edelstein SL, Knowler WC, Bain RP, et al. Predictors of progression
from impaired glucose tolerance to NIDDM: an analysis of six
prospective studies. Diabetes 1997; 46: 701–10.
28 Foster-Powell K, Holt SH, Brand-Miller JC. International table of
glycemic index and glycemic load values. Am J Clin Nutr 2002; 76:
5–56.
29 Jarvi AE, Karlström BE, Granfeldt YE, Björck IE, Asp N-GL,
Vessby BOH. Improved glycemic control and lipid profile and
normalized fibrinolytic activity on a low-glycemic index diet in
type 2 diabetic patients. Diabetes Care 1999; 22: 10–18.
30 Frost G, Leeds A, Trew G, Margara R, Dornhorst A. Insulin
sensitivity in women at risk of coronary heart disease and the effect
of a low glycemic index diet. Metabolism 1998; 47: 1245–51.
31 Salmeron J, Ascherio A, Rimm EB, et al. Dietary fiber, glycemic
load, and risk of NIDDM in men. Diabetes Care 1997; 20: 545–50.
32 Salmeron J, Manson JE, Stampfer MJ, Colditz GA, Wing AL,
Willett WC. Dietary fiber, glycemic load, and risk of non-insulin-
dependent diabetes mellitus in women. JAMA 1997; 277: 472–77.
33 American Diabetes Association. Evidence-based nutrition
principles and recommendations for the treatment and
prevention of diabetes related complications. Diabetes Care 2002;
25: S50–60.
