(Posted August 15, 1997 ? Issue 14; archived September 5, 1997)
Abstract
Do obese people extend their life span by loosing weight? The question
sounds simple. But getting a valid answer is by no means straightforward. The authors
provide a historical perspective on research in obesity and mortality. They
discusse the difficulties in determining both what is too fat and what is too
thin. Then they argue that understanding the influence of obesity and weight
loss on mortality can only come from studies designed to directly address the
question, rather than by extracting the answer from research designed for
other purposes.
Readers should note this issue's Meeting Brief, which reviews
a workshop on the genetics of obesity.
How much do you weigh? Virtually every adult in the United States can answer
this question with some accuracy. We may not know our cholesterol levels, for
whom we recently voted, or where the children are after 10 P.M., but we do
know how much we weigh. Although weight has considerable social and cosmetic
importance, our interest in the matter also comes from the belief that body
weight may be a determining factor of longevity. Most people believe that
thinner is better. As the Duchess of Windsor noted, "you can never be
too rich or too thin." But is this really true?
The assumed association between body weight, health, and longevity dates back
to antiquity. In Buddha: His Life and Teachings, an obese man told of
being advised to "exercise self-control at thy meals. In following this
advice thou wilt prolong thy life," for "he who indulges in the
satisfaction of his appetites works his own destruction." In
Shakespeare's Henry IV, Henry directs Falstaff to "leave
gormandizing; know that the grave doth gape/For thee thrice wider than for
other men." In the twentieth century, the insurance industry formalized
these perceptions. Based on experience with scores of policyholders, insurers
published data indicating that "overweight" people died earlier than
people of average weight. Tables of ideal body weight were developed in which
"ideal" was defined as that body weight associated with the greatest
longevity. Among the earliest and best was the 1959 table of ideal weight
ranges produced by the Metropolitan Life Insurance Company. For years, these
were the standards for defining the "ideal" weight. Obesity was
defined by reference to these tables. Specifically, obesity was considered as
20% above ideal weight where ideal weight was defined according to the 1959
Metropolitan Life Insurance Company table.
However, interest in the scientific study of body weight began much earlier.
In the 1800s, Francis
Galton ran a most active anthropometric laboratory. He collected
thousands of observations with apparently great precision, including the body
weights of thousands of his countrymen. During that same century, Adolphe
Quetelet, the Belgian astronomer, statistician, and epidemiologist,
developed the first widely used index of relative body weight. Quetelet
observed that, among adults, weight increased roughly in proportion to the
square of height. This remains the most widely used index of relative body
weight. It is generally expressed as weight in kilograms divided by the
square of height in meters. This index is referred to as either Quetelet's
index or body mass index (BMI). To gain perspective on this index, consider
the figure below.
A BMI of approximately 10 to 12 seems to be the lower limit capable of
sustaining adult life. Studies of hunger strikers and others who died of
starvation revealed that death generally comes when a BMI of approximately 10
to 12 is reached. Among the modern U.S. adult population, the tenth
percentile of BMI is around 19 to 21, depending on the age, race, and sex
group examined. According to most authoritative sources, obesity begins at a
BMI between 27 and 30. By this definition, approximately one-third of the
adult U.S. population is obese. Finally, the term morbid obesity is
applied to those individuals with BMIs greater than or equal to 40.
In the last 50 years, many studies have assessed the relationship between BMI
(or other indices of relative weight) and mortality. The vast majority have
observed a U- or J-shaped relationship, i.e., a plot of BMI on the x-axis and
mortality rate on the y-axis of a Cartesian plane yields a U- or J-shaped
curve. Although these data clearly show that elevated BMI, or obesity,
correlates with decreased longevity, they also show a rather puzzling and
unexpected outcome: namely that unusually low BMI, or thinness, is also
associated with decreased longevity. This non-monotonic relationship flies in
the face of the Duchess of Windsor's quotation. It also defies the desire of
many investigators and public health officials for simple conceptions.
The complexity of the BMI-mortality association was further noted in a paper
by Dr. Reuben Andres of the National Institute of
Aging published during the early 1980s. Andres reanalyzed data from the
major Build and Blood Pressure Study of the Metropolitan Life Insurance
Company. [1] For each age group
considered, the relationship between BMI and mortality was U- or J-shaped, and
the nadir of the curve increased with increasing age. Thus, he argued, the
BMI associated with minimum mortality was lower for young adults and far
higher for older adults than typically thought. This observation implied that
it is healthiest to gain weight slowly as one ages. This notion is comforting
since it is exactly what most adults do.
However, among obesity researchers the sense of comfort was short-lived. In a
1987 paper, Joanne Manson and colleagues argued that previous studies of body
weight and mortality contained serious methodological flaws that led to
erroneous conclusions about the actual relationship between BMI and mortality.
[2] The flaws they identified included
insufficient sample size, no control for cigarette smoking (as smokers tend to
die earlier and be thinner), mistakenly controlling for mediators of the BMI
mortality relationship (e.g., high blood pressure, dislipidemias, glucose
intolerance), and failing to control for preexisting occult disease associated
with thinness and short lives. This last issue is especially thorny since
occult disease, by definition, is unobservable and therefore not easily
incorporated into statistical analyses.
To overcome this difficulty Manson and colleagues recommended eliminating
subjects who die within a few years of follow-up in longitudinal cohort
studies of the BMI-mortality relationship. They reasoned that by excluding
such subjects, one is probably eliminating most people with serious diseases;
this should reduce the bias due to occult disease. Manson et al. suggested
that large studies without these design flaws would probably reveal a
monotonic increase between BMI and mortality with age.
These suggestions and conjectures were quickly adopted by the obesity and
epidemiologic communities. Two studies in particular seemed to bolster Manson
et al.'s propositions. The first was a report by Lindsted et al. studying
male Seventh Day Adventists because the lifestyle of these individuals makes
them relatively free from confounding factors. [3] Specifically, the consumption of meat, alcohol, and
tobacco is markedly reduced in this population. Lindsted et al. found in
their sample a monotonic relationship between BMI and mortality, as Manson et
al. predicted. Moreover, they observed no increase in BMI associated with
minimum mortality with aging. A second noteworthy study, by Manson and
colleagues, analyzed data from nearly 100,000 white, female nurses. [4] The authors reported that although the
relationship between BMI and mortality was initially somewhat J-shaped, it
increased monotonically after controlling for the previously mentioned
variables, and most dramatically after eliminating subjects with substantial
weight fluctuation. Based on these results, the authors concluded that BMIs
below 19 were associated with the lowest mortality. By inference, anyone with
a BMI above 19 is overweight. The Duchess of Windsor seems to have been right
after all.
The findings and conclusions of Lindsted et al., Manson et al., and others
did not go unchallenged. Most notably, in 1996 Richard Troiano et al.
published an impressive meta-analysis of the association between BMI and
mortality. [5] This meta-analysis is
the most quantitative, comprehensive, and objective study of published data
addressing this question to date. Troiano's results were quite clear.
Regardless of whether subjects dying early were excluded, or whether smoking
was controlled for, and so forth, the results showed that the relationship
between BMI and mortality was U- or J- shaped.
Moreover, recent analyses from our own group at the St. Luke's/Roosevelt
Hospital Obesity Research Center questioned some of the very logic of the
methods initially proposed by Manson et al. in 1987. Specifically, we showed
that excluding subjects who die early may not reduce the bias due to
preexisting occult disease, if such bias exists. Moreover, excluding subjects
who die early can even exacerbate that bias. These final observations suggest
that if occult disease confounds the results of BMI-mortality association
studies, then the methods currently used to control for that confounding may
have been quite inadequate and such confounding may remain. This leaves open
the question of whether the U- or J-shaped relationship is real or due to
confounders from occult disease.
There are still other complexities in this arena. The U-shaped relationship
implies that one can be too thin. Still, obesity researchers are quick to
note that obesity refers to excess adipose tissue, not excess weight per se.
Hence, while BMI is highly correlated with adiposity, it is not equivalent to
adiposity. In fact, BMI correlated well with the amount of both fat and lean
body tissue an individual has. Excess fat is the component of excess weight
that is considered the risk factor among obese individuals. With this in
mind, my colleagues and I explored the differential relationships one could
obtain between body fat and mortality and BMI and mortality. We found that
even if both the fat mass-to-mortality relationship and the BMI-to-percent
body fat increase monotonically while the lean tissue-to-mortality
relationship decreases monotonically, the relationship between BMI and
mortality can be U-shaped. Thus, the epidemiologic observation that one can
be too thin is not necessarily inconsistent with the clinical wisdom that one
can never be too lean. It points out that leanness and thinness are not
equivalent. This final analysis also suggests that future studies will need
to measure body composition, not merely body weight, to understand fully the
relationship between adiposity and mortality.
There are more puzzling data still. Some of these data come from the world of
laboratory rodents. On the one hand, the "thinner is better"
argument is supported by the studies of calorically restricted animals. In
tightly controlled experimental conditions, with laboratory rodents living in
pathogen-free environments, the variable that most powerfully and consistently
extends longevity is caloric restriction. [6] Such results make it tempting to conclude that
being very thin
might greatly extend longevity. However, even in these studies among groups of
equally restricted or ad libitum fed animals, it is not always clear that the
lighter animals live longer. Sometimes the opposite holds true. That is, for
any degree of caloric restriction, the animals with the higher body weights
tend to live longer.
A final quandary comes from the area of human weight loss. Clinical data show
that weight loss produces a reduction in morbidity. When obese people go on
medically supervised diets and lose weight, their blood pressure decreases,
their dislipidemias alleviate, and their glucose tolerance improves. [7] However, epidemiologic observations of
the association between weight loss and mortality have yielded very different
results. Specifically, in large prospective cohort studies, subjects who lose
weight tend to die earliest. These data are quite sobering when one is faced
with a patient who wants to lose weight and asks whether they are likely to
live longer if they so. While these data deserve serious attention, in many
such epidemiologic studies, the cause of weight loss is unknown. It is
unclear whether the majority or even a substantial proportion of individuals
who show weight reduction in epidemiologic studies are obese individuals,
losing weight intentionally through medically recommended procedures. A
landmark paper by David Williamson and colleagues highlighted this point. [8] Williamson's is the first study of
reasonable quality to suggest that among obese people with co-morbidities,
weight loss may extend life.
What does the future hold? There is almost universal agreement that above some
level obesity is associated with increased mortality. But above what level?
That is less clear. And are very thin people also at an increased risk of
mortality within a defined period of time? That also is unclear. Finally,
will weight loss among obese individuals increase their longevity? Again, we
can speculate, but the answers simply aren't yet in.
Not only are there several unanswered questions, but in our opinion,
we may not answer them using current strategies. Sophisticated statistical
manipulations will not make bad data good. If we want to draw solid
conclusions we must collect solid data. There are, at least, three important
changes we can make to the kind of data we collect that will hopefully yield
more solid conclusions.
First, as a field we must move beyond studying BMI and begin studying the
association of mortality with carefully measured body composition. Methods to
assess body composition have been well established and deserve to be applied.
Second, appeals to latent confounding variables such as preexisting occult
disease should yield to careful and thorough measurements of health status by
a complete physical examination at baseline. This should markedly reduce or
eliminate the confounders due to unknown diseases. Finally, the old saw that
correlation does not necessarily imply causality is as true here as anywhere.
Collecting observational data in this context must be replaced by conducting
randomized controlled clinical trials. If the specific question to be
addressed is "Do obese people who lose weight live longer?", we must
conduct controlled clinical trials in which obese people are randomized to
lose or not lose weight and observe the effects. Such studies will be
difficult, time-consuming, and expensive to conduct. But the one-third of our
population that is obese deserves no less.
David B. Allison is
Associate Research Scientist at the Obesity
Research Center at St. Luke's/Roosevelt Hospital and Assistant
Professor of Psychology in Psychiatry at the Columbia University College
of Physicians and Surgeons, New York City. Daisy N. Siemon is
currently involved in various projects at the Obesity Research Center at St.
Luke's/Roosevelt Hospital Center in New York City.
Send us your comments and ideas for future articles.
American Journal of Clinical Nutrition - full
text of recent issues and supplements, as well as journal abstracts, are available online
for free, and full online journal subscriptions are available for a fee.
Obesity Surgery - "an
international surgical journal for research and treatment of massive obesity."
Free abstracts and full-text PDFs of sample issues.
The assumed association between body weight, health, and longevity dates back
to antiquity. In Buddha: His Life and Teachings, an obese man told of
being advised to "exercise self-control at thy meals. In following this
advice thou wilt prolong thy life," for "he who indulges in the
satisfaction of his appetites works his own destruction." In
Shakespeare's Henry IV, Henry directs Falstaff to "leave
gormandizing; know that the grave doth gape/For thee thrice wider than for
other men." In the twentieth century, the insurance industry formalized
these perceptions. Based on experience with scores of policyholders, insurers
published data indicating that "overweight" people died earlier than
people of average weight. Tables of ideal body weight were developed in which
"ideal" was defined as that body weight associated with the greatest
longevity. Among the earliest and best was the 1959 table of ideal weight
ranges produced by the Metropolitan Life Insurance Company. For years, these
were the standards for defining the "ideal" weight. Obesity was
defined by reference to these tables. Specifically, obesity was considered as
20% above ideal weight where ideal weight was defined according to the 1959
Metropolitan Life Insurance Company table.
In the last 50 years, many studies have assessed the relationship between BMI
(or other indices of relative weight) and mortality. The vast majority have
observed a U- or J-shaped relationship, i.e., a plot of BMI on the x-axis and
mortality rate on the y-axis of a Cartesian plane yields a U- or J-shaped
curve. Although these data clearly show that elevated BMI, or obesity,
correlates with decreased longevity, they also show a rather puzzling and
unexpected outcome: namely that unusually low BMI, or thinness, is also
associated with decreased longevity. This non-monotonic relationship flies in
the face of the Duchess of Windsor's quotation. It also defies the desire of
many investigators and public health officials for simple conceptions.
However, among obesity researchers the sense of comfort was short-lived. In a
1987 paper, Joanne Manson and colleagues argued that previous studies of body
weight and mortality contained serious methodological flaws that led to
erroneous conclusions about the actual relationship between BMI and mortality.
[
The findings and conclusions of Lindsted et al., Manson et al., and others
did not go unchallenged. Most notably, in 1996 Richard Troiano et al.
published an impressive meta-analysis of the association between BMI and
mortality. [
There are still other complexities in this arena. The U-shaped relationship
implies that one can be too thin. Still, obesity researchers are quick to
note that obesity refers to excess adipose tissue, not excess weight per se.
Hence, while BMI is highly correlated with adiposity, it is not equivalent to
adiposity. In fact, BMI correlated well with the amount of both fat and lean
body tissue an individual has. Excess fat is the component of excess weight
that is considered the risk factor among obese individuals. With this in
mind, my colleagues and I explored the differential relationships one could
obtain between body fat and mortality and BMI and mortality. We found that
even if both the fat mass-to-mortality relationship and the BMI-to-percent
body fat increase monotonically while the lean tissue-to-mortality
relationship decreases monotonically, the relationship between BMI and
mortality can be U-shaped. Thus, the epidemiologic observation that one can
be too thin is not necessarily inconsistent with the clinical wisdom that one
can never be too lean. It points out that leanness and thinness are not
equivalent. This final analysis also suggests that future studies will need
to measure body composition, not merely body weight, to understand fully the
relationship between adiposity and mortality.
A final quandary comes from the area of human weight loss. Clinical data show
that weight loss produces a reduction in morbidity. When obese people go on
medically supervised diets and lose weight, their blood pressure decreases,
their dislipidemias alleviate, and their glucose tolerance improves. [
First, as a field we must move beyond studying BMI and begin studying the
association of mortality with carefully measured body composition. Methods to
assess body composition have been well established and deserve to be applied.
Second, appeals to latent confounding variables such as preexisting occult
disease should yield to careful and thorough measurements of health status by
a complete physical examination at baseline. This should markedly reduce or
eliminate the confounders due to unknown diseases. Finally, the old saw that
correlation does not necessarily imply causality is as true here as anywhere.
Collecting observational data in this context must be replaced by conducting
randomized controlled clinical trials. If the specific question to be
addressed is "Do obese people who lose weight live longer?", we must
conduct controlled clinical trials in which obese people are randomized to
lose or not lose weight and observe the effects. Such studies will be
difficult, time-consuming, and expensive to conduct. But the one-third of our
population that is obese deserves no less.
