International Journal of Obesity (1998) 22, 89±96
ß 1998 Stockton Press All rights reserved 0307±0565/98 $12.00
Predictors of weight change in men: Results
from The Health Professionals Follow-Up Study
EH Coakley1,4, EB Rimm1,2,4, G Colditz1,2, I Kawachi1,3 and W Willett1,2,4
1
Channing Laboratory, Department of Medicine, Harvard Medical School and Brigham and Women's Hospital, Boston, and 2 Department
of Epidemiology, Health and Social Behavior, and Nutrition, Harvard School of Public Health, Boston, MA, USA
OBJECTIVE: Since the prevalence of adult obesity is increasing in the United States, we examined the effect of
changing common habits (exercise, TV viewing, smoking and eating habits) on four year change in body weight.
DESIGN: A prospective cohort study of US male health professionals with follow-up from 1988±1992. Participants
were 19 478 men aged 40±75 in 1986, who were free of cancer, coronary heart disease, stroke and diabetes.
METHODS: Multiple regression was used to determine the association between four year change in body weight
(from 1988±1992) and common habits, after adjusting for baseline age, hypertension and hypercholesterolemia.
RESULTS: For middle aged men, vigorous activity was associated with weight reduction and TV=VCR viewing and
eating between meals with weight gain. Quitting smoking and a history of voluntary weight loss prior to the study
period were consistently related to weight increase. Recently being on a diet was more strongly associated with
weight loss among older men. Over the four year follow-up period, middle-aged men who increased their exercise,
decreased TV viewing and stopped eating between meals, lost an average weight of 71.4 kg (95% con®dence interval
(CI)71.6 ± 71.1 kg), compared to a weight gain of 1.4 kg among the overall population. The prevalence of obesity
among middle-aged men was lowest among those who maintained a relatively high level of vigorous physical activity,
compared to those who were relatively sedentary.
CONCLUSION: These data suggest that improvement in the mix of health habits, particularly increasing vigorous
activity, as well as decreasing TV use and changing eating habits, results in weight maintenance or a modest weight
loss over four years.
Introduction
Increased lifetime risks of heart disease,1,2,3,10 diabetes4,5, osteoarthritis,6,7 hypertension8,9 and other
illnesses have been associated with increased levels
of body mass. These risks appear to exist at levels
below the currently accepted de®nition of obesity
(BMI of 27.8 kg=m2 for men11).
The prevalence of obesity in adults has steadily
increased in the United States over the past 20 y.11
According to estimates based on NHANES III data11,
33% of men (similar estimates for black and white
men) meet the de®nition of obesity. Furthermore,
there is a substantial prevalence of obesity in children12,13 and since weight in adolescence correlates
with adult weight,14,34 the high prevalence of obesity
could continue into the next adult generation. Therefore, identifying mechanisms that will help prevent or
reverse weight gain is critical.
Given the prevalence of obesity in the adult population, it is not surprising that the prevalence of
reported attempts to lose weight at a given point in
time is also high ± approximately 25% of adult US
Correspondence: Eugenie Coakley, Channing Laboratory,
181 Longwood Ave, Boston, MA 02115, USA.
Received 20 May 1997; revised 8 September 1997; accepted
23 September 1997
men.15 A recent review of the literature of weight
cycling16 concluded, however, that the preponderance
of evidence suggests, that repeated weight losses and
regains does not affect the success of future weight
loss attempts.
The purpose of the present study is to describe the
impact that ordinary lifestyle factors such as exercise,
smoking, TV=VCR viewing, dieting and eating habits
have on weight change over four years in a cohort of
middle to older aged male health professionals.
Because of the prospective nature of our study cohort,
we are able to control for diseases that in¯uence weight,
as well as for age, baseline weight and height. We also
study whether a history of voluntary weight loss is an
independent predictor of weight change and whether
it modi®es the effect of the other factors that may
in¯uence weight change.
Methods
Study Group
The Health Professionals Follow-up Study is a prospective investigation of 51 529 male health professionals aged 40±75 y in 1986. The cohort includes
29 683 dentists, 10 098 veterinarians, 4185 pharmacists, 3745 optometrists, 2218 osteopathic physicians
and 1600 podiatrists. The study began in 1986, when
the participants completed a detailed questionnaire on
Predictors of weight change in men
EH Coakley et al
90
diet, exercise and medical history. We mailed follow
up questionnaires in 1988, 1990 and 1992, to update
information on exposures and to ascertain events
related to newly diagnosed disease.17
Every two years, up to three mailings of the main
study questionnaire are sent to participants, which
contains speci®c questions on weight, smoking
status and physical activity. If men do not respond
to the main questionnaire, we send a short questionnaire primarily to document clinical endpoints and not
other characteristics, such as weight. Of the 36 353
men who returned a main questionnaire in 1992, we
excluded 9345 men who had developed cancer, heart
disease, stroke or diabetes prior to December 1992.
These cases were excluded because these diseases
could lead to change in weight, activity or eating
habits. An additional 7530 men were excluded
because they were missing information on key variables, such as body weight. This analysis is based on
19 478 men for whom we had a complete set of
predictor and outcome information for the study
period 1988±1992.
Most cases were excluded for reasons presumably
unrelated to the reporting of body weight. However,
there may be biased non-response18 among the men
who completed long questionnaires, but skipped the
questions regarding body weight. To try to address
this issue, we compared the 1988 body weight of the
analysis group to the group who were excluded for
missing the 1992 body weight. The average 1988
weight in the analysis group was 80.7 kg (Standard
Deviation (s.d.) ˆ 11.2; Inter-quartile Range (IQR) ˆ
72.6±86.2 kg). For the 3297 men who reported weight
in 1988 but not in 1992, their mean 1988 weight was
81.0 kg (s.d. ˆ 11.6; IQR ˆ 73.0±87.5 kg). Thus, there
was not a substantial difference in baseline (1988)
weight between men in the analysis group and men
who were excluded due to incomplete weight data at
follow-up (1992).
Variables of interest
Body weight in pounds was self-reported on both the
1988 (study baseline) and 1992 questionnaires. Height
was ascertained on the 1986 questionnaire and
assumed to be constant throughout the study period.
In 1990, Rimm et al19 found self-reported weight to be
highly accurately assessed when compared with standardized measurements by a technician in a subset of
this cohort. The corrected Pearson correlation between
self reported and technician measured body weight
was 0.97. Self-reported weight was, on average, 1.0 kg
lower than technician measured weight.
Previous research in this cohort has shown that
there are age-related patterns of obesity, as well as
age-related differences in anthropometric risk factors
for disease.20±22 For example, body mass index (BMI,
kg=m2) peaks among men aged 60±65 y, and declines
thereafter.20 To explore whether there are also agerelated differences in factors associated with weight
change, our analyses are strati®ed by baseline age
group: 45±54 y, 55±64 y and 65 y.
The level of recreational physical activity was
measured on both the 1988 and 1992 questionnaires.
The instrument assesses the average number of hours
spent per week over the past year, engaged in seven
common activities: jogging, running, lap swimming,
bicycling and rowing (including stationary machines),
calisthenics and racquet sports. Vigorous, as opposed
to total, physical activity was used because recall and
report of vigorous activity is better than for light to
moderate activity.23
The average amount of time (h) per week, over the
past year, that respondents spent watching TV=VCR
was ascertained on both the 1988 and 1992 questionnaires. Hours of TV=VCR use, is our measure of the
level of recreational sedentary activity. Smoking
status was obtained on the 1988, 1990 and 1992
questionnaires. We classi®ed men as smokers, nonsmokers and those attempting to quit during this
period.
In 1992, men were asked if and when they ate
between meals. They were also asked whether or not
they used a diet (that is, `restricted caloric intake',
`skipped meals' or `used a commercial weight loss
program') to voluntarily lose ten or more pounds
(4.5 kg) in the four year period from 1988 to 1992
(Figure 1). Fat intake as measured by the food
frequency questionnaire in 1990, was used to estimate
typical fat intake during 1988±1992.
The fat composition of diet, rather than absolute
intake, has been shown to be a predictor of weight
gain in younger men and women in one study.24
Energy-adjusted fat intake was computed according
to Willett's method.25 Adjusted intake (g=d) was
computed as the residual of a regression model with
total caloric intake as the independent variable and
total fat intake as the dependent variable, plus a
constant. The constant was the predicted fat intake
using the mean calories for the entire cohort.
Four-year (1988±1992) and 20-year (1972±1992)
history of voluntary weight loss was measured in 1992
(Figure 1). For each period, the number of times ®ve
or more pounds (2.3 kg) was voluntarily lost, as well
as an estimate of the amounts of each loss, was
collected.
The number of voluntary weight loss episodes for
the 16 y prior to the baseline period (1972±1987) was
computed by subtracting the number of episodes
during the four-year period from the number of
episodes during the twenty-year period. Finally, selfreport of physician-diagnosed high blood pressure and
high cholesterol was ascertained from the 1988 questionnaire.
Statistical methods
We investigated the unadjusted relationship between
health habits and self-reported body weight by age
group (45±54 y, 55±64 y and 65 y). Overall differ-
Predictors of weight change in men
EH Coakley et al
91
43. Between the ages of 18±30, how many times did you purposely lose 10 or more pounds (excluding illness)?
s 0 times
s 1±2 times
s 3±4 times
44. Within the last 20 years (exclude illness):
5±9 pounds:
s 0 times
10±10 pounds:
s 0 times
20±49 pounds:
s 0 times
50‡
s 0 times
s 7‡ times
s
s
s
s
1±2
1±2
1±2
1±2
times
times
times
times
s
s
s
s
3±4
3±4
3±4
3±4
times
times
times
times
s
s
s
s
5±6
5±6
5±6
5±6
times
times
times
times
s
s
s
s
7‡
7‡
7‡
7‡
times
times
times
times
45. Within the last 4 years (exclude illness):
a. What was your: Minimum weight ________ lbs. Maximum weight ________ lbs.
b. How many times did you lose each of the following amounts of weight on purpose (exclude illness):
5±9 pounds:
s 0 times
s 1±2 times
s 3±4 times
s 5±6 times
10±10 pounds:
s 0 times
s 1±2 times
s 3±4 times
s 5±6 times
20-49 pounds:
s 0 times
s 1±2 times
s 3±4 times
s 5±6 times
50‡
s 0 times
s 1±2 times
s 3±4 times
s 5±6 times
s
s
s
s
7‡
7‡
7‡
7‡
times
times
times
times
c. If you lost 10 or more pounds, what primary method(s) did you use for your most recent weight loss
(®ll in all that apply)
s Did not lose 10 or more pounds s Weight loss was unintentional (e.g. illness, unusual stress, depression)
s Low calorie diet s Skipped meals/fasted s Increased exercise s Diet pills
s Commercial weight loss program s Gastric surgery/intestinal bypass s Other
Figure 1 Series of questions on voluntary weight loss, 1992 Questionnaire.
ences across age strata were tested using one-way
ANOVA for continuous variables and the chi-square
test for dichotomous variables.26
Multivariate regression analyses were performed to
determine if health habits in 1992 and the 16 y history
of prior voluntary weight loss (1972±1987) were
predictive of the 4 y weight change (1988±1992).
The 1992 health habits were: vigorous physical activity, TV=VCR viewing, eating between meals, adjusted
fat intake and indicator variables for being a continuing or quitting smoker during the study period. All
models controlled for baseline (1988) weight, height,
exercise and TV=VCR viewing, blood pressure and
cholesterol. Models were strati®ed by three levels of
age; actual age was also included in each of the three
models to control for residual confounding. The actual
outcome variable was weight in 1992, but by controlling for baseline weight, it was equivalent to modeling
weight change during the study period as the outcome.
Thus, results are presented in terms of four year
average weight loss or gain. Since exercise level and
TV=VCR viewing in both 1988 and 1992 were covariates, we were modelling the effect of a 4 y change
in these variables on weight change.
In order to determine whether a history of voluntary
weight loss weakened the association between these
lifestyle factors and weight change, we further strati®ed the models by (yes=no) history of loss. We
compared the regression coef®cients for the lifestyle
factors and the results were similar for those with and
without a history; this was true for all three age strata.
If anything, the effect of exercise was somewhat
stronger among those who had a voluntary weight
loss history. Thus, only age-strati®ed models, controlling for the number of voluntary (16 y) weight losses,
are presented. Sixteen year weight loss history was
used as a predictor, rather than the 4 y or 20 y
measures, because the latter two measures include
weight loss between 1988 and 1992, a component of
the outcome.
Since body weight is typically skewed toward
higher values, regression models were also run using
log transformed weight in 1992 as the outcome.
However, since the results were equivalent to those
using untransformed weight, models based on
untransformed weight are presented. The goodness
of ®t of the models was evaluated by examining
extreme outliers and points with high leverage
values.26 All analyses were performed in release
6.12 of SAS.27
Results
Sample description
The 19 478 men comprising this cohort were middleto older-aged (average age 54.6 y), primarily nonsmoking (5.1% current smokers) and weighed an
average of 81.5 kg in 1992, having gained an average
of 0.8 kg since 1988 (Table 1). The group averaged
about 2 h vigorous activity per week, in both 1988 and
1992.
Categorical analysis
Men differed signi®cantly in weight and health habits
when classi®ed by age category (Table 1). Body
weight and vigorous activity levels declined with
age, whereas TV=VCR viewing increased with age.
Men aged 65 y were less likely to diet (12% vs 23%
for those aged 45±54 y) or to eat between meals (45%
vs 59%, respectively). Energy-adjusted fat intake
averaged about 70 g=d in each age group. Older men
were less likely than younger men to have experi-
Predictors of weight change in men
EH Coakley et al
92
Table 1 Characteristics [mean and (standard deviation)] according to age category for 19 478 US men in the Health Professionals
Follow±Up Study
Age Group
Overall
(n ˆ19 478)
Weight
1992 (kg)*
1988 (kg)*
Age 1988 (y)*
Vigorous activity (h=week)*
1992
1988
TV=VCR viewing (h=week)*
1992
1988
Adjusted fat intake 1990 (g=d)*
No. Vol. Weight Losses*
Hypertension 1988(%)*
Hypercholesterolaemia 1988 (%)*
Current smokers(%)*
Quit smoking (%)
On diet last 4 y (%)*
Eat Between Meals(%)*
44^54 Yrs
(n ˆ10 272)
55^64 Yrs
(n ˆ 5729)
65 ‡ Yrs
(n ˆ 3477)
81.5 (11.7)
80.7 (11.2)
54.6 (9.2)
82.6 (12.0)
81.2 (11.4)
47.1 (3.8)
81.4 (11.5)
80.9 (11.0)
59.2 (2.9)
78.4 (10.8)
78.6 (10.5)
69.0 (3.4)
1.9 (3.0)
1.8 (2.7)
2.1 (2.9)
2.0 (2.8)
1.8 (3.0)
1.6 (2.6)
1.5 (3.0)
1.3 (2.5)
8.9 (7.7)
11.3 (8.4)
69.6 (13.8)
1.6 (2.4)
16.4%
18.8%
5.1%
3.1%
19.2%
54.1%
8.2 (7.0)
10.3 (7.9)
70.1 (13.8)
1.9 (2.6)
11.1%
17.6%
5.1%
3.2%
22.9%
58.6%
9.3 (7.9)
11.7 (8.5)
69.5 (13.9)
1.5 (2.3)
19.5%
20.5%
5.7%
3.3%
17.4%
51.5%
10.2 (9.3)
13.2 (9.4)
68.4 (13.8)
1.0 (1.8)
26.9%
19.8%
4.1%
2.2%
11.7%
45.1%
For continuous variables, one±way random effects ANOVA was used to test overall differences across age groups. Chi-square test
used for dichotomous variables.
*P 0.001.
**P 0.01.
enced a substantial ( 4.5 kg) voluntary weight loss
during the 16 y prior to 1988. Hypertension and
hypercholesterolaemia were more prevalent with
increasing age. The prevalence of smoking, as well
as quitting smoking, were low across ages, especially
among older men.
Multivariate models
For all age groups, weight in 1988 was the single most
important predictor of weight in 1992. However,
health habits in 1992 such as vigorous physical
activity, TV=VCR viewing, eating habits and quitting
smoking were signi®cantly related to weight change,
even after controlling for baseline weight, height and
baseline vigorous physical activity and TV=VCR use
(Table 2).
Among men in the younger (45±54 y) age group,
increased vigorous activity was signi®cantly related to
the average 4 y weight loss. Speci®cally, every
1.5 h=week average increase in activity over four
years, was associated with an average 0.2 kg weight
loss. Smoking, being on a diet to lose 4.5 kg and age
were mildly related to weight loss. On the other hand,
eating between meals, quitting smoking and a history of
weight losses (in the previous 16 y) were all positively
related to weight gain, as was a 4 y net increase in
TV=VCR viewing. For example, a 10 h=week increase
in TV=VCR viewing and eating between meals, were
each associated with approximately a 0.2 kg weight gain
over four years. Quitting smoking, however, had the
most serious effect on average weight gain (1.2 kg). The
level of fat in the diet (g=d) was positively associated
with weight gain (0.1 kg increase for 10 g=d of fat).
Table 2 Regression coef®cients (s.e.m.) for the prediction of four year weight change, strati®ed by age, among 19 478 men in the
Health Professionals Follow-Up Studya
Age 45^54 y
(n ˆ10 272)
Vigorous activity 1992 (m=week)
TV=VCR use 1992 (h=week)
Eat Between Meals '92
Used diet to lose 10 lbs, 1988±1992
Current smoker
Tried to quit smoking (1988±1992)
Age in 1988 (y)
No. vol. weight losses
Adjusted fat (10 g=d) intake (1990)
a
70.16* (0.02)
0.02** (0.01)
0.25** (0.09)
70.23*** (0.11)
70.44*** (0.19)
1.19* (0.23)
70.02*** (0.01)
0.17* (0.02)
0.10* (0.003)
Age 55^64 y
(n ˆ 5729)
70.09* (0.02)
0.01 (0.01)
0.31** (0.11)
70.47** (0.15)
70.63** (0.24)
0.62*** (0.31)
70.08* (0.02)
0.07** (0.03)
0.10* (0.004)
Age 65 y
(n ˆ 3477)
70.02 (0.02)
0.00 (0.01)
70.01 (0.12)
70.64** (0.20)
71.02** (0.31)
2.23* (0.42)
70.09* (0.02)
0.23* (0.04)
0.10 (0.005)
All models controlled for baseline (1988): weight, height, vigorous activity, TV=VCR viewing, high blood pressure and high cholesterol.
Each regression coef®cient represents weight change (in kg) for a one unit change in the predictor variable.
*p < 0.001.
**p 0.01.
***p 0.05.
Predictors of weight change in men
EH Coakley et al
Table 3 The Effect of changing vs maintaining vigorous activity and TV=VCR viewing on
weight change
Average 4 y weight change
(compared to referent)
Vigorous Activity Pattern:
Increase activity (to 1.5 h=week)
Maintain high level (at 1.5 h=week)
Maintain low level (at < 1.5 h=week)
Decrease activity (to < 1.5 h=week)
TV=VCR Pattern:
Decrease viewing (to 14 h=week)
Maintain low level (at 14 h=week)
Maintain high level (at > 14 h=week)
Increase viewing (to > 14 h=week)
95% CI
70.9 kg
70.3 kg
hreferenti
0.6 kg
(71.2, 70.6)
(70.5, 0.1)
70.1 kg
70.1 kg
hreferenti
1.2 kg
(70.5, 0.3)
(70.4, 0.2)
(0.3, 0.8)
(0.4, 2.0)
CI ˆ con®dence interval.
Values, based on a regression model, are for a man aged 50 y, 1.8 m, 77 kg (in 1988), nonsmoking, non-hypertensive and non-hypercholesterolemic (in 1988), who did not diet during
the study period. Controlling for these covariates, men in the referent category of low activity
and high TV viewing gained, on average, 1.1 kg (95% CI: (0.7, 1.4)).
For men in the middle (55±64 y) age group, age,
increased vigorous activity, dieting and smoking were
associated with weight loss. The effect of exercise was
lower than for the younger age group; every 1.5 h=week
increase in vigorous activity was associated with a
0.1 kg weight loss over four years. Energy-adjusted fat
intake, eating between meals, history of weight loss and,
especially quitting smoking, were all positively associated with weight gain. TV=VCR use was not a signi®cant predictor of weight gain.
The model for men in the oldest age group ( 65 y)
differed from those of the middle and younger age
groups. Age, recent dieting and being a smoker, were
signi®cant negative predictors. Their effects were
stronger on weight loss than for the younger age
group. Positive predictors of weight gain were quitting
smoking, history of weight loss and fat intake. Neither
physical activity, nor TV=VCR use, nor eating
between meals, were related to the outcome in this
age group.
There were many similiarities among the three agespeci®c models, in terms of the direction and magnitude of the effects of covariates on weight change.
However, the exercise effect was signi®cantly different across ages and there were interesting trends in the
effects of the other covariates that justi®ed the age
strati®cation.
These regression models indicate that behavior
change for relatively modi®able health habits such
as vigorous activity, TV=VCR viewing, fat composition of the diet and eating between meals, would have
the greatest effect among non-smoking men aged 45±
54 y (Table 2). This age group was also, on average,
heavier and had greater 4 y weight gains (Table 1).
Using the regression model for this age group, we can
predict* that increasing exercise (from 0 h=week in
1988 to 5 h=week in 1992), decreasing TV=VCR use
(from 21 h=week in 1988 to 3.5 h=week in 1992) and
*Other regression values set to their mean values: age ˆ 50 y,
height ˆ 1.78 metres, baseline (1988) weight ˆ 77 kg, non-smoker,
no baseline hypertension nor hypercholesterolaemia, no dieting
during study period (1988±1992).
eliminating eating between meals, would result in an
average 1.4 kg weight reduction over four years (95%
con®dence interval (CI) ˆ 71.7±71.1 kg), among
men without a history of voluntary weight loss.
Among men with a history of two 4.5 kg voluntary
weight losses, the weight loss would still be signi®cant, but attenuated, at 71.1 kg (95% CI ˆ 71.4±
70.7 kg).
An issue not resolved by hese analyses, is whether
change in health habits, if maintained over time,
would result in sustained weight loss. Although not
the focus of our analysis, we tried to address this issue
with a re-parameterization of activity and TV/VCR
viewing in our regression models. Again, for simplicity, we focus on the model for younger (45±54 y)
men.
We replaced the continuous variables for vigorous
physical activity and TV=VCR viewing in 1988 and
1992 with indicator variables describing the 4 y level
of activity and inactivity: increased from lower to
higher level, decreased from higher to lower level,
maintained higher level and maintained lower level.
The referent group were men who maintained a low
level of vigorous activity and a high level of TV
viewing. This sedentary group gained an average
1.1 kg over four years (Table 3). Men who increased
exercise to a higher level and those who maintained a
higher level, tended to gain less weight. These results
did not change when we used a higher cut-off point
for vigorous activity. Men who increased their level of
TV=VCR viewing (without increasing exercise)
increased their weight (Table 3).
The results for weight gain and weight loss, may be
somewhat attenuated because there may be substantial
change in activity or viewing among those classi®ed
as `maintainers'.
Another approach to describing the long term effect
of exercise is given in ®gure 2, which shows the
prevalence of obesity (BMI 27.8 kg=m2), over time,
for men aged 45±54 y, with the activity patterns
described in Table 3. The group maintaining a level
of vigorous activity of 1.5 h=week showed a much
93
Predictors of weight change in men
EH Coakley et al
94
Figure 2 Prevalence of obesity (body mass index) 27.8 kg/m2 ) over time for different patterns of recreational vigorous physical
activity. This chart is based on 3666 men aged <55 years (in 1986), non-smoking, non-hypertensive and non-hypercholesterolemic.
Use of BMI, controls for height's effect on weight. Fat intake and history of weight loss was not controlled.
lower prevalence of obesity over time, and the group
that increased to this activity level by 1992 showed a
leveling-off in the rate of increase of obesity prevalence. These results are consistent with a stable or
lower rate of increase in average weight over time
among men achieving or maintaining a high level of
physical activity.
Discussion
In our cohort of middle- and older-aged professional
men, multivariate models showed that predictors of
weight change differed somewhat by age. Increased
vigorous physical activity was associated with weight
loss or maintenance in all age groups, but had a
stronger effect among younger men (45±54 y). History
of previous weight loss and quitting smoking, were
consistently associated with weight gain, but history
of previous weight loss did not modify the effect of
the other factors on weight change. Energy-adjusted
fat intake was a modest predictor of weight gain
before the age of 65 y. Age was an important predictor
of weight loss after age 55 y. Similarly, dieting was
not a signi®cant predictor of weight loss until after age
55 y.
Of the methods examined for men aged 45±54 y, a
change in the mix of health habits, such as a combination of increased activity, decreased TV=VCR use and
abstention from eating between meals, was associated
with a modest average weight loss of approximately
1.4 kg. This offset the general trend in the 45±54 y age
group to gain an average of 1.4 kg over the same period.
These results are consistent with those of Blair28 and
Williamson et al29. The apparent effect of these factors
was smaller (about 70.8 kg) in men with a history of
two weight losses of 4.5 kg, perhaps in part because
data at only two time points characterizes long term
trends less well among those whose weights ¯uctuate
substantially.
Furthermore, younger, healthy, non-smoking men
with relatively high vigorous physical activity levels
had lower annual prevalences of obesity, as well as a
slower rate of increase in obesity over time than
comparable, but sedentary, men. If this type of sustained effect also exists in the general population of
such men, this dampening effect on obesity rates
would be of public health and economic signi®cance.
These ®ndings are limited to middle- to older-aged
men of relatively high socio-economic status (SES).
However, it is unlikely the direction of the effects
would change due to SES, but the list of candidate
recreational activities may need to be modi®ed for
different SES groups. The magnitude of effects might
differ for men aged < 45 y.
This research focused on the effects of vigorous
recreational activity, such as running and swimming,
because past research showed cardiovascular bene®t,
as well as better respondent recall of these types of
activities. Walking, however, is a much more common
form of recreational activity. In regression models that
replaced vigorous activity with walking, increased
hours of walking per week was a statistically signi®cant predictor of weight loss among younger men
( < 65 y). However, the average weight loss was much
lower than for vigorous activity.
In this population-based cohort of free-living men,
it is not possible to control why health habits change.
For example, exercise level may increase in response
to weight gain, attenuating the effect of exercise on
weight regulation. Furthermore, increased activity
levels may not be sustained once weight loss has
been achieved.
The timing of measurement of key variables may
have also affected the magnitude of the absolute
differences in weight change. Vigorous activity and
Predictors of weight change in men
EH Coakley et al
TV=VCR viewing were average amounts per week
over the past year. Averaging activity over a year
smoothes out variation due to shorter sustained periods of activity (for example, during the summer) and
periods of inactivity (for example during the winter).
Measurements were taken two years apart, so it is not
known when during the interval activity, or inactivity,
levels changed, nor for how long the actual change
was maintained. These factors could all attenuate the
effects of exercise and TV=VCR viewing on weight.
TV=VCR use was a proxy measure for sedentary
activity and may not capture the full amount of
inactivity. However, TV viewing is a major source
of recreation in the US, where the average adult
watches 29 h=week.30 Furthermore, TV use has been
associated with snacking and is relatively more sedentary compared to other hobbies.31±33
Two variables served as proxies for eating habits,
eating between meals and energy-adjusted fat intake.
Unfortunately, data on these measures were available
at only one point in time during the study period, so
only cross-sectional, not longitudinal, effects were
studied. It is also important to note that not all
eating between meals leads to excess caloric intake.
Adjustment of fat intake for the total caloric (energy)
intake allowed us to investigate the effects of fat
composition of diet. A diet with a high fat composition has been inconsistently linked to obesity in other
observational studies.14,24,37
Neither the accuracy of 4 y and 20 y weight loss
episodes nor dieting, has been validated in this population. It is unclear whether relatively lighter men
would be more likely to remember a substantial
voluntary loss, since these events are rare and presumably important. Alternatively, perhaps heavier
men with a history of weight cycling may recall
losses more frequently, if they more closely monitor
their weight. Similarly, the effect of dieting was seen
only among older men. Perhaps this is a re¯ection of
a higher dieting success rate due to concern about
chronic disease. In the same vein, more effective
dieting in this age group may imply more reliability
in reporting voluntary weight loss episodes.
We did not control for caloric intake. Other studies
have suggested it is not a signi®cant predictor of
weight change,14 in part because it re¯ects increased
physical activity.35 We also tested 1990 alcohol consumption in the regression models and it was not a
signi®cant predictor of weight. Excluding it from the
models did not affect the coef®cients of other variables.
In summary, we have demonstrated that there are
independent associations of changes in vigorous activity and TV=VCR use, as well as eating and smoking
habits with average weight change in a large cohort of
men. These results imply that changes in these health
habits, particularly regular physical activity, over
time, may lead to weight maintenance or modest
weight loss in adulthood, and hence the prevention
of major chronic illness.
Acknowledgements
This study was funded in part by NIH grants CA55075
and HL35464, and by the Boston Obesity=Nutrition
Center, Epidemiology Core, a collaborative agreement with the Centers for Disease Control (DK
46200).
References
1 Willett WC, Manson JE, Stampfer MJ, Colditz GA, Rosner B,
Speizer FE, Hennekins CH. Weight, Weight Change, and
Coronary Heart Disease in Women. JAMA 1995; 273(6):
461±465.
2 Manson JE, Colditz GA, Stampfer MJ, Willett WC,
Rosner B, Monson RR, Speizer FE, Hennekins CH. A
Prospective Study of Obesity and Risk of Coronary Heart
Disease in Women. New Eng J of Med 1990; 332(13): 882±
889.
3 Kannel WB, Cupples LA, Ramaswami R, Stokes J, Kreger
BE, Higgins M. Regional Obesity and Risk of Coronary
Vascular Disease: The Framingham Study. Clin Epidemiol
1991; 44(2): 183±190.
4 Colditz GA, Willett WC, Rotnitsky AG, Manson JE. Weight
Gain as a Risk Factor for Clinical Diabetes in Women. Ann Int
Med 1995; 122(7): 481±486.
5 Lipton RB, Liao Y, Cao G, Cooper RS, McGee D. Determinants of Incident NIDDM Among Blacks and Whites in a
National Sample. Am J Epidemiol 1993; 138(10): 826±839.
6 Anderson JJ, Felson DT. Factors Associated with Osteoarthritis of the Knee in the First National Health and Nutritional
Examination Survey (HANES I): Evidence for an Association
with Overweight, Race, and Physical Demands of Work. Am J
Epidemiol 1988; 128(1): 179±189.
7 Felson DJ, Anderson JJ, Naimark A, Walker AM, Meenan RF.
Obesity and Knee Osteoarthritis. Annals of Int Med 1988;
109(1): 18±24.
8 DeFronzo RA, Ferrannini E. A Multifaceted Syndrome
Responsible for NIDDM, Obesity, Hypertension, Dyslipidemia, and Atherosclerotic Coronary Vascular Disease. Diabetes
Care 1991; 14(3): 173±191.
9 Kaplan NM. The Deadly Quartet. Arch Intern Med 1989; 149:
1514±1520.
10 Lauer MS, Anderson KM, Kannell WD, Levy D. The Impact
of Obesity on Left Ventricular Mass and Geometry. JAMA
1991; 266(2): 231±236.
11 Kuczmarski RJ, Flegal KM, Campbell SM, Johnson CL.
Increasing Prevalence of Overweight Among US Adults.
JAMA 1994; 272(3): 205±211.
12 Gortmaker SL, Dietz WH, Sobol A, Wehler CA. Increasing
Pediatric Obesity in the United States. Am J Dis Children
1987; 141(5): 535±540.
13 Gortmaker SL, Dietz WH, Cheung LW. Inactivity, Diet, and
the Fattening of America. J Am Dietetic Assoc 1990; 90(9):
1247±1252.
14 Colditz GA, Willett WC, Stampfer MJ, London SJ, Segal MR,
Speizer F. Patterns of Weight Change and Their Relation to
Diet in a Cohort of Healthy Women. Am J Clin Nutr 1990; 51:
1100±1105.
15 Williamson DF, Serdula MK, Anda RF, Levy A,
Byers T. Weight Loss Attempts in Adults: Goals, Duration,
and Rate of Weight Loss. Am J Public Health 1992; 82: 1251±
1257.
16 National Task Force on the Prevention and Treatment of
Obesity. Weight Cycling. JAMA 1994; 272(15): 1196±2002.
17 Rimm EB, Stampfer MJ, Colditz GA, Giovannucci E, Willett
WC. Effectiveness of Various Mailing Strategies Among
Nonrespondents in a Prospective Cohort Study. Am J Epidemiol 1990; 131: 1068±1071.
18 Little RJA, Rubin DB. Statistical Analysis with Missing Data.
John Wiley and Sons: New York, 1987, pp 3±19.
95
Predictors of weight change in men
EH Coakley et al
96
19 Rimm EB, Stampfer MJ, Colditz GA, Chute CG, Litin LB,
Willett WC. Validity of Self-Reported Waist and Hip Circumferences in Men and Women. Epidemiology 1990; 1(6): 466±
473.
20 Rimm EB, Stampfer MJ, Giovannucci E, Ascherio A, Spiegelman D, Colditz GA, Willett WC. Body Size and Fat Distribution as Predictors of Coronary Heart Disease Among Middleand Older- Aged U.S. Men. Am J Epidemiol 1995; 141(12):
1117±1125.
21 Giovannucci E, Rimm EB, Chute CG, Kawachi I, Colditz GA,
Stampfer MJ, Willett WC. Obesity and Benign Prostatic
Hyperplasia. Am J Epidemiol 1994; 140: 989±1002.
22 Chan JM, Rimm EB, Colditz GA, et al. Fat Distribution and
Weight Gain as Risk Factors for Clinical Diabetes in Men.
Diabetes Care 1994; 17: 961±969.
23 Chasan-Taber S, Rimm EB, Stampfer MJ, Spiegelman D,
Colditz GA, Giovannucci E, Ascherio A, Willett WC. Reproducability and Validity of a Self-Administered Physical Activity Questionnaire for Male Health Professionals. Epidemiology
1996; 7(1): 81±86.
24 Klesges RC, Klesges LM, Haddock CK, Eck LH. A Longitudinal Analysis of the Impact of Dietary Intake and Physical
Activity on Weight Change in Adults. Am J Clin Nutr 1992;
55: 818±822.
25 Willett WC. Nutritional Epidemiology. Oxford University
Press: New York, 1990, pp. 261±265.
26 Kleinbaum DG, Kupper LL, Muller KE. Applied Regression
Analysis and Other Multivariable Methods (2nd edn) PWSKent: Boston, 1988.
27 SAS. SAS Proprietary Software Release 6.12, SAS Institute
Inc: Cary NC, USA.
28 Blair SN. Evidence for Success of Exercise in Weight Loss
and Control. Annals of Int Med 1993; 119(7, part 2): 702±706.
29 Williamson DF, Madans J, Anda RF, Kleinman JC, Kahn HS,
Byers T. Recreational Physical Activity and Ten Year Weight
Change in a U.S. National Cohort. Int J Obes 1993; 17(5):
279±286.
30 Ching PLYH, Willett WC, Rimm EB, Colditz GA, Gortmaker S,
Stampfer MJ. Activity Level and Risk of Overweight in Male
Health Professionals. Am J Pub Health 1996; 86(1): 25±30.
31 Kaufman L. Prime Time Nutrition. J Commun 1980; 30: 37±
46.
32 Story M, Faulkner P. The Prime Time Diet: A Content
Analysis of Eating Behavior and Food Messages in TV
Program Content and Commercials. Am J Pub Health 1990;
80: 738±740.
33 Galst JP, White MA. The Unhealthy Persuader: The Reinforcing Value of TV and Children's Purchase-In¯uencing
Attempts at the Supermarket. Child Dev 1976; 47: 1089±1096.
34 Guo SS, Roche AF, Chumlea WC, Gardner JD, Siervogel RM.
The Predictive Value of Childhood Body Mass Index Values
for Overweight at Age 35 y. Am J Clin Nutr 1994; 59: 810±
819.
35 Leibel RL, Rosenbaum M, Hirsch J. Changes in Energy
Expenditure Resulting From Altered Body Weight. New Eng
J Med 1995; 332(10): 621±628.
37 Willett WC. Is Dietary Fat a Major Determinant of Body Fat?
Am J Clin Nutr. In Press.