prospective associations between sedentary lifestyle and bmi in midlife*
TRANSCRIPT
Prospective Associations between SedentaryLifestyle and BMI in MidlifeLaust H. Mortensen,*† Ilene C. Siegler,† John C. Barefoot,†‡ Morten Grønbæk,* and Thorkild I.A. Sørensen‡
AbstractMORTENSEN, LAUST H., ILENE C. SIEGLER, JOHN C.BAREFOOT, MORTEN GRØNBÆK, AND THORKILDI.A. SØRENSEN. Prospective associations betweensedentary lifestyle and BMI in midlife. Obesity. 2006;14:1462–1471.Objective: A strong positive cross-sectional relationshipbetween BMI and a sedentary lifestyle has been consistentlyobserved in numerous studies. However, it has been ques-tioned whether high BMI is a determinant or a consequenceof a sedentary lifestyle.Research Methods and Procedures: Using data from fourfollow-ups of the University of North Carolina AlumniHeart Study, we examined the prospective associations be-tween BMI and sedentary lifestyle in a cohort of 4595middle-aged men and women who had responded to ques-tionnaires at the ages of 41 (standard deviation 2.3), 44(2.3), 46 (2.0), and 54 (2.0).Results: BMI was consistently related to increased risk ofbecoming sedentary in both men and women. The oddsratios of becoming sedentary as predicted by BMI were 1.04(95% confidence limits, 1.00, 1.07) per 1 kg/m2 from ages41 to 44, 1.10 (1.07, 1.14) from ages 44 to 46, and 1.12(1.08, 1.17) from ages 46 to 54. Controlling for concurrentchanges in BMI marginally attenuated the effects. Seden-tary lifestyle did not predict changes in BMI, except whenconcurrent changes in physical activity were taken intoaccount (p � 0.001). The findings were not confounded by
preceding changes in BMI or physical activity, age, smok-ing habits, or sex.Discussion: Our findings suggest that a high BMI is adeterminant of a sedentary lifestyle but did not provideunambiguous evidence for an effect of sedentary lifestyle onweight gain.
Key words: physical activity, cohort study, University ofNorth Carolina Alumni Heart Study, BMI, adults
IntroductionWith obesity reaching epidemic proportions, knowledge
of modifiable risk factors for weight gain is pivotal forpublic health interventions to successfully target this prob-lem. A candidate for one such risk factor is lack of physicalactivity, i.e., sedentary lifestyle. A strong inverse cross-sectional relationship between BMI and a sedentary lifestylehas been consistently observed in numerous studies. Thishas predominantly been attributed to the effects of physicalactivity on BMI (1–4). However, observational studies haveyielded mixed results, which may, to some extent, be attrib-uted to different conceptual and methodological ap-proaches. Recently, two observational studies have sug-gested that a high BMI is a determinant rather than aproduct of a sedentary lifestyle, which could serve as analternative explanation to the cross-sectional associationbetween BMI and sedentary lifestyle (5,6). If a high BMIcan be considered both a determinant and a product ofsedentary lifestyle, then the relationship between BMI andsedentary lifestyle is an instance of bidirectional causation.That the temporal sequence is unknown implies that wecannot determine whether or not concurrent changes in onevariable are possible causes of changes in the other. Theimplications of bidirectional causation on statistical model-ing in the study of obesity and sedentary lifestyle have beendiscussed previously. Petersen et al. and Bak et al. (5,6)have argued that changes in physical activity between twopoints in time should not be included in the regressionanalysis because changes may have occurred after thechange in the dependent variable (i.e., BMI or obesity) and
Received for review April 11, 2005.Accepted in final form June 5, 2006.The costs of publication of this article were defrayed, in part, by the payment of pagecharges. This article must, therefore, be hereby marked “advertisement” in accordance with18 U.S.C. Section 1734 solely to indicate this fact.*Research Programme on Health Behavior, Lifestyle, and Living Conditions, NationalInstitute of Public Health, Centre for Health and Society, Copenhagen, Denmark; †Behav-ioral Medicine Research Center, Department of Psychiatry and Behavioral Sciences, DukeUniversity, Durham, North Carolina; and ‡Danish Epidemiology Science Centre, Institute ofPreventive Medicine, Copenhagen University Hospital, Centre for Health and Society,Copenhagen, Denmark.Address correspondence to Laust H. Mortensen, National Institute of Public Health, ØsterFarimagsgade 5, 1399 Copenhagen K, Denmark.E-mail: [email protected] © 2006 NAASO
1462 OBESITY Vol. 14 No. 8 August 2006
may, in fact, even be caused by changes in the dependentvariable, leading to reverse causality. For that reason, theauthors argue that a valid interpretation of temporal se-quence requires that only measures of physical activityassessed before the changes in weight occurs are included.However, in these studies, the time interval between base-line and follow-up were so long that the associations be-tween physical activity and BMI changes may have beendiluted by changes in both, including regression to the meantendencies. Moreover, changes in both before the baselinemay have created spurious associations due to such regres-sion to the mean tendencies. The aim of this study was toexamine the prospective associations between BMI and asedentary lifestyle over fairly close follow-up waves in acohort of middle-aged adults and to examine to what extentcontrolling for preceding and concurrent changes in physi-cal activity and BMI influences the results.
Research Methods and ProceduresThe University of North Carolina Alumni Heart Study
(UNCAHS)1 is a longitudinal study of 4595 men andwomen who attended University of North Carolina inChapel Hill during the period 1964 to 1966. The UNCAHScohort was born primarily from 1940 to 1949. The cohortwas followed up with regards to a variety of lifestyle andpsychosocial variables in yearly and biyearly waves from1987 onwards. This paper will use data collected from foursurveys, collected in 1987, 1989, 1992, and 2001. Theresponse rate to the 1987 questionnaire was 75.9%. Thecorresponding response rates for the 1989, 1992, and 2002questionnaires were 76.3%, 71.2%, and 60.4%. Data anddata collection have been described in greater detail else-where (7).
MeasurementsInformation on body mass and leisure time physical ac-
tivity (ltpa) was assessed four times: at the approximateages 41 (standard deviation: 2.3), 44 (2.3), 46 (2.0), and 54(2.0). BMI (kilograms per meter squared) was calculatedfrom self-reported height and weight. At age 41 (1987),information on ltpa was assessed by the question: Howmany hours a week, on average, do you exercise or playsports for fun or to keep in shape, not counting job orhousework activities? At age 44 (1989), the question was:On average, how many hours of exercise do you get a week?Respondents were asked to report the number of hours spentweekly on sports and physical fitness activities and theweekly number of hours spent on normal activities at workand household chores. For comparability, only hours spenton sports and physical activity were included. At ages 46
(1991) and 54 (2000), the scale for rating physical activityfor use with the Houston Non-exercise VO2 Test was ad-ministered (8). The scale is measured on an eight-categoryordinal scale, with zero representing no physical activity. Toensure comparability across waves, each of these variableswas recoded into a binary variable, reflecting a sedentarylifestyle. At ages 41 and 44, respondents who reported 0hours of ltpa per week were classified as sedentary. At ages46 and 54, having a sedentary lifestyle was defined asbelonging to the category: do not participate regularly inprogrammed recreation sport or heavy physical activity.Covariates include smoking status (current, former, never),age, and sex.
Statistical AnalysisThe analytical strategy was to compare the observed
effects of BMI and ltpa as assessed in two different analyt-ical models, termed baseline and concurrence. The baselinemodel does not include changes in the independent vari-ables during the time interval where changes in the outcomevariable are recorded. The concurrence model adjusts forchanges in the independent variables that occur concur-rently with changes in the outcome variable. The models areschematically represented in Figure 1.
In analysis of the effect of sedentary lifestyle on BMI,change in BMI from one follow-up to the next was used asoutcome in linear regression models; for example, change inBMI from first to second assessment was calculated as:[BMI at the second assessment] � [BMI at the first assess-ment] and so forth. Distributional assumptions, linearity,and model fit of linear regressions were assessed by exam-ining the residuals.
1 Nonstandard abbreviations: UNCAHS, University of North Carolina Alumni Heart Study;ltpa, leisure time physical activity.
Figure 1: Schematical representation of the analytic models: base-line and concurrence.
Sedentary Lifestyle and BMI, Mortensen et al.
OBESITY Vol. 14 No. 8 August 2006 1463
Analyses of dichotomous outcome variables sedentarylifestyle or obesity were carried out by logistic regressionmodels, for which we excluded respondents who were al-ready sedentary or obese, respectively, at the beginning ofthe interval in which the risk (expressed as odds ratios) ofbecoming sedentary or obese was estimated; for example, asubject was considered to have become sedentary at thethird assessment only if he had been non-sedentary at thesecond assessment and sedentary at the third assessment.
When several BMI assessments were used as independentvariables, we entered BMI at the beginning at the interval inwhich the risk of change in ltpa was assessed as a contin-uous variable into the regression equation. Other measuresof BMI were entered as change in BMI from one assessmentto the next. This strategy was chosen to avoid problems ofcollinearity between BMI measures. For example, in thelogistic regression analysis where a dichotomous variableindicating whether or not a subject became sedentary fromthe second to the third assessment is used as outcome
variable, two continuous BMI variables were included: BMIat the second assessment and change in BMI from first tosecond assessment (i.e., [BMI at second assessment] �[BMI at first assessment]). Previous assessments of BMIand ltpa were included as potentially confounding or effect-modifying variables where such assessments were avail-able; for example, change in BMI from ages 46 to 54 wasadjusted for ltpa and BMI at ages 41, 44, and 46. Statisti-cally significant refers to inference based on statistic tests orcorresponding confidence intervals with a 5% type I errorrate per test.
ResultsTable 1 shows the sample characteristics at each of the
four assessments. Table 2 shows the cross-sectional associ-ations between sedentary lifestyle and BMI at each of thefour points of assessment. As is evident from Table 2,sedentary lifestyle and BMI are strongly and significantlyassociated at all four assessments.
Table 1. Characteristics at ages 41, 44, 46, and 56; the UNCAHS
Variable
Age 41 (N �
841, women;N � 3754,
men)
Age 44 (N �
734, women;N � 3016,
men)
Age 46 (N �
678, women;N � 2583,
men)
Age 54 (N �
516, women;N � 1946,
men)
Mean SD Mean SD Mean SD Mean SD
WomenAge 40.68 3.01 43.19 3.06 45.14 2.92 54.23 3.14BMI (kg/m2) 22.76 4.34 23.14 4.35 23.95 4.78 25.39 5.32Change in BMI since last assessment 0.47 1.77 1.02 1.78 1.46 2.46Obese (%) 6.54 7.22 9.10 14.66Becoming obese since last assessment (%) 2.32 3.05 6.85Sedentary lifestyle (%) 18.31 21.39 30.92 26.95Becoming sedentary since last assessment (%) 13.22 18.81 12.91Current smokers (%) 15.81 15.12 14.45 15.89Former smokers (%) 34.36 34.33 32.74 33.53
MenAge 40.47 1.92 42.98 1.96 44.96 1.87 53.93 1.70BMI 25.21 3.27 25.45 3.36 25.87 3.47 26.97 4.01Change in BMI since last assessment 0.35 1.31 0.54 1.49 1.09 1.99Obese (%) 7.33 8.06 10.85 18.15Becoming obese since last assessment (%) 2.39 4.46 8.93Sedentary lifestyle (%) 13.21 17.22 17.26 22.51Becoming sedentary since last assessment (%) 9.75 10.13 14.04Current smokers (%) 17.50 15.82 16.18 15.11Former smokers (%) 38.01 38.86 38.64 40.24
UNCAHS, University of North Carolina Alumni Heart Study; SD, standard deviation.
Sedentary Lifestyle and BMI, Mortensen et al.
1464 OBESITY Vol. 14 No. 8 August 2006
Table 3 shows the regression analysis of BMI as pre-dicted by sedentary lifestyle. The baseline models show thatthere was no association of sedentary lifestyle at the begin-ning of the interval with subsequent gain in BMI. Pointestimates range from 0.05 to 0.07 kg/m2 and are all non-significant. However, including concurrent changes into theregression equation significantly changes the estimates.Compared with respondents who were non-sedentarythroughout the interval, respondents who were sedentarythroughout the interval experienced an excess mean gain of0.43 kg/m2 from ages 41 to 44, an excess mean gain of 0.47kg/m2 from ages 44 to 46, and an excess mean gain of 0.73kg/m2 from ages 46 to 54. This corresponds to mean yearlyexcess gains of 0.14, 0.23, and 0.09 kg/m2. Respondentswho became sedentary had excess mean gains in BMI of0.18, 0.55, and 0.48 kg/m2 (mean yearly excess gains of0.06, 0.28, and 0.06 kg/m2) when compared with those werenon-sedentary throughout the interval. Respondents whobecame non-sedentary experienced a smaller gain in BMIthan those who were non-sedentary throughout the interval.The excess gains were �0.36, �0.07, and �0.33 kg/m2 inthe three age intervals (mean yearly excess gains of �0.12,�0.03, and �0.04 kg/m2). Adjustment for putative con-founders and preceding changes in ltpa and BMI did notchange these results. Table 4 shows the corresponding anal-ysis using obesity as an outcome. In these analyses, seden-
tary lifestyle was only borderline statistically significantlyassociated with obesity from ages 44 to 46, where a seden-tary lifestyle was associated with 1.71-fold [95% confi-dence limits, 1.02, 2.84] in odds. Although all other esti-mates were insignificant, point estimates generallysuggested that a sedentary lifestyle was a risk factor for thedevelopment of obesity.
Table 5 shows the results of the logistic regression anal-ysis of sedentary lifestyle as predicted by BMI. When usingthe baseline only approach, BMI was associated with in-creased risk of becoming sedentary in all three age intervals.An elevation of 1 kg/m2 in BMI at age 41 conferred astatistically significant increase in odds of becoming seden-tary from ages 41 to 44 of 4% (95% confidence interval, 0%to 7%). The corresponding increases in odds were 10% (6%to 14%) from ages 44 to 46 and 13% (9% to 18%) from ages45 to 54. Including concurrent changes in BMI did notchange the effect of BMI at a particular age. However,increase in BMI was independently associated with risk ofbecoming sedentary. An increase of 1 kg/m2 in BMI be-tween assessments was associated with increases in odds ofbecoming sedentary of 9% (1% to 18%) from ages 41 to 44,21% (13% to 31%) from ages 44 to 46, and 12% (5% to20%) from ages 46 to 54. Adjustment for putative con-founders and preceding changes in ltpa and BMI did notchange these results. The same pattern was observed when
Table 2. Cross-sectional associations between sedentary lifestyle and BMI (kg/m2), 95% confidence limits inparentheses; the UNCAHS
Variable N Mean 95% confidence limits p*
BMI at age 41Non-sedentary 3945 24.63 (24.52, 24.73)Sedentary 650 25.60 (25.24, 25.97)Difference in BMI �0.98 (�1.28, �0.68) �0.001
BMI at age 44Non-sedentary 3226 24.84 (24.72, 24.96)Sedentary 524 25.93 (25.52, 26.34)Difference in BMI �1.09 (�1.42, �0.75) �0.001
BMI at age 46Non-sedentary 2543 25.17 (25.04, 25.30)Sedentary 632 26.66 (26.27, 27.05)Difference in BMI �1.49 (�1.82, �1.16) �0.001
BMI at age 54Non-sedentary 1805 26.11 (25.94, 26.29)Sedentary 552 28.37 (27.90, 28.83)Difference in BMI �2.25 (�2.66, �1.85) �0.001
UNCAHS, University of North Carolina Alumni Heart Study.* Test of the null hypothesis: difference in BMI equal to zero.
Sedentary Lifestyle and BMI, Mortensen et al.
OBESITY Vol. 14 No. 8 August 2006 1465
Tab
le3.
Lin
ear
regr
essi
onof
sede
ntar
ylif
esty
leas
apr
edic
tor
ofex
cess
gain
inB
MI
(kg/
m2)
from
ages
41to
44,4
4to
46,a
nd46
to54
usin
gco
ncur
renc
ean
dba
selin
eap
proa
ches
,95
%C
Ls
inpa
rent
hese
s;th
eU
NC
AH
S
Bas
elin
eC
oncu
rren
ce
Var
iabl
e�
(95%
CL
)�
/yr
(95%
CL
)p
Var
iabl
e�
(95%
CL
)�
/yr
(95%
CL
)p
Age
s41
to44
†(n
�37
40)
Sede
ntar
y0.
05(�
0.08
,0.1
8)0.
02(�
0.03
,0.0
6)0.
47Se
dent
ary
0.43
(0.2
6,0.
60)
0.14
(0.0
9,0.
20)
�0.
0001
�0.
0001
*N
on-s
eden
tary
0(r
efer
ence
)B
ecom
eno
n-se
dent
ary
�0.
36(�
0.55
,�0.
17)
�0.
12(�
0.18
,�0.
06)
�0.
001
Bec
ome
sede
ntar
y0.
18(0
.03,
0.32
)0.
06(0
.01,
0.11
)0.
02N
on-s
eden
tary
0(r
efer
ence
)A
ges
44to
46‡
(n�
2998
)Se
dent
ary
0.05
(�0.
10,0
.20)
0.03
(�0.
05,0
.10)
0.50
Sede
ntar
y0.
47(0
.26,
0.68
)0.
23(0
.13,
0.34
)�
0.00
01�
0.00
01*
Non
-sed
enta
ry0
(ref
eren
ce)
Bec
ome
non-
sede
ntar
y�
0.07
(�0.
25,0
.12)
�0.
03(�
0.13
,0.0
6)0.
48B
ecom
ese
dent
ary
0.55
(0.3
8,0.
72)
0.28
(0.1
9,0.
36)
�0.
0001
Non
-sed
enta
ry0
(ref
eren
ce)
Age
s46
to54
§(n
�20
70)
Sede
ntar
y0.
07(�
0.17
,0.3
2)0.
01(�
0.02
,0.0
4)0.
55Se
dent
ary
0.73
(0.4
0,1.
07)
0.09
(0.0
5,0.
13)
�0.
0001
�0.
0001
*N
on-s
eden
tary
0(r
efer
ence
)B
ecom
eno
n-se
dent
ary
�0.
33(�
0.65
,�0.
01)
�0.
04(�
0.08
,0.0
0)0.
04B
ecom
ese
dent
ary
0.48
(0.2
1,0.
75)
0.06
(0.0
3,0.
09)
�0.
001
Non
-sed
enta
ry0
(ref
eren
ce)
CL
,co
nfid
ence
limit;
UN
CA
HS,
Uni
vers
ityof
Nor
thC
arol
ina
Alu
mni
Hea
rtSt
udy.
*p
valu
eof
likel
ihoo
dra
tiote
st(3
df).
†A
djus
ted
for
smok
ing
habi
tsan
dse
x.‡
Adj
uste
dfo
rsm
okin
gha
bits
,se
x,se
dent
ary
lifes
tyle
atag
e41
,an
dch
ange
inB
MI
from
ages
41to
44.
§A
djus
ted
for
smok
ing
habi
tsan
dse
x,se
dent
ary
lifes
tyle
atag
es41
and
44,
and
chan
gein
BM
Ifr
omag
es41
to44
and
44to
46.
Sedentary Lifestyle and BMI, Mortensen et al.
1466 OBESITY Vol. 14 No. 8 August 2006
Tab
le4.
Log
istic
regr
essi
onof
sede
ntar
ylif
esty
leas
apr
edic
tor
ofob
esity
(BM
I�
30kg
/m2)
from
ages
41to
44,4
4to
46,a
nd46
to54
usin
gco
ncur
renc
ean
dba
selin
eap
proa
ches
,95
%C
Ls
inpa
rent
hese
s;th
eU
NC
AH
S
Bas
elin
eC
oncu
rren
ce
Var
iabl
eO
R(9
5%C
L)
OR
/yr
(95%
CL
)p
Var
iabl
eO
R(9
5%C
L)
OR
/yr
(95%
CL
)p
Age
s41
to44
†(n
�34
86,
even
ts�
89)
Sede
ntar
y1.
51(0
.80,
2.76
)1.
15(0
.93,
1.41
)0.
20Se
dent
ary
1.97
(0.8
3,4.
35)
1.25
(0.9
5,1.
65)
0.10
Non
-sed
enta
ry1
(ref
eren
ce)
Bec
ome
non-
sede
ntar
y1.
26(0
.51,
2.85
)1.
08(0
.81,
1.44
)0.
59B
ecom
ese
dent
ary
1.35
(0.5
9,2.
84)
1.11
(0.8
5,1.
43)
0.44
Non
-sed
enta
ry1
(ref
eren
ce)
0.41
*A
ges
44to
46‡
(n�
2792
,ev
ents
�12
5)Se
dent
ary
1.71
(1.0
2,2.
84)
1.31
(1.0
1,1.
69)
0.04
Sede
ntar
y2.
17(1
.00,
4.56
)1.
47(1
.01,
2.15
)0.
05N
on-s
eden
tary
1(r
efer
ence
)B
ecom
eno
n-se
dent
ary
1.53
(0.8
1,2.
81)
1.24
(0.9
1,1.
69)
0.18
Bec
ome
sede
ntar
y1.
12(0
.46,
2.48
)1.
06(0
.70,
1.60
)0.
79N
on-s
eden
tary
1(r
efer
ence
)0.
18*
Age
s46
to54
§(n
�18
67,
even
ts�
172)
Sede
ntar
y0.
8(0
.47,
1.35
)0.
97(0
.91,
1.04
)0.
41Se
dent
ary
1.42
(0.7
0,2.
84)
1.05
(0.9
6,1.
14)
0.32
Non
-sed
enta
ry1
(ref
eren
ce)
Bec
ome
non-
sede
ntar
y0.
6(0
.28,
1.20
)0.
94(0
.86,
1.03
)0.
16B
ecom
ese
dent
ary
1.58
(0.9
4,2.
64)
1.06
(0.9
9,1.
13)
0.08
Non
-sed
enta
ry1
(ref
eren
ce)
0.07
*
CL
,co
nfid
ence
limit;
UN
CA
HS,
Uni
vers
ityof
Nor
thC
arol
ina
Alu
mni
Hea
rtSt
udy.
*p
valu
eof
likel
ihoo
d-ra
tiote
st(3
df).
†A
djus
ted
for
smok
ing
habi
tsan
dse
x.‡
Adj
uste
dfo
rsm
okin
gha
bits
,se
x,se
dent
ary
lifes
tyle
atag
e41
,an
dch
ange
inB
MI
from
ages
41to
44.
§A
djus
ted
for
smok
ing
habi
tsan
dse
x,se
dent
ary
lifes
tyle
atag
es41
and
44,
and
chan
gein
BM
Ifr
omag
es41
to44
and
44to
46.
Sedentary Lifestyle and BMI, Mortensen et al.
OBESITY Vol. 14 No. 8 August 2006 1467
Tab
le5.
Log
istic
regr
essi
onof
BM
I(k
g/m
2)
aspr
edic
tor
ofse
dent
ary
lifes
tyle
from
ages
41to
44,4
4to
46,a
nd46
to54
usin
gco
ncur
renc
ean
dba
selin
eap
proa
ches
,95
%C
Ls
inpa
rent
hese
s;th
eU
NC
AH
S
Bas
elin
eC
oncu
rren
ce
Var
iabl
eO
R(9
5%C
L)
OR
/yr
(95%
CL
)p
Var
iabl
eO
R(9
5%C
L)
OR
/yr
(95%
CL
)p
Age
s41
to44
†(n
�32
16,
even
ts�
390)
BM
I1.
04(1
.00,
1.07
)1.
01(1
.00,
1.02
)0.
03B
MI
1.04
(1.0
1,1.
07)
1.01
(1,1
.02)
0.02
Gai
nin
BM
I1.
09(1
.01,
1.18
)1.
03(1
,1.0
6)0.
02A
ges
44to
46‡
(n�
2452
,ev
ents
�35
5)B
MI
1.10
(1.0
6,1.
14)
1.05
(1.0
3,1.
07)
�0.
0001
BM
I1.
08(1
.04,
1.12
)1.
04(1
.02,
1.06
)�
0.00
01G
ain
inB
MI
1.21
(1.1
3,1.
31)
1.10
(1.0
6,1.
14)
�0.
0001
Age
s46
to54
§(n
�16
79,
even
ts�
280)
BM
I1.
13(1
.09,
1.18
)1.
02(1
.01,
1.02
)�
0.00
01B
MI
1.12
(1.0
7,1.
17)
1.01
(1.0
1,1.
02)
�0.
0001
Gai
nin
BM
I1.
12(1
.05,
1.20
)1.
01(1
.01,
1.02
)�
0.00
1
CL
,co
nfid
ence
limit;
UN
CA
HS,
Uni
vers
ityof
Nor
thC
arol
ina
Alu
mni
Hea
rtSt
udy.
*p
valu
eof
likel
ihoo
d-ra
tiote
st(3
df).
†A
djus
ted
for
smok
ing
habi
tsan
dse
x.‡
Adj
uste
dfo
rsm
okin
gha
bits
,se
x,se
dent
ary
lifes
tyle
atag
e41
,an
dch
ange
inB
MI
from
ages
41to
44.
§A
djus
ted
for
smok
ing
habi
tsan
dse
x,se
dent
ary
lifes
tyle
atag
es41
and
44,
and
chan
gein
BM
Ifr
omag
es41
to44
and
44to
46.
Sedentary Lifestyle and BMI, Mortensen et al.
1468 OBESITY Vol. 14 No. 8 August 2006
Tab
le6.
Log
istic
regr
essi
onof
obes
ity(B
MI
�30
kg/m
2)
aspr
edic
tor
ofse
dent
ary
lifes
tyle
from
ages
41to
44,
44to
46,
and
46to
54us
ing
conc
urre
nce
and
base
line
appr
oach
es,
95%
CL
sin
pare
nthe
ses;
the
UN
CA
HS
Bas
elin
eC
oncu
rren
ce
Var
iabl
eO
R(9
5%C
L)
OR
/yr
(95%
CL
)p
Var
iabl
eO
R(9
5%C
L)
OR
/yr
(95%
CL
)p
Age
s41
to44
†(n
�32
16,
even
ts�
390)
Obe
se1.
51(0
.98,
2.26
)1.
15(1
.00,
1.32
)0.
06O
bese
1.64
(1.0
3,2.
52)
1.18
(1.0
2,1.
37)
0.03
0.21
*N
otob
ese
1(r
efer
ence
)1
(ref
eren
ce)
Bec
ome
not
obes
e1.
04(0
.31,
2.68
)1.
01(0
.71,
1.44
)0.
93B
ecom
eob
ese
1.25
(0.6
0,2.
37)
1.08
(0.8
6,1.
35)
0.52
Not
obes
e1
(ref
eren
ce)
1(r
efer
ence
)A
ges
44to
46‡
(n�
2452
,ev
ents
�35
5)O
bese
2.23
(1.4
6,3.
34)
1.49
(1.2
2,1.
83)
�0.
001
Obe
se2.
38(1
.53,
3.64
)1.
54(1
.25,
1.91
)�
0.00
01�
0.00
01*
Not
obes
e1
(ref
eren
ce)
1(r
efer
ence
)B
ecom
eno
t-ob
ese
1.81
(0.4
0,5.
87)
1.34
(0.7
0,2.
54)
0.37
Bec
ome
obes
e2.
29(1
.38,
3.69
)1.
51(1
.18,
1.93
)�
0.00
1N
otob
ese
1(r
efer
ence
)1
(ref
eren
ce)
Age
s46
to54
§(n
�16
79,
even
ts�
280)
Obe
se1.
83(1
.13,
2.9)
1.08
(1.0
2,1.
14)
0.01
Obe
se2.
34(1
.41,
3.81
)1.
11(1
.05,
1.18
)�
0.00
1�
0.00
1*N
otob
ese
1(r
efer
ence
)1
(ref
eren
ce)
Bec
ome
not
obes
e0.
73(0
.11,
2.61
)0.
96(0
.80,
1.16
)0.
68B
ecom
eob
ese
2.05
(1.3
2,3.
14)
1.09
(1.0
4,1.
15)
�0.
01N
otob
ese
1(r
efer
ence
)1
(ref
eren
ce)
CL
,co
nfid
ence
limit;
UN
CA
HS,
Uni
vers
ityof
Nor
thC
arol
ina
Alu
mni
Hea
rtSt
udy.
*p
valu
eof
likel
ihoo
d-ra
tiote
st(3
df).
†A
djus
ted
for
smok
ing
habi
tsan
dse
x.‡
Adj
uste
dfo
rsm
okin
gha
bits
,se
x,se
dent
ary
lifes
tyle
atag
e41
,an
dch
ange
inB
MI
from
ages
41to
44.
§A
djus
ted
for
smok
ing
habi
tsan
dse
x,se
dent
ary
lifes
tyle
atag
es41
and
44,
and
chan
gein
BM
Ifr
omag
es41
to44
and
44to
46.
Sedentary Lifestyle and BMI, Mortensen et al.
OBESITY Vol. 14 No. 8 August 2006 1469
obesity was considered as a determinant of becoming sed-entary, although dichotomization of the BMI variable likelyreduced power (see Table 6).
DiscussionIn this study, we found BMI to be consistently related to
increased risk of becoming sedentary regardless of whetherconcurrent or preceding changes in BMI were included ascovariates. Sedentary lifestyle was not associated with in-creased gains in BMI when assessed prospectively in thebaseline model but was associated when changes in seden-tary lifestyle occurring between baseline and follow-upwere also included in the concurrent model. With regards tothe findings of this study, our analysis of the effect of asedentary lifestyle on BMI confirms the results of the ex-isting studies of this association. The studies conducted byBak et al. and Petersen et al. (5,6) were carried out in Danishstudy populations, and this study extends the findings toanother, but similar, population, i.e., middle-aged whiteAmericans. The effect of BMI at baseline was not substan-tially changed by whether or not concurrent changes in BMIwere taken into account. The findings of this study suggestthat a high BMI is a determinant of a sedentary lifestyle butdid not provide unambiguous evidence for an effect ofsedentary lifestyle on weight gain.
This study was conducted using a prospective observa-tional design with four follow-ups over a 14-year period.Compared with the previous studies, the multiple follow-ups and the short time period between follow-ups are majorstrengths of this study. A potentially important weakness ofthis study is the change in the method of assessment ofphysical activity across waves. These changes increase therisk of misclassification of physical activity, which may biasour findings toward inconsistency, but generally the resultsappeared to be almost independent of whether the measure-ments of physical activity were highly or less highly com-parable across the waves being evaluated. On the otherhand, the consistent findings despite these methodologicaldifferences may also be seen as strength of the study. Themultiple follow-ups allow for testing the hypothesis at sev-eral points in time and the relatively short follow-up time isan advantage when studying effects with short latencytimes. This design would generally be considered highlyvalid for studying the effect of risk factors over time.However, in some cases of bidirectional causation, obser-vational studies may have inherent weaknesses. Althoughwe recognize these methodological issues, we will contendthat these phenomena can be studied this way. Becauseeither analytical model can be contested and might intro-duce bias, a balanced interpretation requires a presentationof the results from each of the models. As noticed earlier,the findings are only suggestive as to the size of the trueeffect of BMI on sedentary lifestyle and vice versa due tolimitations in design. It is also important to recognize that
sedentary lifestyle and BMI are not measured with the sameprecision and that the asymmetry in our findings may, inpart, be due to this.
The apparently conflicting findings with regards to theeffect of sedentary lifestyle on BMI are also in line with theexisting findings of the literature. From studies using phys-ical activity at baseline as predictor of change in BMI orweight results are mixed. Several have reported no associ-ation (5,6,9), whereas others have reported protective ef-fects of physical activity at baseline (10,11). A study byBild et al. (12) found reported increased weight gain in thehighly physically active compared with less active studyparticipants. In studies controlling for concurrent changes inphysical activity, increases in physical activity are predom-inantly associated with decreases in weight (13–20), andthis is in line with our findings. We suggest that part of theexplanation for the positive findings when the concurrenceapproach is used is bias caused by reverse causation.
Because the relationship between BMI and a sedentarylifestyle is of great importance to public health, we suggestthat this association be studied in further detail. Anthropo-metric measures other than weight and height should beincluded, e.g., waist circumference, waist-to-hip ratio, andfat-free body mass. These measures may be more appropri-ate in describing the physiological effect of a sedentarylifestyle on body composition. Unfortunately, randomizedtrials addressing the concerns of external validity do notseem to be feasible. Compliance to the high amount ofexercise to possibly prevent weight gain over a substantialfollow-up period (21) may be difficult to achieve. The sizeof the study populations of such trials and the necessaryduration of follow-up for making inferences to weight gainwould make the trials very costly. Therefore, observationalepidemiological studies with even better control over puta-tive confounders and the interplay between preceding andsubsequent changes in both physical activity and bodyweight and composition are worthwhile.
ConclusionOur findings suggest that a high BMI is a determinant of
a sedentary lifestyle but do not unambiguously support thatsedentary lifestyle has an effect on later BMI changes.
AcknowledgmentsThis work was supported, in part, by The Danish National
Institute of Public Health and by Grants R01 HL54780 (toJ.C.B.) and R01 HL55356 (to I.C.S.) from the NationalHeart, Lung, and Blood Institute with co-funding from theNational Institute on Aging. The Danish National ResearchFoundation established the Danish Epidemiology ScienceCentre.
Sedentary Lifestyle and BMI, Mortensen et al.
1470 OBESITY Vol. 14 No. 8 August 2006
References1. Grundy S, Blackburn G, Higgins M, Lauer R, Perri M,
Ryan D. Physical activity in the prevention and the treatmentof obesity and its comorbidities. Med Sci Sports Exerc. 1999;31(Suppl 11):S502–8.
2. Hardeman W, Griffin S, Johnston M, Kinmonth A, Ware-ham N. Interventions to prevent weight gain: a systematicreview of psychological models and behaviour change meth-ods. Int J Obes Relat Metab Disord. 2000;24:131–43.
3. Haapanen-Niemi N, Miilunpaol S, Pasanen M, Vuori I,Oja P, Mallmberg J. Body mass index, physical inactivityand low level of physical fitness as determinants of all-causeand cardiovascular disease mortality: 16 y follow-up of mid-dle-aged and elderly men and women. Int J Obes Relat MetabDisord. 2000;24:1465–74.
4. Fogelholm M, Kukkonen-Harjula K. Does physical activityprevent weight gain: s systematic review. Obes Rev. 2000;1:95–111.
5. Petersen L, Schnohr P, Sørensen TIA. Longitudinal study ofthe long-term relation between physical activity and obesity inadults. Int J Obes Relat Metab Disord. 2004;28:105–12.
6. Bak H, Petersen L, Sørensen TIA. Physical activity inrelation to development and maintenance of obesity in menwith and without juvenile onset obesity. Int J Obes RelatMetab Disord. 2004;28:99–104.
7. Siegler IC, Peterson BL, Barefoot JC et al. Using collegealumni populations in epidemiologic research: the UNCAlumni Heart Study. J Clin Epidemiol. 1992;45:1243–50.
8. Jackson AS, Blair SN, Mahar MT, Wier LT, Ross RM,Stuteville JE. Prediction of functional aerobic capacity with-out exercise testing. Med Sci Sports Exerc. 1990;33:863–70.
9. Parker DR, Gonzalez S, Derby CA, Gans KM, LasaterTM, Carleton RA. Dietary factors in relation to weightchange among men and women from two southeastern NewEngland communities. Int J Obes. 1997;21:103–9.
10. Klesges RC, Klesged LM, Haddock CK, Eck LH. A longi-tudinal analysis of the impact of dietary intake and physicalactivity on weight change in adults. Am J Clin Nutr. 1992;55:818–22.
11. Kahn HS, Tatham LM, Rodriguez C, Calle EE, Thun MJ,Heath CW Jr. Stable behaviors associated with adults’ 10-year change in body mass index and likelihood of gain at thewaist. Am J Public Health. 1997;87:747–54.
12. Bild DE, Sholinsky P, Smith DE, Lewis CE, Hardin JM,Burke GL. Correlates and predictors of weight loss in youngadults: the CARDIA study. Int J Obes. 1996;20:47–55.
13. Owens JF, Matthews KA, Wing RR, Kuller LH. Can phys-ical activity mitigate the effect of aging in middle-agedwomen? Circulation. 1992;85:1265–70.
14. Williamson DF, Anda R, Kleiman J, Kahn H, Byers T.Recreational physical activity and ten-year weight change in aUS national cohort. Int J Obes Relat Metab Disord. 1993;17:279–86.
15. Taylor CB, Jatulis DE, Winkleby MA, Rockhill BJ, Krae-mer HC. Effects of life-style on body mass index change.Epidemiol. 1994;5:599–603.
16. Haapanen N, Miilunpalo S, Pasanen M, Oja P, Vuori I.Association between leisure time physical activity and 10-yearbody mass change among working-aged men and women. IntJ Obes. 1997;21:288–96.
17. Coakley EH, Rimm EB, Colditz G, Kawachi I, Willett W.Predictors of weight change in men: results from The HealthProfessionals Follow-up Study. Int J Obes. 1998;22:89–96.
18. Fogelholm M, Mannisto S, Vartiainen E, Pietinen P. De-terminants of energy balance and overweight in Finland 1982and 1992. Int J Obes Relat Metab Disord. 1996;20:1097–104.
19. Sternfeld B, Wang H, Quesenberry CP. Jr et al. Physicalactivity and changes in weight and waist circumference inmidlife women: findings from the Study of Women’s HealthAcross the Nation. Am J Epidemiol. 2004;160:912–22.
20. Littman AJ, Kristal AR, White E. Effects of physical ac-tivity intensity, frequency, and activity type on 10-y weightchange in middle-aged men and women. Int J Obese. 2005;29:524–33.
21. Saris WH, Blair SN, van Baak MA et al. How muchphysical activity is enough to prevent unhealthy weight gain?Outcome of the IASO 1st Stock Conference and ConsensusStatement. Obes Rev. 2003;4:101–14.
Sedentary Lifestyle and BMI, Mortensen et al.
OBESITY Vol. 14 No. 8 August 2006 1471