American Journal of Epidemiology
Copyright © 2004 by the Johns Hopkins Bloomberg School of Public Health
All rights reserved
Vol. 160, No. 2
Printed in U.S.A.
DOI: 10.1093/aje/kwh195
PRACTICE OF EPIDEMIOLOGY
Relative Risk of Mortality in the Physically Inactive Is Underestimated Because of
Real Changes in Exposure Level during Follow-up
Lars Bo Andersen1,2
1
2
Norwegian University of Sport and Physical Education, Oslo, Norway.
Institute for Exercise and Sport Sciences, University of Copenhagen, Copenhagen, Denmark.
Received for publication May 22, 2003; accepted for publication February 27, 2004.
Relative risk among exposure groups in prospective cohort studies is based on the assumption that all subjects
are exposed at the level recorded at baseline throughout the study. Changes in risk behavior during follow-up will
dilute the relative risk. This prospective cohort study in Copenhagen, Denmark, between 1964 and 1994 included
30,640 men and women; 19,149 were examined twice, with an interval of 6.7 (standard deviation, 3.4) years.
Relative risks calculated from baseline measurements for moderately active and sedentary groups compared
with the highly active group were 1.11 (95% confidence interval: 1.05, 1.18) and 1.64 (95% confidence interval:
1.53, 1.75), respectively. The relative risk between the highly active group and the sedentary group decreased
with increasing follow-up time. When intraindividual changes in physical activity level during follow-up were taken
into account, the relative risk of physical inactivity was 24–59% higher compared with the relative risk estimated
from baseline measurements. The risk of a sedentary lifestyle is underestimated when it is calculated from one
baseline measurement in prospective studies, because subjects change behavior during follow-up.
behavior; exercise; follow-up studies; mortality
Abbreviation: SD, standard deviation.
Leisure time physical activity is a complex behavior that is
difficult to assess. Powell et al. (1) found that better studies
were more likely than poorer studies to report an inverse
association between leisure time physical activity and the
incidence of cardiovascular disease. A poor assessment
method will dilute the estimated relative risk because of
misclassification of subjects. Another very important source
of invalidity is real changes in behavior during the followup. Some studies have used two examinations of exposure
level and found that those who increased their physical
activity level between the examinations had a subsequent
lower risk of disease, and that subjects who decreased their
physical activity level had a higher risk (2–7). In all of these
studies, it was possible to calculate the effect of a change in
physical activity level, which was possible because a
substantial number of subjects changed behavior between
the two examinations. Further, the changes in physical
activity level were related to changed risk of disease, which
means that at least part of the observed changes was not due
to assessment error. In the studies by Lissner et al. (3) and
Wannamethee et al. (6), almost half of the population
changed exposure level during 6 years and during 13 years,
respectively. In the study by Paffenbarger et al. (8), even
more reported changes, but they used quite narrow physical
activity definitions. Therefore, it seems that physical activity
is a behavior that changes over time in many persons. In
most prospective studies where the level of leisure time
physical activity has been related to subsequent disease or
death, only one baseline measurement of exposure has been
used, and the period after the baseline, when the changes in
behavior found in the former studies occurred, is part of the
follow-up period.
The aim of this study was to analyze the effect of changes
in exposure level during follow-up on the estimated relative
Correspondence to Dr. Lars Bo Andersen, Norwegian University of Sport and Physical Education, Box 4014, Sognsveien 220, Ullevaal
Stadion, 0806 Oslo, Norway (e-mail: [email protected]).
189
Am J Epidemiol 2004;160:189–195
190 Andersen
TABLE 1. Association of baseline characteristics with level of leisure time physical activity, Copenhagen, Denmark, 1964–1994*
ptrend†
Sedentary
Moderately active
Vigorously active
All
Total no. (%)
3,235 (24.7)
7,437 (56.7)
2,444 (18.6)
No. of deaths (%)
919 (33.6)
1,413 (51.6)
406 (14.8)
Age at baseline (years)
51.0 (12.8)
50.2 (12.2)
49.0 (13.0)
0.001
50.2 (12.5)
Body mass index (kg/m2)
25.0 (5.0)
24.5 (4.4)
23.9 (3.9)
0.001
24.5 (4.4)
Systolic blood pressure (mmHg)
131.8 (24.1)
131.2 (22.6)
130.6 (21.8)
Not significant
131.2 (22.8)
Cholesterol (mmol/liter)
6.00 (1.65)
6.23 (1.33)
6.06 (1.32)
<0.001
6.22 (1.34)
Triglyceride (mmol/liter)
1.41 (0.90)
1.35 (0.79)
1.30 (0.84)
<0.001
1.36 (0.83)
Years in school‡
1.53 (0.61)
1.67 (0.65)
1.75 (0.68)
<0.001
1.65 (0.65)
Smokers (%)
60.5
53.9
53.3
<0.001
56.4
Total no. (%)
3,024 (20.5)
7,909 (53.5)
3,843 (26.0)
14,776
No. of deaths (%)
1,190 (25.5)
2,448 (52.4)
1,034 (22.2)
4,672
Age at baseline (years)
51.7 (12.2)
50.1 (10.9)
48.1 (12.5)
<0.001
49.7 (11.7)
Body mass index (kg/m2)
26.0 (4.0)
25.6 (3.4)
25.3 (3.4)
<0.001
25.6 (3.5)
Systolic blood pressure (mmHg)
136.5 (21.3)
136.2 (20.4)
134.9 (19.7)
Not significant
135.9 (20.4)
Cholesterol (mmol/liter)
5.91 (1.56)
6.05 (1.19)
5.92 (1.20)
<0.001
6.03 (1.22)
Triglyceride (mmol/liter)
1.96 (1.56)
1.83 (1.30)
1.72 (1.32)
<0.001
1.83 (1.37)
Years in school
1.59 (0.67)
1.70 (0.68)
1.75 (0.70)
<0.001
1.69 (0.69)
Smokers (%)
73.4
67.8
61.6
<0.001
67.3
Women
13,116
2,738
Men
* Values are the mean and standard deviation for all baseline characteristics with the exceptions of total, deaths, and smokers.
† Age distribution differed among groups. Therefore, the p values for trend were calculated after adjustment for age.
‡ Three educational levels: level 1, 0–7 years; level 2, 8–11 years; and level 3, >11 years of school education.
risk for leisure time physical activity calculated from the
baseline measurement. To elucidate this problem, relative
risk was calculated from 1) baseline levels of exposure
(leisure time physical activity level), 2) baseline levels of
exposure but with decreasing follow-up time, 3) changed
exposure from the first to the second examination, and 4)
stable exposure at both examinations. The estimated relative
risk of leisure time physical activity was used to calculate
population attributable risk with and without dilution.
MATERIALS AND METHODS
Study populations
This study is based on pooled data from three epidemiologic surveys of the population of Copenhagen, Denmark:
the Copenhagen City Heart Study, the Glostrup Population
Studies, and the Copenhagen Male Study. In both the Glostrup Population Studies and the Copenhagen City Heart
Study, subjects were randomly selected from the Danish
Central Population Registry. The Copenhagen Male Study
included all middle-aged male employees in 14 public and
private companies. Data were collected from 1964 through
1994, and the participation rates in the different cohorts were
between 78 percent and 87 percent. A description of key
variables in relation to physical activity levels is given in
table 1 (9).
At baseline, 13,375 females and 17,265 males were examined. The relative risk of physical inactivity among these
subjects was used as the referent to evaluate dilution. At the
second examination, 9,969 males were eligible for calculations, with 527 males excluded because of a diagnosis of
cardiovascular disease. Female participants at the second
examination numbered 8,426, and 194 were excluded
because of cardiovascular disease prior to the second examination. The age at the first examination was 20–93 years,
with a mean of 47.6 (standard deviation (SD), 10.0) years.
The time between the first and second examinations was 6.7
(SD, 3.4) years, and the subjects who participated in the two
examinations were followed for another 10.2 (SD, 3.8) years
after the second examination. Knowledge of chronic
diseases before the first examination was available from the
questionnaires, the examination by the physician, and
hospital records for 6,919 subjects. Baseline analyses were
conducted with these subjects both included and excluded.
No difference was found, and we decided to include subjects
with chronic disease in the baseline calculation. In the analysis where changes in physical activity were used, subjects
with cardiovascular disease (International Classification of
Diseases diagnosis codes 410–414) prior to the second
examination were excluded, because the disease may have
caused the behavioral change. Other diseases also may cause
behavioral changes, but the analyses were similar whether
they were included or not.
Am J Epidemiol 2004;160:189–195
Mortality Risk in Physically Inactive Persons 191
Assessment
At baseline, self-report questionnaires were used to assess
the subjects’ leisure time physical activity and educational
status. The subjects’ height, weight, and blood pressure were
measured, and a venous blood sample was drawn for analysis of serum cholesterol and triglyceride.
Physical activity during leisure time was classified into
four categories by means of questions originally constructed
and evaluated by Saltin and Grimby (10). Some differences
existed in the phrasing of the questions among the cohorts,
but four categories were used in all the cohorts, and no
difference was found in the age-specific distributions of
leisure time physical activity among cohorts. However, as
few subjects belonged to the highest category of leisure time
physical activity, analyses were performed with levels three
and four combined as one group. The three groups are
referred to as sedentary, moderately active, and highly
active.
Blood pressure was measured on the upper arm with the
use of a mercury sphygmomanometer with subjects in the
sitting position having rested for at least 5 minutes. The
venous blood sample was drawn following a 12-hour fast
and analyzed for total serum cholesterol by conventional
methods (11). In the Copenhagen City Heart Study, subjects
were nonfasting, and blood lipids were analyzed from
plasma.
Endpoints
Information on mortality between the second examination
and December 31, 1994, was obtained. All subjects were
traced by means of the Danish Central Population Registry.
Person-years were calculated from the date of the first examination until December 31, 1994, or to the date of emigration, death, or disappearance.
Statistics
All data were analyzed using Intercooled Stata version 5
software (Stata Corporation, College Station, Texas). Relative risks were calculated from Cox proportional hazards
models. The date of the first examination was used as the
entry time in the baseline model, but in all models where two
examinations were used, the date of the second examination
was used. Categorized risk factors were entered into the Cox
models for adjustment since a nonlinear relation was found
between mortality and the risk factors body mass index and
serum cholesterol. All analyses were done both as crude
analyses (age adjusted) for each sex separately and afterwards with adjustment for other risk factors, but these adjustments did not alter any comparisons among the different
methods used to calculate dilution caused by intraindividual
variation during follow-up. Therefore, it was decided to
present age- and sex-adjusted calculations, except for the
original baseline estimates where all data are presented.
In figure 1, the relative risk among exposure groups is
plotted as a function of follow-up time. The relative risk was
calculated first using a maximum of 30 years of follow-up.
Subjects were censored when one of the following events
Am J Epidemiol 2004;160:189–195
FIGURE 1. Relative risks and confidence intervals of mortality in
the sedentary group compared with those in the highly active group,
Copenhagen, Denmark, 1964–1994. Relative risks are shown as a
function of length of follow-up. In the analyses with long follow-up
times, cohorts examined so late that they could not fulfill the required
follow-up time were included.
occurred: 1) death, 2) emigration, 3) 30 years of follow-up,
and 4) December 31, 1994. In the next calculation, subjects
were censored after death, emigration, 25 years of follow-up,
or December 31, 1994, and so on. This means that the same
subjects are included in all the calculations, but the calculation after 30 years of follow-up includes subjects who have
been followed less. This approach was chosen, because it
gives the most conservative result and the best estimation of
the effect of follow-up time. In the figure, data are included
only until 20 years of follow-up, because no further change
in relative risk was found.
Population attributable risk (PAR) is the percentage of
deaths that theoretically can be prevented if all subjects were
at the lowest level of risk. It was calculated as follows:
PAR = ΣPexp(i) × (RRi – 1) × 100/1 + ΣPexp(i) × (RRi – 1),
where RRi was the relative risk in group i compared with the
group having the lowest risk, and Pexp(i) was the proportion of
subjects belonging to group i.
RESULTS
Four methods were used to calculate the relative risk of
leisure time physical activity.
Method 1: analysis using baseline levels of exposure
If the knowledge from the second examination had not
existed, the relative risk of mortality related to exposure
(leisure time physical activity level) would be calculated
using all subjects who participated in the first baseline
measurements. Data for each sex are presented in table 2.
192 Andersen
TABLE 2. Baseline analysis with full follow-up time, displaying age-adjusted and multivariate-adjusted
relative risks of mortality in relation to leisure time physical activity, with the highly active as referent
group, Copenhagen, Denmark, 1964–1994*
Leisure time physical activity levels at the first examination
Sedentary
Age adjusted
Moderately active
Multivariate
adjusted†
Age adjusted
Multivariate
adjusted†
Highly active
Women
Total no.
3,235
7,437
No. of deaths
919
1,413
Relative risk (95% CI‡)
1.81 (1.61, 2.03)
1.75 (1.55, 1.98)
1.15 (1.03, 1.29)
2,444
406
1.13 (1.01, 1.27)
1.00
Men
Total no.
3,024
7,909
No. of deaths
1,190
2,448
Relative risk (95% CI)
1.55 (1.42, 1.68)
1.49 (1.36, 1.63)
1.10 (1.02, 1.18)
3,843
1,034
1.04 (0.96, 1.13)
1.00
* Calculations include 13,116 women and 14,776 men from the first examination.
† Adjusted for total cholesterol, systolic blood pressure, educational level, smoking, and body mass index.
‡ CI, confidence interval.
Method 2: analysis using baseline levels of exposure but
with decreasing follow-up time
The number of subjects changing exposure level is
supposed to increase with increasing follow-up time. The
relative risk among exposure groups may therefore depend
on the length of follow-up, because more subjects may be
misclassified when the follow-up time is long. The relative
risks were calculated for leisure time physical activity and
with decreasing follow-up time starting with 30 years (figure
1). All subjects were followed for the specific number of
years or until censoring by death, emigration, or December
31, 1994. As seen from figure 1, the relative risk of the
sedentary group compared with the highly active group
decreases with increasing follow-up time.
Method 3: analysis using changed exposure from the
first to the second examination
Leisure time physical activity was assessed twice, with an
interval of 5.5 (SD, 1.4) years, for 7,154 females. During
10.8 (SD, 2.8) years of follow-up after the second examination, 932 females died, and 194 females were excluded from
the analysis because of cardiovascular disease prior to the
second examination. Leisure time physical activity was also
assessed twice, with an interval of 7.8 (SD, 4.2) years, for
7,666 males who were followed for another 9.6 (SD, 4.4)
years, resulting in 1,490 deaths (527 males were excluded
because of cardiovascular disease). Changes in leisure time
physical activity from the first to the second examination are
described in table 3. The Spearman correlation coefficient
between leisure time physical activity at the first and the
second examination was 0.34.
The mortality rates were calculated for the subjects who
changed behavior between the two examinations (table 4).
From the prevalence of subjects who changed behavior and
their actual mortality rate, we calculated that the combined
rate for all sedentary subjects at the first examination was
only 0.85 of those who stayed sedentary. For the highly
active group from the first examination, a combined rate of
1.20 was found compared with those who were highly active
at both examinations.
This means that the relative risk calculated from baseline
measurements is underestimated 1.20/0.85 times (40
percent) between the sedentary group and the highly active
group and 1.20/1.02 times (18 percent) between the moderately active group and the highly active group. Therefore,
instead of the calculated relative risk of 1.64 for the sedentary group in relation to the highly active group, a more realistic estimate of the real difference in mortality would be
1.64 × 1.40 = 2.30. Instead of a relative risk of 1.11 for the
moderately active group compared with the highly active
group, a more realistic difference would be 1.31.
Method 4: analysis using stable exposure at both
examinations
Many subjects have not been exposed to the amount of
leisure time physical activity that they reported at the first
examination during the whole period of follow-up since the
first examination, and they were therefore misclassified if
only baseline data were used. Some of the misclassification
caused by true change in exposure level can be excluded if
the relative risk between activity groups is calculated
including only the subjects who reported the same leisure
time physical activity level in the questionnaire at both
examinations. In this analysis, only 54 percent of those who
participated in the first examination were included. The
difference in mortality rates between groups increased after
exclusion of subjects who were misclassified (table 5).
Population attributable risk
Population attributable risk was calculated using all four
methods of estimating the relative risk among leisure time
Am J Epidemiol 2004;160:189–195
Mortality Risk in Physically Inactive Persons 193
TABLE 3. Leisure time physical activity levels in women and men participating in two examinations,
Copenhagen, Denmark, 1964–1994
Leisure time physical activity levels at the first examination
Leisure time physical activity
levels at the second
examination by gender
Sedentary
No.
Moderately active
%
No.
%
Highly active
No.
%
Total no. at
the second
examination
Women
Sedentary
630
43.2*
Moderately active
681
46.6
149
10.2
Highly active
Total at the first examination
1,460
587
2,738
917
13.8
116
8.0
1,333
64.6*
654
45.0
4,073
682
47.0*
1,748
21.6
4,242
1,452
7,154
Men
Sedentary
500
38.8*
Moderately active
555
43.1
2,285
53.5*
Highly active
234
18.1
1,492
35.0
Total at the first examination
1,289
491
11.5
4,268
146
6.9
1,137
760
36.0
3,600
57.0*
2,929
1,203
2,109
7,666
* Row percentages are shown to illustrate how many at a given physical activity level stayed at that level at the
second examination.
physical activity groups (table 5). As the mortality rate of the
highly active group was the lowest, rates of the sedentary
and moderately active groups were expressed in relation to
this group (table 5). Population attributable risk is highly
dependent on the mortality rate in the referent group (highly
active), and using the rate of the combined leisure time physical activity groups 3 and 4 will underestimate population
attributable risk but narrow the confidence interval considerably.
DISCUSSION
The present study analyzes a methodological problem
related especially to physical activity and inherited in all
published prospective studies estimating the benefit of physical activity from one baseline assessment and subsequent
follow-up. All the studies we are aware of in the literature,
where physical activity level has been assessed more than
once with some years between, have found changes similar
to those of the present study. The fact that physical activity
levels change in many subjects during the study causes an
underestimation, which can be quantified, but only a few
studies have data to do it. A decrease in physical activity
level could be caused by disease, which could increase
mortality and therefore introduce bias. However, all subjects
were examined by a physician at both the first and the second
examinations, and complete hospital records were available.
Those who had cardiovascular disease, which might affect
their participation in physical activity, were excluded from
the analysis. In addition, half of the dilution was caused by
subjects who increased their physical activity level, and this
part of the dilution cannot be a bias.
Most prospective cohort studies have only one baseline
measurement of leisure time physical activity, and they can
therefore calculate mortality rates based on only this
measurement, often with a follow-up time of 10 or more
years. In the present study, we compared the estimated relative risk calculated from one baseline measurement of
leisure time physical activity and from the knowledge of
changes in exposure between two measurements with an
interval of 6.7 (SD, 3.4) years. Spearman correlation coefficients between the first and second measurements were 0.34
for leisure time physical activity. In both the first and the
second assessments of the individual leisure time physical
TABLE 4. Age- and sex-adjusted relative risk of mortality in those who changed leisure time physical activity between the two
examinations related to those who stayed at the same level, Copenhagen, Denmark, 1964–1994
Leisure time physical activity level at the second examination
Leisure time physical
activity level at the
first examination
Sedentary
% of
subjects
Relative
risk
Moderately active
95% CI*
% of
subjects
Relative
risk
95% CI
% of
subjects
Relative
risk
95% CI
Combined
relative risk
0.61, 0.89
14
0.74
0.58, 0.95
0.85
28
0.94
0.83, 1.06
53
1
Sedentary
41
1
45
0.74
Moderately active
13
1.31
1.12, 1.52
59
1
7
2.03
1.55, 2.03
40
1.31
Highly active
* CI, confidence interval.
Am J Epidemiol 2004;160:189–195
Highly active
1.09, 1.57
1.02
1.20
194 Andersen
TABLE 5. Calculation of population attributable risk using four different methods of estimating the relative risk of mortality among
leisure time physical activity groups, Copenhagen, Denmark, 1964–1994*
Sedentary
Pexp(1)† (%) Relative risk
95% CI
Pexp(3+4)†
(%)
Population
attributable
risk (%)
Moderately active
95% CI†
Pexp(2)† (%) Relative risk
Relative risk from baseline leisure time physical
activity
22.4
1.64
1.53, 1.75
55.0
1.11
1.05, 1.18
22.5
16.9
Relative risk from baseline leisure time physical
activity, 2-year follow-up
22.4
2.60
1.93, 3.50
55.0
1.70
1.28, 2.26
22.5
42.6
Relative risk adjusted for changed mortality
18.6
2.30
2.15, 2.47
57.4
1.31
1.24, 1.39
24.1
32.2
Relative risk in groups with stable leisure time
physical activity
14.5
2.04
1.70, 2.46
64.4
1.28
1.10, 1.49
21.2
25.2
* The mortality rate in the combined leisure time physical activity groups 3 + 4 is used as referent.
† Pexp(3+4), proportion of highly active; Pexp(1), proportion of sedentary; CI, confidence interval; Pexp(2), proportion of moderately active.
activity level, some misclassification may exist caused by an
inaccurate assessment method and variation about the true
mean. The analyses were not corrected for this type of variability, and the dilution found from the different methods is
therefore an underestimation of the real dilution. We have
treated both assessments of leisure time physical activity as
if they were the true means of habitual leisure time physical
activity levels in the individual at two different times of his
or her life, and the dilution may be more severe than these
calculations suggest. The lower mortality rates in subjects
who increased their leisure time physical activity level and
the higher rates in those who decreased their physical
activity level, compared with those who stayed at the same
level (table 4), were evidence of true changes, even if it was
not possible to quantify the true changes compared with the
error variation from assessment error.
Calculated dilution
The present study calculated the relative risks of leisure
time physical activity levels using four different methods in
order to evaluate the influence of changes in behavior after
the first baseline assessment of the relative risk. The relative
risk calculated from a usual baseline measurement was
compared with the relative risk estimated with a short
follow-up period, where fewer changes in true behavior
occur, and with two methods taking changes into account.
Calculations after 2 years of follow-up were possible
because more than 500 deaths occurred within the first 2
years.
Baseline. An estimation of the relative risk of mortality
from all subjects participating in the first examination was
calculated as a referent value. It could be argued that the
referent calculation should be conducted with values from
the first examination but including only the subjects participating in both examinations. However, this analysis
increased the dilution considerably, and we chose the former
for comparison as the more conservative solution and
because these data would have been used if a second examination had not existed. In the analysis of the complete baseline group, a relative risk of 1.11 was found for the
moderately active group, and a relative risk of 1.64 was
found for the sedentary group compared with the most active
group. These values are comparable with those of most other
studies using one baseline measurement of leisure time physical activity related to subsequent all-cause mortality with a
long follow-up time (12–15). Subsequently, follow-up time
was gradually shortened by censoring subjects after 20
years, 15 years, etc., until a follow-up time of 2 years. Below
this level, the number of endpoints became critical. The
reason for the inclusion of subjects examined so late that
they could not fulfill a longer follow-up time was to include
the same subjects in all the analyses included in figure 1, and
this inclusion made the difference in relative risk only less
between long and short follow-up times. The calculated
population attributable risk was surprisingly high, and physical inactivity accounted for more than 40 percent of all
deaths.
Two methods using two examinations. In the analysis of
subjects reporting the same leisure time physical activity
level at both examinations, the only dilution of the estimates
is in the assessment of leisure time physical activity at the
two examinations and possible fluctuations between the two
examinations. Thus, if the true mean of leisure time physical
activity could be assessed, the estimate would be very close
to the real difference between the physically active and the
sedentary groups, with the exception that changes in leisure
time physical activity after the second examination and
during the follow-up would not be taken into account. The
analysis excluded all subjects who had not participated in
two examinations and who had changed leisure time physical activity level, and therefore only one fourth of the original cohort was included (7,806 subjects experiencing 1,199
deaths). The population attributable risk in the analysis of the
subjects with a stable leisure time physical activity level was
25.2 percent or 55 percent higher than the referent population attributable risk.
In the second analysis, we used the knowledge of the real
mortality rates of all subjects who had participated twice and
belonged to a certain leisure time physical activity level at
the first examination regardless of whether they had changed
leisure time physical activity level or not. As some of the
sedentary subjects at the first examination had become active
with a decreased mortality rate, the mean rate of the whole
Am J Epidemiol 2004;160:189–195
Mortality Risk in Physically Inactive Persons 195
group was lower than that in subjects who stayed sedentary.
The true difference in mortality rates between the sedentary
and the vigorously active groups was 40 percent higher, and
the calculated population attributable risk was 191 percent
higher than in the referent analysis. This is a substantial
difference that should affect preventive strategies. Powell
and Blair (16) have estimated that 35 percent of deaths from
cardiovascular disease, 32 percent of deaths from colon
cancer, and 35 percent of deaths from diabetes could theoretically be prevented if everyone were vigorously active.
Haapanen-Niemi et al. (17) calculated that physical inactivity was responsible for between 22 percent and 39 percent
of all cardiovascular disease cases. However, these values
are calculated from the prevalence of a risk factor in a certain
exposure group and the relative risk of disease in this group
compared with the group with the lowest risk, and calculations are based on one baseline measurement and subsequent
mortality rates in baseline exposure groups. Therefore, these
calculations may include a substantial amount of dilution.
It could be argued that unrecognized disease could have
caused the changes in exposure level. However, dilution is
experienced only with physical activity and not with
smoking (data not shown), and completely different
approaches to calculate the dilution in physical inactivity
estimates gave similar conclusions.
Other studies have used two examinations and estimated
the effect of changes in leisure time physical activity on
mortality or cardiovascular disease (3, 6, 8, 18). In these
studies, the data have not been used to analyze the dilution of
the association between leisure time physical activity and
mortality caused by changes in leisure time physical activity
between the first and second examinations, but just as many
subjects changed physical activity level as in the present
study. We therefore believe that other prospective studies
include as many subjects changing physical activity level
after baseline assessment, but they just do not have the data
to analyze it. As dilution is a problem primarily in studies
analyzing the relation between disease and a complex and
unstable health behavior, the problem might be just as
important in analyses of nutrition, which also involves a
complex health behavior.
We have tried different approaches to elucidate the
problem of the analysis of an unstable health behavior. All
these approaches pointed in the same direction. Most
prospective studies have long follow-up time because the
number of cases is the limiting factor in the statistical analyses, and it is more expensive to increase the size of the
cohort than to increase follow-up time.
ACKNOWLEDGMENTS
The study was supported with grants from the Danish
Heart Foundation and the Danish Medical Research Council.
Many thanks to the Copenhagen Center for Prospective
Population Studies for providing raw data.
Am J Epidemiol 2004;160:189–195
REFERENCES
1. Powell KE, Thompson PD, Caspersen CJ, et al. Physical activity and the incidence of coronary heart disease. Annu Rev Public Health 1987;8:253–87.
2. Erikssen G, Liestøl K, Bjørnholt J, et al. Changes in physical
fitness and changes in mortality. Lancet 1998;352:759–62.
3. Lissner L, Bengtsson C, Björkelund C, et al. Physical activity
levels and changes in relation to longevity. A prospective study
of Swedish women. Am J Epidemiol 1996;143:54–62.
4. Paffenbarger RS, Hyde RT, Wing AL, et al. The association of
changes in physical-activity level and other lifestyle characteristics with mortality among men. N Engl J Med 1993;328:538–
45.
5. Paffenbarger RS. Influence of adopting a physically active lifestyle on mortality rates of middle-aged and elderly men. J Int
Counc Health Phys Educ Recreat Sport Dance 1994;30:5–10.
6. Wannamethee G, Shaper AG, Walker M. Changes in physical
activity, mortality, and incidence of coronary heart disease in
older men. Lancet 1998;351:1603–8.
7. Blair SN, Kohl HW, Barlow CE, et al. Changes in physical fitness and all-cause mortality. JAMA 1995;273:1093–8.
8. Paffenbarger RS, Kampert JB, Lee IM, et al. Changes in physical activity and other lifeway patterns influencing longevity.
Med Sci Sports Exerc 1994;26:857–65.
9. Andersen LB, Schnohr P, Schroll M, et al. All-cause mortality
associated with physical activity during leisure time, work,
sports, and cycling to work. Arch Intern Med 2000;160:1621–
8.
10. Saltin B, Grimby G. Physiological analysis of middle-aged and
old former athletes: comparison with still active athletes of the
same ages. Circulation 1968;38:1104–15.
11. Epidemiology of chest pain and angina pectoris, with special
reference to treatment needs. Acta Med Scand Suppl 1983;682:
1–120.
12. Lapidus L, Bengtsson C. Socioeconomic factors and physical
activity in relation to cardiovascular disease and death. A 12
year follow up of participants in a population study of women
in Gothenburg, Sweden. Br Heart J 1986;55:295–301.
13. Leon AS, Connett J, Jacobs DR, et al. Leisure-time physical
activity levels and risk of coronary heart disease and death. The
Multiple Risk Factor Intervention Trial. JAMA 1987;258:
2388–95.
14. Lindsted KD, Tonstad S, Kuzma JW. Self-report of physical
activity and patterns of mortality in Seventh-day Adventist
men. J Clin Epidemiol 1991;44:355–64.
15. Sherman SE, D’Agostino RB, Cobb JL, et al. Physical activity
and mortality in women in the Framingham Heart Study. Am
Heart J 1994;128:879–84.
16. Powell KE, Blair SN. The public health burdens of sedentary
living habits: theoretical but realistic estimates. Med Sci Sports
Exerc 1994;26:851–6.
17. Haapanen-Niemi N, Vuori I, Pasanen M. Public health burden
of coronary heart disease risk factors among middle-aged and
elderly men. Prev Med 1999;28:343–8.
18. Hein HO, Saudicani P, Sørensen H, et al. Changes in physical
activity level and risk of ischaemic heart disease. A six year follow-up in the Copenhagen Male Study. Scand J Med Sci Sports
1994;4:57–64.