Anda di halaman 1dari 8

Lifetime Weight Patterns in Male Physicians:

The Effects of Cohort and Selective Survival


Bethany B. Barone,* Jeanne M. Clark,* Nae-Yuh Wang, Lucy A. Meoni, Michael J. Klag,* and
Frederick L. Brancati*

Abstract
BARONE, BETHANY B., JEANNE M. CLARK, NAEYUH WANG, LUCY A. MEONI, MICHAEL J. KLAG,
AND FREDERICK L. BRANCATI. Lifetime weight
patterns in a cohort of male physicians: an analysis of cohort
effects and selective survival in the Johns Hopkins
Precursors Study. Obesity. 2006;14:902908.
Objective: The natural history of lifetime weight change is
not well understood because of conflicting evidence from
cross-sectional and longitudinal studies. Cross-sectional
analyses find that adult weight is highest at 60 years of
age and lower thereafter. Longitudinal analyses have not
found this pattern. Our objective was to test whether cohort
effects and selective survival may explain the differences
observed between cross-sectional and longitudinal studies.
Research Methods and Procedures: We analyzed data on
white men from the Johns Hopkins Precursors Study (n
1197). Weight and height were measured at enrollment
during medical school. The Precursors Study collected subsequent weight measurements by self-report and follows all
participants for mortality.
Results: In preliminary analyses that ignored cohort and
survival effects, average weight increased 0.16 kg/yr to age
65 (p 0.001) and declined 0.10 kg/yr thereafter (p
0.002). When controlling for differing rates of weight
change by cohort and survival group, the apparent decline
after 65 years of age was mostly explained.
Discussion: These data suggest that, in white men, weight
increases steadily until age 65 and then plateaus. These

Received for review February 1, 2005.


Accepted in final form March 7, 2006.
The costs of publication of this article were defrayed, in part, by the payment of page
charges. This article must, therefore, be hereby marked advertisement in accordance with
18 U.S.C. Section 1734 solely to indicate this fact.
*Department of Epidemiology, Department of Medicine, Department of Biostatistics, and
Department of Health Policy and Management, The Johns Hopkins Bloomberg School of
Public Health and The Johns Hopkins University School of Medicine, Baltimore, Maryland.
Address correspondence to Jeanne M. Clark, 2024 E. Monument Street, Suite 2-600,
Baltimore, MD 21287.
E-mail: jmclark@jhmi.edu
Copyright 2006 NAASO

902

OBESITY Vol. 14 No. 5 May 2006

findings emphasize the necessity of longitudinal rather than


cross-sectional data to describe lifetime weight patterns.
Key words: aging, body weight changes, cohort effect,
mortality, prospective studies

Introduction
Obesity is one of the most significant public health problems facing the United States today (13). High BMI is
associated with greater risk of cardiovascular diseases (2),
type 2 diabetes (4), cancer (5), and all-cause mortality (6,7).
Although reducing obesity was a major goal of Healthy
People 2000 (information regarding Healthy People 2000
is available at http://www.cdc.gov/nchs/about/otheract/
hp2000/hp2k.htm), the prevalence of obesity increased in
every state during the 1990s (3). By 2000, the prevalence of
overweight (BMI 25 kg/m2) in the United States had
reached 64%, and the prevalence of obesity (BMI 30
kg/m2) had reached a staggering 30% (8). Understanding
the natural history of obesity is important for developing
effective treatment and prevention programs.
Cross-sectional evidence from the National Health and
Nutritional Examination Survey (NHANES)1 as well as
other studies has suggested that the natural pattern of weight
change is an increase in weight until the seventh decade of
life followed by a decline in weight thereafter (9 11). This
pattern was also observed when NHANES repeated weight
measurements 10 years after the first weight measurement
(12). However, several longitudinal studies of weight patterns have found sustained weight gain past the seventh
decade of life as opposed to a decline (13,14). We hypothesized that the inverted U trajectory in body weight observed in cross-sectional analyses is not the true lifetime
pattern of weight change but rather the result of several
epidemiological effects. First, increases in the prevalence of
obesity over the past few decades may be creating a birth

1
Nonstandard abbreviation: NHANES, National Health and Nutritional Examination Survey.

Lifetime Weight Patterns in Male Physicians, Barone et al.

cohort effect where individuals in earlier cohorts have lower


average weight (8). Thus, a cross-sectional analysis would
show a lower weight among the elderly group relative to
younger groups. Second, the relationship between obesity
and all-cause mortality may be creating a survival effect in
which those who die earlier are heavier, on average, than
those who live longer (11). Therefore, the mean weight at
successive ages would decrease as a greater proportion of
the heavier individuals are removed from the sampling
group. A final possibility is that disease-related weight loss
causes a decrease in average weight among the elderly (15).
To our knowledge, no other analysis has looked at both
cross-sectional weight patterns and the contribution of longitudinal epidemiological effects in the same population.

Research Methods and Procedures


Setting and Sample Description
Caroline Bedell Thomas, MD, initiated the Johns Hopkins Precursors Study in 1947 to prospectively study precursors of cardiovascular disease. The study recruited a total
of 1337 Johns Hopkins medical students who graduated
between 1948 and 1964. Baseline information thought to be
relevant to cardiovascular disease was measured at the time
of enrollment by a thorough medical history and physical
exam. Baseline height was measured using a stadiometer,
and weight was measured using a balance beam scale.
Subsequent collection of body weight data occurred intermittently throughout follow-up by self-report on mailed
questionnaires (every 5 years after graduation from 1948 to
1965, annually from 1966 to 1985, and in 1988, 1989, 1993,
1998, 2001, 2002, and 2003). The Precursors Study continues today with annual questionnaires and follow-up for
disease outcomes and death, including National Death Index searches. Vital status is known for 99% of the study
population. The study has excellent retention, with response
rates of at least 85% for every 5-year follow-up interval.
Further information about the Johns Hopkins Precursors
Study can be found in more than 150 publications (16). All
study procedures were approved by the Institutional Review
Board (Johns Hopkins Medical Institution, Baltimore, MD).
The study population for this analysis includes white,
male participants of the Johns Hopkins Precursors Study
Cohort (n 1197). Women (n 121) and non-white men
(n 19) were not included because these subgroups were
too small to analyze independently. Of the remaining participants, those without a baseline height or weight measurement (n 23) were excluded. Participants with baseline
weight measurements but no subsequent reported weights
(n 38) were also excluded because we were specifically
interested in studying weight trajectories over time. Participants in the final study population had a variable number of
follow-up weight responses ranging from 1 to 25, with an
average of 15. The distribution of weight responses was the

result of a function of individual response rate, year of


enrollment in the study, and the design of weight data
collection by the Precursors Study. The year for each weight
observation was recorded as the year the individual was
measured at baseline or the year a questionnaire was returned by the participant to the Precursors Study. The final
study population consisted of 1136 (95% of the entire
cohort) white men with both baseline anthropometric measurements and at least one follow-up weight observation.
Statistical Methods
Descriptive statistics (mean, standard deviation, median),
including birth year, baseline age, baseline BMI, and number of weight observations per individual, were calculated
for the study population. All-cause mortality was also calculated. The full study population was separated by birth
year into three cohorts of approximately equal sample size,
and the same statistics were calculated for each cohort.
Average length of follow-up was calculated. The crosssectional, running mean of BMI on age using unweighted
kernal smoothing (17) was plotted for the calendar years
1970, 1985, and 2002 to establish the cross-sectional pattern
of BMI with age seen in our study population. These plots
included all members of the Johns Hopkins Precursors
Study who reported a weight in the given calendar year, and
a given individual could contribute to the plot for up to 3
calendar years. These calendar years were arbitrarily chosen
to show cross-sectional plots across different age ranges of
our entire study population.
Generalized estimating equations multiple linear regression models with robust estimation of variance were used to
test the association of age, birth cohort, and survival with
attained weight adjusted for height squared. We chose attained weight as the outcome measure and adjusted for
height squared as an independent variable in all models. We
felt that changes in weight would be easier to interpret than
changes in BMI. We fit three models. In Model 1, we used
continuous age with a linear spline at age 65, adjusting for
height squared as a predictor of attained weight. This model
represents evaluating the data without considering birth
cohort effects and survival effects. The purpose of the linear
spline was to determine whether weight declines after this
age as suggested by cross-sectional analyses (9 11). In
Model 2, we used continuous age with spline, birth cohort
(1, 2, and 3), and survival status (alive/dead) as predictors of
attained weight, adjusting for height squared. Survival status was categorized by the reported condition (alive/dead)
of an individual in December 2002. Our second model
added the main effects of cohort and survival status to the
model. In Model 3, we added birth cohort-by-age and survival status-by-age interactions. Our third model studied
whether rates of weight change differed by cohort and
survival status. Because illness preceding death may also
cause weight loss in the elderly, we repeated all models
OBESITY Vol. 14 No. 5 May 2006

903

Lifetime Weight Patterns in Male Physicians, Barone et al.

cohort effect where individuals in earlier cohorts have lower


average weight (8). Thus, a cross-sectional analysis would
show a lower weight among the elderly group relative to
younger groups. Second, the relationship between obesity
and all-cause mortality may be creating a survival effect in
which those who die earlier are heavier, on average, than
those who live longer (11). Therefore, the mean weight at
successive ages would decrease as a greater proportion of
the heavier individuals are removed from the sampling
group. A final possibility is that disease-related weight loss
causes a decrease in average weight among the elderly (15).
To our knowledge, no other analysis has looked at both
cross-sectional weight patterns and the contribution of longitudinal epidemiological effects in the same population.

Research Methods and Procedures


Setting and Sample Description
Caroline Bedell Thomas, MD, initiated the Johns Hopkins Precursors Study in 1947 to prospectively study precursors of cardiovascular disease. The study recruited a total
of 1337 Johns Hopkins medical students who graduated
between 1948 and 1964. Baseline information thought to be
relevant to cardiovascular disease was measured at the time
of enrollment by a thorough medical history and physical
exam. Baseline height was measured using a stadiometer,
and weight was measured using a balance beam scale.
Subsequent collection of body weight data occurred intermittently throughout follow-up by self-report on mailed
questionnaires (every 5 years after graduation from 1948 to
1965, annually from 1966 to 1985, and in 1988, 1989, 1993,
1998, 2001, 2002, and 2003). The Precursors Study continues today with annual questionnaires and follow-up for
disease outcomes and death, including National Death Index searches. Vital status is known for 99% of the study
population. The study has excellent retention, with response
rates of at least 85% for every 5-year follow-up interval.
Further information about the Johns Hopkins Precursors
Study can be found in more than 150 publications (16). All
study procedures were approved by the Institutional Review
Board (Johns Hopkins Medical Institution, Baltimore, MD).
The study population for this analysis includes white,
male participants of the Johns Hopkins Precursors Study
Cohort (n 1197). Women (n 121) and non-white men
(n 19) were not included because these subgroups were
too small to analyze independently. Of the remaining participants, those without a baseline height or weight measurement (n 23) were excluded. Participants with baseline
weight measurements but no subsequent reported weights
(n 38) were also excluded because we were specifically
interested in studying weight trajectories over time. Participants in the final study population had a variable number of
follow-up weight responses ranging from 1 to 25, with an
average of 15. The distribution of weight responses was the

result of a function of individual response rate, year of


enrollment in the study, and the design of weight data
collection by the Precursors Study. The year for each weight
observation was recorded as the year the individual was
measured at baseline or the year a questionnaire was returned by the participant to the Precursors Study. The final
study population consisted of 1136 (95% of the entire
cohort) white men with both baseline anthropometric measurements and at least one follow-up weight observation.
Statistical Methods
Descriptive statistics (mean, standard deviation, median),
including birth year, baseline age, baseline BMI, and number of weight observations per individual, were calculated
for the study population. All-cause mortality was also calculated. The full study population was separated by birth
year into three cohorts of approximately equal sample size,
and the same statistics were calculated for each cohort.
Average length of follow-up was calculated. The crosssectional, running mean of BMI on age using unweighted
kernal smoothing (17) was plotted for the calendar years
1970, 1985, and 2002 to establish the cross-sectional pattern
of BMI with age seen in our study population. These plots
included all members of the Johns Hopkins Precursors
Study who reported a weight in the given calendar year, and
a given individual could contribute to the plot for up to 3
calendar years. These calendar years were arbitrarily chosen
to show cross-sectional plots across different age ranges of
our entire study population.
Generalized estimating equations multiple linear regression models with robust estimation of variance were used to
test the association of age, birth cohort, and survival with
attained weight adjusted for height squared. We chose attained weight as the outcome measure and adjusted for
height squared as an independent variable in all models. We
felt that changes in weight would be easier to interpret than
changes in BMI. We fit three models. In Model 1, we used
continuous age with a linear spline at age 65, adjusting for
height squared as a predictor of attained weight. This model
represents evaluating the data without considering birth
cohort effects and survival effects. The purpose of the linear
spline was to determine whether weight declines after this
age as suggested by cross-sectional analyses (9 11). In
Model 2, we used continuous age with spline, birth cohort
(1, 2, and 3), and survival status (alive/dead) as predictors of
attained weight, adjusting for height squared. Survival status was categorized by the reported condition (alive/dead)
of an individual in December 2002. Our second model
added the main effects of cohort and survival status to the
model. In Model 3, we added birth cohort-by-age and survival status-by-age interactions. Our third model studied
whether rates of weight change differed by cohort and
survival status. Because illness preceding death may also
cause weight loss in the elderly, we repeated all models
OBESITY Vol. 14 No. 5 May 2006

903

Lifetime Weight Patterns in Male Physicians, Barone et al.

Table 1. Selected characteristics of white men in the full study population and by birth cohort in the Johns
Hopkins Precursors Study, Baltimore, MD, 1948 to 2002
Cohort

N
Median birth year*
Mean baseline age (years)*
Mean baseline BMI (kg/m2)
Median number of weight observations
Cumulative mortality by December 2002 (%)*

Overall

1902 to 1927

1928 to 1933

1934 to 1941

1136
1930
23.1 (2.6)
23.1 (2.6)
16 (6)
19

431
1925
24.7 (3.3)
23.2 (2.7)
16 (7)
34

363
1930
22.3 (1.4)
23.1 (2.5)
16 (6)
15

342
1936
21.9 (1.0)
23.1 (2.5)
15 (6)
6

* Characteristic differed significantly (p 0.05) when compared across cohorts using ANOVA.
Values are means (standard deviation).

excluding any weight data reported in the 3 years before an


individuals death. In all models, age was centered at the
median baseline age (22 years) rather than age zero to allow
for main effects to be compared at a baseline age of 22
rather than a baseline age of 0. A random effects model was
used for all analyses to allow for individual variability.
Variograms of the model residuals vs. age were generated
for a random sample of the study population to determine
whether the exchangeable correlation structure was appropriate.
The STATA 7 Statistical Package (StataCorp LP., College Station, TX) was used for data analysis.

Results
Selected characteristics of the study population are reported for all participants and by cohort (Table 1). At
enrollment, the mean age was 23 years, and the average
BMI was 23 kg/m2. By December of 2002, approximately
one fifth of the cohort was reported to have died. Median
birth year, age at enrollment, and mortality differed significantly across birth cohorts. There was no statistical difference in number of follow-up observations by cohort. Baseline BMI did not differ significantly by the number of
follow-up observations. Average length of follow-up was
42 11 (standard deviation) years.
Cross-sectional plots of BMI by age in 3 different calendar years showed that average BMI increased and then
decreased as age increased (Figure 1). The average BMI of
the study population at a given age was generally greater in
later calendar years.
In Model 1 (Table 2), we observed the expected result of
a yearly increase in average weight (0.16 kg/yr, p 0.001)
before the spline at age 65 and a yearly decrease in average
weight (0.10 kg/yr, p 0.002) thereafter. This analysis, in
which cohort and survival effects are not considered, shows
a basic longitudinal weight trajectory, the inverted U.
904

OBESITY Vol. 14 No. 5 May 2006

In Model 2 (Table 3), which added birth cohort and


survival status, the average rate of weight change did not
change from Model 1. At age 22 (where age was centered),
average weight did not differ significantly by cohort. However, average weight was 1.78 kg higher (p 0.05) at age
22 in men who died by the end of follow-up compared with
men who survived through 2002. This analysis evaluated
simple overall effects of cohort and survival group but was
not optimal because it did not allow for the cohort and
survival effects to vary with age (before and after the spline
set at age 65), as these types of effects likely would in a
trajectory.
In Model 3 (Table 4), we added age-by-birth cohort and
age-by-survival status interactions to assess whether the
survival and cohort effects were better modeled as rates of
change with age rather than just a one-time effect. In this
model, the reference group included men in the first cohort
who were alive at the end of follow-up. Their lifetime

Figure 1: Cross-sectional plot of running mean of BMI on age


using unweighted kernel smoothing in 1970, 1985, and 2002 in the
Johns Hopkins Precursors Study.

Lifetime Weight Patterns in Male Physicians, Barone et al.

Table 2. Model 1: rate of weight change by age in


the Johns Hopkins Precursors Study
Covariate
(age)

Coefficient*

95% confidence
interval

Up to age 65
After age 65

0.16 kg/yr
0.10 kg/yr

0.15,0.18
0.16,0.04

* Age is centered at the median baseline age (22 years), and all
coefficients are adjusted for other covariates in the table and
height2.
p 0.001.
p 0.01.

weight pattern was described by the coefficients seen under


age. This pattern continued to be a highly significant rate of
weight gain with age up to age 65 (0.14 kg/yr, p 0.001).
However, we no longer observed a significant decline in
weight beyond age 65. There was no significant difference
in weight at age 22 by survival status or cohort. The main
effects of cohort and survival status alone are not statistically significant in this model after adjusting for the interaction terms. However, we did identify two important interaction effects in our study population. First, we found a
cohort-age effect, in that individuals in the most recent birth
cohort had a statistically greater rate of weight gain up to
age 65 compared with the oldest birth cohort. Second, we

found a survival-age effect, in that there was a highly


significant rate of decline in average weight among men
over age 65 who died by December 2002 compared with
men alive at the end of follow-up. The main or simple
effects of cohort and survival status were attenuated in the
third model because the significant effects were rates of
change with age (the interaction terms). Model 3 better
adjusted for the cohort and survival effects and confirmed
the hypothesis that the weight decline observed in crosssectional studies after age 65 is significantly attenuated once
properly modeled survival and cohort effects are removed.
To understand how weight varies with age in the groups of
men other than the reference group for Model 3, the coefficient
of weight change for age must be combined with the interaction terms applicable to the group of interest. For example, to
understand how weight changes up to age 65 for men who are
deceased by the end of follow-up in Cohort 3, the up to age
65 coefficient (0.14 kg/yr) would be combined with the
age-by-Cohort 3 coefficient (0.05 kg/yr) and the up to age
65 deceased coefficient (0.02 kg/yr).
Similar results were observed when all models were
repeated after excluding any weight measurements in the 3
years before an individuals death (data not shown).

Discussion
The results observed in the final and best model (Model
3) showed that, contrary to the impression created by crosssectional studies such as NHANES III (9 11), body weight

Table 3. Model 2: rate of weight change by age and weight differences by cohort and survival status in the Johns
Hopkins Precursors Study
Covariate
Age
Up to age 65
After age 65
Birth cohort
1 (1902 to 1927)
2 (1928 to 1933)
3 (1934 to 1941)
Survival status
Alive on December 31, 2002
Deceased before December 31, 2002
* Age
p
p
p

Coefficient*

95% confidence
interval

0.16 kg/yr
0.10 kg/yr

0.15, 0.18
0.16, 0.04

Reference
0.05 kg
0.71 kg

1.19, 1.29
0.65, 2.07

Reference
1.78 kg

0.24, 3.32

is centered at the median baseline age (22 years), and all coefficients are adjusted for other covariates in the table and height2.
0.001.
0.01.
0.05.

OBESITY Vol. 14 No. 5 May 2006

905

Lifetime Weight Patterns in Male Physicians, Barone et al.

Table 4. Model 3: rate of weight change by age, weight differences by cohort and survival status, and
age-by-cohort and age-by-survival status interaction terms in the Johns Hopkins Precursors Study
Covariate*
Age
Up to age 65
Age 65 and after
Birth cohort
1 (1902 to 1927)
2 (1928 to 1933)
3 (1934 to 1941)
Survival status
Alive in 2002
Deceased before end of 2002
Age-by-cohort
Up to age 65cohort 1
Up to age 65cohort 2
Up to age 65cohort 3
Age 65 and aftercohort 1
Age 65 and aftercohort 2
Age 65 and aftercohort 3
Age-by-survival status
Up to age 65alive
Up to age 65deceased
Age 65 and afteralive
Age 65 and afterdeceased

Coefficient*

95% confidence interval

0.14 kg/yr
0.06 kg/yr

0.12, 0.17
0.14, 0.02

Reference
0.01 kg
0.42 kg

1.17, 1.19
1.67, 0.83

Reference
1.36 kg

0.02, 2.74

Reference
0.00 kg/yr
0.05 kg/yr
Reference
0.11 kg/yr
0.39 kg/yr
Reference
0.02 kg/yr
Reference
0.36 kg/yr

0.04, 0.03
0.01, 0.09
0.04, 0.23
0.16, 0.94

0.03, 0.07
0.53, 0.18

* Age is centered at the median baseline age (22 years), and all coefficients are adjusted for other covariates in the table and height2.
p 0.001.
p 0.05.

does not generally decrease in individuals after age 65.


Rather, the apparent decline is largely explained by two
epidemiological phenomena: 1) a birth cohort effect where
average weight increases at a greater rate in the youngest
cohort compared with older cohorts and 2) a survival effect
by which heavier individuals are removed from the cohort
as they die, and, thus, the average weight decreases over
time, especially after age 65. As was expected, the cohort
effect is observed as a significantly greater increase in
average weight among the youngest cohort up to age 65
(age-by-cohort covariate, up to age 65Cohort 3), and a
similar but non-significant trend is seen in men over age 65
in more recent cohorts (age-by-cohort covariate, age 65 and
afterCohort 2, and age 65 and afterCohort 3). The
significantly greater decrease in average weight observed in
men after age 65 who died explains much of the observed
decrease in average weight after age 65 (age-by-survival
status covariate, age 65 and after deceased). This is likely
an effect of the heaviest individuals being removed from the
906

OBESITY Vol. 14 No. 5 May 2006

group and average weight going down as a result. Diseaserelated weight loss may also contribute to this effect. Our
analysis showed the complex nature of the cohort and
survival effects in that they need to be evaluated longitudinally, rather than the cross-sectionally, to truly understand
lifetime weight patterns. Furthermore, we must adjust for
cohort and survival effects as interaction terms with aging
(age-by-cohort, age-by-survival status) and not just simple,
main effect covariates (age, cohort, survival status). Only
when using these interaction terms, as in Model 3, do we see
significant effects by both cohort and survival status, as well
as the attenuation of the weight loss effect after age 65.
The greatest strength of our analysis was the prospective
design of the Johns Hopkins Precursors Study, which has
collected information on weight and has followed all participants for mortality for more than 50 years. This rare
collection of longitudinal data is ideal for analyses of body
weight patterns over time because it allows for the assessment of selective mortality and limits variability by the use

Lifetime Weight Patterns in Male Physicians, Barone et al.

of repeated measures. Another strength of this analysis is


the fairly large, homogenous sample of white, educated
men, which reduces variability caused by race, sex, and
socioeconomic status.
Our longitudinal results are generally consistent with the
few previous studies that have been done, although none has
had such a long follow-up. The Tromso Study (13) and the
Fels Longitudinal Study (14) both observed that BMI did
not decrease with age, even among elderly white men. The
Tromso Study also found a faster rate of weight gain among
more recent cohorts. However, the Tromso Study could not
investigate the effects of selective survival or weight loss
preceding death because participation was limited to individuals attending all follow-up visits (13). Among men who
were still alive at the end of follow-up, we observed a
non-significant weight decline of 0.06 kg/yr after age 65
after adjusting for cohort and interaction terms. This result
is less than the 0.1-kg weight loss per year observed in a
cohort of elderly, independent, healthy white men (18).
While we were surprised to find a slower rate of weight loss
among elderly men in our analysis, the men who were alive
at the end of follow-up were likely to be a sample of the
healthiest individuals in the Precursors Study cohort.
Despite its strengths, several limitations of this study
warrant discussion. First, most data on body weight were
self-reported. However, these data were prospectively collected, and the excellent correlation ( 0.98) of selfreported weight to measured weight has been previously
shown in this cohort (19). A second potential for bias was
that individuals excluded because they lacked a baseline
weight observation (n 38) had a statistically higher mortality rate. However, most deaths in the excluded group
occurred in young adulthood and, thus, are not likely to be
related to obesity.
Variable follow-up (the number of weight observations
ranged from 2 to 26 per person) may also be a source of
bias. One possibility is that individuals with cyclical weight
patterns may be less likely to report a weight when at a peak
and more likely to report a weight when in a valley. Therefore, some individuals may be misclassified as having a
lower average weight.
Additionally, the benefit of a homogenous population
comes at the cost of generalization. White, educated, male
physicians may be thinner and less likely to gain weight
than the general population. This possibility is supported by
the comparatively slow rate of weight gain we observed
(14,20). However, we do not believe that this quantitative
difference would explain the pattern we found.
Despite a range in birth year of 39 years, the difference
between median birth years by cohort was 6 years. The
distribution of baseline age and attained age also differed by
cohort, mainly in that the oldest cohort has a greater range
than the two younger cohorts. Both factors may have reduced the ability to observe the main effects differences by

cohort that have been shown in other analyses (13,20 22),


although we did observe a significant age-by-cohort interaction effect.
Increasing longevity may be another source of bias in this
analysis because we know that later cohorts are heavier but
life expectancy has increased. While the cohort effects we
adjusted for may incorporate this effect, a more complex
model (e.g., including a different spline point for each
cohort) would be needed to take into account lifespan increases with time. Furthermore, increasing longevity by
cohort may also have prevented us from observing a more
distinct cohort effect.
Finally, information on voluntary vs. involuntary diseaserelated weight loss, body composition, and body fat distribution was not available for this analysis. We did attempt to
remove the effects of unintentional weight loss immediately
preceding death by excluding weight measurements within
3 years of death. The results of this analysis were similar,
suggesting that our inferences about cohort and survival
effects were not distorted by weight loss immediately before
death.
Our findings have several implications for future research. The first is that, even among affluent, well-educated,
medically sophisticated men, BMI continues to increase
across the lifespan at least to age 65 without evidence of
significant decline in healthy adults thereafter. Furthermore,
BMI increases at a greater rate in successive cohorts, and
the decline of average weight in the older ages seems to be
a survival effect. Therefore, we should not be deceived by
cross-sectional studies that suggest that weight naturally
declines after age 65 in individual adults, because this
phenomenon is explained mostly by cohort effects and
selective survival. This knowledge offers the opportunity to
continue efforts and counseling for weight management
across the healthy adult lifetime. The effects of voluntary
and involuntary disease-related weight loss on expected
lifetime weight trajectories among healthy individuals, the
role of body composition and body fat distribution in longitudinal weight patterns with aging, and the potential benefit of voluntary weight loss in the elderly await further
research.

Acknowledgments
This work was supported by NIH Grants AG01760 and
DK02856. F.B. was supported by National Institute of Diabetes & Digestive & Kidney Diseases Grant K24
DK62222-01. The authors thank Drs. Brad Astor, Greta
Bunin, and Lawrence Cheskin for helpful suggestions on
earlier versions of the manuscript.
References
1. Visscher TLS, Seidell JC. The public health impact of obesity. Annu Rev Public Health. 2001;22:35575.
OBESITY Vol. 14 No. 5 May 2006

907

Lifetime Weight Patterns in Male Physicians, Barone et al.

2. Must A, Spadano J, Coakley EH, Field AE, Colditz G,


Dietz WH. The disease burden associated with overweight
and obesity. JAMA. 1999;282:15239.
3. Nelson DE, Bland S, Powell-Griner E, Klein R, Wells HE,
Hogelin G, Marks JS. State trends in health risk factors and
receipt of clinical preventive services among US adults during
the 1990s. JAMA. 2002;287:2659 67.
4. Brancati FL, Wang NY, Mead LA, Liang KY, Klag MJ.
Body weight patterns from 20 to 49 years of age and subsequent risk for diabetes mellitus. Arch Intern Med. 1999;159:
957 63.
5. Calle EE, Rodriguez C, Walker-Thurmond K, Thun MJ.
Overweight, obesity, and mortality from cancer in a prospectively studied cohort of U.S. adults. N Engl J Med. 2003;348:
162538.
6. Calle EE, Thun MJ, Petrelli JM, Rodriguez C, Heath CW
Jr. Body mass index and mortality in a prospective cohort of
U.S. adults. N Engl J Med. 1999;341:1097105.
7. Peeters A, Barendregt JJ, Willekens F, Mackenbach JP, Al
Mamun A, Bonneux L. Obesity in adulthood and its consequences for life expectancy: a life-table analysis. Ann Intern
Med. 2003;138:24 32.
8. Flegal KM, Carroll MD, Ogden CL, Johnson CL. Prevalence and trends in obesity among US adults, 1999 2000.
JAMA. 2002;288:17237.
9. National Health and Nutritional Examination. Published
Tables.
http://www.cdc.gov/nchs/about/major/nhanes/data
tblelink.htm (Accessed May 2, 2006).
10. Fogelholm M, Kujala U, Kaprio J, Sarna S. Predictors of
weight change in middle-aged and old men. Obes Res. 2000;
5:36773.
11. Seidell JC, Visscher TLS. Body weight and weight change
and their health implications for the elderly. Eur J Clin Nutr.
2000;54:S339.

908

OBESITY Vol. 14 No. 5 May 2006

12. Williamson DF. Descriptive epidemiology of body weight


and weight change in U.S. adults. Ann Intern Med. 1993;119:
646 9.
13. Jacobsen BK, Njolstad I, Thune I, Wilsgaard T, Lochen
ML, Schirmer H. Increase in weight in all birth cohorts in a
general population. Arch Intern Med. 2001;161:466 72.
14. Guo SS, Zeller C, Chumlea WC, Siervogel RM. Aging,
body composition, and lifestyle: the Fels Longitudinal Study.
Am J Clin Nutr. 1999;70:40511.
15. Losonczy KG, Harris TB, Cornoni-Huntley J, Simonsick
EM, Wallace RB, Cook NR, Ostfeld AM, Blazer DG. Does
weight loss from middle age to old age explain the inverse
weight mortality relation in old age? Am J Epidemiol. 1995;
141:31221.
16. Thomas CB. Observations on some possible precursors of
essential hypertension and coronary artery disease. Bull Johns
Hopkins Hosp. 1951;89:419 41.
17. Stone CJ. Consistent nonparametric regression. Ann Stat.
1977;5:595 625.
18. Chumlea WC, Garry PJ, Hunt WC, Rhyne RL. Distributions of serial changes in stature and weight in a healthy
elderly population. Human Biol. 1988;60:91725.
19. Klag MJ, He J, Mead LA, Ford DE, Pearson TA, Levine
DM. Validity of physicians self-reports of cardiovascular
disease risk factors. Ann Epidemiol. 1993;3:4427.
20. Lewis CE, Jacobs DR Jr, McCreath H, et al. Weight gain
continues in the 1990s: 10-year trends in weight and overweight from the CARDIA Study. Am J Epidemiol. 2000;151:
1172 81.
21. McTigue KM, Garrett JM, Popkin BM. The natural history
of the development of obesity in a cohort of young U.S. adults
between 1981 and 1998. Ann Intern Med. 2002;136:857 64.
22. Lahti-Koski M, Jousilahti P, Pietinen P. Secular trends in
body mass index by birth cohort in eastern Finland from 1972
to 1997. Int J Obes Relat Metab Disord. 2001;25:72734.