ORIGINAL INVESTIGATION
Active Commuting and Cardiovascular Disease Risk
The CARDIA Study
Penny Gordon-Larsen, PhD; Janne Boone-Heinonen, PhD; Steve Sidney, MD, MPH;
Barbara Sternfeld, PhD; David R. Jacobs Jr, PhD; Cora E. Lewis, MD
Background: There is little research on the associa-
tion of lifestyle exercise, such as active commuting (walk-
ing or biking to work), with obesity, fitness, and cardio-
vascular disease (CVD) risk factors.
Methods: This cross-sectional study included 2364 par-
ticipants enrolled in the Coronary Artery Risk Develop-
ment in Young Adults (CARDIA) study who worked out-
side the home during year 20 of the study (2005-2006).
Associations between walking or biking to work (self-
reported time, distance, and mode of commuting) with
body weight (measured height and weight); obesity (body
mass index [BMI], calculated as weight in kilograms di-
vided by height in meters squared,�30); fitness (symp-
tom-limited exercise stress testing); objective moderate-
vigorous physical activity (accelerometry); CVD risk
factors (blood pressure [oscillometric systolic and dia-
stolic]); and serum measures (fasting measures of lipid,
glucose, and insulin levels) were separately assessed by
sex-stratified multivariable linear (or logistic) regres-
sion modeling.
Results: A total of 16.7% of participants used any means
of active commuting to work. Controlling for age, race,
income, education, smoking, examination center, and
physical activity index excluding walking, men with any
active commuting (vs none) had reduced likelihood of
obesity (odds ratio [OR],0.50; 95% confidence interval
[CI], 0.33-0.76), reduced CVD risk: ratio of geometric
mean triglyceride levels (trigactive)/(trignonactive)=0.88 (95%
CI, 0.80 to 0.98); ratio of geometric mean fasting insu-
lin (FIactive)/(FInonactive)=0.86 (95% CI, 0.78 to 0.93); dif-
ference in mean diastolic blood pressure (millimeters of
mercury) (DBPactive) − (DBPnonactive) = −1.67 (95% CI,
−3.20 to −0.15); and higher fitness: mean difference in
treadmill test duration (in seconds) in men (TTactive)
−(TTnonactive)=50.0 (95% CI, 31.45 to 68.59) and women
(TTactive)−(TTnonactive)=28.77 (95% CI, 11.61 to 45.92).
Conclusions: Active commuting was positively associ-
ated with fitness in men and women and inversely asso-
ciated with BMI, obesity, triglyceride levels, blood pres-
sure, and insulin level in men. Active commuting should
be investigated as a modality for maintaining or improv-
ing health.
Arch Intern Med. 2009;169(13):1216-1223
B ECAUSE OF ITS FLEXIBILITYand accessibility,1,2 walk-ing is generally reported asthe most popular leisure-time physical activity for
adults3-5 and has been specifically pro-
moted as a targeted activity to achieve na-
tional physical activity recommenda-
tions.1,6 For most adults, walking 60
minutes per day at a brisk pace is suffi-
cient to meet the Institute of Medicine’s
physical activity guidelines for avoiding
weight gain.7,8 One potentially effective
means of increasing physical activity is
through alternative, nonleisure forms of
physical activity such as active commut-
ing (walking or biking to work).
Selected research has suggested an in-
verse relationship between leisure-time
walking and adiposity9,10 and cardiovas-
cular disease (CVD) risk factors.11-14 A re-
cent meta-analysis15 showed modest re-
ductions in cardiovascular outcomes
related to active commuting, but most of
these studies were conducted in Scandi-
navian samples. The frequency of active
commuting is likely to vary by region. Fur-
thermore, the prevalence of obesity and
CVD risk factors is higher in the United
States than in Scandinavia. In addition,
many previous analyses have not in-
cluded detailed behavioral and clinical con-
trol variables that would permit investi-
gations of whether active commuting has
an independent effect on cardiovascular
health. Thus, research is needed in US
population–based cohorts with rich be-
havioral data and clinically measured CVD
risk factors.
We used population-based data from the
Coronary Artery Risk Development in
Young Adults (CARDIA) study to exam-
Author Affiliations:
Department of Nutrition,
School of Public Health,
University of North Carolina
at Chapel Hill
(Drs Gordon-Larsen and
Boone-Heinonen);
Epidemiology and Prevention
Section, Division of Research,
Kaiser Permanente, Oakland,
California (Drs Sidney and
Sternfeld); Division of
Epidemiology and Community
Health, University of
Minnesota, Minneapolis
(Dr Jacobs); Department
of Nutrition, University of Oslo,
Oslo, Norway (Dr Jacobs); and
Division of Preventive
Medicine, University of
Alabama at Birmingham
(Dr Lewis).
(REPRINTED) ARCH INTERN MED/ VOL 169 (NO. 13), JULY 13, 2009 WWW.ARCHINTERNMED.COM
1216
©2009 American Medical Association. All rights reserved.
Downloaded From: http://archinte.jamanetwork.com/ on 09/13/2013
ine the association between active commuting (defined as
walking or biking to work) with obesity, fitness, and CVD
risk factors (blood pressure [BP] and lipid, blood glucose,
and insulin levels) to understand whether active commut-
ing is a feasible target for maintaining or improving health.
We hypothesized that active commuting is positively as-
sociated with lower obesity, higher fitness, and a favor-
able CVD risk factor profile.
METHODS
SETTING AND PARTICIPANTS
The CARDIA study is a population-based prospective epide-
miologic study of the determinants and evolution of CVD risk
factors among young adults. At baseline (1985-1986), 5115 eli-
gible participants, aged 18 to 30 years, were enrolled with bal-
ance by race, sex, education (high school or less and more than
high school), and age (18-24 years and 25-30 years) from the
populations of Birmingham, Alabama; Chicago, Illinois; Min-
neapolis, Minnesota; and Oakland, California. Specific recruit-
ment procedures are described elsewhere.16 Six follow-up ex-
aminations were conducted over 20 years. We used data from
the year-20 (2005-2006) examination, with the year-20 reten-
tion rate for surviving cohort members of 72%.
From the initial 3549 study participants at year 20, we ex-
cluded 1 transgendered respondent (n=1) and women who were
pregnant at the time of examination (n=6). We further ex-
cluded participants who reported that they did not work outside
of the home (n=507) or for whom data on work outside of the
home were missing (n=567) and those missing outcome or co-
variate data (n=104). The final analysis sample included 2364
individuals with complete exposure, outcome, and covariate data.
Among those meeting inclusion criteria, white individuals, non-
smokers, and those with high income, education, and physical
activity levels were more likely to have complete data and thus
were included in analysis. Missing data also varied by study site,
with those in Minneapolis less likely to have complete data. This
secondary data analysis was approved by the CARDIA steering
committeeand the institutional reviewboardofUniversityofNorth
Carolina at Chapel Hill (UNC-CH).
EXPOSURE MEASURE: ACTIVE COMMUTING
At the year-20 examination, participants reported (in minutes
and miles) how long it takes to get from home to their place of
work and the percentage of the trip taken by car, public trans-
portation (bus, train, subway), walking, or bicycling. Active com-
muting was defined as any walking or biking during the trip
from home to work.
OUTCOME MEASURES
BMI and Obesity
Measurements of weight and height, with participants in light
clothing and without shoes, were obtained according to stan-
dardized protocol described previously.17 Body mass index (BMI)
was calculated as weight in kilograms divided by height in me-
ters squared, and obesity was classified as a BMI of at least 30.0.
Leisure Time and Occupational Physical Activity
At each examination, self-reported physical activity was ascer-
tained by an interviewer-administered questionnaire designed
for CARDIA. Participants were asked about the frequency of
participation in 13 different physical activity (PA) categories
(8 vigorous and 5 moderate [VPAs and MPAs, respectively])
of recreational sports, exercise, leisure, and occupational ac-
tivities over the previous 12 months. The VPAs included run-
ning, racquet sports, bicycling faster than 10 miles per hour,
swimming, vigorous exercise classes, sports (eg, basketball, foot-
ball), heavy lifting, carrying or digging on the job, and home
activities such as snow shoveling or lifting heavy objects. The
MPAs included nonstrenuous sports (eg, softball), walking,
bowling or golf, home maintenance (eg, gardening or raking),
and calisthenics. Because participants were not asked explic-
itly about duration of activity, PA scores are expressed in ex-
ercise units (EU), from which duration can be estimated.18 Scores
were computed by multiplying the intensity of the activity by
the number of months of participation, weighted by a factor
proportional to lesser or greater frequency and duration. Sepa-
rate scores were obtained for VPAs and MPA. The 2 subscores
were summed for a total PA score. As an example, a score of
100 EU is roughly equivalent to participation in a VPA 2 or 3
hours per week for 6 months of the year, calculated as [6 METs�
(3�6 months of high volume activity)], where MET indi-
cates metabolic equivalent. The reliability and validity of the
instrument is comparable with other activity questionnaires.18
Using the PA scoring algorithm, we created 2 PA mea-
sures. First, we created a specific leisure-walking score de-
rived from walking items in the PA questionnaire as de-
scribed. We used the continuous walking score, ranging from
0 to 144 EU, to categorize 12-month walking patterns at 3 METs
defined as multiples of the resting metabolic rate: none (0 EU),
intermittent (1-143 EU), and regular (144 EU, approximating
walking�4 h/wk over a 12-month period) to capture partici-
pants with no, moderate, and high levels of walking. Second,
we created a PA score that excluded walking, which was di-
chotomized into low (below the median) and high. This “non-
walking” activity variable was used as a control variable in our
multivariable regression models to statistically control for PAs
other than walking for transit or leisure in models using active
transit PA exposures.
Accelerometer-Measured PA
Total daily minutes of MPA and VPA were obtained from at least
4 days of accelerometer recordings. The MPA cut points were es-
tablished during a treadmill walking session using Freedson cut
points (1952-5725 counts/min). Participants were instructed to
wear the accelerometer (model 7164; ActiGraph, Pensacola,
Florida) around the waist for 7 days, except when sleeping, bath-
ing, or engaging in water activities. The epoch was set at 1 minute,
and periods of nonwear were identified by 60 or more consecu-
tive zero counts. At least 4 days of valid data (�720 minutes of
inactive time) were required for inclusion in analyses. The MPA
minutes per day were dichotomized (�24.0 and�24.1 min/d),
equivalent to the recommended 5 MPA bouts/wk and examined
as a CVD-related health behavior (outcome). The mean acceler-
ometer-measured minutes per day of VPA were dichotomized into
meeting (vs not meeting) VPA recommendations and used to ex-
clude those meeting VPA recommendations (n=102 men and 98
women of those with valid accelerometer data) in models exam-
ining accelerometer-measured MPA. The rationale for the exclu-
sion was to tie findings directly with the recommendation for
MPA.19,20
Treadmill Fitness Test Duration
A symptom-limited maximal Graded Exercise Test was admin-
istered using a modified Balke protocol,21 including nine
(REPRINTED) ARCH INTERN MED/ VOL 169 (NO. 13), JULY 13, 2009 WWW.ARCHINTERNMED.COM
1217
©2009 American Medical Association. All rights reserved.
Downloaded From: http://archinte.jamanetwork.com/ on 09/13/2013
2-minute stages of increasing difficulty with participants en-
couraged to exercise to exhaustion, followed by a recovery pe-
riod at a speed of 3.2 km/h at 0% grade. Fitness was indicated
by the treadmill test duration in seconds. Primary exclusion
criteria for exercise testing included a resting systolic or dia-
stolic BP (SBP or DBP) measurement greater than 160 or greater
than 100 mm Hg, respectively, or being febrile at time of ex-
amination.
Lipid, Glucose, and Insulin Measurements
Samples of blood lipids, glucose, and insulin were collected ac-
cording to standardized CARDIA protocols and were pro-
cessed at central laboratories as described previously.22-25 In-
dividuals fasting for less than 8 hours were excluded from these
analyses. Insulin was measured by radioimmunoassay.24 We cre-
ated the following measures: high-density lipoprotein choles-
terol (HDL-C), low-density lipoprotein cholesterol (LDL-C),
and triglyceride levels (all 3 measures reported in milligrams
per deciliters) and exclude participants reporting cholesterol-
lowering medications (n=194), and fasting glucose and fast-
ing insulin levels (both measures reported in milligrams per
deciliters) and exclude participants reporting diabetes medi-
cations (n=85).
BP Measurements
Three SBP and DBP measurements were obtained by a trained
technician using a standard automated BP measurement moni-
tor (model HEM907XL; Omron, Bannockburn, Illinois) after
a 5-minute seated rest. The mean of the second and third mea-
surements was used for analysis. Participants were asked to fast
for at least 12 hours and not to smoke or engage in heavy PA
for at least 2 hours prior to the measurement. We used SBP and
DBP measurements calibrated to be comparable with random-
zero sphygmomanometers used in prior CARDIA examina-
tion periods (SBP calibrated to the random zero level was es-
timated as 3.74�0.96� the Omron value; DBP as 1.30�0.97�
the Omron value) and reported in milligrams per deciliter and
excluding participants reporting use of BP-lowering medica-
tions (n=374).
Control Variables
Sociodemographic and behavioral characteristics were mea-
sured by self- and interviewer-administered questionnaires. Age
(years), race (black or white), income tertiles (�$50 000,
$50 000-$99 999, or�$100 000), years of education (high school
or less, any college, graduate school or professional training),
and clinic site (Birmingham, Chicago, Minneapolis, or Oak-
land) were used as control variables in all statistical models.
Smoking status was classified as never smoker, former smoker,
or current smoker, and alcohol intake was classified as no
consumption, 12 mL/d or less (sample median), or more than
12 mL/d).
STATISTICAL ANALYSIS
Statistical analyses were conducted using Stata software (ver-
sion 9.2; StataCorp, College Station, Texas). Descriptive sta-
tistics were computed for commuting patterns, nonwalking PA,
smoking, alcohol consumption, and sociodemographic fac-
tors and presented by active vs nonactive commuting (any walk-
ing or biking during the trip from home to work) and sex. Per-
centages were calculated for categorical variables. Continuous
variables were calculated either as means and standard errors
or median and interquartile range (for skewed measures).
Associations between walking or biking to work and BMI,
fitness, and CVD risk factors were separately assessed by sex-
stratified multivariate regression (linear, logistic, or multino-
mial logistic) modeling. If necessary, outcome variables were
transformed or categorized based on their sample distribu-
tion. Skewed variables were natural log transformed to achieve
approximate normality or categorized into ordinal variables if
transformation was not adequate. Leisure walking and accel-
erometer-measured leisure MPA were examined as categori-
cal variables to explicitly examine policy-relevant categories of
PA. All models adjusted for sociodemographics (age, race, in-
come, years of education, and examination center). Leisure-
time walking models also adjusted for nonwalking PA score (to
hold all nonwalking PAs constant). Accelerometer-measured
MPA models excluded those meeting accelerometer-
measured VPA recommendations (�8 min/d). Two sets of fit-
ness models—obesity and BMI—were conducted: model 1 ad-
justed for sociodemographics and model 2 adjusted for
sociodemographics and health-related behaviors (alcohol con-
sumption, smoking, and nonwalking PA score). In addition,
models for lipid levels, BP, and fasting glucose and insulin mea-
surements controlled for BMI to examine BMI as a potential
mediator.
Measures of effect varied across models, depending on the
outcome measure. For categorical outcomes, adjusted odds ra-
tios were used. For continuous natural log–transformed out-
comes, we calculated the ratio of the outcome in its reported
scale for those who actively commute relative to those who do
not. For continuous untransformed outcome, we calculated the
difference in outcome between those who actively commute
vs those who do not.
Interactions between active commuting and MPAs to VPAs
other than walking were tested by including the appropriate
cross-product terms in the model and assessing likelihood ra-
tio tests (P� .10). Final models were stratified by sex. Vari-
ables were retained in models if backward elimination re-
sulted in a greater than 10% change in the estimated effect
measures or if variables were conceptually relevant (eg, con-
trol for clinic site).
RESULTS
DESCRIPTIVE CHARACTERISTICS
Of the 2364 respondents who worked outside of the home,
16.7% of the sample (men, 18.0%; women, 15.6%) used
any means of active commuting to work. In both sexes,
active commuters were generally of higher education lev-
els, with variation across examination center (particu-
larly low active commuting was found in Birmingham).
Among women, active commuting was higher among
whites and those with higher nonwalking PA levels.
Among men, active commuting was higher in those with
greater alcohol intake (Table 1).
Patterns of commuting behavior, shown in Table 2,
were reported for the total trip to work (eg, participants
reported the percentage of trip made by walking, bike, car,
and public transportation). Average miles and minutes of
the commute to work varied between active and nonac-
tive commuters, with medians of 5 miles (for men and wom-
en) and 20 and 17 minutes (for men and women, respec-
tively) for those who actively commuted to work (distance
and minutes of commuting may not correspond owing to
combined modes of transportation). Considerably higher
proportions of participants used walking vs biking for their
(REPRINTED) ARCH INTERN MED/ VOL 169 (NO. 13), JULY 13, 2009 WWW.ARCHINTERNMED.COM
1218
©2009 American Medical Association. All rights reserved.
Downloaded From: http://archinte.jamanetwork.com/ on 09/13/2013
active commuting. There was variation across modes of tran-
sit, even for those who used active means of commuting,
with the highest proportions of commuters using cars for
some portion of their commute. Of note are low overall rates
of active commuting.
Likelihood of leisure walking was positively related
to active commuting, with strongest association seen for
the regular walker vs nonwalker comparison (Table 3).
Similarly, among those meeting VPA recommenda-
tions, accelerometer-measured MPA was positively re-
lated to active commuting, although statistically signifi-
cant for women only (P=.002).
Treadmill fitness test duration (in seconds) was higher
among men and women who actively vs nonactively com-
muted to work in models with adjustment for sociode-
mographics only (model 1) and then adding smoking,
alcohol, and leisure PA, excluding walking (model 2)
(Table 4). Similarly, BMI and likelihood of obesity were
lower among men who were active (vs nonactive) com-
muters. When analyses were restricted to those living
within 2 miles of their place of work, results were simi-
lar (with the exception of women for BMI and obesity).
Table 5 contrasts associations between active com-
muting and CV