© International Epidemiological Association 1999
International Journal of Epidemiology 1999;28:1073–1080
Printed in Great Britain
Change in lifestyle factors and their influence
on health status and all-cause mortality
Sven-Erik Johanssona,c and Jan Sundquistb,c
Background The purpose of this study was to analyse both cross-sectional associations and
how longitudinal changes in lifestyle factors from one state in 1980–1981 to
another in 1988–1989 influence self-reported health status. Another aim was to
estimate the hazard ratios for all-cause mortality for the changes in lifestyle
factors and self-reported hypertension during the same period of time.
Method
The cross-sectional and the longitudinal analyses are based on the same simple
random sample of 3843 adults, aged 25–74, interviewed in 1980–1981 and 1988–
1989 and is part of the Swedish Annual Level-of-Living Survey. About 85% of
the respondents in the first interview participated in a second interview in 1988–
1989. Cross-sectional odds ratios, based on a marginal model, were estimated
using the generalized estimating equations. The transitional models were analysed using unconditional logistic regression. A proportional hazard model was
applied to investigate the influence of lifestyle transitions on mortality.
Results
Physical inactivity, being a current or former smoker and obesity (women only)
were strong risk factors for poor health either as main effects and/or combined
(interactions). There was a strong interaction between physical activity and
smoking, and for women, also between body mass index (BMI) and physical
activity. Smoking, physically inactive and obese women had about a ten times
higher risk of poor health status than non-smoking, physically active, and
normal-weight women. The corresponding risk for men was about five times
higher. Physically active, but smoking and obese individuals showed only moderately increased risks for poor health status. The transitional model showed that
those who were physically inactive in 1980–1981, but did exercise in 1988–1989,
improved their health after adjustments for sociodemographic and other lifestyle
factors. Continuing to smoke or being physically inactive or having hypertension
at both points in time were all associated with higher hazard ratios for all-cause
mortality (1.6, 1.9 and 1.8, respectively) than those who reported that they were
in good status at both points in time.
Conclusions We found that physical activity protects against poor health irrespective of
an increased BMI and smoking. The major clinical implications are the longstanding benefits of physical activity and not smoking.
Keywords
Self-reported health status, mortality, lifestyle, sociodemographic factors
Accepted
19 May 1999
Several surveys have provided cross-sectional analyses of selfreported health outcomes.1–4 All indicate a moderate to strong
relationship between lifestyle, marital status, socioeconomic
a Department of Welfare and Social Statistics, Statistics Sweden, Box 24300,
SE-104 51 Stockholm, Sweden. E-mail: [email protected]
b Stanford Center for Research in Disease Prevention, Stanford University
School of Medicine, Palo Alto, California, USA.
c Lund University, Department of Community Medicine, Malmö University
Hospital, Malmö, Sweden.
status (SES), country of birth and poor health. Cross-sectional
data have some limitations with regard to inferences from lifestyle or sociodemographic variables concerning health.
During the 1980s, mortality declined in Sweden5 as in most
Western countries.6–8 A substantial part of the decline in mortality reflects the decrease in the rates of cardiovascular disease
(CVD) and stroke.9 A number of studies have recently identified
lifestyle and behavioural risk factors that could explain at least
part of the decline in mortality from CVD.10 Studies focusing
on mortality and lifestyle show that physical inactivity and
1073
1074
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
smoking are associated with increased risks compared with
physical activity and non-smoking.11–13 On the other hand,
overweight showed an ambiguous association with mortality,
while obesity showed an increased mortality risk.
Physical activity protects the individual from CVD,11,14
diabetes,15 and obesity.16 Smoking cessation decreases the risk
of CVD17 and stroke,10 although the decrease in stroke incidence appears to be slow and the risk remains on a high level.18
In addition, other studies have found that prevention of obesity
protects against the risk of CVD, at least among women,19,20
and stroke.21 However, there are some inconsistencies in previous longitudinal studies focused on lifestyle and stroke. The
Copenhagen City Heart Study showed that physical inactivity
was a potent risk factor for stroke while the body mass index
(BMI) was not found to have any effect on stroke.22
Trends in American society show that reduced fat and calorie
intake and frequent use of low-calorie food products are associated with a paradoxical increase in the prevalence of obesity.23
The dramatic decrease in total physical activity could explain
the paradox. Moreover, Swedish men who became physically
inactive had a higher increase in BMI than those who were
physically active on a regular basis between 1980–1981 and
1988–1989 when adjustments were made for smoking, SES
and country of birth.24 A Finnish study showed that only
dietary habits improved during the 1980s among males in
Finland while physical inactivity and daily smoking remained
unchanged.25 Unhealthy behaviour was more common in
lower educational groups, especially among middle-aged men
and divorced men.
Most studies have been limited in that they have been
restricted to a single geographical area or to a segment of the
working population. It is also uncommon to control for SES and
country of birth. In this study, we had the opportunity to
examine whether there were differences existed in both selfreported health status and mortality in a longitudinal sample
representative of Sweden.
The first aim of the present study was to analyse crosssectional associations between lifestyle factors, such as physical
inactivity, smoking, and BMI, and self-reported health status,
with adjustments for marital status, educational status and
country of birth. We expected cross-sectional associations
between physical inactivity, daily smoking, obesity and poor
self-reported health status.
The second aim was to analyse how transitions in lifestyle
influenced self-reported health status between 1980–1981 and
1988–1989. We expected to find poor health status among
those who changed their lifestyle to a more sedentary type, i.e.
from physical activity to physical inactivity, non-smoking to
smoking and non-obesity to obesity.
The third aim of the present study was to investigate the
influence of changes in lifestyle factors and self-reported hypertension on all-cause mortality. We expected to find increased
mortality among those who remained in a poor status or changed
their lifestyle in a negative way, or who developed hypertension
during the study period.
Method
Both the cross-sectional and the longitudinal analyses are based
on the same representative, simple random sample of 3843
adults aged 25–74 in the Swedish population. The data were
collected by Statistics Sweden, the Swedish Annual Levelof-Living Survey.26 They were interviewed face-to-face in
1980–1981 and the same respondents were reinterviewed using
the same questionnaire in 1988–1989. Data on mortality from
interview up to 31 December 1995 were obtained from the
Cause-of-Death Register based on the individual Swedish national
registration number.
The non-response rate was about 15% in 1980–1981 and also
about 15% of those who participated in 1980–1981 did not
participate in 1988–1989. Refusal is the dominating reason
not to respond. Slightly more than two-thirds of the nonresponse consists of refusals, one-fifth of the non-respondents
were not found and one-tenth were ill. Mortality, as a measure
of ill-health among both respondents and non-respondents
based on the complete 1980–1981 samples, has been analysed
in a proportional hazard model adjusted for sex, age, marital
status and region. It was found that those who refused had
the same mortality as the respondents, but the other two
groups had significantly higher mortalities. Owing to nonresponse, ill-health in the population will probably be underestimated to some extent. Relative risks will probably be less
influenced by non-response than absolute measures such as
prevalence.
Outcome variables
Health status was based on the question: ‘How would you
describe your general health?’ Is it ‘good’, ‘poor’ or ‘anywhere
between good and poor’. Those who answered that their health
status was ‘poor’ or ‘anywhere between good and poor’ are
counted as having poor health status.
All-cause mortality. Person-years at risk were calculated from
the date of the second interview until death or, for those who
survived, until 31 December 1995.
Independent variables
Age (at baseline) was categorized into the following groups:
25–34, 35–44, 45–54, 55–64 and 65–74.
Physical activity comprised five categories of leisure physical
activity, which were dichotomizd in the analysis into: being
physically inactive or occasionally active versus regular physical
activity at least once a week.
Smoking habits: Tobacco consumption (grams per day) was
obtained by adding the number of cigarettes smoked per day
(one cigarette corresponding to one gram), the number of
cigarillos (2 grams) per day, of cigars (5 grams) per day and
grams of pipe tobacco per day. The reference category was
‘Never smoked’ and the other categories were ‘Former smoker’
and ‘Daily smoker, 1–14 (1–10 for females) grams per day’ and
‘more than 14 (10) grams per day’.
Body mass index (BMI) was calculated as weight/height 2
(kg/m2) and comprised four categories: underweight (BMI , 19
for females and BMI , 20 for males), normal weight (19 < BMI
, 23.8 for females and 20 < BMI , 25 for males), overweight
(23.8 < BMI , 28.6 for females and 25.0 < BMI , 30.0 for males)
and obesity (BMI > 28.6 for females and > 30.0 for males).
Marital status comprised two groups: single and married/
cohabiting.
Country of birth was defined as either indigenous (Sweden) or
foreign-born.
CHANGE IN LIFESTYLE AND HEALTH
The respondents were classified into three groups according
to their educational status: (1) primary school level, <9 years of
education; (2) at least 2 years of high school, 10–11 years of
education; and (3) 3 years of high school or university studies,
.11 years of education (reference group).
Longitudinal effect of age, expressed in years: Age at second
interview minus age at first interview.
Self-reported hypertension included as an independent variable
in the model with mortality: those who said they used antihypertensive drugs or reported hypertension (diagnoses 401–405
according to ICD-8 and ICD-9) as a long-standing disease and all
others.
The reliability of the dependent and independent variables,
with the exception of country of birth and marital status, was
analysed in 1989 in reinterviews (test-retest method) about
4 weeks after the main interview of a random sample of 410
respondents (response rate 88.4%) for the following variables
included in the present study: health status, physical activity;
smoking; weight and height (BMI); educational status giving
kappa coefficients between 0.7 and 0.9 indicating a high level of
reliability.27
Statistical Methods
1075
were included in the model for the following variables: physical
activity, BMI and smoking habits, with adjustments for all other
variables. Adjusting for transitions in marital status gave
essentially the same results as only including marital status from
the second interview. Therefore, the simpler model was chosen.
A transition model takes into account both the regression
objective and the within-subject correlation simultaneously.28
The conditional expectation of the last response depends on the
independent variables and the past response.
Proportional hazard models
All-cause mortality was analysed using a proportional hazard
model including sex, age (continuous), marital status, education,
physical activity, smoking and obesity.29 To estimate the changes
in lifestyle factors (physical activity, smoking, obesity and mortality), four categories were defined (remained at high risk,
changed from low to high risk, changed from high to low risk
and remained at low risk). The mortality risks in each of the categories were analysed, adjusting for marital status, educational
status and country of birth. Changes in marital status did not
influence the risks. The results are presented as hazard ratios
(HR) with 95% CI. The proportional hazard assumption was
analysed by inspecting log (–log) survival curves for parallelism.
All of the included variables approximately met the assumption.
Marginal model
A marginal model was used to estimate cross-sectional odds ratios
for poor self-reported health separately for men and women
because we found several interactions between sex and other
variables. The model included age, marital status, educational
status, country of birth, physical activity, smoking habits and
BMI as independent variables. The regression coefficients were
estimated by generalized estimating equations (GEE).28 There
were significant interactions between physical activity and
smoking, and between physical activity and BMI (women only).
The fit of the model was strongly improved when the interactions
were included. The regression and within-subject correlations
are modelled separately. This method is more efficient than a
cross-sectional analysis with the same number of subjects.
The efficiency depends on the size of the correlation between
the measurements. Another advantage is that the number of
repeated observations may vary among subjects without
changing the interpretation of the coefficients, but there should
be at least two. The exponentiated regression coefficients are
interpreted in the same way as OR from a cross-sectional analysis,
which is sometimes called a ‘population-averaged’ interpretation.28 They compare the odds of disease in the populations
with and without a risk factor. The working correlation matrix
was exchangeable (compound symmetry), which is always
applicable when there are two measurements. The results are
shown as OR with 95% confidence intervals (CI) and are interpreted in the same way as OR from cross-sectional analyses.
Longitudinal transition models
The transitional model was analysed by applying unconditional
logistic regression including health status, sex, age and country
of birth from baseline (1980–1981) and marital status and
educational status from the second interview (1988–1989) as
independent variables.28 Self-reported health status in 1988–
1989 was the dependent variable. In order to estimate transitions
from one state to another, measurements from both interviews
Results
The age-standardized distribution of poor health status in
the age range 25–74 years in the sociodemographic and the lifestyle variables by sex and interview year are shown in Table 1.
The interviews were completed by 1871 men and 1972 women
on both occasions. For all variables, the prevalence of selfreported poor health status is higher for 1988–1989 than for
1980–1981; women and those who had a more sedentary lifestyle generally had higher prevalences than their counterparts.
Single and foreign-born people and those with decreasing
educational status have consistently higher prevalences of poor
health status than their counterparts (Table 1).
Table 2 presents estimated OR with 95% CI for self-reported
poor health status for a full model including all variables, by sex.
Being physically inactive, current or former smoker or obese
(women only) was strongly associated with an increased risk of
poor health. Even if the main effect of smoking was not significant in the final model, it is an important variable combined
with physical activity (interaction). The overall fit of the model
was significantly improved when the interactions between
physical activity and smoking and between physical activity and
BMI (women only) were included (Table 2). The regression
coefficients with P-values are shown in Table 2 and the combined effects of the lifestyle variables in Table 3, as OR. Being
single or being born abroad and having an educational status
<9 years or 10–11 years were risk factors for poor self-reported
health status.
Physical activity is a strong protective factor for poor health
status (Table 3). Physically active and obese current smokers or
former smokers have an OR of about 1.50 for poor health
status. In contrast, physically inactive and smoking women
show a similar gradient for poorer health status with increasing
BMI. The OR were about 3 for normal weight, 4 for overweight
and 9 for obesity (Table 3).
1076
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Table 1 Age-standardized prevalences (percentages, direct standardization) of poor health status for the different demographic and lifestyle
variables by sex and year in the age group 25–74, n = 1972 (females), n = 1871 (males)
1980–1981
Variable
Category
Totals
Physical activity
Smoking habits
BMI
Marital status
Country of birth
Education
1988–1989
Males
Females
Males
Females
All
20.7
25.2
25.4
31.9
No
37.7
46.2
43.3
58.3
Yes
18.5
22.5
20.6
26.6
Never smoked
18.3
23.6
18.4
29.5
Smoked earlier
20.9
23.9
25.7
31.2
1–14 (10a) g/day
22.7
22.8
24.3
35.7
.14 (10a) g/day
24.6
31.4
30.9
35.7
Underweight
29.9
32.5
25.6
39.9
Normal
18.2
23.2
22.2
29.7
Overweight
24.7
25.6
25.5
27.5
Obesity
25.4
38.9
33.4
46.9
Married/cohabiting
19.5
24.4
23.2
29.9
Single
27.8
29.8
27.5
37.2
Sweden
20.0
25.1
22.5
30.7
Foreign-born
37.2
28.6
46.0
37.7
<9 years
25.7
38.8
26.7
37.7
10–11 years
20.3
24.9
25.1
30.7
.11 years
14.5
19.3
19.2
21.6
a Women.
Table 2 Estimated regression coefficients (βi) with P-values for self-reported poor health status in a full marginal model (GEE) adjusted for age,
age group 25–74 years, by sex. The reference groups are not shown. n = 1972 (females), n = 1871 (males)
Females
Variable
Category
Males
βi
P-value
βi
P-value
0.001
Main effects
Physical activity
No
0.496
0.001
0.473
Smoking habits
Smoked earlier
0.001
ns
0.211
ns
1–14 (10a) g/day
0.053
ns
–0.052
ns
BMI
.14 (10a) g/day
0.096
ns
0.199
ns
Underweight
0.341
ns
0.189
ns
Overweight
ns
–0.092
ns
0.115
Obesity
0.376
0.009
0.269
ns
Marital status
Single
0.232
0.012
0.360
0.001
Country of birth
Foreign-born
0.347
0.025
0.915
0.001
Education
<9 years
0.555
0.001
0.463
0.001
10–11 years
0.382
0.003
0.313
0.009
1988–1989
0.349
0.001
0.281
0.001
No and smoked earlier
0.675
0.022
0.686
0.001
No and 1–14 (10a) g/day
0.506
ns
1.065
0.001
0.025
Year
Interactions
Physical activity and smoking
Physical activity and BMI
No and .14 (10a) g/day
0.438
0.032
0.531
No and underweight
1.060
0.018
–
No and overweight
0.470
0.009
–
No and obesity
0.783
0.001
–
a Women.
Regression coefficients and P-values from the final transitional model are presented in Table 4. The final model shows
the main effects and interactions of the independent variables.
Self-reported health status, sex, and country of birth are
indicated from baseline and all others from the second
interview. The final model was tested for interactions. Only
one was discovered, health status with education (<9 years),
which resulted in a significantly better fit. Self-reported health
status from baseline was a very strong predictor of health status
at the second interview (Table 4). The other variables were
CHANGE IN LIFESTYLE AND HEALTH
1077
Table 3 Estimated odds ratio for self-reported poor health status in a full marginal model (GEE) adjusted for age, age group 25–74, by sex and
adjusted for the other variables in Table 2
Body mass index
Sex
Smoking habits
Females
Males
Physical activity
Normal
Overweight
Obesity
No
Yes
1
0.91
1.46
Smoked earlier
Yes
1.00
0.91
1.46
1–10 g/day
Yes
1.05
0.96
1.54
.10 g/day
Yes
1.10
1.00
1.60
No
No
1.64
1.50
5.22
Smoked earlier
No
3.23
4.71
10.28
1–10 g/day
No
2.87
4.19
9.14
.10 g/day
No
2.80
4.09
8.92
No
Yes
1
1.12
1.31
Smoked earlier
Yes
1.23
1.39
1.62
1–14 g/day
Yes
0.95
1.07
1.24
.14 g/day
Yes
1.22
1.37
1.60
No
No
1.60
1.80
2.10
Smoked earlier
No
3.93
4.41
5.15
1–14 g/day
No
4.42
4.96
5.79
.14 g/day
No
3.33
3.73
4.35
Table 4 Estimated regression coefficients, standard errors (SE) and P-values for poor health status at the second interview in a transitional model
(logistic regression) adjusted for age (not shown), n = 3843. The reference groups are not shown. All variables except ‘Health status, baseline’ are
from the second interview
Final model
Variable
Category
Intercept
βi
SE
–2.00
0.15
0.0001
0.0001
P-value
Health status, baseline 1980/1981
Poor
2.27
0.13
Sex
Male
–0.23
0.09
0.008
Physical activity
No
1.00
0.11
0.0001
0.015
Smoking habits
BMI
Smoked earlier
0.26
0.10
1–14 (10a) g/day
0.13
0.16
ns
.14 (10a) g/day
0.28
0.12
0.022
Underweight
0.24
0.21
ns
Overweight
–0.06
0.09
ns
0.0007
Obesity
0.49
0.14
Marital status
Single
0.11
0.10
ns
Country of birth
Foreign-born
0.55
0.15
0.0002
0.0001
Education
<9 years
0.65
0.13
10–11 years
0.32
0.12
0.009
–0.69
0.18
0.0002
Health status by education <9 years
Change in –2 log likelihoodb
d.f.
Hosmer-Lemeshow test
14.3
1
0.39
a Women.
b Compared to main effects model.
associated with about the same risks as in the marginal models
in Table 2.
The transition matrices for physical inactivity, smoking and
BMI were estimated by the final model in Table 4. Going from
‘physical activity’ to ‘physical inactivity’ resulted in an increased
risk for poor health status compared with the reference group
‘physical active’ at both interviews (Table 5).
Those who stopped smoking during the 8-year period had
an OR of 1.29 (CI : 1.01–1.67) and those who remained daily
smokers also had an increased risk with an OR of 1.61
(CI : 1.18–2.19), compared with never-smokers on both
occasions (not shown in Table).
Going from overweight to obesity gave an OR for poor health
status of 1.61 (CI : 1.09–2.43) and remaining obese an OR of
1078
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
Table 5 Odds ratios for poor health status with 95% confidence
intervals and frequency of transitions from first interview to second
interview for physical activity adjusted for all other variables in the
transitional model according to Table 4
Physical activity (1988–1989)
Variable and year
Yes
No
Totals
1 (Reference)
2.07 (1.59–2.64)
1734
1053
681
1.23 (0.94–1.62)a
2.25 (1.79–2.82)
681
1428
1734
2109
Physical activity
(1980–1981)
Yes
No
Totals
2109
3843
a If 1 is included in the confidence interval, the odds ratio is not significant
(P . 0.05).
1.48 (CI : 1.04–2.17), compared with those who were of normal
weight on both occasions. All other transitions were nonsignificant (not shown in Table).
The risk of death from all causes was examined according to
changes in lifestyle and hypertension and is shown in Table 6.
From the second interview in 1988–1989 to 31 December 1995,
there were 357 deaths during 25 588 person-years. Respondents
who decreased their physical activity between 1980–1981 and
1988–1989 also increased their mortality risk compared with those
who were physically active on both occasions. Respondents who
were smokers on both occasions and those who stopped smoking
by the second interview had a higher mortality than respondents who had never smoked on either one of the occasions.
Overweight was only a risk factor for increased mortality when
there was a change from being overweight to not being overweight during the 1980s. Respondents who remained or became
hypertensive by the second interview had a higher mortality risk
compared with those who were normotensive on both occasions.
Discussion
This study, based on a large representative sample of the whole
Swedish population, demonstrates the associations between
changes in physical activity, smoking and BMI when simultaneous adjustments are made for sex, age and socioeconomic
variables, such as marital status and educational status, and
country of birth. We demonstrated, as expected, cross-sectional
associations between lifestyle factors, socioeconomic environment, and health status by analysing marginal models using
GEE. Furthermore, we demonstrated that physical activity
seems to be a strong protective factor against poor health status
among obese/overweight and smoking individuals. Moreover,
we found that poor health status and increased mortality are
associated with a change in lifestyle to a more sedentary type.
Those who developed hypertension during the study period
showed increased mortality rates. The transitional matrices of
the Markov regression models predicted changes in health risks
on changing from one state to another.
The finding that the transition from not being physically
active to being physically active was associated with a nondeteriorating health status partly agreed with a longitudinal study
from Alameda County focused on 6-year predictors of successful
ageing for men and women in the age bracket 65–95 years.30
They were studied prospectively in 1984 and followed through
1990 and walking for exercise was prospectively associated with
successful ageing. Multiple logistic regression was used in the
Alameda Study in which successful ageing in 1990 was regressed
on the baseline predictors, as well as on successful ageing in
1984, which had a strong association with follow-up successful
ageing.30 Physical activity was associated with good self-reported
health status, which was consistent with previous findings that
exercise affects both morbidity and mortality.31–34 In community prevention trials that targeted multiple CVD risk factors
using multiple intervention modalities, regular physical activity
Table 6 Death rates (per 10 000 per year) and hazard ratios (HR) with 95% confidence interval (CI) of all-cause mortality by change in lifestyle
factors and hypertension between 1980–1981 and 1988–1989 adjusted for sex, age, marital status, education and the other variables, n = 3843
Variable
1980–1981
1988–1989
Prevalence of
person-years
(%)
Physical activity
Daily smoking
No. of
deaths
Death rate
(per 10 000
per year)
HR (CI)a
1.87 (1.36–2.57)
No
No
36.4
191
205
Yes
No
17.7
63
139
1.73 (1.19–2.53)
No
Yes
17.9
54
118
1.39 (0.94–2.05)
Yes
Yes
28.0
49
68
1
Yes
Yes
23.0
82
139
1.60 (1.23–2.09)
Yes
No
8.5
34
157
1.70 (1.18–2.46)
No
Yes
3.2
11
134
1.73 (0.94–3.20)
No
No
65.3
230
138
1
Overweight
Yes
Yes
33.4
133
155
0.85 (0.67–1.09)
BMI .25 (males)/BMI .23.8 (females)
Yes
No
5.2
57
425
1.59 (1.16–2.18)
No
Yes
13.6
30
86
0.92 (0.61–1.37)
No
No
47.8
203
112
1
Yes
Yes
6.5
70
423
1.75 (1.32–2.31)
Hypertension
Yes
No
2.1
24
445
1.45 (0.94–2.24)
No
Yes
6.2
39
244
1.66 (1.17–2.36)
No
No
85.2
224
103
1
a If 1.0 is included in the confidence interval, the hazard ratio is not significant (P . 0.05).
CHANGE IN LIFESTYLE AND HEALTH
was one of the four ‘cornerstone’ approaches to securing cardiac
health.32–34 The benefit of regular physical activity in providing
long-standing protection against the manifestations of coronary
heart disease has been demonstrated.35–36
The finding of a strong interaction between some lifestyle
factors, e.g. that physically inactive obese women, regardless of
whether they were smokers earlier or are current smokers, had
a more than five times increased risk of poor health status
compared with physically active obese smokers is new, but it
should be interpreted with caution because it is a cross-sectional
association. The corresponding risk for men was increased
about three times. Another interesting finding is that those who
had smoked earlier also had an increased risk, which is partly in
accord with a longitudinal follow-up in the National Health and
Nutrition Examination Survey (NHANES I), which also showed
that smoking, above all in middle-aged and older men, was a
predictor of survival time.13 Smokers also exhibited lower levels
of health consciousness, e.g. low fibre ingestion and not taking
exercise.37–38
A prospective 8-year follow-up study of more than 115 000
women found that even mild-to-moderate overweight increased
the risk of coronary heart disease.19 We found that overweight
subjects have about the same risk as the normal group, a finding
in line with a review of mortality among obese individuals.39
Being single is a well-known risk factor for poor health status
and increased mortality and was confirmed in this study.40–42
The finding that foreign-born individuals had a higher risk for
poor health than Swedes after adjustment for sociodemographic
and lifestyle factors agreed with other studies.4,43 There was a
graded association between education and poor health which
agreed in part with other studies reporting that a low-tointermediate educational level was a risk factor for severe longterm illness44 and mortality.45–48
Mortality and changes in lifestyle behaviour involving physical
activity and smoking, as well as hypertension, show, as expected,
that remaining in a negative lifestyle status or transiting from a
good to a negative lifestyle status results in higher risks than
remaining in a good lifestyle status. However, changes in BMI
(overweight) present an ambiguous picture, which might be
due to the fact that we do not know if the weight losses were
voluntary or not. Cross-sectional studies often show a strong
relationship between obesity and mortality.19–20,49 The findings
in this study are in perfect accord with those of Davis et al.13
Strengths and limitations
The advantages of this study, a large random sample of the
Swedish population and repeated measurements, allowed us to
analyse how the change in a lifestyle factor influenced selfreported health status and mortality. The marginal model, using
GEE, was effective in analysing the meaning of being single,
being physically inactive and being a smoker, while the transitional models were most useful in analysing the change, with
some individuals experiencing recovery from poor health and
others experiencing a decline in health.
The present study does have some limitations, however, one
disadvantage is that all changes that might have occurred during
the 8-year period are not known, but only the status at the
interview. A change in self-reported health might have taken
place at any point in time between the two assessments. It is not
known whether the change in lifestyle preceded or followed
1079
the onset of a change in health. A similar problem occurs when
using mortality as the outcome. Furthermore, the use of selfreported data and the non-response factor might introduce a
bias. One example is the opinion that self-rated data are a subjective and not an objective measure of health status like, for
instance, standardized mortality rates or diagnoses obtained from
health examinations. But longitudinal Swedish50 and American
studies51–53 showed that self-reported poor health status, used
as a dependent variable in the present study, was a strong predictor of death. This type of survey has a long tradition, and the
questions were well validated. A recently published American
study used data from the National Health and Nutrition Examination Survey (NHANES) III from 1988–1991 and found that
self-reported hypertension could be used for the surveillance of
hypertension trends.54
In summary, the present study confirms the cross-sectional
associations between smoking, obesity, and physical inactivity
and poor health status. Furthermore, this study revealed the importance of a sound lifestyle, especially regular physical activities,
for maintaining good health and achieving low mortality.
Acknowledgements
This work was supported by a travel grant to Jan Sundquist
from the Swedish Medical Research Council, Grant No. K9721P-11333-01A, the Swedish Council for Social Research,
Grant No. 94-0048:2B, and the Swedish Society of Medicine.
The authors thank Dr Michaela Kiernan at the Stanford Center
for Research in Disease Prevention for her insightful comments
on an earlier draft. There are no financial or other conflicts of
interest for any author.
References
1 Sundquist J, Johansson S-E. Long-term illness among indigenous and
foreign-born people in Sweden. Soc Sci Med 1997;44:189–98.
2 Blaxter M. Evidence of inequality in health from a national survey.
Lancet 1987;i:30–33.
3 Sorlie PD, Backlund E, Keller JB. US mortality by economic, demo-
graphic, and social characteristics: the national longitudinal mortality
study. Am J Public Health 1995;85:949–56.
4 Sundquist J. Country of birth, social class and health. A population-
based study on the influence of social factors on self-reported illness
in 223 Latin American refugees, 333 Finnish and 126 South European
labour migrants and 841 sex-, age- and education-matched Swedish
controls. Soc Sci Med 1995;40:777–87.
5 Statistics Sweden. Cause of Death 1997. Stockholm: Statistics Sweden, 1997.
6 Winkleby MA, Fortmann SP, Rockhill B. Trends in cardiovascular
disease risk factors by educational level: the Stanford Five-City Project.
Prev Med 1992;21:592–601.
7 Luepkker R, Murray D, Jacobs D et al. Community education for
cardiovascular disease prevention: risk factor changes in the Minnesota Heart Health Program. Am J Public Health 1994;84:1383–93.
8 Shea S, Basch C, Lantigua R, Wechsler H. The Washington Heights-
Inwood Healthy Heart Program: a 6-year report from a disadvantaged
urban setting. Am J Public Health 1996;86:166–71.
9 Tuomilehto J, Bonita R, Stewart A, Nissinen A, Salonen JT.
Hypertension, cigarette smoking and the decline in stroke incidence
in eastern Finland. Stroke 1991;22:7–11.
10 Kawachi I, Colditz GA, Stampfer MJ. Smoking cessation and
decreased risk of stroke in women. JAMA 1993;269:232–36.
1080
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
11 Paffenbarger RS, Hyde RT, Wing AL, Hsieh CC. Physical activity, all-
cause mortality, and longevity of college alumni. N Engl J Med 1986;
314:605–13.
12 Blair SN, Kohl HW, Paffenberger RS, Clark DG, Cooper KH, Gibbons
LW. Physical fitness and all-cause mortality. A prospective study of
healthy men and women. JAMA 1989;262:2395–401.
13 Davis MA, Neuhaus JM, Moritz DJ, Lein D, Barcaly JD, Murphy SP.
Health behaviors and survival among middle-aged and older men and
women in the NHANES I epidemiologic follow-up study. Prev Med
1994;23:369–76.
14 Leon AS, Connett J, Jacobs DR, Rauramaa R. Leisure time physical
activity levels and risk of coronary heart disease and death: The
Multiple Risk Factor Intervention Trial. JAMA 1987;258:2388–95.
15 Helmrick SP, Ragland DR, Leung RW, Paffenberger RS Jr. Physical
activity and reduced occurrence of non-insulin dependent diabetes
mellitus. N Engl J Med 1991;325:147–52.
16 Pavlou KN, Krey S, Steffee WP. Exercise as an adjunct to weight loss and
maintenance in moderately obese. Am J Clin Nutr 1989;49:1115–23.
community intervention programme. Int J Epidemiol 1993;22:
1026–37.
35 Willich SN, Lewis M, Löwel H, Arntz H-R, Schubert F, Schröder R.
Physical exertion as a trigger of acute myocardial infarction. N Engl J
Med 1993;329:1684–90.
36 Berlin JA, Colditz GA. A meta-analysis of physical activity in the
prevention of coronary heart disease. Am J Epidemiol 1990;132:
612–28.
37 Midgette AS, Baron JA, Rohan TE. Do cigarette smokers have diets
that increase their risks of coronary heart disease and cancer? Am J
Epidemiol 1993;137:521–29.
38 Näslund GK, Fredriksson M, Hellénius M-L, de Faire U. Effect of diet
and physical exercise intervention on coronary heart disease risk in
smoking and non-smoking men in Sweden. J Epidemiol Community
Health 1996;50:131–36.
39 Sjöström LV. Mortality of severely obese subjects. Am J Clin Nutr
1992;55:516S–523S.
40 Joung IAM, Stronks K, van de Mehren H, Mackenbach JP. Health
mortality from coronary heart disease due to biological and behavioural factors. Scand J Soc Med 1996;1:67–76.
behaviours explain part of the differences in self-reported health
associated with partner/marital status in the Netherlands. J Epidemiol
Community Health 1995;49:482–88.
18 Shinton R. Lifelong exposures and the potential for stroke prevention:
41 Berkman LF, Syme SL. Social network interaction and mortality.
17 Qvist J, Johansson S-E, Johansson LM. Multivariate analyses of
the contribution of cigarette smoking, exercise, and body fat. J Epidemiol
Community Health 1997;51:138–43.
A six-year follow-up study of Alameda county residents. Am J Epidemiol
1979;109:186–204.
19 Manson JE, Colditz G, Stampffer MJ et al. A prospective study of
42 Orth-Gomér K, Johnson JV. Social network interaction and mortality.
obesity and risk of coronary heart disease in women. N Engl J Med
1990;322:882–89.
A six-year follow-up study of a random sample of the Swedish
population. J Chron Dis 1987;40:949–57.
20 Manson JE, Willett WC, Stampfer MJ et al. Body weight and mortality
43 Sundquist J. Migration and health. Epidemiological studies in Swedish
among women. N Engl J Med 1995;333:677–85.
21 Shinton R, Sagar G, Beevers G. Body fat and stroke: unmasking the
hazards of overweight and obesity. J Epidemiol Community Health 1995;
49:259–64.
22 Lindenstrom E, Boysen G, Nyboe J. Lifestyle factors and risk of
cerebrovascular disease in women. Stroke 1993;24:1468–72.
23 Heini AF, Wiensier RL. Divergent trends in obesity and fat intake
patterns: the American paradox. Am J Med 1997;102:259–64.
24 Sundquist J, Johansson S-E. The influence of socioeconomic status,
ethnicity and lifestyle on increase in the body mass index 1980–81 to
1988–89. Int J Epidemiol 1998;27:57–63.
25 Prättälä R, Karisto A, Berg MA. Consistency and variation in
unhealthy behaviour among Finnish men, 1982–1990. Soc Sci Med
1994;39:115–22.
26 Statistics Sweden. The Swedish Survey of Living Conditions. Design and
Method. Appendix no 16. Stockholm: Statistics Sweden, 1996.
27 Wärneryd B. Levnadsförhållanden. Återintervjustudie i undersökningen av
levnadsförhållanden (ULF) 1989. (Living conditions. Reinterview in ULF
1989). (In Swedish.) Appendix 12. Stockholm: Statistics Sweden, 1990.
28 Diggle PJ, Liang KY, Zeger SL. Analysis of Longitudinal Data. Oxford
Science Publications. Oxford: Clarendon Press, 1994.
29 Kleinbaum DG. Survival Analysis. New York: Springer, 1995.
30 Strawbridge WJ, Cohen RD, Shema SJ, Kaplan GA. Successful aging:
predictors and associated activities. Am J Epidemiol 1996;144:135–41.
31 Stewart AL, Hays RD, Wells KB et al. Long-term functioning and well-
primary health care (dissertation). University of Lund: Studentlitteratur
Lund, 1994.
44 Sundquist J, Johansson S-E. Indicators of socio-economic position
and their relation to mortality in Sweden. Soc Sci Med 1997;45:
1757–66.
45 Sundquist J, Johansson S-E. The influence of country of birth on
mortality from all-causes and cardiovascular disease in Sweden
1979–85. Int J Epidemiol 1997;26:279–87.
46 Pekkanen J, Uutela A, Valkonen T, Vartiainen E, Tuomilehto J, Puska P.
Coronary risk factor levels: differences between educational groups in
1972–1987 in eastern Finland. J Epidemiol Community Health 1995;
49:144–49.
47 Valkonen T. Adult mortality and level of education: A comparison of
six countries. In: Fox J (ed.). Health Inequalities in European Countries.
Aldershot: Gower, 1989, pp.142–62.
48 Elo IT, Preston SH. Educational differentials in mortality: United
States, 1979–85. Soc Sci Med 1996;42:47–57.
49 Hoffmans MDAF, Kromhout D, de Lezenne Coulander C. The impact
of body mass index of 78,612 18-year old Dutch men on 32-year
mortality from all-causes. J Clin Epidemiol 1988;41:749–56.
50 Sundquist J, Johansson S-E. Self-reported poor health and low
educational level predictors for mortality: a population-based followup study of 39,156 people in Sweden. J Epidemiol Community Health
1997;51:35–40.
51 Kaplan GA, Camacho T. Perceived health and mortality: a nine-year
follow-up of the Human Population Laboratory cohort. Am J Epidemiol
1983;117:292–304.
being outcomes associated with physical activity and exercise in
patients with chronic conditions in the medical outcomes study. J Clin
Epidemiol 1994;47:719–30.
52 Idler EL, Angel RJ. Self-rated health and mortality in the NHANES-I
32 Farquar J. Community-based model of lifestyle intervention trials.
53 McGee DL, Liao Y, Cao G, Cooper RS. Self-reported health status
Am J Epidemiol 1978;108:103–11.
33 Farquar JW, Fortmann SP, Maccoby N et al. The Stanford five-city
project: design and methods. Am J Epidemiol 1985;122:323–34.
34 Brännström I, Weinehall L, Pearson L, Wester P, Wall S. Changing
social patterns of risk factors for cardiovascular disease in a Swedish
—epidemiologic follow-up study. Am J Public Health 1990;80:446–52.
and mortality in a multiethnic US cohort. Am J Epidemiol 1999;149:
41–46.
54 Vargas CM, Burt VL, Gillum RF, Pamuk ER. Validity of self-reported
hypertension in the National Health and Nutrition Examination
Survey III, 1988–1991. Prev Med 1997;26:678–85.