University of Southern California Dissertations and Theses

The effect of body weight on mortality: different countries and age groups

(USC Thesis Other)

The effect of body weight on mortality: different countries and age groups

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

THE EFFECT OF BODY WEIGHT ON MORTALITY:
DIFFERENT COUNTRIES AND AGE GROUPS

by

Jihye Yeom
_____________________________________________________________

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(GERONTOLOGY)

May 2009

Copyright 2009 Jihye Yeom
ii

Dedication

Dedicated to my father,

Yeom, Hung-Kyu ( 염흥규: 1935-2001)

iii
Acknowledgements
I would like to express my deep and sincere gratitude to all those who
have made this dissertation possible. First of all, I cannot help but thank my
committee members: Dr. Eileen Crimmins, Dr. Kate Wilber, and Dr. Timothy
Biblarz. I am grateful to each and every one of my committee members for
believing in me and for their generosity and mentorship. I am honored to have
had the opportunity to learn from such remarkable scholars. I specially
appreciate Dr. Eileen Crimmins, my advisor for this dissertation, whose
intellectual guidance and insightful comments have been most valuable
throughout writing my dissertation. She taught me how to develop a thesis
and analyze data through gerontological eyes.
I would also like to extend my special thanks to my friends. Jung Ki
Kim always exposed me to advanced statistical methods and analyses. Aaron
Hagedorn, my co-worker, encouraged me to finish my dissertation. They have
provided me tremendous support and confidence during my years of study at
the University of Southern California. There are many others who merit my
equal gratitude in Los Angeles, namely at the Buddhist Society, JTS. Special
thanks to Myung K. Lee, a member of JTS, for her continued support since I
got to Los Angeles in 2001.
I despair of adequately acknowledging the devotion of my family. First
of all, I appreciate my parents-in-law for always praying for me. I regret that

iv
my mother-in-law’s passing in the spring of 2007, still wishing to see me
complete my Ph.D.
I cannot find the words to express my heartfelt gratitude for my parents,
who have shaped and made me who I am, encouraged me to complete my
studies, and always instilled me with invaluable confidence. I regret that my
father’s passing right after I started Ph.D. program in 2001. Even though he
had never received a higher education, he was the exemplar of the wisest
father and he taught me how to live with other in harmony. Also, I would like
to thank my sister Jisook, her husband Hakjoon, and niece Soyoung for their
encouragement and continuing support in my work.
Last but not least, this acknowledgement would not be complete unless
I express my deepest gratitude to my two years old baby daughter, Yeonwoo
and my husband, Chan. My daughter, Yeonwoo, literally means lotus in the
rain, and she is the joyful source of my inspiration as well as the crucial
motivation of my study. As a philosopher, Chan is my intellectual comrade
with whom I have pondered many gerontological questions. I am deeply
grateful my husband for his true love and endless support in every possible
way. Without him, this would not be possible.

v
Table of Contents

Dedication

ii
Acknowledgements

iii
List of Tables

vi
List of Figures

x
Abstract

xi
Chapter I: Introduction

1
Chapter II: Literature Review

8
Chapter III: Data and Methods

24
Chapter IV: Characteristics of Samples

55
Chapter V: Factors Affecting Body Weight and the Correlates of It
With Health Outcomes in Two Different Countries

72
Chapter VI: The Effect of Body Mass Index on Mortality in Different
Countries and Age Groups

89
Chapter VII: The Effect of Body Mass Index on Mortality Through
Biomarkers, Nutritional Factors, and Health Behaviors
in the NHANES III

111
Chapter VIII: Conclusions and Discussion

175
Bibliography

191

vi
List of Tables

Table 2.1: The Prevalence of Overweight and Obesity in the U.S. by
Gender (Age-adjusted 20-74y)

9

Table 2.2:

The Prevalence of Overweight and Obesity in Japan by
Gender (Age-adjusted 20-74y)

9
Table 2.3:

The Prevalence of Overweight and Obesity in the U.S. by
Gender and Age Group in 2003-2004

11
Table 2.4:

The Prevalence of Overweight and Obesity in Japan by
Gender and Age in 2001

11
Table 3.1: Study Variables and Definition of the NUJLSOA and the
LSOA II in Chapter V

36
Table 3.2: Study Variables and Definition of the NUJLSOA, the LSOA
II, and HRS in Chapter VI

41
Table 3.3: Age, Gender, Weight, Death, and Biomarker Variables and
Definition for the NHANES III in Chapter VII

45
Table 3.4: Indicators of Diets and Health Behavioral Variables and
Definition for the NHANES III in Chapter VII

46
Table 4.1: Characteristics of Older Adults in Japan and the United
States (70+): Percents and Means (SD)

56
Table 4.2: Characteristics of the Three Samples: Percents and
Means (SD)

60
Table 4.3: Characteristics of the Three Samples

for Never Smokers:
Percents and Means (SD)

65
Table 4.4: Percentage of Population 40 and Older at High Risk for
Biological Markers by WHO BMI Category in the NHANES
III Controlling for Age and Gender

67
Table 4.5: Percentage of Population 40 and Older at High Risk for
Behavioral Factors by WHO BMI Category in the NHANES
III Controlling for Age and Gender
69

vii

Table 5.1:

Odds Ratios From Multinomial Logistic Models of Three
BMI Levels for Males Including Demographic and
Socioeconomic Factors

73
Table 5.2: Odds Ratios (95% Confidence Intervals) From Multinomial
Logistic Models of Three BMI Levels for Males Adding
Health Behaviors in Model1

74
Table 5.3: The Overall Model Fit Test for Japanese and American
Men

75
Table 5.4: Odds Ratios (95% Confidence Intervals) From Multinomial
Logistic Models of Three BMI Levels for Females Including
Demographic and Socioeconomic Factors

78
Table 5.5: Odds Ratios (95% Confidence Intervals) From Multinomial
Logistic Models of Three BMI Levels for Females Adding
Health Behaviors in Model

79
Table 5.6: The Overall Model Fit Test for Japanese and American
Women

80
Table 5.7: Odds Ratios (95% Confidence Intervals) for Effect of
Underweight (BMI<18.5) and Overweight (BMI>=25.0) on
Health Outcomes Among Japanese and American Older
Men 70 and Over

82
Table 5.8: Odds Ratios (95% Confidence Intervals) for Effect of
Underweight (BMI<18.5) and Overweight (BMI>=25.0) on
Health Outcomes Among Japanese and American Older
Women 70 and over

84
Table 6.1: Odds Ratios and 95% Confidence Intervals (C.I.) of World
Health Organization (WHO) BMI Category on the
Likelihood of Dying on Three Samples

91
Table 6.2: Odds Ratios and 95% Confidence Intervals (C.I.) of World
Health Organization (WHO) BMI Category on the
Likelihood of Dying on Three Samples

Controlling for
Health Status and Behaviors

93

viii

Table 6.3:

Odds Ratios and Confidence Intervals (C.I.) of Behavior
and Health on the Likelihood of Dying on Three Samples

for Underweight People

98
Table 6.4: Odds Ratios and Confidence Intervals (C.I.) of Behavior
and Health on the Likelihood of Dying on Three Samples

for Overweight or Obese People

100
Table 6.5: Odds Ratios and Confidence Intervals (C.I.) of World
Health Organization (WHO) BMI Category

on the
Likelihood of Dying on Three Samples

of Only Never
Smokers

104
Table 6.6: Odds Ratios and Confidence Intervals (C.I.) of World
Health Organization (WHO) BMI Category on the
Likelihood of Dying on Three Samples

Controlling for
Behavior and Health for Only Never Smokers

105
Table 7.1: Percentage of Population 40 and Older at High Risk for
Biological Markers by Three BMI Categories in the
NHANES III Controlling Age and Gender

112
Table 7.2: Percentage of Population 40 and Older at High Risk for
Behavioral Factors by Three BMI Categories in the
NHANES III Controlling for Age and Gender

114
Table 7.3: Multivariate Analyses Using Cox Proportional Hazard
Models Indicating Effect of Low BMI for People with Low
and Normal Weight Aged 40 and Older: All Cause
Mortality

116
Table 7.4: Multivariate Analyses Using Cox Proportional Hazard
Models Indicating Effect of High BMI for People With
Normal or Overweight Aged 40 and Older: All Cause
Mortality

120
Table 7.5: Multivariate Analyses Using Cox Proportional Hazard
Models Indicating Effect of for Low BMI People With Low
or Normal Weight Aged 40 to 64: All Cause Mortality

124

ix

Table 7.6:

Multivariate Analyses Using Cox Proportional Hazard
Models Indicating Effect of Low BMI People With Low or
Normal Weight Aged 65 and Older: All Cause Mortality

127
Table 7.7: Multivariate Analyses Using Cox Proportional Hazard
Models Indicating Effect of High BMI for People With
Normal or Overweight Aged 40 to 64: All Cause Mortality

130
Table 7.8: Multivariate Analyses Using Cox Proportional Hazard
Models Indicating Effect of High BMI for People With
Normal or Overweight Aged 65 and Older: All Cause
Mortality

132
Table 7.9: Odds Ratios From Multinomial Logistic Models of Specific
Cause of Death Limiting the Sample to Those Who are 40
Years and Older: Cancer

137
Table 7.10: Odds Ratios From Multinomial Logistic Models of Specific
Cause of Death Limiting the Sample to Those Who are 40
to 64 Years Old: Cancer

143
Table 7.11: Odds Ratios From Multinomial Logistic Models of Specific
Cause of Death Limiting the Sample to Those Who are 65
Years and Older: Cancer

149
Table 7.12: Odds Ratios From Multinomial Logistic Models of Specific
Cause of Death Limiting the Sample to Those Who are 40
Years and Older: Cardiovascular Disease

155
Table 7.13: Odds Ratios From Multinomial Logistic Models of Specific
Cause of Death Limiting the Sample to Those Who are 40
to 64 Years Old: Cardiovascular Disease

161
Table 7.14: Odds Ratios From Multinomial Logistic Models of Specific
Cause of Death Limiting the Sample to Those Who are 65
Years and Older: Cardiovascular Disease

165
Table 8.1: Summary of Datasets and Statistical Analyses by Chapter

175

x
List of Figures

Figure 3.1:

Total Sample Construction (NUJLSOA, 65+)

27

Figure 3.2:

Selected Sample for Analyses in Chapter V and VI
(NUJLSOA, 70+)

27
Figure 3.3:

Sample Construction (LSOAII)

29
Figure 3.4:

Sample Construction (HRS)

31
Figure 3.5:

Sample Construction (NHANES III)

34
Figure 4.1:

BMI in NUJLSOA, LSOA II, and HRS

59
Figure 4.2:

Percent Mortality by WHO BMI Category

61
Figure 4.3:

Percent Mortality by WHO BMI Category for Never
Smokers

63
Figure 7.1:

Proportion by Causes of Death in Age Group

135
xi
Abstract
The goals of this dissertation were to examine differences in Body mass
index (BMI) distribution among Japanese older adults, American older adults,
and the U.S. middle-aged; to investigate which factors predict being
underweight, overweight or obese among Japanese older adults and
American older adults and how weight is associated with chronic diseases and
ADL functioning difficulties; to determine how BMI is associated with all-cause
mortality in Japanese older adults and the U.S. middle-aged and older adults;
to investigate how different levels of BMI are related to all cause mortality and
mortality from specific-causes considering the role of biomarkers indicating
physiological risk, nutrition, and health behaviors.
I found that Japanese older adults (the Nihon University Japanese
Longitudinal Study of Aging) have a lower weight distribution with less
dispersion than both American middle-aged (the Health and Retirement Study)
and older adults (the Longitudinal Study on Aging). Second, education and
income did not have an effect on BMI for older men in both countries.
However, as years of education increased, both American and Japanese older
women are less likely to be overweight. Also, it is interesting that underweight
men and women are more likely to have ADL difficulties in the U.S. Third, for
total Japanese older adults, being underweight has greater risk of death than
being normal weight. For total American older adults, being overweight was
associated with a reduced likelihood of dying although most of the indicators of
xii
poor health and behaviors are related to being overweight or obese. For
American middle-aged persons, overweight or obese people are less likely to
die than normal weight. Last, I found that overweight people had significantly
higher levels of risk in many biomarkers than normal weight people (the Third
National Health and Nutrition Examination Interview Survey). However,
underweight people have significantly worse nutritional status and health
behaviors than normal weight people. Underweight people who are 65 years
and older have an increased hazard of death from all causes and an
increased odds of death from cardiovascular disease than normal weight
people.
1
Chapter I: Introduction
1.1 Background
An epidemic of obesity is occurring all over the world. Substantial
literature has generally linked higher weight to poorer health outcomes
including illnesses and functional limitations. The increase in obesity is likely
to adversely affect health status in the near future in all countries and across
the age range (Himes, 2000; Must et al., 1999; Tsugane, Sasaki, & Tsubono,
2002; Song & Sung, 2001).
The purpose of this study is to investigate how body weight is
associated with mortality in different countries and at different ages. This
study also attempts to explore the mechanisms through which weight affects
mortality in different age groups.
Until recently there was general acceptance of the negative effect of
weight on health outcomes. However, the relation of obesity and mortality is
currently less clear because of inconsistent findings in recent research. While
some studies have found that higher body mass index (BMI) is linked to
premature mortality among the middle-aged, particularly death from
cardiovascular disease (Diehr et al., 1998; Seidell, Verschuren, van Leer, &
Kromhout, 1996); others have shown that there is no linear relation of BMI to
mortality among the elderly (Baik et al., 2000; Grabowski & Ellis, 2001) or
have shown that the association between BMI and mortality attenuates in later
life (Bender, Jockel, Trautner, Spraul, & Berger, 1999; Stevens et al., 1998).
2
In fact, recent studies appear to find that low weight is associated with
an even higher risk of mortality than normal weight (Allison, Gallagher, Heo,
Pi-Sunyer, & Heymsfield, 1997; Flegal, Graubard, Williamson, & Gail, 2005).
The unexpected relationship of higher mortality associated with low weight has
been found in more than one country. For example, among the middle-aged
in Japan and Korea, people with lower weight have been shown to have a
significantly higher risk of death than normal weight people (Hayashi et al.,
2005; Kuriyama et al., 2004; Song & Sung, 2001; Tsugane et al., 2002).
These findings call into question what we consider acceptable guidelines for
healthy weight (Willett, 1999). Given these inconsistent findings, it is
interesting to further examine the association of BMI with mortality in different
cultural settings to see how universal this is and what factors might explain
this. Asian countries, which tend to have a different weight distribution and
different diet from western countries, provide an especially salient comparison
group.
In addition, in order to reveal the mechanisms linking BMI across its
range and mortality, it is crucial to take into account additional markers of
physiological status which are potential indicators of the mechanisms by which
weight affects mortality (Manson, Stampfer, Hennekens, & Willett, 1987). Not
only overweight but also underweight is related to many other biomarkers
including blood pressure, lipid levels, and markers of inflammation (Bostom et
al., 1999; Reuben et al., 2002). These are all risk factors for poor health
3
outcomes. Nevertheless, few studies have attempted to disentangle the
biological mechanisms through which BMI might affect mortality.
The first purpose of this study is therefore to examine BMI distributions
for different countries and age groups to see how they differ. Thus, we
explore differences in the factors which predict being overweight or obese,
normal weight, and underweight in Japanese older adults and among older
adults in the U.S. For this, analyses are performed separately for men and
women because of evidence for sex differences in both BMI levels and
correlates of BMI. Then, the study will investigate the link between BMI levels
and mortality in those countries and age groups. In addition to mortality, BMI
will be related to a set of demographic, social, health behavioral indicators,
and biomarkers. Therefore, this study further examines which social,
behavioral, and biological factors are mediators of the effect of BMI on
mortality to uncover the pathways by which BMI affects mortality.
1.2 Importance and Contributions of Studying the Effect of BMI on Mortality
The importance of this study derives from the fact that obesity is a
worldwide problem. While the prevalence of obesity is lower in Asian than
western countries due to different diets, both Asian countries and the U.S.
have shown an increased prevalence of overweight and obesity over the past
decade (Flegal, Carroll, Ogden, & Johnson, 2002; Yoshiike, Kaneda, &
Takimoto, 2002; Yoshiike, Seino, et al., 2002). However, the rate of change
differs by countries and ages. It is important for this study to provide current
4
BMI levels in Japan and the U.S. for different age groups in order to establish
a foundation for studies of the BMI-mortality association.
Second, the association of weight, particularly low weight, with high
mortality may not be true across age groups and countries. If the pattern of
the relationship between BMI and mortality were the same in different
countries and age groups with very different weight distributions, this would
indicate a finding that is more universal and not specific to the observed
cohorts in specific countries. In addition, since Asian people are less likely to
be overweight, but more likely to be underweight and to have higher body fat
than Caucasians, and older adults are more likely to have higher weight than
younger adults, investigating the pattern of BMI and mortality in different
countries and age groups will contribute to the understanding of how BMI
affects mortality in various populations.
The third contribution of this study is to build a conceptual framework
that integrates demographic, social, behavioral, psychological, and biological
approaches as it attempts to reveal the true causal paths of BMI on mortality in
different age groups by considering mediators such as biomarkers and
diseases and confounders like smoking, drinking, and other health behaviors.
From this conceptual framework, this study will attempt to uncover the
underlying mechanisms through which BMI affects overall mortality and
specific causes of death. Therefore, this study contributes to our
understanding of which levels of BMI affect mortality and furthermore, how
5
BMI is linked to mortality by presenting a number of pathways from BMI to
mortality in different ages. This will be a foundation of theoretical development
in the BMI-mortality association for future research.
In addition to providing pathways linking BMI and mortality, this study
contributes to information for health practitioners including physicians,
dietitians, and nursing home directors, to better understand appropriate weight
control among older adults as well as young adults. It is expected that the
findings will be clinically relevant for health practitioners in linking weight to
risks of mortality and other biomarkers and will be applicable to patients.
1.3 Research Questions
Because of inconsistent findings and controversy regarding the
relationship between BMI and cause-specific mortality as well as overall
mortality, this study seeks to answer the following research questions.
(1) How does BMI vary in different countries and age groups? Is the
pattern of BMI distribution among Japanese older adults the same as
that of American older adults? Is the distribution of BMI for older adults
the same as for younger adults in the U.S.? If it is not, how different are
the distributions by country and age?
(2) What factors determine being overweight, obese, or underweight
among Japanese older adults, and U.S. older adults? Are determinants
of weight the same in Japan and the U.S. by sex? How are the factors
affecting weight different between Japanese and American individuals
6
of the same sex? Are underweight and overweight associated with
health outcomes differently?
(3) How is BMI associated with all-cause mortality in different countries and
age groups? What is the optimal BMI, relative to mortality, in Japanese
older adults and the U.S. middle-aged and older adults? Are the
associations of BMI categories with all-cause mortality the same in
those countries and at all adult ages?
(4) What is the relationship of BMI to other risk factors? Does the inclusion
of biological risk for thin and overweight people differ from that of
persons with normal weight? Is BMI related to specific-cause mortality
as well as all-cause mortality the same way when other biomarkers
known to be risk factors for mortality are controlled? When other risks
such as smoking, exercise, and diet are introduced, is this relationship
between BMI and mortality affected differently at different ages?
1.4 Overview of Remaining Chapters
This study begins with a description of trends in BMI in the U.S. and
Japan by age and sex over past decades. Chapter II will discuss the
determinants of BMI in different countries and among various age groups.
Then, an overview of the association of BMI and mortality will be provided,
followed by a review of the literature linking biomarkers with BMI.
Confounders in the association between BMI and mortality will be reviewed to

7
clarify how those confounders play a role in the mechanisms through which
weight affects mortality.
Chapter III describes the research design, sample selection process of
each dataset, measures, and analysis plans used in this study. Chapter IV
describes BMI distributions in Japanese older adults and U.S. middle-aged
and older adults, as well as the characteristics of the samples in greater detail.
Chapter V, VI, and VII report the results of this study. Chapter V
explores what factors are the determinants of being overweight or obese,
normal weight, or underweight in different countries including Japan and the
U.S. In addition, this study examines how those BMI levels are associated
with health outcomes such as diabetes, heart disease, stroke, and so on. In
chapter VI, the effect of BMI on all-cause mortality, controlling for variables
related to weight, is examined in Japanese adults aged 70 and over, U.S.
adults aged 70 and over, and U.S. adults aged 51 to 61. Chapter VII focuses
on whether biomarkers, health behaviors, and diet known to be confounders
for both underweight and obesity play a role as mediators when low and high
BMI are related to all-cause mortality and cause-specific mortality in adults in
three age groupings: 40 and older, 40 to 64, and 65 and older.
Chapter VIII summarizes and discusses the findings of this study.
Furthermore, limitations of the study as well as future directions for research in
this area are suggested.

8
Chapter II: Literature Review
2.1 Body Mass Index by Country and Age Group
2.1.1 Prevalence and Trends of Overweight and Obesity in the U.S. and Japan
The distribution of the body mass index (BMI) differs across countries
and age groups. The prevalence of obesity is lower in Asian than western
countries, which may be largely due to different diets. However, in recent
decades, BMI has increased all over the world.
Table 2.1 shows that two-thirds of adults in the U.S. were overweight or
obese in 2003-2004 (Ogden et al., 2006) although the prevalence of
overweight has little increased from 1960s to 1990s (Flegal, Carroll,
Kuczmarski, & Johnson, 1998). However, for men and women aged 20 to 74
years old in the U.S., the prevalence of obesity increased dramatically from
the 1960s through 2003-2004 (Flegal et al., 1998). The prevalence of obesity
among U.S. men significantly increased from 12% in 1976-1980, to 20% in
1988-1994 to 28% in 1999-2000 and 31% in 2003-2004. During the period
between 1976-1980 and 1999-2000, obesity increased from 17% to 33%
among women in the U.S. (Flegal et al., 2002; Ogden et al., 2006).
Asian countries also experienced increasing weight although the
prevalence of obesity in Asian countries is much lower than in western
countries. The number of those who are above the normal range of weight
has increased recently in Asian countries, too. Table 2.2 shows that in Japan,

9
the prevalence of obesity was 2.9% and 3.4% for men and women
respectively in 2001.

Table 2.1
The Prevalence of Overweight and Obesity in the U.S. by Gender (Age-adjusted 20-74y)

Overweight
(25.0 ≤ BMI ≤ 29.9)
Obesity
(BMI ≥ 30.0)
Year All Men Women All Men Women
1960-1962 30.5 37.8 23.6 12.8 10.4 15.0
1971-1974 32.0 41.1 23.6 14.1 11.8 16.2
1976-1980 31.5 39.1 24.3 14.5 12.3 16.5
1988-1994 32.0 39.4 24.7 22.5 20.0 24.9
1999-2000
a
34.0 39.7 28.5 30.5 27.5 33.4
2003-2004
a
34.1 39.7 28.6 32.2 31.1 33.2

Note: Combined table from different sources (Flegal et al., 1998;
a
Ogden et al., 2006)

Table 2.2

The Prevalence of Overweight and Obesity in Japan by Gender (Age-adjusted 20-74y)

Overweight
(25.0 ≤ BMI ≤ 29.9)
Obesity
(BMI ≥ 30.0)
Year All Men Women All Men Women
1976-1980 15.2 14.5 15.7 1.7 0.8 2.3
1981-1985 15.9 16.5 15.5 1.8 1.1 2.3
1986-1990 16.1 18.0 14.6 1.9 1.5 2.2
1991-1995 17.3 20.5 14.7 2.2 2.0 2.3
2001
a
21.2 25.1 18.2 3.2 2.9 3.4

Note: Combined table from different sources (
a
Sakamoto, 2006; Yoshiike, Seino, et al., 2002).

Compared to statistics from 1991-1995, obesity increased by 1% in
both sexes. While obesity is low, almost one-fourth of Japanese men and
one-fifth of Japanese women have weight higher than the normal range of
weight in 2001. This results from the remarkable increase in overweight over
recent decades, particularly for men (Yoshiike, Kaneda, et al., 2002; Yoshiike,
Seino, et al., 2002). In both Japan and the U.S., while the prevalence of
10
overweight is higher for men than for women, the prevalence of women being
obese is higher than that of men.
The increased prevalence in those who are overweight appears even
greater in Korea. Almost one-third of Korean adults have weight above the
normal range in 2001 (Kim, Ahn, & Nam, 2005). Although the prevalence of
overweight among older adults (65 or older) in Korea is lower (13.7%, 20.2%
for men and women, respectively) than that of U.S. counterparts, it also has
increased recently. Since the mid 1990’s, the dramatic increase in the
prevalence of overweight is thought to be due to shifts in diet from
carbohydrates to animal fat in Korea (Kim et al., 2005). Japan also has
experienced these types of changes in diet since the 1970s (Sakamoto, 2006).
2.1.2 Prevalence of Overweight and Obesity in the U.S. and Japan by Age
Group
In 2003-2004, almost two-thirds of adults aged 20 to 39 years, and
three-fourths of persons aged 40 and older in the U.S. were above the normal
range of weight (see Table 2.3) (Ogden et al., 2006). As Americans get older,
they tend to have a higher prevalence in overweight, but the highest
prevalence of obesity in both men and women occurs among middle-aged
Americans.
The prevalence of obesity in Japan is about 10 times less than that in
the U.S. at each age (see Table 2.4). In Japan, the highest prevalence of
obesity is among the oldest of the three age groups. Overall, the percent of
11
overweight increases with age in Japan, but this is not true for men who are
more likely to be overweight in 40-59 age group.

Table 2.3
The Prevalence of Overweight and Obesity in the U.S. by Gender and Age Group in 2003-2004
Overweight
(25.0 ≤ BMI ≤ 29.9)
Obesity
(BMI ≥ 30.0)
Age All Men Women All Men Women
20-39 28.6 34.2 22.8 28.5 28.0 28.9
40-59 36.3 43.4 29.3 36.8 34.8 38.8
≥ 60 y 40.0 43.3 37.4 31.0 30.4 31.5

Note: This table was made by recalculating prevalence from different tables of Ogden et al. (2006).

Table 2.4
The Prevalence of Overweight and Obesity in Japan by Gender and Age in 2001
Overweight
(25.0 ≤ BMI ≤ 29.9)
Obesity
(BMI ≥ 30.0)
Age All Men Women All Men Women
20-39 14.2 20.8 9.1 2.9 3.7 2.2
40-59 22.8 28.9 17.9 3.3 2.9 3.5
≥ 60 y 24.4 22.9 25.6 3.9 3.6 4.1

Note: This table was made by recalculating prevalence from a table of Sakamoto (2006).

2.2 Determinants of Body Mass Index
While the prevalence of overweight and obesity has increased during
the last few decades all over the world, there have been differences in the
increase of obesity by country. Weight has a strong link to cultural factors
such as diet and life style, which differ remarkably between the U.S. and
Japan. While the U.S. and Japan are the most affluent countries, they differ in
terms of health and life expectancy for the elderly. Japan has both the highest
12
life expectancy and active life expectancy in the world (Crimmins, Saito, &
Ingegneri, 1997; World Health Organization [WHO], 2007). Meanwhile, the
U.S. is not even listed among the top 10 countries in either average life
expectancy or active life expectancy (WHO, 2007). Determining what factors
are associated with body weight affecting poor health and how those are
different in Japan and the U.S. will give us a clue to weight-control for
Japanese and Americans.
2.2.1 Link Between Body Mass Index and Health Outcomes
Being overweight and obese is linked to greater risk for a number of
diseases including ischemic heart disease, strokes, diabetes, hypertension,
atherosclerosis, some types of cancer (National Heart, Lung, and Blood
Institute, 1998), and osteoarthritis (Felson, Zhang, Anthony, Naimark, &
Anderson, 1992) in the U.S. In particular, obesity is closely related to
cardiovascular disease risk factors such as high C-reactive protein, high
interleukin6, hypertension, and low level of HDL (high-density lipoprotein)
cholesterol (Escobar-Morreale, Villuendas, Botella-Carretero, Sancho, & San
Millan, 2003; Soteriades et al., 2005). Also, being overweight and obese is an
independent predictor of type 2 diabetes (Weinstein et al., 2004).
In addition, being overweight and obese is associated with inactivity,
resulting in lower aerobic capacity and less muscle strength (Andersen,
Franckowiak, Christmas, Walston, & Crespo, 2001), and greater potential for

13
disability and functioning loss (Davison, Ford, Cogswell, & Dietz, 2002; Himes,
2000; Reynolds, Saito, & Crimmins, 2005).
Although weight may not play the same role in the Japanese population
in causing adverse health outcomes as in the U.S. population due to lower
levels of overweight and obesity, researchers have found strong associations
between higher weight and heart disease, as well as hypertension,
hyperlipidemia, and diabetes in older Japanese adults (Chei, Iso, Yamagishi,
Inoue, & Tsugane, 2007; Ito et al., 2003; Krause et al., 2002).
Meanwhile, underweight as well as overweight, has been linked to
higher mortality among middle aged and older people (Allison et al., 1997;
Grabowski, & Ellis, 2001; Hayashi et al., 2005; Miyazaki et al., 2002).
Therefore, understanding the role of body weight in causing poor health
outcomes in Japan and the U.S., as well as the factors associated with body
weight in the two countries should improve our understanding of this serious
international public health issue as well as the determinants of differential
health in the two countries.
2.2.2 Body Mass Index and Socio-Demographic and Economic Factors
In the U.S., higher BMI has been generally associated with older age
(Ogden et al., 2006) although, as shown above, this may no longer be true at
least in prevalence of obesity. While being overweight is more common
among men, obesity is more common for women (Flagal et al., 1998).
Research has shown that the prevalence of overweight and obesity is higher
14
among women and rural residents in Japan (Yoshiike, Kaneda, et al., 2002;
Yoshiike, Seino et al., 2002).
Married adults, particularly men, have higher rates of overweight and
obesity than adults in other marital statuses. Never married people are the
least likely to be overweight or obese (Schoenborn, 2004).
Numerous studies in the U.S. report BMI to be inversely associated with
socioeconomic status (SES) (Jeffery, French, Forster, & Spry, 1991; Smith,
1999; Zhang & Wang, 2004a) although the relationship between BMI and
education has weakened over time (Zhang & Wang, 2004b). There appears
to be an interaction among demographic factors. Among women, there is a
stronger inverse association between SES and obesity than among men.
There is also evidence of greater socioeconomic inequality among middle-
aged adults compared to other age groups (Zhang & Wang, 2004a).
It has been thought that socioeconomic differentials in health outcomes
were almost nonexistent in Japan due to the relatively equal income
distribution (Marmot & Smith, 1989; Wilkinson, 1994). An analysis of Gini
coefficients in income inequality in the U.S. and Japan reveals the lower level
of income inequality in Japan than the U.S. (Xiaobin, Li, & Kelvin, 2004).
2.2.3 Body Mass Index and Behavioral Factors
The inverse associations between SES and BMI in the U.S. may be
caused by differences in health behaviors. As Jeffery et al. (1991) have found,
higher SES was also associated with health behaviors that contribute
15
importantly to energy balance. Those who have high SES are more likely to
report a lower fat diet, more exercise, and a higher prevalence of dieting to
control weight. In addition, higher SES men and women are less likely to
smoke.
Some research has linked smoking cessation to increased body mass
index (John, Meyer, Rumpf, Hapke, & Schumann, 2006; Kadowaki et al.,
2006). A higher rate of smoking tends to be closely linked to lower rates of
overweight and obesity in Japan (Mizoue, Kasai, Kubo, & Tokunaga, 2006).
It has also been found that lower physical activity and sedentary
lifestyle are strongly associated with higher BMI in many countries (Kruger,
Venter, Vorster & Margetts, 2002; Martinez-Gonzalez, Martinez, Hu, Gibney, &
Kearney, 1999; Monda & Popkin, 2005; Sharma, 2007). Life style may have
become more sedentary in rich countries because of increasing television
viewing, computers, videos, labor-saving devices, decreased use of public
transportation and increased automobile use (Sharma, 2007). Thus, the
increase in overweight and obese people may result from insufficient physical
activity and/or excess calorie intake (Ravussin, Fontveille, Swinburn, &
Bogardus, 1993).
Examining what factors are associated with BMI may provide some
clues in understanding why people become underweight, overweight or obese,
how socioeconomic status is associated with health outcomes, and how
people could control body weight in different countries and age groups.
16
2.3 Body Mass Index and Mortality in Different Countries and Age Groups
2.3.1 Body Mass Index and Mortality by Different Countries and Age Groups
Conventional knowledge regarding the link between body weight and
mortality has been that obesity is problematic regardless of age. This is
because research has shown that obesity raised mortality risk for adults of all
ages. Calle, Thun, Petrelli, Rodriguez, & Health (1999) showed that heavier
men and women in all age groups had an increased risk of death. Also,
Peeters et al. (2003) have shown that obesity in later life predicted mortality,
especially from ischemic heart disease.
However, there have been other studies in which the positive linear
association between higher BMI and mortality gradually faded out at the older
ages. Stevens et al. (1998) have shown that higher BMI is related to higher
mortality both from all-causes and from cardiovascular disease in adults up to
75 years of age. However, the relative risk associated with higher BMI
declined with age. The findings of Bender et al. (1999) were consistent with
those of Stevens et al. which is that while the relative risk of death increased
with higher BMI, the relationship declined with age.
Other research has also shown that there is no linear relationship
between higher BMI and mortality in later life (Baik et al., 2000; Grabowski &
Ellis, 2001). According to Thorpe and Ferraro (2004), the effect of obesity on
mortality is about twice as strong for persons younger than 50 years of age,
compared to those who are 50 years of age or older. Furthermore, Diehr et al.
17
(1998) have shown that the linkage of higher BMI to mortality found in the
middle-aged was not observed in older adults.
Moreover, recent research has emphasized that not only obesity but
underweight is associated with increased mortality relative to normal weight for
older adults in the U.S. (Allison et al., 1997; Flegal et al., 2005). It also has
been shown for both the middle-aged and older adults in Japan (Hayashi et
al., 2005; Kuriyama et al., 2004). Low weight is related to a greatly increased
risk for all-cause mortality among Japanese who are middle-aged and elderly
(Miyazaki et al., 2002; Ohta, Aoki, Takeuchi, Yosiak, & Suzuki, 2001).
2.3.2 Methodological Issues in Studies of Body Mass Index and Mortality
The inconsistent findings in the relationship between BMI and mortality
seem to occur from methodological factors that may complicate observed
relationships. According to Manson et al. (2007), the most important problem
in studies of body weight and mortality is reverse causation, i.e., both death
and being underweight are caused by some other factors. For instance, the
reason why underweight people are more likely to die than normal or heavier
people is because researchers did not control antecedent diseases. Among
the older population weight loss can possibly result from underlying diseases
rather than leanness causing diseases. Related to this issue, some studies of
body weight and mortality fail to eliminate early mortality from the analysis,
due to short follow-up periods, which may affect the relationship (Manson et al.,
1987).
18
Another issue is failure to control for cigarette smoking because
smoking is both more common among underweight people and is a strong risk
factor for cardiovascular diseases and smoking-related cancer (Manson et al.,
1995). Furthermore, smoking is an important factor increasing the risk of
cancer-related mortality among underweight individuals (Krueger, Rogers,
Hummer, & Boardman, 2004). This association may produce an artificially
higher rate of mortality among underweight people. Thus, Manson et al. (2007)
suggests that analysis be restricted to never smokers to provide the best
estimates of the link between weight and mortality.
A third issue is that inappropriate control of the effects of obesity may
attenuate the relationship between obesity and mortality. Controlling for
hypertension, dyslipidemia, and hyperglycermia, which are biological
consequences of obesity, reduces the strength of the causal pathway by which
obesity causes mortality, although it is rarely totally eliminated (Manson et al.,
1987).
Manson et al. (2007) emphasized several methodological factors that
should be considered in studies of the elderly: 1) the high prevalence of
comorbidity, preexisting diseases, and weight loss related to illness, 2) BMI as
a less reliable marker of adiposity due to differential loss of muscle, and 3)
higher risk of death at baseline and unexpected effects of individual risk
factors. The exclusion of deaths occurring in the first few years of follow-up

19
and whether the number of deaths was sufficient to predict the effect of BMI
on mortality are also factors to be considered.
Previous analyses have generally not considered the entire age range
due to the limitation of most survey samples. Therefore, to date there is little
known about the relationship between BMI and mortality across countries and
ages.
2.4 Mechanisms Through Which Body Mass Index Affects Mortality: Pathway
Through Biomarkers, Nutrition, and Health Behaviors
2.4.1 The Link of Body Mass Index to Mortality and Biomarkers
Until recently, it had been widely accepted that overweight or obesity
results in poor health outcomes including functional disabilities, onset of
diseases, and mortality (Adams et al., 2006; Burke et al., 1999; Colin, Adair, &
Popkin, 2002; Must et al., 1999). However, recent research has shown that
people with low weight are more likely to die than normal weight or heavier
people (Allison et al., 1997; Flegal et al., 2005; Grabowski & Ellis, 2001). This
has been shown in both younger and older people (Flegal et al., 2005;
Krueger et al., 2004).
Thin people who have smoked for a while may have particularly high
risk because of associated physiological changes. They may have elevated
levels of inflammatory biomarkers: white blood cell, fibrinogen, C-reactive
protein (CRP), and interleukin-6 (IL-6) (Hammett et al., 2007). It has been
shown that people with both elevated IL-6 and CRP are associated with a
20
greater risk of overall death (Harris et al., 1999; Reuben et al., 2002). Low
albumin, which is sometimes used as an indicator of inflammation, has also
been shown to be significantly and independently associated with mortality
(Fried et al., 1998; Reuben et al., 2002). Low albumin may also be an
indicator of poor nutritional status that might characterize underweight persons
(Liu, Bopp, Roberson, & Sullivan, 2002). Underweight may also be related to
low blood pressure, anemia, low HDL cholesterol, and low systolic blood
pressure (Fried et al., 1998). Low diastolic blood pressure (Goldman et al.,
2006) and low total cholesterol (Reuben et al., 2002) are related to mortality
(Fried et al., 1998). It is also plausible that thin people have low pulse. Links
between being underweight and levels of biomarkers might provide a clue as
to why thin people have a higher risk of death than normal weight or heavier
people; it is also important to look at the link between overweight/obesity and
risk factors of death.
Another reason why overweight or obese people have had relatively
lower mortality over the last decade, although they are assumed to have more
cardiovascular disease (CVD) than normal weight people, is because CVD risk
factors, i.e., high cholesterol, high blood pressure, and smoking became less
prevalent among overweight or obese persons over time (Gregg et al., 2005).
These changes in secular trends in risk factors were also related to increases
in antihypertensive and lipid-lowering medication use, particularly among
overweight or obese people (Gregg et al., 2005). Thus, both lower prevalence
21
of CVD risk factors among the overweight or obese and medication use may
have led to lower mortality among overweight or obese people than two or
three decades ago. However, it is also true that there remains some
controversy: some research has found that the presence of hypertension in
overweight people is the key factor in higher CVD mortality (Thomas et al.,
2005).
2.4.2 The Link of Body Mass Index to Health Behaviors, Antioxidants,
and Dietary Intake
Research regarding the association between body weight and mortality
needs to also consider how behavioral, nutritional, and physical factors affect
mortality and which age groups are more affected by which factors in order to
investigate why recent research has shown that thin people are more likely to
die than people with normal or heavier.
Body weight is a reflection of diet and energy expenditure in some way.
Thin people should differ in one or both of these characteristics from heavier
people. It is thus crucial to investigate whether there are mechanisms through
which diet and exercise link specific body weight levels and mortality. For
instance, low weight persons may have diets low in antioxidants (Michelon et
al., 2006; Ray et al., 2006). Being overweight is related to both diet and
exercise. Sedentary life style is strongly associated with high BMI, and
reduced energy expenditure during leisure time may be one of the
determinants of obesity (Kruger et al., 2002; Martinez-Gonzalez et al., 1999).
22
Lower levels of physical activity within the same categories of BMI were
related to higher levels of CRP, HDL cholesterol, LDL cholesterol, total
cholesterol, and fibrinogen (Mora, Lee, Buring, & Ridker, 2006; Schumacher et
al., 2006). Low cardiorespiratory fitness is strongly associated with increased
mortality (Farrell, Braun, Barlow, Cheng, & Blair, 2002). In addition,
increasing levels of physical fitness continuously decrease CRP levels
(Aronson et al., 2004).
Interesting enough, thin men who do not exercise had a higher risk of
all-cause mortality and CVD mortality than men who are obese but fit. Also,
among thin men, the unfit had a two times higher risk of all-cause mortality
than fit men (Lee, Blair, & Jackson, 1999). Therefore, physical exercise is an
important predictor of death regardless of weight.
Smoking tends to contain many oxidants which result in lipids and DNA
damage (Mizoue et al., 2006). Particularly, smoking is closely associated with
lower levels of serum antioxidants including vitamin C, vitamin E, selenium, α-
carotene, and β-carotene because smoking may be linked directly to poor diet
and tooth loss (Hanioka, Ojima, Tanaka, & Aoyama, 2006; Rock, Jacob, &
Bowen, 1996; Northrop-Clewes, & Thurnham, 2007; Wei, Kim, & Boudreau,
2001). Older adults as well as smokers have lower levels of serum
antioxidants which are obtained from consumption of fruits and vegetables
which may be lower among the socially isolated and those with missing teeth
(Sahyoun, Zhang, & Serdula, 2005). Alcohol consumption is related to being
23
overweight and high alcohol consumption increases risk of mortality (Baglietto,
English, Hopper, Powles, & Giles, 2006; Nakaya et al., 2004). Furthermore,
exposure to both smoking and high levels of alcohol consumption increased
the risk of gastric cancer nearly 5 times (Sjodahl et al., 2007).
Many studies have shown the positive effects of antioxidants against
cancers, cardiovascular diseases, and lung function (Block, 1991; Comstock
et al., 1997; Hu & Cassano, 2000). In addition, high vegetable and fresh fruit
intake is inversely associated with CVD mortality (Harriss et al., 2007;
McEligot, Largent, Ziogas, Peel, & Anton-Culver, 2006) and an important
factor in survival from cancer (Pierce et al., 2007). Whole-grain intake is also
related to reduced mortality from CVD among older adults (Sahyoun, Jacques,
Zhang, Juan, & McKeown, 2006). Vitamin and carotenoid deficiencies are
related to frailty in older women (Michelon et al., 2006). Therefore, it is
thought to be important to have a diet characterized by intake of low levels of
dietary fat and high levels of fiber, vegetables, fruits, and other nutrients
including folate, vitamin C, and carotenoid in order to reduce the risk of death.
In summary, biomarkers, behaviors including smoking, drinking, and
exercise, serum antioxidants, and dietary intake should be considered when
we explore the mechanisms through which BMI affects mortality.

24
Chapter III: Data and Methods
3.1 Data
This study uses four different data sets: The Nihon University Japanese
Longitudinal Study of Aging (NUJLSOA), the Longitudinal Study on Aging
(1994 Second Supplement on Aging, LSOA II), the Health and Retirement
Study (HRS), and the National Health and Nutrition Examination Interview
Survey III (NHANESIII). NUJLSOA and LSOA II are used to reveal what
factors are determinants of being overweight or obese, normal weight, and
underweight in chapter V; the NUJLOSA, LSOA II, and HRS are nationally
representative samples of different countries and age groups and used to
examine which BMI levels are associated with mortality in chapter VI. This
study also uses NHANES III to investigate the link between BMI and health
behaviors as well as biomarkers of risk and the effect of BMI on all-cause
mortality and specific causes of death when considering biomarkers and
health behaviors among the various age groups of U.S. adults in chapter VII.
3.1.1 The Nihon University Japanese Longitudinal Study of Aging (NUJLSOA)
The NUJLSOA is a nationally representative sample which includes a
total of 4,997 Japanese older adults (2,114 men and 2,883 women) who were
aged 65 and over at the baseline 1999 survey. The first wave of data
collection was conducted in November 1999 (75% response rate), the second
in November 2001, and the third in November 2003. The sample was
refreshed with younger members at each wave to retain a representative
25
sample of Japanese older adults 65 years of age and over. The NUJLSOA
sample was selected using a multistage stratified sampling method, stratifying
the 47 prefectures throughout Japan into 11 regions. Approximately 3,000
municipalities were then stratified by population size within the regions and
prefectures, and a systematic sampling method was used to select 340
primary sampling units (PSUs). The residents chosen for the NUJLSOA in
298 PSUs were selected based on the National Residents Registry of Japan,
the most up to date source of personal data in Japan. In another 41 units,
sample members were drawn from the Eligible Voters List, which is based on,
but updated not as regularly as the National Residents Registry. From each of
the 340 PSUs, 6-11 persons aged 65-74 and 8-12 persons at age 75 and over
were selected for the sample. The population aged 75 and older were
oversampled, and appropriate weighting systems are available and used in
this analysis to ensure that the total sample was nationally representative of
those 65 and older. More information is available at the home page of
USC/UCLA Center of Biodemography and Population Health (USC/UCLA
Center of Biodemography and Population Health, 2004).
The primary purpose of NUJLSOA was to investigate the health status
of the Japanese elderly and changes in health status over time. In addition, it
was designed to explore the impact of long-term care insurance systems on
the use of services by the Japanese elderly and to investigate relationship
between co-residence and the use of long term care. The special aim of the
26
NUJLOSA survey is to provide data comparable to that collected in the United
States and other countries. The format of questions in the initial questionnaire
was determined with attention to the format of the U.S. LSOA II and the Assets
and Health Dynamics Among the Oldest Old (AHEAD) sample of the Health
and Retirement Study (HRS). Similar questions were asked at the three
waves with the addition of a number of questions related to long-term care in
the second wave.
The survey followed up the survival status of all subjects from 1999
through 2003. A total of 327 people died within the first 2 years, the interval of
the first wave and the second wave, from 1999 through 2001. Another 370
people among the subjects in the first wave died between the second wave
and the third wave, from 2001 through 2003. Thus, 697 people among
subjects from the first wave in 1999 died between the first and the third wave.
To compare with U.S. older adults (LSOA II, 70+) in chapter V and VI,
this study selected Japanese older adults aged 70 and over at baseline. The
total number in the selected sample is 3,833 persons. In the selected sample,
a total of 302 people died within the first 2 years, from 1999 through 2001 (see
Figure 3.1). Another 333 people among subjects of the first wave died
between the second wave and the third wave, from 2001 through 2003. Thus,
635 people among subjects from the first wave of 1999 died between the first
and the third wave (see Figure 3.2).

27
Figure 3.1
Total Sample Construction (NUJLSOA, 65+)

Figure 3.2
Selected Sample for Analyses in Chapter V and VI (NUJLSOA, 70+)

Wave 1 (1999)

Total: 4,997
Wave 2 (2001)

Total: 4,267
Wave 3 (2003)

Total: 3,418
Decedents between W1 and W2:
327 persons
Missing: 403 persons
Decedents between W2 and W3:
370 persons
Missing: 479 persons
Total
decedents:

697 persons
Wave 1 (1999)

Total: 3,833
Wave 2 (2001)

Total: 3,227
Wave 3 (2003)

Total: 2,519
Decedents between W1 and W2:
302 persons
Missing: 303 persons
Decedents between W2 and W3:
333 persons
Missing: 376 persons
Total
decedents:

635 persons
28
3.1.2 The Longitudinal Study on Aging II (1994 Second Supplement on Aging)
LSOA II is a nationally representative sample of the U.S. population 70
years of age and older conducted by the National Center for Health Statistics
(NCHS) and the National Institute on Aging (NIA) beginning in 1994. This
study was designed to provide information necessary for analyzing important
temporal changes in health and functioning among community dwelling older
Americans, including background demographic characteristics, health
behaviors, attitudes, preexisting illness, and social and environment support.
Also, LSOA II aimed to provide information on the sequence and
consequences of health events, including utilization of health care and
services for assisted community living, along with the physiological
consequences of disability such as pain and fatigue.
The sample for the LSOA II was drawn from individuals who
participated in the 1994 NHIS core interview regardless of disability status,
and completed additional questions for the Second Supplement on Aging
(SOA II). These participants were followed through two follow up surveys that
became known as the Longitudinal Study of Aging II.
The self response rate for LSOA II was 84.5 percent and the proxy
response rate was 14.4 percent. In about 1 percent of the sample,
interviewers failed to indicate who responded to the interview. Details of
sampling and design for LSOA have been published elsewhere, and will not
be repeated here (Center for Disease Control and Prevention [CDC], 2007).
29
The sample consists of 9,447 older Americans who were at age 70
years and over at the time of the first LSOA II interview. A total of 1,625
persons died within four years of the first interview (see Figure 3.3).

Figure 3.3
Sample Construction (LSOAII)

3.1.3 The Health and Retirement Study (HRS)
The HRS is a nationally representative sample of U.S. adults with an
initial sample of 12,652 persons aged 51 to 61 from the 1931-1941 birth cohort
and their spouses in 7,600 households in 1992. The HRS sample was
selected using a multi-stage area probability sample design. The sample
included four distinct selection stages. The primary stage of sampling involves
W1 (1994)

Total: 9,447
The number of
persons alive in 4
years of initial
interview

Total: 7,882
Decedents reported within 4
years of initial interviews
1,625 persons
30
probability proportionate to size (PPS) selection of U.S. Metropolitan Statistical
Areas (MSAs) and non-MSA counties. Then a second stage sampling of area
segments (SSUs) within sampled primary stage units (PSUs) is conducted.
The third sampling stage is a systemic selection of housing units from all
housing unit listings from the sample SSUs. The fourth and final stage in the
multi-stage design is the selection of the household financial unit within a
sample HU (Heeringa & Connor, 1995). The original sample consists of in-
home, face-to-face interviews and includes oversamples of Hispanics, Blacks,
and residents of the state of Florida.
The survey topics include demographic background, employment status
and job history, health insurance and pensions, health and cognitive
conditions, family structure and transfers, retirement plans and perspectives,
and income and net worth.
The response rate is 81.7%, 89.1%, and 86.3% in wave 1 in 1992,
wave 2 in 1994, and wave 3 in 1996, respectively. The response rate includes
exit interviews for deceased sample members provided by proxies, as well as
core interviews.
This study uses the first three waves for the analysis from 1992 to
1996. Three hundred and twelve persons died by 1996, which is relatively a
small number of people who died in 4 years because respondents are
relatively young compared to the others (see Figure 3.4).

31
Figure 3.4
Sample Construction (HRS)

Selecting people who are aged from 51 to 61

3.1.4 The Third National Health and Nutrition Examination Interview Survey
(NHANESIII)
NHANES III was conduced by the National Center for Health Statistics
from October 1988 through October 1994 with mortality follow-up through
2000. As a nationally representative sample, NHANES III is based on
complex, multi-stage, stratified, clustered samples of civilian,
noninstitutionalized populations. The age range for the survey was two

Wave 1
(1992)

Total:
12,652
Wave 2
(1994)

Total: 8,744
Wave 3
(1996)

Total: 8,550
Decedents
between W1 and
W2: 118 persons
Decedents
between W2 and
W3: 194 persons
Total
decedents
: 312
persons
Wave 1
(1992)

Total: 8,862
32
months and older, and this is the first NHANES without an upper age limit
(CDC, 1994).
The home examination option was employed for the first time in order to
obtain examination data for very young children and for the elderly who were
unable to visit the mobile examination center (MEC). Because the full MEC
examination would have been difficult to collect in a home setting, only a
subset of the components of the full MEC examination were included in the
home examination (CDC, 1996).
NHANES III was conducted in two phases, each of which consisted of a
national probability sample. The first phase was conducted from October
1988 through October 1991 at 44 locations. The second phase was
conducted from September 1991 through October 1994 at 45 different
locations. NHANES III includes a household survey and a medical
examination. In NHANES III, 39,695 persons were selected over the six years
and 33,994 (86%) were interviewed in their homes. All interviewed persons
were invited to the MEC for a medical examination. Seventy-eight percent
(30,818) of the selected persons were examined in the MEC and an additional
493 persons were given a special, limited examination in their homes (CDC,
1996).
The household surveys consist of adult questionnaires and family
questionnaires. The adult questionnaires include demographic backgrounds,
socioeconomic status, health conditions, physical functioning, and health
33
behaviors such as exercise and smoking. The family questionnaires are
comprised of questions on education, occupation, income, health insurance
coverage, and recorded characteristics of the house itself.
A variety of physical examinations were part of the medical
examination: blood pressure, body measurements, physical functioning,
cognitive functioning, bone density, and biological measurements tested from
urine and blood samples. Blood and urine specimen tests were collected from
subjects aged one year and older at the mobile examination center (MEC).
This study uses selected people who are aged 40 and over to examine the link
between weight and mortality through health behaviors and biomarkers among
both the middle-aged and older persons.
The nutritional assessment component of NHANES III provided
important resources to estimate total nutrient intake, nutrient intake from
specific foods, and problems and factors related to insufficient food intake, so
that it formed a comprehensive evaluation of nutritional status (CDC, 1994).
The NHANES III dietary components comprised 24-hour dietary recall, food
frequency questionnaire, nutrition-related interview, and anthropometry,
conducted by the NHANES III Dietary Data Collection (DDC) system. This
data was collected during subjects’ MEC visit by trained dietitians. Subjects
quantified foods and beverages using food-specific units, abstract food
models, special charts, and measurement aids such as cups and spoons. The
dietary interviewers listed the specific types of foods and amounts of
34
beverages, including alcoholic beverages, in a private interview during the 24-
hour recall. The NHANES III interview files were sent to the National Center
for Health Statistics (NCHS) headquarters which coded the NHANES III DDC
interview files by using the U.S. Department of Agriculture (USDA) Survey
Nutrient Data Base (SNDB). This study used some dietary indicators coded
by USDA Survey Nutrient Data Base (SNDB). Figure 3.5 shows sample
construction of NHANES III used in this study.

Figure 3.5
Sample Construction (NHANES III)

3.2. Measures
3.2.1.

In summary, NUJLSOA selected for people aged 70 and above and
LSOA II are used in chapter V to examine the correlates of overweight or
obesity, normal weight, and underweight with demographic, socioeconomic,
1988-1994

9,623

The number of
persons alive in 6
years of initial
interview

8,078
Decedents reported within 6
years of initial interviews:
1,545 persons
35
and health behavioral factors. After including HRS, those three datasets are
used to examine the association between BMI and all-cause mortality over a
four-year interval in different countries and age groups in chapter VI.
NHANES III is analyzed to investigate the link of BMI and mortality mediating
health behavioral factors and biomarkers.
3.2 Measures
3.2.1 Measures of the NUJLSOA and the LSOA II for Chapter V
NUJLSOA and LSOA II are used to examine the correlates of
overweight or obesity with demographic and socioeconomic factors, and
health behaviors, and how BMI levels are associated with health outcomes in
chapter V. All measures used in chapter V are shown in Table 3.1.
3.2.1.1 Weight Categorizations
Weight and height are collected in the NUJLSOA and LSOA II based on
respondent self-reports. BMI, calculated by weight (kg) divided by height
squared (m
2
), is divided into categories based on current World Health
Organization (WHO) standards: underweight (BMI < 18.5), normal weight
(18.5 ≤ BMI ≤ 24.9), overweight (25.0 ≤ BMI ≤ 29.9), and obesity (30.0 ≤ BMI)
(WHO Expert Consultation, 2004). In chapter V, overweight and obesity are
combined to one category, defining overweight as BMI 25.0 and above, due to
the low level of obesity among Japanese older adults.
Three weight categories – underweight, normal weight, and overweight-
are used as a dependent variable first and will be independent variables to

36
Table 3.1
Study Variables and Definition of the NUJLSOA and the LSOA II in Chapter V
Variable Definition
Dependent Variables
BMI* BMI= weight/(height)2 (kg/m2), underweight (BMI<18.5), normal (18.5<=BMI<=24.9), overweight (25.0<=BMI)

Health Outcomes
ADL functioning difficulties 0 = Having no difficulty in any of six ADLs; 1 = having difficulty in at least one of six ADLs
Cancer 0 = no; 1 = yes
Stroke 0 = no; 1 = yes
Diabetes 0 = no; 1 = yes
Heart disease 0 = no; 1 = yes
Arthritis 0 = no; 1 = yes

Independent Variables
Sociodemographic factors
Age Continuous variable: 70-99
Gender 0 = male; 1 = female
Marital status 0 = not married (widowed, divorced, separated, never married); 1 = married (married, married-spouse not in household)
Education Years of education completed: continuous variable
Income Annual income, assigned with midpoint of categories, replaced missing with mean value
Rural/Urban area 0 = living in urban area; 1 = living in rural area
Health behaviors
Smoking 0 = never smoked or former smoker; 1 = current smoker
Drinking 0 = never drunk or light drinker (drinks per day <=2); 1 = heavy drinker (3+ drinks per day)
Exercise 0 = active (NUJLSOA-walked 3+ per week or engaged in any sports or exercise, LSOA II-regular exercise); 1 = inactive
Note: Three weight categories will be a dependent variable in multinomial logistic regression first, and then underweight and overweight are going to be
independent variables (reference category: normal weight) to predict six health outcomes in logistic regression.

37
predict health outcomes including ADL functioning difficulties, cancer, heart
disease, diabetes, and arthritis.
3.2.1.2 Demographic and Socioeconomic Factors
Age was used as a continuous variable from both surveys. Japanese
older adults aged 99 and over were recoded as 99, comparable to LSOA II.
For gender, female was recoded as 1 and the reference category is male (=0).
For marital status, married (=1) included those separated from a
spouse due to hospitalization, institutionalization, or living in another area for
business reasons in the Japanese survey. In the U.S. survey, married
included both spouse in household and not in household. In both surveys,
widowed, divorced, separated, and never married were recoded as not
married, reference category (=0). With respect to living in a rural or urban
area, a question “what sort of community do you currently live in?” was used
and city/suburbs was recoded as the reference category (=0) and living in a
farming and fishing village was coded as living in rural area (=1) in the
NUJLSOA. In the LSOA II, living in a Metropolitan Statistical Areas (MSA)
including outside the central city, was recoded as reference (=0) and not living
in an MSA was coded as living in a rural area (=1).
Information on socioeconomic status – education and annual income-
are available in both surveys. In the LSOA II, respondents are asked to
indicate the total number of years of formal education completed. In the
NUJLSOA, education was reported using 7 categories in the first wave and by

38
single years in the second wave, collected in 2001. This study used the Wave
2 years of education when available, and where the Wave 2 question was not
answered (n=857), the years of education for respondents in the Wave 1
category is assigned. To be comparable to the U.S. survey, 18 years and over
in the years of education completed was recoded as 18 in the Japanese
survey.
Japanese income is converted into US dollars using the exchange rate
for November 15, 1999. In both the NUJLSOA and the LSOA II, income is
reported in categories and the categorical midpoint is assigned. The number
and range of the income categories differs; the US has 26 categories with top
coding at $50,000 and the Japanese has 13 categories with top coding at
$140,000. This study tested the sensitivity of the results to the range of
income categories by collapsing the Japanese top categories to match the top
US category and the results were robust, so we use the available detail on
income. In addition, we control all regression models for respondents for
whom income information is missing.
3.2.1.3 Indicators of Health and Behaviors
Health outcomes investigated include the presence of functioning
problems as indicated by having difficulty in at least one activity of daily living
(ADL): dressing, eating, toileting, walking, bathing, and transferring in or out of
bed. People who are missing in any of those 6 ADL functioning difficulties
were coded as missing. Questions on functioning in the two surveys were

39
designed to be similar and took the format of “do you find it difficult to [bath]
due to your health or physical state” in Japan, and “because of a health or
physical problem, do you have any difficulty [bathing]” in the U.S.
Heart disease was coded as a dichotomous variable when a positive
answer was provided to the following questions, “have you ever experienced
[had] or are currently experiencing [tachychardin from angina, myocardial
infarction, or other forms of heart disease]” in both NUJLSOA and LSOA II.
Other diseases including cancer, stroke, diabetes, and arthritis were also
indicated by whether respondents agreed with similar statements regarding
these diseases.
In addition to health outcomes, this study examines three additional
health behaviors related to weight. Smoking (=1) is designated as “being a
current smoker,” and former smoker or never smoker was coded as reference
category (=0). Heavy drinking is consuming 3+ drinks per day in LSOA II. In
the NUJLSOA, consuming 3+ drinks per day included beer, wine, shochu, and
whisky.
For the LSOA II, activity was determined by reports of whether the
respondent engaged in regular exercise. In order to make the comparisons as
close as possible across surveys, a variable was constructed in the NUJLSOA
that approximates regular exercise. In the NUJLSOA, two questions were
used to create physical activity. Respondents were asked how often during
the week they went for a walk. Another question asked whether the

40
respondent engaged in sports or exercise recently. The respondents were
coded as active (=0) if the respondent walked 3 times or more per week, or
engaged in any sports or exercise; otherwise inactive (=1).
3.2.2 Measures of the NUJLSOA, the LSOA II, and the HRS for Chapter VI
The aim of chapter VI is to examine the link between BMI levels and
mortality in different countries and age groups. Some measures of the
NUJLSOA and the LSOA II were the same as those used in chapter V,
including age, gender, cancer, stroke, heart disease, diabetes, smoking,
drinking, and physical activity. However, other variables differ slightly in terms
of coding, for instance, BMI and ADL functioning difficulties. Depression was
added in chapter VI because depression may be a factor linked to being
underweight. Meanwhile, marital status, living area (rural/urban area),
education, income, and arthritis were removed in the model because those are
not directly related to mortality. Thus, I will focus here on the description of
measures that are different from those in chapter V and added in chapter VI.
Also, this section will explain how measures of HRS are comparable to those
of NUJLSOA and LSOA II in greater detail (see Table 3.2).
3.2.2.1 Weight Categorization and Death Variable
Body Mass Index (BMI) was categorized by using the current WHO BMI
standard and used as four levels in chapter VI: underweight (BMI < 18.5),
normal weight (18.5 ≤ BMI ≤ 24.9), overweight (25.0 ≤ BMI ≤ 29.9), and

41
Table 3.2
Study Variables and Definition of the NUJLSOA, the LSOA II, and HRS in Chapter VI
Variable Definition
Dependent Variables
Death 0 = people who did not die within the 4 year interval after first interview; 1 = people who died within same interval

Independent Variables
Weight categorization
BMI BMI= weight/(height)2 (kg/m2)
underweight (BMI<18.5), normal (18.5<=BMI<=24.9), overweight (25.0<=BMI<=29.9), obesity (30.0<=BMI)
Sociodemographic factors
Age Continuous variable: 70+ (NUJLSOA, LSOA II), 51-61 (HRS)
Gender 0 = male; 1 = female
Health Outcomes
ADL functioning difficulties 1 = having difficulty in at least one of four ADLs (eating, dressing, taking a shower, getting in and out of a bed or chair)
Cancer 0 = no; 1 = yes (NUJLSOA, LSOA II- Have you ever experienced (had) ..., HRS- Has a doctor ever told you…)
Stroke 0 = no; 1 = yes
Diabetes 0 = no; 1 = yes
Heart Disease 0 = no; 1 = yes
Depression 0 = not depressed (NUJLSOA- rarely and HRS- none or almost none during the past week;
LSOA II- a little of the time or none of the time during the past 12 months); 1 = depressed
Health behaviors
Former smoker 0 = never smoker or current smoker; 1 = former smoker
Current smoker 0 = never smoker or former smoker; 1= current smoker
Exercise 0 = active (NUJLSOA & HRS- walked 3+ per week or engaged in any sports or exercise, LSOA II- regular exercise);
1 = inactive

42
obesity (30.0 ≤ BMI). All three datasets collected weight and height based on
respondent self-reports.
Dependent variable is whether subjects died within four years from
Wave 1 (1999) to Wave 3 (2003) in the NUJLSOA, four years of the initial
interview in LSOA II, and from Wave 1 (1992) to Wave 3 (1996) in the HRS.
People who died were coded as death (=1); otherwise (=0).
3.2.2.2 Health and Behavioral Factors
Control variables include health conditions and behavioral factors
thought to be related to weight. Health conditions include having ADL
disability, having cancer other than skin cancer, heart disease, stroke,
diabetes, and depression. Among those health conditions, cancer, heart
disease, stroke, and diabetes are categorized almost the same way in HRS.
For example, heart disease was coded as a dichotomous variable by
agreement with the questions, “has a doctor ever told you that you had (a
heart attack, coronary heart disease, angina, congestive heart failure, or other
heart problems)?” for the HRS. Other diseases including cancer, stroke, and
diabetes were also indicated by whether respondents agreed with same
statements regarding these diseases.
In all three samples, ADL disability was coded to reflect either no
difficulties (=0) with four ADL items they reported or one or more difficulties
(=1): eating, dressing, taking a shower, and getting in and out of a bed or chair.
The NUJLSOA asked respondents a question, “do you find it difficulty to [bath]

43
due to your health or physical state”, and in the LSOA II, “because of a health
or physical problem, do you have any difficulty [bathing]”. Responses were
“yes” or “no” in both surveys. The format of HRS about ADL difficulties was
“how much difficulty do you have to [dress] without help?” Responses of “a
little difficult,” “somewhat difficult,” and “very difficult/can’t do” were coded as
having difficulty (=1); “not at all difficult” as reference (=0). The reason why
chapter VI uses only four ADL functioning difficulties is because there were
only four ADL difficulties in HRS that are matched with the ADL difficulty
questions in the other two surveys.
In the Japanese sample (NUJLSOA) and the younger US sample
(HRS), depression was indicated by agreement with the statement, “during the
past week, I felt depressed.” Responses of “sometimes” or “often” were coded
as depressed (=1) and the other response of “rarely” was designated to not
depressed (=0) in the NUJLSOA. In the HRS, responses of “none or almost
none” were coded not depressed (=0); otherwise depressed (=1). In the
LSOA II, respondents indicated “how often [they] felt sad or depressed in the
past 12 months.” Responses of “all of the time” and “some of the time” were
coded depressed (=1). The other responses of “a little of the time” and “none
of the time” were coded as not depressed (=0).
Health behaviors included smoking and an indicator of physical activity.
Smoking was coded in a different way in all three samples for chapter VI:
former, current, and never smoker. This is because past smoking is also

44
related to mortality so that it needs to be a separate category. Since models
include both former (=1) and current smoker (=1) simultaneously, never
smoker (=0) is a reference category. As described in 3.2.1.3, activity was
determined by reports of whether the respondent engaged in regular exercise
for the LSOA II. In order to make a comparable variable in the HRS, a
variable was created to approximate regular exercise. In the HRS, two
questions were used, indicating how many times per week the respondent
engaged in light physical activity (e.g., walking, dancing, gardening, or
bowling), and in heavy physical activity (e.g., aerobics, running, swimming, or
bicycling). Respondents who engaged in light exercise at least 3 times per
week or engaged in any heavy physical activity, were coded as active (=0);
otherwise inactive (=1).
3.3.3 Measures of the NHANES III for Chapter VII
The NHANES III is used to investigate whether the link between low
and high BMI and mortality changes when biomarkers, health behaviors, and
diets are introduced in the models in chapter VII and to examine the link
between weight and biomarkers. To do this, this study first examined
differences by BMI in the percentage of the population with measured values
above or below clinically or conventionally defined “high risk” levels for
biomarkers that have been shown to affect the risk for mortality, indicators of
diet, and behavioral factors. Tables 3.3 and 3.4 show measures from the
NHANES III. As Table 3.3 shows, age is 40 and over, but people who are 90

45
Table 3.3
Age, Gender, Weight, Death, and Biomarker Variables and Definition for the NHANES III in Chapter VII
Variable Definition
Dependent Variables
Death 1 = people who died within 6 year interval of the initial interview;
0 = people who did not die within same interval
Independent Variables
Weight categorization
BMI BMI= weight/(height)
2
(kg/m
2
)
underweight (BMI<18.5), normal (18.5<=BMI<=24.9), overweight (25.0<=BMI)
Demographic factors
Age Continuous variable: 40 – 90+
Gender 1 = female; 0 = male
Biomarkers
Diastolic blood pressure I (DBP I) 1 = high risk (>= 90 mmHg); otherwise, 0
Diastolic blood pressure II (DBP II) 1 = high risk (<= 60 mmHg); otherwise, 0
Systolic blood pressure

1 = high risk (>= 140 mmHg); otherwise, 0
Total cholesterol I 1 = high risk (>= 240 mg/dl); otherwise, 0
Total cholesterol II 1 = high risk (<= 160 mg/dl); otherwise, 0
High-density lipoprotein (HDL) cholesterol 1 = high risk (< 40 mg/dl); otherwise, 0
Fibrinogen 1 = high risk (> 400 mg/dl); otherwise, 0
C-reactive protein (CRP) 1 = high risk (3.0 < <= 10 mg/dl); otherwise, 0
Glycosylated hemoglobin 1 = high risk (> 6.4%); otherwise, 0
White blood cell count 1 = high risk (> 10.8 x10
3
/uL); otherwise, 0
Low-density lipoprotein (LDL) cholesterol 1 = high risk (>= 160 mg/dl); otherwise, 0
Triglycerides 1 = high risk (>= 200 mg/dl); otherwise, 0

46
Table 3.4. Indicators of Diets and Health Behavioral Variables and Definition for the NHANES III in Chapter VII
Variable Definition
Independent Variables
Indicators of Diets
Albumin 1 = high risk (< 3.8 mg/dl); otherwise, 0
Serum Homocysteine 1 = high risk (>= 15 umol/l); otherwise, 0
Malnutrition 1 = high risk (any deficiency in iron, B12, and folate); 0 = no deficiency in iron, B12, and folate
Anemia 1 = high risk (< 120 g/L for men, < 130 g/L for women); otherwise, 0
Antioxidants 1 = high risk (being in the bottom quartile for 3+ indicators among serum vitamine A, C, E,
selenium, and lycopene); 0 = being in the bottom quartile for less than 3 indicators)
% of daily intake kilocalorie from carbohydrate 1 = high risk (> 65%); otherwise, 0
% of daily intake kilocalorie from fat 1 = high risk (> 40%); otherwise, 0
% of daily intake kilocalorie from protein 1 = high risk (< 10%); otherwise, 0

Health Behaviors
Smoking 1 = current smoker; 0 = never smoked or former smoker
Drinking 1 = heavy drinker (drinks per day 3+); 0 = never drunk or light drinker (drinks per day <=2)
Exercise 1 = no regular exercise (people who do not any type of exercise out of 9);
0 = regular exercise (people who engaged in any type of exercise out of 9)

47
years and older are redefined as 90+ by the NHANES III to maintain
confidentiality. Gender was coded as female (=1) and male (=0).
3.3.3.1 Weight Categorization and Death Variable
Body Mass Index (BMI) was categorized by using the WHO BMI
standard, but combining overweight with obesity in chapter VII: low weight
(BMI < 18.5), normal weight (18.5 ≤ BMI ≤ 24.9), overweight (BMI ≥ 25.0).
Since I attempted to examine whether the association of low weight with
mortality was different from that of overweight with mortality when other
confounders were introduced in the models, BMI was divided into three
categories with normal weight as the reference category as shown above.
Unlike the NUJLOSA, LSOA II, and HRS, weight and height were
measured in the mobile examination center (MEC) using standardized
procedures and equipment in the NHANES III. Persons aged under one year
or 20 years and over who were frail or unable to come to the MEC were
scheduled for home examination. Weight and height were measured at home
by a trained examiner using equipment and procedures comparable to those
used in the MEC (CDC, 1996).
The dependent variable is dichotomous indicating whether subjects
died within six years of initial interview conducted between 1988 and 1994.
People who died were coded as death (=1); otherwise coded as 0.

48
3.3.3.2 Biomarkers
High-risk cut points used in this chapter are shown in Table 3.3. I used
10 biomarkers (high-risk cut point); diastolic blood pressure (DBP) ( ≥ 90
mmHg), systolic blood pressure (SBP) ( ≥ 140 mmHg), total cholesterol ( ≥ 240
mg/dl), high-density lipoprotein (HDL) cholesterol (< 40 mg/dl), fibrinogen (>
400 mg/dl), C-reactive protein (CRP) (3.0 < CRP ≤ 10 mg/dl), glycosylated
hemoglobin (> 6.4%), white blood cell count (> 10.8 X 10
3
/ μL), fasting low-
density lipoprotein (LDL) cholesterol ( ≥ 160 mg/dl), and fasting triglycerides ( ≥
200 mg/dl) (USC/UCLA Center of Biodemography and Population Health,
2005). These biomarkers, except for DBP and SBP, were assayed from blood
samples collected at the MEC.
Since low values of both DBP and total cholesterol, outside the normal
range, may also pose additional risk, I also constructed variables indicating
high-risk in low DBP ( ≤ 60 mmHg) and low total cholesterol ( ≤ 160 mg/dl).
People who have low DBP and low total cholesterol were coded as DBP II =1
and total cholesterol II= 1.
3.3.3.3 Indicators of Diet
This study used 8 indicators of having a diet that would contribute to
higher mortality (high-risk cut points); albumin (< 3.8 mg/dl), serum
homocysteine ( ≥ 15 μmol/l), malnutrition, anemia, antioxidants, and % of daily
kilocalorie intake from fat, carbohydrate, and protein. The diet indicators other

49
than the distribution of calorie intake were derived from blood samples
collected at the MEC.
Malnutrition was defined as having deficiency in levels of iron, B12, or
folate. Iron deficiency was created by transferrin saturation (< 15 %), serum
ferritin concentration (<12 ng/mL), and erythrocyte protoporphyrin
concentration ( ≤ 1.24 μmol/l). If a person has two or more levels of these
three indicators at a high-risk level, (s)he has an iron deficiency. High-risk
levels of vitamin B12 (< 147.56 pmol/L) and folate (red blood cell folate <
232.49 nmol/L or serum folate < 5.89 nmol/L) were coded as B12 deficiency
and folate deficiency. Then, malnutrition was defined by counting the number
of deficiencies in iron, B12, and folate. People with any deficiency among the
three were defined as malnourished (=1).
The high-risk cut-off point for anemia differs by sex: hemoglobin<120
g/L for men (=1) and hemoglobin<130 g/L for women (=1). Antioxidants
included vitamin A, C, D, selenium, and lycopene. The relative level of
antioxidants was indicated by counting the number of measures of serum
antioxidants out of 5 for which the level was in the bottom quartile of the
sample. High-risk level of antioxidants (=1) was defined as being in the
bottom quartile for three or more indicators out of 5.
As described in 3.1.4, percentages of daily kilocalorie intake from fat,
carbohydrate, and protein were provided by automated coding of foods of the
U.S. Department of Agriculture (USDA) Survey Nutrient Data Base (SNDB).

50
For instance, percentage of kilocalories from carbohydrate was calculated as
follows:
% of Kcal from carbohydrate =
[(intake carbohydrate)*4 Kcal/gm carbohydrate)/food energy from total
intake]*100
Percentages of daily kilocalorie intake from fat and protein were
calculated in the same way. Conventionally, high fat, high carbohydrate and
low protein intake are problematic so that high-risks in % of daily kilocalorie
intake are more than 65% from carbohydrate (=1), more than 40% from fat
(=1), and less than 10% from protein (=1).
3.3.3.4 Health Behaviors
For health behaviors, smoking, drinking, and no exercise variables were
constructed. Two questions were used for smoking: (1) have you smoked at
least 100 cigarettes during your entire life (approximately 5 packs) and (2) do
you smoke cigarettes now. People who answered “yes” on the second
question were coded as current smokers (=1); people who answered “no” on
the first question (never smoked) or “yes” on the first question, but “no” on the
second question (former smokers) were coded as the reference category (=0).
For defining heavy drinker, respondents were asked three questions:
(1) have you had at least 12 drinks of alcohol in your life, (2) have you had at
least 12 drinks in last 12 months, and (3) the number of drinks per day on a
drinking day. In the third question, if a person answered three or more drinks

51
per day on a drinking day, he or she was coded as a heavy drinker (=1).
Meanwhile, answers of “no” on the first and second questions or less than 3
drinks on the third question were coded as the reference category of not being
a heavy drinker (=0).
People who did not participate in any type of exercises including
jogging, riding a bicycle, swimming, aerobics, dancing, calisthenics, gardening,
weighting, and physical active hobbies were coded as no regular exercise (=
1); otherwise, regular exercise (=0).
In summary, chapter V used the NUJLSOA (70+) and the LSOA II (70+)
whose variables included demographic, socioeconomic, health, and behavioral
variables. Chapter VI used those two datasets plus the HRS (51-61) to
examine the relationship between BMI and mortality controlling demographic,
health and behavioral factors. Depression was added in chapter VI and only
four ADL functioning difficulties were used for ADL functioning variable due to
lack of other two questions regarding ADLs in the HRS. In chapter VII,
demographic variables, biomarkers, indicators of diets, and behaviors were
used to examine whether the relationship between BMI and mortality has been
changed by the mediation of biomarkers, diets, and health behaviors. In
chapter V and VII, BMI was divided into three categories by combining
overweight with obesity. However, chapter VI used four BMI categories by
following WHO BMI standard. Death variable was constructed from decedents
with four-year interval in chapter VI and within 6 years in chapter VII.

52
3.3 Statistical Analyses for Research Questions
This study uses different statistical methods for specific research
questions. For the first research question, the analysis is descriptive in
clarifying the pattern of BMI distributions in Japanese older adults and older
and the middle-aged Americans. The chapter IV also describes the
characteristics of samples in each dataset.
Multinomial logistic regression is used for research question 2 to
examine correlates of overweight or obesity, normal weight, and underweight
in different countries in chapter V. The first model includes demographic and
socioeconomic variables such as age, gender, education, income, marital
status, living in rural or urban area, and the second model adds smoking,
drinking into the first model. This study then uses binary logistic regression to
relate being overweight or obese, separately to functional disabilities and
diseases such as heart diseases, diabetes, arthritis, and cancer. Finally, I
interpret the regression coefficients linking weight and chronic conditions and
disability in the U.S., compared to Japan.
The study uses logistic regression for research question 3 again in
chapter VI. Each sample’s respondents were followed over a four-year
interval. Deceased subjects within the four-year period and people who are
alive are categories of the dependent variable in logistic regressions, all of
which are controlled for age, gender, and BMI in the WHO categories. As our
concerns are with under-, overweight, and obese people, we omit normal

53
weight as the reference category in all regressions. Then, other independent
variables are added: health measures related to weight, (e.g., ADL functioning
difficulties, presence of diseases, depression), and health behaviors (e.g.,
smoking and exercise). The introduction of these controls should result in a
reduction in the observed relationship between weight and mortality if these
are the mediating mechanisms. In all cases, odds-ratios and 95% confidence
intervals are presented.
For the last research question in chapter VII, we examined the
percentages with high-risk value of all biomarkers, diet indicators, and health
behaviors to see whether those with low BMI and overweight BMI are
significantly different from that of normal BMI using the chi-square test. Then,
mortality is linked to weight using Cox proportional hazard regression models
controlling age and gender because dependent variable has two categories,
dead or alive within the 6-year of the initial interview, and people have different
exposure times. Eight biomarkers including diastolic and systolic blood
pressures, total cholesterol, HDL cholesterol, glycosylated hemoglobin,
fibrinogen, white blood cell count, and inflammation marker such as CRP, 7
indicators of diet such as antioxidants, malnutrition, albumin, anemia, % of
daily intake kilocalorie from carbohydrate, fat, and protein and 3 health
behaviors, i.e., tobacco use, alcohol use, and doing regular exercise, are
added to the model to examine how this affects the weight indicators and how
these indicators play a role as mediators in the relationship between body

54
weight and mortality. Change in the effect of weight indicates that the addition
of the variables explains at least part of the relationship between weight and
mortality. Since three measures of LDL cholesterol, triglycerides, and
homocysteine have a smaller number of respondents in the NHANES III, those
three measures were excluded in running regression models. Those
regression analyses were conducted using both SAS and SUDAN to reflect
complex, multi-stage, stratified, clustered samples of NHANES III. Other
chapters including chapter IV, V, and VI use only SAS for analyses.

55
Chapter IV: Characteristics of Samples
This chapter focuses on the characteristics of the samples analyzed in
each chapter. Japanese older adults and American older adults are used in
both chapter V and VI, and chapter V analyzes those datasets by sex. Thus,
this chapter describes the datasets used in the analysis by chapter rather than
by individual dataset.
4.1 Characteristics of the NUJLSOA (70+) and the LSOA II (70+) by Sex
The characteristics of older men and women in the Japanese data
(NUJLSOA) and in the American data (LSOA II) are presented in Table 4.1.
While over two-thirds of older men have normal weight in Japan, more than
one half of American older men are overweight. However, the prevalence of
underweight American men is very small (2.4%), but about 15% of Japanese
men fall into the underweight category. The weight distribution of females
does not differ much from that of men in both countries. The prevalence of
underweight among Japanese women (14.8%) is more than two times higher
than that among American women (6.3%), but the prevalence of overweight is
the other way around (Japanese women: 19.1%, American women: 45.1%).
This study uses 6 health outcomes which will be dependent variables at
the end of chapter: ADL functioning difficulties, cancer, stroke, diabetes, heart
disease, and arthritis. For both men and women, all health outcomes show a
higher prevalence in the U.S. than in Japan except for stroke. The prevalence
of arthritis in American men (49.5%) and women (63.8%) is more than three

56
Table 4.1
Characteristics of Older Adults in Japan and the United States (70+)
a
: Percents and Means (SD)
Male Female
NUJLOSA LSOA II NUJLOSA LSOA II
N= 1532 N=3774 N=2301 N=5703
Dependent variable
Weight categorization
b

Underweight (<18.5) 14.5 2.4 14.8 6.3
Normal weight (18.5 to 24.9) 71.5 44.5 66.1 48.6
Overweight (25.0+) 14 53.1 19.1 45.1

Health outcomes
ADL functioning difficulties 12.1 23.6 13.7 31.5
Cancer 5.3 23.4 3.7 16.7
Stroke 13.4 10.4 8.0 7.9
Diabetes 10.0 12.9 9.3 11.5
Heart disease 18.2 30.2 18.3 25.0
Arthritis 12.1 49.5 20.9 63.8

Independent variables
Demographic and Socioeconomic factors
Age -mean (SD) 76.3 (4.8) 76.7 (5.5) 77.1 (4.9) 77.7 (5.9)
Female n.a n.a. 59.4 59.9
Married 83.8 77.4 35.9 37.9
Living in rural area 40.1 27.8 38.8 26.8
Education -mean (SD) 9.2 (2.7) 11.4 (3.8) 8.1 (2.2) 11.0 (3.3)
Income -mean (SD)
28,652
(20,013)
29,058
(19,569)
20,261
(14,983)
24,667
(17,871)
Missing income 16.9 22.6 22.4 25.5
Health behaviors
Current smoker 28.6 11.1 5.0 9.2
Heavy drinker 27.5 6.0 9.0 1.2
No regular exercise 21.2 57.0 25.7 62.8
a
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA, 70+),
US Longitudinal Study on Aging II (LSOA II)
b
Three weight categories will be a dependent variable in multinomial logistic regression first, and then
underweight and overweight are going to be independent variables (reference category: normal
weight) to predict six health outcomes in logistic regression.

57
times higher than that in Japanese men (12.1%) and women (20.9%),
respectively. The prevalence of cancer is also much higher in the U.S. in both
sexes (men: 23.4%, women; 16.7%) than in Japan (men: 5.3%, women: 3.7%).
Also, heart disease occurs more in American men (30.2%) than in Japanese
men (18.2%). For women, it is also higher in American women (25.0%) than
in Japanese women (18.3%). However, Japanese men (13.4%) show higher
prevalence in stroke, compared to American men (10.4%). For women, the
prevalence of heart disease is almost the same in Japan (8.0%) and the U.S
(7.9%).
Mean age is slightly lower in the Japanese sample than in the American
sample for both sexes. The prevalence of females in both samples is almost
same (Japan: 59.4%, U.S.: 59.9%). However, the prevalence of married men
in Japan (83.8%) is higher than that in American men (77.4%), but for women,
American women (37.9%) are slightly more likely to be married than Japanese
women (35.9%) in these samples. In terms of living area, Japanese men and
women are more likely to be living in a rural area than American men and
women.
This sample contains two indicators of socioeconomic status: education
and income. The mean years of completed education are higher in the U.S.
than in Japan in both men and women. The mean education of American men
(11.4 years) is almost two years higher than that of Japanese men (9.2 years).
The mean education difference is almost three years between American

58
women (11.0 years) and Japanese women (8.1 years). Mean income is also
higher in the U.S. than in Japan in both sexes. Mean family incomes of
American men and Japanese men do not differ by much, but mean income of
American women is more than $4,000 higher than that of Japanese women.
These mean incomes were calculated after replacing missing values of
income with the mean as described in chapter IV. Since missing incomes
were over 20% except for that of Japanese men, missing income variables will
be controlled in the equation.
While Japanese men are more likely to be current smokers (28.6%)
than American men (11.1%), American women (9.2%) have a higher
prevalence of current smokers than Japanese women (5.0%). About 28% of
Japanese men are heavy drinkers, but 6% of American men are heavy
drinkers in the samples. Only about 1% are heavy drinkers among American
women, but among Japanese women heavy drinking (9.0%) is almost three
times higher than among their American counterparts. As expected, the
prevalence of American men (57%) and women (62.8%) who did not regularly
exercise is much higher than that of Japanese men (21.2%) and women
(25.7%), which may be related to the prevalence of overweight.
Overall, American men and women have higher education and higher
income, but poorer health compared to Japanese men and women.

59
4.2 Characteristics of the NUJLSOA (70+), the LSOA (70+), and
the HRS (51-61)
The weight distributions (BMI) of three samples are presented in Figure
4.1. Japanese older people are lighter than either group of Americans as
indicated by the leftward shifting of their weight distribution. In addition, there
is less dispersion in the BMI of Japanese older people than either group of
Americans. In the U.S., persons 70 years and older have somewhat lower
weight than those aged 51-61.

Figure 4.1
BMI in NUJLSOA, LSOA II, and HRS
0
2
4
6
8
10
12
14
12
15
18
21
24
27
30
33
36
39
42
45
48
51
54
57
60
76
BMI
%
NUJLSOA (70+)
LSOA II (70+)
HRS (51-61)

60
Table 4.2 presents the characteristics of three samples. Very few
people in the 51-61 age group are underweight (1.3%) in the U.S., but quite a
lot of Japanese older adults are underweight (14.7%) compared to the two
American groups. Meanwhile, very few Japanese older adults are obese
(1.4%), but a fourth of people aged 51-61 in the U.S. are obese (21.9%). The
prevalence of being overweight in the 51-61 age group and among older
Americans is about 41% and 36%, respectively, which means about half of
older Americans and two-thirds of middle-aged Americans are overweight or
obese.
Table 4.2
Characteristics of the Three Samples
a
: Percents and Means (SD)
NUJLSOA (70+) LSOA II (70+) HRS (51-61)
N= 3833 N=9447 N=8862
Dependent variable
% death 17.8 17.1 3.2

Independent variable
Age -mean (SD) 76.7 (4.9) 77.3 (5.8) 55.9 (3.1)
Female 59.059.9 52.5

Underweight (<18.5) 14.7 4.7 1.3
Normal weight (18.5 to 24.9) 68.4 46.9 36.1
Overweight (25 to 29.9) 15.6 35.5 40.7
Obesity (30+) 1.4 12.8 21.9

ADL difficulties 12.4 18.7 8.7
Cancer 4.419.4 5.5
Stroke 10.28.9 2.6
Diabetes 9.612.1 10.0
Heart disease 18.2 27.1 12.9
Depression 16.9 23.9 28.8

Current smoker 14.7 9.9 27.0
No regular exercise 23.9 60.5 26.9
a
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA),
US Longitudinal Study on Aging II (LSOA II), Health and Retirement Study 1994 (HRS)

61
Figure 4.2. shows the percentage dying by four WHO BMI categories in
each sample. The highest mortality of all three groups occurs among people
with underweight, but the lowest mortality in all three groups occurs among the
overweight, although the rate is almost same for the normal, overweight, and
obese people aged 51-61. It is interesting that the percentage dying among
the underweight older Japanese (33%) is even higher than that in underweight
older Americans (30%).

Figure 4.2
Percent Mortality by WHO BMI Category
0
5
10
15
20
25
30
35
Underweight Normal Overweight Obesity
BMI
Percent Dying
NUJLSOA (70+)
LOSA II (70+)
HRS (51-61)

62
The total percentage dying in the four-year interval in older Japanese
(17.8%) is almost same as that in older Americans (17.1%), but it is much
lower in the American group 51-61 years old (3.2%) (see Table 4.2).
Mean age is higher in the older American sample (77.3) than in the
older Japanese sample (76.7). Mean age for middle-aged Americans is 55.9.
The proportion of females is almost the same among older Japanese (59%)
and older Americans (60%), but the middle-aged group of American has about
53% female.
All health variables of older Americans including depression are higher
in prevalence than those of older Japanese except for stroke. For the middle-
aged group, the prevalence of one or more ADL difficulties, cancer, stroke,
and heart disease is lower than for the other two older groups. This is
because the middle aged American group is younger than the other two
groups. However, middle-aged Americans have an even higher prevalence of
diabetes than older Japanese and the prevalence of depression among middle
aged Americans is highest among those three groups.
The proportions currently smoking are 9.9%, 14.7%, and 27.0% for
older Americans, older Japanese, and the middle-aged Americans,
respectively. A lot of older Americans do not regularly exercise (60.5%) and
the proportion of people who do not regularly exercise is the lowest in the
older Japanese group (23.9%), followed by the middle-aged Americans
(26.9%).

63
Figure 4.3 presents percent mortality by WHO BMI category only for
never smokers to test sensitivity regarding the link between underweight and
smoking when predicting death. Basically, reversed J-shapes in relationship
between BMI and percentage dying for older Japanese and older Americans
are the same as Figure 4.2 although the percentage dying among underweight
people became lower than that in Figure 4.2. Also, the lowest mortality occurs
among overweight people except for the middle-aged Americans. For the
middle-aged Americans who have never smoked, the lowest mortality occurs

Figure 4.3
Percent Mortality by WHO BMI Category for Never Smokers
0
5
10
15
20
25
30
35
Underweight Normal weight Overweight Obesity
BMI Categories
Percent of Dying
NUJLSOA (70+)
LOSA II (70+)
HRS (51-61)

64
among people with normal weight and the percentage dying among
underweight people is even less than 5%, which is quite different from that in
Figure 4.2, including people who have ever and never smoked.
Table 4.3 shows characteristics of the three samples only for never
smokers. The total percentages of decedents among older Japanese, older
Americans, and middle-aged Americans are 16.5%, 14.8%, and 1.5%,
respectively. These percentages dying are slightly lower than those for three
samples including ever and never smokers in Table 4.2. Mean age for never
smokers is a little bit higher than that in each sample including all people.
Also, the proportions of females for older Japanese, older Americans, and the
middle-aged Americans are about 85%, 74%, and 66%, respectively, which is
quite higher than when all people were included in Table 4.2. This is because
men are more likely to smoke than women, so many of them were not
selected for the sample consisting of never smokers.
Due to selecting never smokers, the prevalence of overweight and
obese people among older Japanese and the prevalence of obese people in
both older and middle-aged Americans have increased about 1 %. Overall,
the prevalence of diseases has decreased a little bit in all three samples.
However, the prevalence of ADL difficulties in older Japanese and older
Americans, the prevalence of depression in older Japanese, and the
prevalence of diabetes in older Americans are higher than those of the total

65
Table 4.3
Characteristics of the Three Samples
a
for Never Smokers : Percents and Means (SD)
NUJLSOA (70+) LSOA II (70+) HRS (51-61)
N= 1926 N=4275 N=3230
Dependent variable
% death 16.5 14.8 1.5

Independent variable
Age -mean (SD) 77.1 (5.0) 78.1 (6.0) 56.0 (3.1)
Female 84.874.265.7

Underweight (<18.5) 14.4 4.7 0.9
Normal weight (18.5 to 24.9) 66.8 47.2 36.4
Overweight (25 to 29.9) 17.0 34.1 39.4
Obesity (30+) 1.8 14.0 23.3

ADL difficulties 12.6 19.3 7.5
Cancer 3.516.35.3
Stroke 8.78.02.1
Diabetes 8.612.98.8
Heart disease 17.4 24.6 9.9
Depression 18.3 23.9 27.7

No regular exercise 25.0 60.2 23.7
a
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA),
US Longitudinal Study on Aging II (LSOA II), Health and Retirement Study 1994 (HRS)

samples. Older Americans who have never smoked do not exercise regularly
(60.2%), as seen in Table 4.3.
Overall, the BMI of the two American groups is higher than that of older
Japanese. However, the highest death rates by four BMI categories occur
among the underweight in all three samples. Interestingly, when selecting
never smokers, the highest death rate by four BMI categories occurs among
people with normal weight for the middle-aged Americans. As indicated by
most health variables, older Americans have poorer health than older

66
Japanese except for stroke, regardless of selecting never smokers. However,
the prevalence of depression in the 51-61 age group of Americans is highest
among the three groups and this group has an even higher prevalence of
diabetes than the older Japanese group, regardless of selecting never
smokers. The highest proportions of current smokers are among the middle-
aged Americans and people not doing regular exercise are among older
Americans.
4.3 Characteristics of the NHANES III (40+)
The percentage of the U.S. population 40 and over with high risk level
for specific biological markers by WHO BMI category in the NHANES III,
adjusting for age and gender, is presented in Table 4.4. High diastolic blood
pressure (DBP I) shows the highest prevalence among obese people (11.8%),
but low diastolic blood pressure (DBP II) indicates the highest prevalence
among underweight people (11.6%), as expected. Overweight and obese
people show high prevalence of high systolic blood pressure (SBP) as 24.3%
and 29.7%, respectively.
Both overweight and obese people show more than 30% prevalence in
high total cholesterol (total cholesterol I), and overweight people have the
highest prevalence (32.4%) among four WHO BMI categories. However, the
highest prevalence of low total cholesterol (total cholesterol II) occurs among
underweight people (16.3%).

67
Table 4.4
Percentage of Population 40 and Older at High Risk for Biological Markers by WHO BMI Category in the NHANES III Controlling for
Age and Gender
% High Risk
Indicator High-Risk Cut Point N Underweight Normal Overweight Obese
Diastolic BP I >= 90 mmHg 9233 2.9 4.8 8.1 11.8
Diastolic BP II <= 60 mmHg 9233 11.6 10.7 8.2 7.1
Systolic BP >= 140 mmHg 9231 22.1 20.8 24.3 29.7
Total cholesterol I >= 240 mg/dl 9233 9.3 23.8 32.4 30.9
Total cholesterol II <= 160 mg/dl 9233 16.3 8.4 7.2 6.0
HDL cholesterol < 40 mg/dl 9155 12.0 15.7 26.2 34.8
Fibrinogen > 400 mg/dl 9673 19.2 14.4 13.9 18.4
C-reactive protein 3.0 < <= 10 mg/dl 8203 12.5 16.0 25.9 38.9
Glycated hemoglobin > 6.4% 9153 8.9 3.9 6.3 14.6
White blood cell count > 10.8 X10
3
/uL 9673 12.5 8.2 7.4 9.7
Fasting LDL cholesterol >= 160 mg/dl 3958 17.7 20.4 28.2 27.3
Fasting triglycerides >= 200 mg/dl 4091 12.8 10.7 22.7 30.6

68
Obese people have the highest prevalence of low HDL cholesterol
(34.8%), high C-reactive protein (38.9%), high glycosylated hemoglobin
(14.6%), and high fasting triglycerides (30.6%). Among those biomarkers, low
HDL cholesterol (26.2%), high C-reactive protein (25.9%), and high fasting
triglycerides (22.7%) are also second highest among overweight people.
Underweight people have the highest prevalence of high fibrinogen (19.2%)
and high white blood cell count (14.3%).
Table 4.5 shows the percentage of the population aged 40 and older at
high risk for behavioral factors by WHO BMI category in the NHANES III,
controlling for age and gender. Interestingly enough, underweight people
have the highest prevalence of adverse levels of albumin (22.9%), malnutrition
(51.2%), anemia (16.4%), antioxidants (31.0%), % daily intake from
carbohydrate and protein (14.2%), current smokers (48.9%), heavy drinkers
(15.6%), and no regular exercise (33.2%). All indicators of poor diet have the
highest prevalence among the underweight except for serum homocysteine, %
daily intake from carbohydrate, and % daily intake from fat, which are related
to poor nutrition. However, even % daily intake from fat show high prevalence
among underweight people with 23.7%. In addition, underweight people are
more likely to be heavy drinkers than people in other BMI categories (15.6%),
which is unexpected because overweight and obese people are usually more
likely to be heavy drinkers in previous research. Meanwhile, about 29% of
obese people take in more than 40% of daily Kilocalories from fat.

69
Table 4.5
Percentage of Population 40 and Older at High Risk for Behavioral Factors by WHO BMI Category in the NHANES III Controlling for
Age and Gender
% High Risk
Indicator High-Risk Cut Point N Underweight Normal Overweight Obese
Albumin < 3.8 mg/dl 9673 22.9 11.1 11.0 15.8
Serum Homocysteine >= 15 umol/l 4009 6.7 12.1 9.4 9.2
Malnutrition Any deficiency from iron, B12,
folate
5751 51.2 31.1 26.5 30.9
Anemia
F<130, M<120
9673 16.4 8.3 6.6 6.4
Antioxidants ≥ 3 low quartiles out of vitamin A,
C, E, selenium, and lycopene
9223 31.0 14.2 13.1 18.5

% daily intake from carbohydrate > 65% 9310 5.4 10.9 9.7 8.8
% daily intake from fat > 40 % 9310 23.7 22.5 22.3 28.9
% daily intake from protein < 10% 9310 14.2 8.2 7.9 6.3

Current smoker Ref- not current 9693 48.9 28.0 20.0 18.0
Heavy drinker Ref- < 3 drinks per day 9366 15.6 14.0 13.5 15.0
No regular exercise Ref- exercise 9693 33.2 23.3 25.3 29.7

70
Table 4.4 and 4.5 have a smaller number of respondents reporting
biomarkers which require fasting LDL cholesterol and triglycerides. These
measures were reported only for participants fasting at least 6 hours before
being tested in the morning. Serum homocysteine also has a smaller N
because it was not measured until the second half of NHANES III (1991-
1994). Therefore, all three measures will be removed from the model.
Overall, obese people are more likely to have high DBP I, high SBP,
low HDL cholesterol, high CRP, high glycasylated hemoglobin, and high
fasting triglycerides. Also, they are more likely to have high fat intake.
Underweight people have a higher prevalence of some high risk biomarkers
and health behaviors related to poor nutrition as mentioned above. In
addition, they are more likely to be current smokers, heavy drinkers, and do
not tend to exercise regularly.
4.4 Conclusions
While quite a few American older men are underweight, more than a
half of American older men are overweight (BMI ≥ 25) in the LOSA II. In
contrast, only about 14% of the Japanese older men are overweight in the
NUJLSOA. This difference in body weight is almost the same for women.
American older men and women have more years of completed education
than Japanese older men and women. Income differences are similar to those
in years of education. American men show poorer health status than
Japanese men except for the prevalence of stroke. Japanese older men are

71
more likely to be current smokers, heavy drinkers, but do more regular
exercise than American older men. However, Japanese older women are less
likely to be current smokers, but heavier drinkers and do more regular exercise
than American older women.
Japanese older adults (NUJLOSA) have a lower weight distribution with
less dispersion than both the 51-61 age group (HRS) and older adults in the
U.S. (LSOA II). Underweight people have the highest death rate and
overweight people have the lowest death rate in all three groups before
selecting never smokers. Meanwhile, for never smokers, people with normal
weight among the middle-aged Americans have the lowest death rate,
although overweight people have the lowest death rate in the other two older
groups. While only a few people are underweight among the middle-aged and
older adults in the U.S., very few Japanese older adults are obese. The
middle-aged Americans have the lowest prevalence of the health problems
examined except for cancer, diabetes, and depression for the whole sample
and for never smokers. Also, they are more likely to be current smokers, but
American older adults do less regular exercise.
In the NHANES III, it should be pointed out that underweight people
have a high risk of low DBP, low total cholesterol, high fibrinogen, high white
blood cell count, and indicators of poor diet. Also, they are more likely to be
current smokers, heavy drinkers, and less likely to exercise regularly.

72
Chapter V: Factors Affecting Body Weight and the Correlates of It With
Health Outcomes in Two Different Countries
This chapter attempts to answer research questions regarding what
factors determine being underweight, normal weight, or overweight among
Japanese older adults, and American older adults by sex and whether
underweight and overweight are differently associated with health outcomes
including ADL functioning difficulties, cancer, stroke, diabetes, heart disease,
and arthritis in these two countries. The NUJLOSA for Japanese older adults
and the LSOA II for the U.S. older adults were analyzed and I ran multinomial
logistic regression because the dependent variable has three BMI categories-
under (BMI<18.5), normal (18.5 ≤ BMI ≤ 24.9: reference), and overweight
(BMI ≥ 25.0). In the second section, those BMI categories (reference: normal
weight) are used as independent variables to predict six health conditions
mentioned above.
5.1 Factors Affecting Body Weight Among Japanese and American Older
Adults 70 and Over by Sex
5.1.1 Factors Affecting Underweight and Overweight for Males
Tables 5.1 and 5.2 present results from the multinomial logistic
regression analyses relating independent variables to the three BMI categories
for men. Results from the baseline model (Model 1) indicated that while older
age was associated with higher odds of being underweight, older age was

73
Table 5.1
Odds Ratios (95% Confidence Intervals) From Multinomial Logistic Models of Three BMI Levels for Males Including Demographic and
Socioeconomic Factors
Model 1
NUJLSOA (n=1413) LSOA II (n=3541)
Underweight
a
Overweight
a
Underweight
a
Overweight
a

Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Demographic and Socioeconomic factors
Age 1.06*** (1.03, 1.09) 0.95* (0.92, 0.99) 1.06** (1.02, 1.09) 0.92*** (0.91, 0.94)
Married 0.97 (0.63, 1.49) 0.90 (0.57, 1.42) 0.80 (0.49, 1.30) 1.25* (1.05, 1.48)
Living in rural area 1.10 (0.79, 1.53) 0.84 (0.59, 1.19) 1.07 (0.65, 1.75) 1.07 (0.91, 1.24)
Education 0.98 (0.92, 1.05) 1.00 (0.94, 1.06) 0.95 (0.90, 1.01) 0.98
+
(0.96, 1.00)
Income 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00
+
(1.00, 1.00)
Missing income 1.22 (0.81, 1.84) 0.83 (0.51, 1.36) 1.16 (0.69, 1.93) 0.96 (0.81, 1.13)
Health behaviors
Current smoker
Heavy drinker
No regular exercise

-2 Log L 1990.18 5419.80

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= normal weight

74
Table 5.2
Odds Ratios (95% Confidence Intervals) From Multinomial Logistic Models of Three BMI Levels for Males Adding Health Behaviors in Model1
Model 2
NUJLSOA (n=973) LSOA II (n=3447)
Underweight
a
Overweight
a
Underweight
a
Overweight
a

Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Demographic and Socioeconomic factors
Age 1.05* (1.01, 1.09) 0.96+ (0.92, 1.01) 1.07*** (1.03, 1.11) 0.91*** (0.90, 0.93)
Married 1.01 (0.60, 1.71) 0.83 (0.47, 1.44) 0.81 (0.48, 1.36) 1.17
+
(0.99, 1.40)
Living in rural area 1.23 (0.82, 1.83) 0.81 (0.52, 1.25) 1.17 (0.71, 1.94) 1.12 (0.95, 1.30)
Education 0.99 (0.92, 1.07) 0.96 (0.89, 1.03) 0.97 (0.91, 1.04) 0.98 (0.96, 1.00)
Income 1.00 (1.00, 1.00) 1.00* (1.00, 1.00) 1.00 (1.00, 1.00) 1.00* (1.00, 1.00)
Missing income 0.98 (0.59, 1.62) 0.86 (0.48, 1.54) 0.93 (0.53, 1.61) 0.96 (0.81, 1.14)
Health behaviors
Current smoker 1.27 (0.82, 1.96) 0.58* (0.35, 0.97) 1.86* (1.01, 3.42) 0.41*** (0.33, 0.52)
Heavy drinker 0.51* (0.31, 0.85) 1.29 (0.85, 1.96) 0.90 (0.30, 2.68) 1.52* (1.13, 2.06)
No regular exercise 1.58* (1.03, 2.44) 0.75 (0.43, 1.29) 2.23** (1.31, 3.79) 1.11 (0.96, 1.28)

-2 Log L 1346.13 5155.17

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= normal weight

75
Table 5.3
The Overall Model Fit Test for Japanese and American Men
NUJLSOA LSOA II
Model fit test Model 1 Model 2 Model 1 Model 2
X
2
(df) 39.32 (12) 52.62 (18) 202.64 (12) 288.39 (18)

LR test Model 2 vs Model 1 (df) 13.30 (6)* 85.75 (6)***
* < .05 ** < .01 *** < .001

related to lower odds of being overweight for both Japanese and American
older men. A 1 year increase in age was associated with approximately 6%
higher odds of being underweight rather than being normal weight.
Meanwhile, with each 1 year age increase, the odds of being overweight
decreased 5% and 8% for Japanese older men and American older men
relative to the likelihood of having normal weight. I also found that being
married was associated with a 25% increase in the odds of being overweight
for American older men, but marital status does not have any effect on weight
for Japanese older men.
Next, in Model 2, health behaviors are added into the Model 1 to
examine the relationship between health behaviors and BMI. As expected,
current smoking is linked to a 42% and a 59% reduction in the odds of being
overweight for Japanese older men and American older men, respectively.
Also, the odds of being underweight are about 1.9 times as high for U.S.
current smokers as nonsmokers. For men, heavy drinkers are less likely to be
underweight in Japan (OR: 0.51; 95% C. I.: 0.31-0.85.), whereas heavy

76
American drinkers are more likely to be overweight (OR: 1.52; 95% C. I.: 1.13-
2.06). People who do not exercise regularly are more likely to be underweight
for both Japanese older men (OR: 1.58; 95% C.I.: 1.03-2.44) and the U.S.
older men (OR: 2.23; 95% C.I.: 1.31-3.79). Overall, poor health behaviors are
more related to being underweight for Japanese men, whereas for American
older men they are linked to both being underweight and overweight.
Finally, I tested the fit of the models by comparing the likelihood ratio to
explore whether adding the health behaviors significantly improved the overall
fit of the model (see Table 5.3.). Expectedly, for Japanese men, adding health
behavior terms in Model 2 improved the model fit as indicated by the fact that
chi-square in the likelihood ratio tests for Model 2 versus Model 1 is significant
at .05 level (likelihood ratio test Model 2 vs Model 1: X
2
(df) =13.30 (6), p>.05).
Therefore, health behaviors for Japanese older adults need to be added in the
model, implying current smoking is associated with being overweight, but
heavy drinking and not exercising regularly are related to being underweight
for Japanese older men. Similar to the Japanese older men, including
behavioral variables such as current smoking, heavy drinking, and not
exercising regularly, improved the overall fit of the model for American older
men (likelihood ratio test Model 2 vs Model 1: X
2
(df) =85.76 (6), p<.001). It is
therefore concluded that whereas heavy drinking is associated with being
overweight, no regular exercise is linked to being underweight among

77
American older men. Current smoking is associated with both being
underweight and being overweight among American older men.
5.1.2 Factors Affecting Underweight and Overweight for Females
Tables 5.4 and 5.5 present the results from multinomial logistic
regression analyses of three BMI levels for females. Table 5.4 is the baseline
model indicating that with 1 year age increase, the odds of being underweight
and being overweight for both Japanese and American older women were
almost the same as what they were for Japanese and American older men in
Table 5.1. The interesting result in Table 5.4, however, is the effect of
education for both countries. A 1 year increase in education was associated
with approximately 6% lower odds of being underweight in the U.S. and being
overweight for both countries.
Next, I considered health behaviors in Model 2 by adding them into
Model 1. The effects of age on being underweight and being overweight in
both countries still remain even after controlling for health behaviors. Also, the
effects of education on being overweight in Japan and on being both
underweight and overweight in the U.S. are almost the same as Table 5.4.
after including health behaviors.
For American women, current smokers are more likely to be
underweight (OR: 2.76; 95% C.I.: 2.00-3.79), but less likely to be overweight
than nonsmokers (OR: 0.56; 95% C.I.: 0.45-0.69). American women not
exercising regularly are more likely to be overweight than those exercising

78
Table 5.4
Odds Ratios (95% Confidence Intervals) From Multinomial Logistic Models of Three BMI Levels for Females Including Demographic and
Socioeconomic Factors
Model 1
NUJLSOA (n=1980) LSOA II (n=5212)
Underweight
a
Overweight
a
Underweight
a
Overweight
a

Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Demographic and Socioeconomic factors
Age 1.07*** (1.04, 1.10) 0.93*** (0.91, 0.96) 1.04*** (1.02, 1.06) 0.95*** (0.94, 0.96)
Married 1.13 (0.82, 1.56) 0.92 (0.70, 1.21) 1.17 (0.90, 1.51) 0.96 (0.85, 1.09)
Living in rural area 1.01 (0.76, 1.35) 0.80 (0.62, 1.05) 0.84 (0.64, 1.10) 1.01 (0.89, 1.15)
Education 1.02 (0.96, 1.09) 0.94* (0.89, 1.00) 0.94*** (0.91, 0.97) 0.93*** (0.91, 0.94)
Income 1.00 (1.00, 1.00) 1.00 (1,00, 1.00) 1.00 (1.00, 1.00) 1.00
+
(1.00, 1.00)
Missing income 1.01 (0.71, 1.43) 1.15 (0.85, 1.57) 0.92 (0.70, 1.20) 0.87* (0.76, 1.00)
Health behaviors
Current smoker
Heavy drinker
No regular exercise

-2 Log L 2951.11 8953.49

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= normal weight

79
Table 5.5
Odds Ratios (95% Confidence Intervals) From Multinomial Logistic Models of Three BMI Levels for Females Adding Health Behaviors in Model 1
Model 2
NUJLSOA (n=1806) LSOA II (n=5118)
Underweight
a
Overweight
a
Underweight
a
Overweight
a

Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Demographic and Socioeconomic factors
Age 1.06*** (1.03, 1.09) 0.94*** (0.91, 0.97) 1.06*** (1.04, 1.08) 0.94*** (0.93, 0.96)
Married 1.14 (0.82, 1.59) 0.87 (0.65, 1.16) 1.22 (0.93, 1.59) 0.93 (0.81, 1.05)
Living in rural area 1.05 (0.78, 1.42) 0.78
+
(0.59, 1.03) 0.84 (0.64, 1.11) 1.01 (0.88, 1.15)
Education 1.03 (0.97, 1.10) 0.94* (0.89, 1.00) 0.95** (0.91, 0.98) 0.93*** (0.92, 0.95)
Income 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00* (1.00, 1.00)
Missing income 0.99 (0.69, 1.43) 1.22 (0.89, 1.68) 0.91 (0.69, 1.20) 0.86* (0.75, 0.99)
Health behaviors
Current smoker 1.32 (0.69, 2.52) 1.14 (0.63, 2.05) 2.76*** (2.00, 3.79) 0.56*** (0.45, 0.69)
Heavy drinker 1.11 (0.67, 1.84) 1.11 (0.73, 1.70) 0.95 (0.33, 2.74) 1.00 (0.59, 1.72)
No regular exercise 1.45* (1.05, 2.00) 0.96 (0.70, 1.33) 1.27
+
(0.98, 1.63) 1.45*** (1.28, 1.63)

-2 Log L 2691.43 8679.59

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= normal weight

80
Table 5.6
The Overall Model Fit Test for Japanese and American Women
NUJLSOA LSOA II
Model fit test Model 1 Model 2 Model 1 Model 2
X
2
(df) 67.91 (12) 66.14 (18) 220.62 (12) 326.61 (18)

LR test Model 2 vs Model 1 (df) -1.77 (6) 105.99 (6)***
* < .05 ** < .01 *** < .001

regularly (OR: 1.45; 95% C.I.: 1.28-1.63). However, for Japanese older
women, exercising regularly, the only related health behavior, is linked to an
increase of 45% in the odds of being underweight.
I tested the overall fit of the models which is presented in Table 5.6.
For Japanese older women, the likelihood ratio test for Model 2 versus Model
1 is not significant. So, unlike Japanese older men, it seems that Model 1
shows the best fit for Japanese older women because chi-squares in the
likelihood ratio tests for Model 2 vs Model 1 is not significant at .05 level,
suggesting health behaviors are not additionally associated with BMI for
Japanese older women (likelihood ratio test Model 2 vs Model 1: X
2
(df) =-1.77
(6)). For American older women, as shown in Table 5.6, including health
behaviors improved the overall fit of the model, suggesting current smoking
and no regular exercise may mediate between cancer and being underweight,
and between stroke and being overweight.

81
5.2 The Correlates of Body Weight With Health Outcomes Among Japanese
and American Older Adults 70 and Over by Sex
5.2.1 The Correlates of Underweight and Overweight With Health Outcomes
for Males
I ran the binary logistic regression of health outcomes including ADL
functioning difficulties, cancer, stroke, diabetes, heart disease, and arthritis to
examine whether underweight and overweight are associated with each health
status when controlling for all other variables, i.e., age, marital status, rural
residence, education, income, missing income, smoking, drinking, and
exercise. Table 5.7 presents the odds ratios of being underweight and
overweight for Japanese and American older men.. As shown in the table,
there are no relationships between BMI and ADL difficulties and presence of
diseases except for diabetes among Japanese older men. Unexpectedly,
being overweight decreases the odds of having diabetes by 58% among
Japanese older men, whereas it increases the odds of having diabetes by only
1% among American older men. In the U.S., the odds of having one or more
ADL difficulties and having heart disease are about 3 times and 1.9 times as
high for underweight men as they are for normal weight men, respectively.
Overweight American older men are more likely to have arthritis than
American older men with normal weight (OR:1.47; 95% C.I.: 1.28-1.70).
As a whole, body weight indicated that BMI does not seem to predict
the presence of diseases except for diabetes among Japanese older men, but

82
Table 5.7
Odds Ratios (95% Confidence Intervals) for Effect of Underweight (BMI<18.5) and Overweight (BMI>=25.0) on Health Outcomes
Among Japanese and American Older Men 70 and Over
Male Male
NUJLSOA LSOA II NUJLSOA LSOA II
OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

ADL functioning difficulties
a
Diabetes
a

Underweight (BMI<18.5) 1.49 (0.86, 2.60) 2.97*** (1.83, 4.83) 0.65 (0.32, 1.30) 0.67 (0.26, 1.71)
Normal weight (18.5<=BMI<=24.9: reference) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI>=25.0) 0.84 (0.39, 1.81) 1.02 (0.86, 1.22) 0.42* (0.19, 0.94) 1.01*** (1.29, 1.99)
-2Log L 522.97 3492.38 559.54 2616.64
N 959 3425 944 3438

Cancer
a
Heart Disease
a

Underweight (BMI<18.5) 1.47 (0.70, 3.13) 1.27 (0.76, 2.14) 0.79 (0.46, 1.34) 1.88** (1.18, 3.01)
Normal weight (18.5<=BMI<=24.9: reference) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI>=25.0) 0.77 (0.30, 2.00) 1.00 (0.85, 1.18) 1.01 (0.61, 1.68) 1.03 (0.88, 1.20)
-2Log L 368.39 3755.5 798.36 4228.1
N 945 3438 946 3421

Stroke
a
Arthritis
a

Underweight (BMI<18.5) 0.71 (0.39, 1.29) 1.39 (0.73, 2.65) 0.57 (0.29, 1.11) 1.10 (0.69, 1.75)
Normal weight (18.5<=BMI<=24.9: reference) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI>=25.0) 1.26 (0.70, 2.26) 0.97 (0.77, 1.22) 1.15 (0.64, 2.07) 1.47*** (1.28, 1.70)
-2Log L 617.74 2240.9 624.87 4687.02
N 949 3434 939 3401
*< .05 **< .01 ***< .001
a
Controlled for age, marital status, rural residence, education, income, missing income, smoking, drinking, and exercise

83
it seems to predict the presence of ADL difficulties, diabetes, heart disease,
and arthritis among American older men.
5.2.2 The Correlates of Underweight and Overweight With Health Outcomes
for Females
Table 5.8 shows odds ratios and 95% confidence interval for effects of
underweight and overweight on same health status variables among Japanese
and American older women.
Overweight women are more likely to have arthritis (OR: 1.68; 95% C.I.:
1.24-2.26) than women with normal weight, which is the only significant
disease in Japan. The relationship between being overweight and arthritis
among American older women is not much different from what happens to
Japanese older women. Unlike in Japan, underweight and overweight women
are more likely to have one or more ADL functioning difficulties than normal
weight women in the U.S. (OR: 1.63; 95% C.I.; 1.26-2.10, OR: 1.71; 95% C.I.:
1.50-1.96, respectively). Also, being overweight is associated with about 2.0
times higher odds of having diabetes than normal weight in American older
women. Being underweight and being overweight have about 1.3 times and
1.2 times higher odds of having heart disease in American older women,
respectively.
Overall, as Japanese men show the odds of diseases and ADL
difficulties in Table 5.8, BMI level is associated with the presence of only one
disease, arthritis, among Japanese women. However, for American women,

84
Table 5.8
Odds Ratios (95% Confidence Intervals) for Effect of Underweight (BMI<18.5) and Overweight (BMI>=25.0) on Health Outcomes
Among Japanese and American Older Women 70 and over
Female Female
NUJLSOA LSOA II NUJLSOA LSOA II
OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

ADL functioning difficulties
a
Diabetes
a

Underweight (BMI<18.5) 1.17 (0.75, 1.82) 1.63*** (1.26, 2.10) 0.54 (0.28, 1.03) 0.61 (0.36, 1.05)
Normal weight (18.5<=BMI<=24.9: reference) 1.00(1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI>=25.0) 1.19 (0.75, 1.89) 1.71*** (1.50, 1.96) 1.47 (0.96, 2.24) 2.02*** (1.68, 2.44)
-2Log L 912.28 5783.02 875.00 3430.02
N 1782 5082 1746 5099

Cancer
a
Heart Disease
a

Underweight (BMI<18.5) 1.42 (0.72, 2.79) 1.29 (0.96, 1.74) 0.84 (0.57, 1.24) 1.34* (1.03, 1.75)
Normal weight (18.5<=BMI<=24.9: reference) 1.00(1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI>=25.0) 1.38 (0.73, 2.61) 1.04 (0.89, 1.22) 1.05 (0.74, 1.49) 1.22** (1.06, 1.40)
-2Log L 512.34 4597.38 1413.08 5584.71
N 1746 5109 1757 5081

Stroke
a
Arthritis
a

Underweight (BMI<18.5) 1.03 (0.61, 1.74) 1.22 (0.83, 1.81) 0.73 (0.50, 1.08) 0.84 (0.66, 1.06)
Normal weight (18.5<=BMI<=24.9: reference) 1.00(1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI>=25.0) 1.13 (0.68, 1.91) 1.01 (0.81, 1.25) 1.68*** (1.24, 2.26) 1.81*** (1.60, 2.05)
-2Log L 764.47 2771.64 1556.48 6451.96
N 1752 5101 1752 5070
*< .05 **< .01 ***< .001
a
Controlled for age, marital status, rural residence, education, income, missing income, smoking, drinking, and exercise

85
being underweight is associated with ADL difficulties and heart disease and
being overweight is related to diabetes and arthritis.
5.3 Discussion
This chapter attempts to explore the relationship between demographic
variables, socioeconomic variables, health behaviors, and BMI for Japanese
and American older adults by sex. Higher BMI has been related to older age
(Ogden et al., 2006), but this chapter found that people are more underweight
and less overweight as age goes up for men and women in both countries.
This is the trend in prevalence observed in section 2.1. Among demographic
variables, marital status and residence area did not have any significant effect
on BMI levels for both countries, which is consistent with the results of
previous research (Schoenborn, 2004).
In addition, education and income were not associated with BMI for
men, although income in Model 2 for both Japanese older men and the U.S.
older men was significant. The odds on income in Model 2 were the same as
the reference (1.00), which means there was almost no income effect on BMI
among Japanese older men and American older men although the odds on
income are significant. With respect to the link of education to BMI, as years
of education goes up, American women are less likely to be underweight and
overweight after controlling for demographic and socioeconomic variables, and
health behaviors. Also, for Japanese women, the same relationship was
found between education and being overweight when including the same

86
variables. This result is consistent with previous research (Zhang & Wang,
2004a), in which there is a stronger inverse association between SES and BMI
among women than among men.
Regarding the relation of health behaviors to BMI levels, no regular
exercise for Japanese women was significant when controlling for
demographic and socioeconomic variables, and health behaviors, but the
overall fit of Model 2 did not improve the model fit compared to Model 1. For
Japanese older men, current smokers are less likely to be overweight, which is
expected given previous research (Mizoue et al., 2006). Heavy drinkers are
less likely to be underweight for Japanese older men. However, Japanese
men who do not exercise regularly are more likely to be underweight, which is
an inconsistent result with previous research (Kruger et al., 2002; Martinez-
Gonzalez et al., 1999; Monda & Popkin, 2005; Sharma, 2007). This is
probably because a lot of Japanese people are too frail to exercise regularly,
which may be explained by the high proportion of underweight (about 15%) in
the NUJLSOA. Therefore, health behaviors including smoking, drinking, and
exercise do not predict BMI among Japanese older women, but predict BMI for
Japanese older men.
For American men and women, including health behaviors in Model 2
significantly improved the model fit compared to Model 1. Current smokers
are expectedly more likely to be underweight and less likely to be overweight,
but heavy drinkers are more likely to be overweight for American men. For

87
American women, current smoking has similar effects predicting being
underweight and overweight. However, men who are not exercising regularly
are more likely to be underweight, but women not exercising regularly are
more likely to be overweight. Probably, the result regarding no regular
exercise for American older men is the same reason as in Japanese older
men.
With respect to the effects of BMI on health outcomes, BMI does not
seem to be a statistically significant predictor of health outcomes in Japan
except for diabetes for Japanese men and arthritis for Japanese women.
Particularly, the fact that overweight Japanese men have unexpectedly lower
odds of having diabetes may imply that the results from previous research
regarding the association between higher BMI and higher prevalence of
diabetes do not appear among Japanese older men. Thus, this work does not
support results from previous research that people with high BMI tend to have
poorer health outcomes (National Heart, Lung, and Blood Institute, 1998). In
addition, this may be because the proportion of overweight among Japanese
men is quite low. Also, unlike American people, diabetes does not occur more
often among heavier people in Japan.
In the U.S., it is interesting that underweight men and women are more
likely to have ADL difficulties and heart disease. As mentioned above, it may
be because American underweight people are very frail, and have difficulties
in performing ADLs. In addition, it should be pointed out that not only heavier,

88
but also thin people may have a higher risk of heart disease compared to
normal weight people in the U.S.

89
Chapter VI: The Effect of Body Mass Index on Mortality in Different
Countries and Age Groups
This chapter focuses on answering the questions about the relationship
between BMI and all-cause mortality among Japanese older adults and U.S.
middle-aged and older adults. Also, I attempt to explore what the optimal BMI
is relative to mortality in those three groups. In addition, whether other
variables such as health status and health behaviors have effects on the
association between BMI and mortality in same way among those three
groups will be addressed, too.
To do this, I ran the logistic regression of death only including age,
female, and BMI, and then added health status and health behaviors in Model
1 to look at whether the effect of BMI on mortality remained even after
controlling for health status and health behaviors. Then, I selected two groups
-underweight people and overweight/obese people- and ran the same model
as I ran before. This is an analysis to investigate whether health status and
health behaviors are associated with death in each group. I performed all
logistic regression for Japanese and American older adults and American
middle-aged including smokers- current and ever- in this section. In the next
section, as discussed in the second chapter, I attempted to control for
cigarette smoking by selecting never smokers for analyses to again examine
the relationship between BMI and mortality.

90
6.1 The Effect of Body Mass Index on Mortality Including Former, Current,
and Never Smokers
6.1.1 The Effect of Body Mass Index on Mortality for Total Samples
Table 6.1 presents the odds ratios and 95% confidence intervals of BMI
categories on the likelihood of dying in the NUJLSOA, the LSOA II, and HRS
controlling for only age and gender. With respect to age, three samples show
that as age increases a year, the odds of dying increase 14% for Japanese
older adults and 9% for both American older adults and the American middle-
aged. The odds of dying are reduced 45%, 42%, and 56% for women relative
to men among Japanese older adults, American older adults, and the U.S.
middle-aged, respectively. Without controlling for any variables other than age
and gender, the odds of dying are about 2.4 times as high for underweight
people as they are for those of normal weight people in Japanese older adults.
Meanwhile, being overweight decreases the odds of dying by 33% compared
to being normal weight among Japanese older adults. The odds of dying for
obese people are about 1.3 times higher than the odds of dying for normal
weight people, however the odds are not significantly higher in Japan.
Similarly, the odds of dying for underweight people are about 1.8 times
higher than the odds of dying for normal weight American older adults.
However, overweight people are less likely to die than underweight people
(OR: 0.80; 95% C. I.: 0.70-0.91). The odds of dying for the obese are not
significantly different from those of normal weight for American older adults.

91
Table 6.1
Odds Ratios and 95% Confidence Intervals (C.I.) of World Health Organization (WHO) BMI
Category
a
on the Likelihood of Dying on Three Samples
b

NUJLSOA (70+) LSOA II (70+) HRS (51-61)
(N= 2870) (N=8929) (N=8818)
Variables Odds Ratio 95% C. I. Odds Ratio 95% C. I. Odds Ratio 95% C. I.
Age 1.14*** (1.11, 1.16) 1.09*** (1.08, 1.10) 1.09*** (1.05, 1.13)
Female 0.55*** (0.44, 0.69) 0.58*** (0.52, 0.65) 0.44*** (0.34, 0.57)
Underweight 2.44*** (1.86, 3.19) 1.84*** (1.46, 2.32) 7.23*** (4.26, 12.29)
Normal 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight 0.67* (0.46, 0.98) 0.80*** (0.70, 0.91) 0.65** (0.49, 0.87)
Obese 1.34 (0.37, 3.51) 0.97 (0.81, 1.17) 0.92 (0.67, 1.27)

-2 Log L 2035.03 7697.88 2401.52
X
2
(df) 256.96 (5) 473.53 (5) 101.29 (5)
a
Underweight: BMI< 18.5; Normal Weight: 18.5<=BMI<=24.9 (the reference category);
Overweight: 25.0<=BMI<=29.9; Obese: BMI >=30.0
b
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA),
US Longitudinal Study on Aging II (LSOA II), Health and Retirement Study 1994 (HRS)
* p<.05 ** p<.01 *** p<.001

Among the U.S. middle-aged, the odds of dying for the underweight is
about 7.2 times higher than the odds of dying for those of normal weight,
which is a very high odds ratio compared to both Japanese and American
older adults. However, being overweight is linked to a 35% reduction in the
odds of dying compared to being normal weight, which is an even larger
reduction in the odds of dying than that for Japanese and American older
adults. As for the other two samples of older adults, obesity is not significantly
associated with the odds of dying in this age range either.
Next, I tried to examine whether the effects of underweight, overweight,
and obesity compared to normal weight on death have changed after

92
controlling for health status and health behaviors (see Table 6.2). First, the
links between age and gender and dying did not change much after adding
health conditions and behaviors in all three samples.
Among Japanese older adults, the odds of dying for underweight and
overweight are not much different from those in Table 6.1, although the odds
of dying for overweight people became insignificant after controlling for health
and behavioral variables. Underweight people are still more likely to die than
normal weight people in Japan (OR: 2.51; 95% C.I.: 1.77-3.55). Having one or
more ADL difficulties is linked to an increase in the odds of dying by a factor of
2.53 compared to no ADL difficulties. As expected, the odds of dying are
about 2.2 times higher for people with cancer than they are for people without
cancer. Also, having diabetes increases the odds of dying by a factor of 1.59.
Stroke, heart disease, and depression are not significantly associated with
death. For Japanese older adults, only no regular exercise is significantly
associated with death. Past and current smoking are not significantly
associated with the likelihood of dying. People who do not exercise regularly
are more likely to die than people who exercise regularly (OR: 1.77; 95% C.I.:
1.26-2.49).
For American older adults, after controlling health conditions and
behaviors, the odds of dying for the underweight were not significantly different
from those of normal weight. This means some of the health conditions and
behavioral variables are associated with underweight, so that controlling for

93
Table 6.2
Odds Ratios and 95% Confidence Intervals (C.I.) of World Health Organization (WHO) BMI
Category
a
on the Likelihood of Dying on Three Samples
b
Controlling for Health Status and Behaviors
NUJLSOA (70+) LSOA II (70+) HRS (51-61)
(N= 2268) (N=7721) (N=8807)
Variables Odds Ratio 95% C. I. Odds Ratio 95% C. I. Odds Ratio 95% C. I.
Age 1.13*** (1.09, 1.16) 1.09*** (1.08, 1.11) 1.07** (1.03, 1.11)
Female 0.63* (0.43, 0.92) 0.61*** (0.53, 0.70) 0.44*** (0.33, 0.57)
Underweight 2.51*** (1.77, 3.55) 1.20 (0.89, 1.63) 3.89*** (2.15, 7.05)
Normal 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight 0.68 (0.42, 1.10) 0.74*** (0.64, 0.86) 0.61** (0.45, 0.82)
Obese 0.72 (0.17, 3.07) 0.82 (0.66, 1.02) 0.62** (0.44, 0.88)

ADL difficulties 2.53*** (1.61, 3.97) 1.73*** (1.47, 2.04) 1.79*** (1.29, 2.49)
Cancer 2.16** (1.20, 3.88) 1.40*** (1.20, 1.63) 2.97*** (2.03, 4.35)
Stroke 1.50 (0.94, 2.39) 1.29* (1.04, 1.59) 1.88** (1.17, 3.03)
Diabetes 1.59* (1.02, 2.48) 1.88*** (1.56, 2.27) 3.02*** (2.23, 4.09)
Heart disease 0.94 (0.65, 1.36) 1.62*** (1.41, 1.86) 1.76*** (1.31, 2.36)
Depression 1.21 (0.84, 1.74) 1.26** (1.08, 1.46) 1.34* (1.03, 1.75)

Former smoker 1.06 (0.69, 1.64) 1.38*** (1.19, 1.61) 1.69** (1.17, 2.43)
Current smoker 1.31 (0.84, 2.03) 2.16*** (1.74, 2.69) 2.85*** (2.00, 4.06)
No regular exercise 1.77*** (1.26, 2.49) 1.53*** (1.33, 1.76) 2.12*** (1.64, 2.74)

-2 Log L 1319.40 5907.22 2121.29
X
2
(df) 204.38 (14) 663.23 (14) 354.64 (14)
a
Underweight: BMI< 18.5; Normal Weight: 18.5<=BMI<=24.9 (the reference category);
Overweight: 25.0<=BMI<=29.9; Obese: BMI >=30.0
b
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA),
US Longitudinal Study on Aging II (LSOA II), Health and Retirement Study 1994 (HRS)
* p<.05 ** p<.01 *** p<.001

those variables reduced the odds ratio of being underweight and resulted in
insignificance. However, the lower odds of dying for the overweight are still
significantly different, and reduced even more from 0.80 to 0.74 by adding
health and behavioral variables. All health variables are associated with a

94
higher death rate for American older adults. Particularly, ADL functioning
difficulties and diabetes have higher odds ratios. The predicted odds of dying
for people with one or more ADL difficulties are about 1.7 times the odds for
people without any ADL difficulties. Having diabetes increases the odds of
dying by a factor of 1.88. Also, the odds of dying for people with heart disease
are 62% higher than the odds of dying for people without heart disease.
Stroke and depression are also associated with high death rate among
American older adults.
In addition, all three health behavioral variables are significantly
associated with death for American older adults. The predicted odds of dying
for former smokers and current smokers are about 1.4 times and 2.2 times the
odds of dying for never smokers, respectively. The odds of dying for people
who do not regularly exercise are 53% higher than the odds of dying for
people who exercise regularly among American older adults.
For the U.S. middle-aged, after controlling for health conditions and
behaviors, the odds of dying for the underweight and overweight still remain
significantly different although those of underweight have been reduced
almost by half, from 7.23 to 3.89. The odds of dying for underweight are about
3.9 times the odds of dying for normal weight. Another point is that the odds
of dying for the obese became significant and even lower than before these
controls were added. However, the odds ratio of dying for being overweight
has not changed much compared to Table 6.1. This means obesity is

95
associated with the health and behavioral variables included among the
middle-aged Americans.
For American older adults, all health and behavioral variables are
significantly related to the likelihood of death. Interesting enough, the odds of
dying for people with cancer and people with diabetes are about 3 times the
odds of dying for people without cancer and people without diabetes, which
are quite high odds ratios compared to those among American older adults.
Also, the odds of dying for people with one or more ADL difficulties, stroke,
heart disease, and depression are 79%, 88%, 76%, and 34 % higher than the
odds of dying for people without ADL difficulties and those diseases,
respectively. The reason why the odds ratios of dying for ADL difficulties and
diseases are quite high compared to the other two groups may be because the
HRS consists of relatively younger people, so that whether middle-aged can
survive to age 70 years may be affected by those health conditions more than
by the survival after that age.
For health behaviors, the odds of dying are about 1.7 times, 2.9 times,
and 2.1 times greater for former smokers, current smokers, and people who
do not regularly exercise than for never smokers and people who exercise
regularly, respectively. As expected, the odds ratio for current smokers is
higher than those for former smokers as the effect of smoking appears to be
reduced after it is stopped.

96
Among older Japanese, as a whole, underweight people are more likely
to die than normal weight people even after controlling for health status and
behavioral variables. However, being overweight and obese are not
significantly related to a higher likelihood of death. Thus, maintaining at least
normal weight is important in avoiding a high risk of death for Japanese older
adults. In addition, only ADL difficulties, cancer, diabetes, and no regular
exercise are associated with higher death rates among the health status and
behavioral variables.
For American older adults, being underweight is not significantly
associated with a higher death rate. However, those overweight are still
significantly less likely to die than normal weight. All health and behavioral
variables are significantly associated with a higher death rate, and particularly,
current smokers have a higher risk of death. Therefore, being overweight
does not appear to negatively affect survival among American older adults.
For the U.S. middle-aged, underweight people are more likely to die
and overweight or obese people are less likely to die than normal weight
people. All poor health conditions and behavioral variables are also
associated with a higher death rate.
6.1.2 The Effect of Health Status and Health Behaviors on Mortality for
Underweight People and Overweight/Obese People
In this section, I attempt to explore whether health conditions and
behavioral variables interacted with being underweight and being

97
overweight/obese when predicting death. To do this, I first combined the
overweight and obese categories and called them overweight (BMI ≥ 25.0).
Then, I selected underweight people and overweight people and ran the same
model for separate samples including health and behavioral variables similar
to those in the previous section.
Table 6.3 presents results of logistic regression on three samples of
underweight people. Regarding age, Japanese and American older adults
who are underweight have an increase of 11% in the odds of dying with each
additional year of age. However, age does not have a significant effect on the
odds of dying for the U.S. middle-aged who are underweight. Regarding
gender, women are significantly less likely to die than men only among
American older adults (OR: 0.40; 95% C.I.: 0.19-0.84).
For underweight Japanese older adults, only ADL difficulties and no
regular exercise are significantly associated with a higher death rate.
Underweight people with one or more ADL difficulties are more likely to die
than underweight people without ADL difficulties (OR: 3.82; 95% C.I.: 1.33-
10.94). The predicted odds of dying for underweight people who do not
regularly exercise are about 2.8 times the odds of dying than for underweight
people who do exercise regularly. Unexpectedly, cancer and diabetes are not
significantly related to higher death rates among the underweight Japanese
older adults.

98
Table 6.3
Odds Ratios and Confidence Intervals (C.I.) of Behavior and Health on the Likelihood of Dying
on Three Samples
b
for Underweight People
a

NUJLSOA (70+) LSOA II (70+) HRS (51-61)
(N= 304/85)
c
(N= 295/67)
c
(N= 119/20)
c

Variables Odds Ratio 95% C. I. Odds Ratio 95% C. I. Odds Ratio 95% C. I.
Age 1.11*** (1.05, 1.19) 1.11*** (1.05, 1.16) 1.00 (0.82, 1.22)
Female 0.81 (0.34, 1.94) 0.40* (0.19, 0.84) 0.31 (0.08, 1.18)

ADL difficulties 3.82* (1.33, 10.94) 1.63 (0.84, 3.15) 1.58 (0.35, 7.19)
Cancer 1.12 (0.25, 5.12) 1.14 (0.56, 2.29) 4.46 (0.95, 20.81)
Stroke 1.02 (0.28, 3.71) 1.60 (0.58, 4.40) 1.23 (0.22, 7.03)
Diabetes 2.13 (0.70, 6.46) 4.73* (1.11, 20.21) 2.02 (0.45, 15.05)
Heart disease 1.33 (0.57, 3.13) 1.55 (0.82, 2.94) 2.60 (0.45, 9.10)
Depression 1.90 (0.90, 4.01) 1.18 (0.62, 2.24) 2.07 (0.63, 6.80)

Former smoker 2.63 (0.97, 7.15) 2.03 (0.96, 4.32) 4.30 (0.31, 59.42)
Current smoker 1.04 (0.38, 2.87) 2.55* (1.10, 5.90) 3.56 (0.43, 29.80)
No regular exercise 2.78** (1.32, 5.84) 1.51 (0.77, 2.96) 0.74 (0.19, 2.82)

-2 Log L 245.90 278.51 82.75
X
2
(df) 53.21 (11) 47.16 (11) 19.11 (11)
a
Underweight: BMI< 18.5
b
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA),
US Longitudinal Study on Aging II (LSOA II), Health and Retirement Study 1994 (HRS)
c
(N= total number of cases/the number of dead people)
* p<.05 ** p<.01 *** p<.001

For underweight American older adults, only diabetes and currently
smoking are associated with an increased likelihood of death. Particularly,
having diabetes increases the odds of dying by a factor of 4.73, which is quite
a large effect on death. Also, current smokers are more likely to die than
never smokers (OR: 2.55; 95% C.I.: 1.10-5.90). The association of these
variables with death among underweight people may have made underweight

99
not significant in Table 6.2.
For the underweight middle-aged in the U.S., nothing is significant in
the relationship between health and behavioral variables and death.
Nevertheless, the reason why the odds ratio reduced from 7.23 to 3.89 after
controlling for health and behavioral variables, and still significant, may be
because the odds ratio of cancer is nearly significant and large, 4.46 (see
Table 6.3).
To examine whether health and behavioral variables affect the odds of
dying, the same regressions are run among overweight people (BMI ≥ 25.0)
(see Table 6.4.). In all three samples, the effect of age on death is similar to
that in the model in Table 6.2. for the whole sample. Regarding gender, the
odds ratio of dying for overweight Japanese older women is not significantly
different than that for men; among the overweight women are less likely to die
than men in the other two American groups.
For overweight Japanese older adults, nothing is significantly related to
death in addition to age. This may mean that the health conditions and
behavioral variables are not related to being overweight in Japan. That is why
the odds ratio of overweight did not really change after adding those variables
in Table 6.2. In addition, the odds ratio of obesity decreased, but not
significantly in Table 6.2.
For overweight American older adults, ADL difficulties, stroke, diabetes,
heart disease, currently smoking, and no regular exercise are associated with

100
Table 6.4
Odds Ratios and Confidence Intervals (C.I.) of Behavior and Health on the Likelihood of Dying
on Three Samples
b
for Overweight or Obese People
a

NUJLSOA (70+) LSOA II (70+) HRS (51-61)
(N= 381) (N= 3834) (N= 5650)
Variables Odds Ratio 95% C. I. Odds Ratio 95% C. I. Odds Ratio 95% C. I.
Age 1.12* (1.03, 1.23) 1.07*** (1.05, 1.09) 1.08** (1.02, 1.14)
Female 1.32 (0.38, 4.56) 0.59*** (0.48, 0.73) 0.47*** (0.33, 0.68)

ADL difficulties 2.34 (0.55, 9.93) 1.71*** (1.35, 2.17) 1.99** (1.32, 3.01)
Cancer 1.05 (0.18, 6.27) 1.06 (0.83, 1.34) 3.10*** (1.86, 5.15)
Stroke 1.50 (0.37, 6.08) 1.79*** (1.32, 2.42) 1.68 (0.89, 3.16)
Diabetes 3.30 (0.96, 11.33) 1.68*** (1.32, 2.15) 3.16*** (2.21, 4.51)
Heart disease 0.43 (0.11, 1.68) 1.60*** (1.30, 1.96) 2.00*** (1.38, 2.90)
Depression 1.07 (0.29, 4.00) 1.21 (0.97, 1.52) 1.15 (0.80, 1.65)

Former smoker 0.34 (0.05, 2.24) 1.12 (0.89, 1.40) 1.37 (0.89, 2.12)
Current smoker 1.34 (0.32, 5.60) 2.07*** (1.47, 2.92) 1.95* (1.24, 3.09)
No regular exercise 1.91 (0.66, 5.51) 1.52*** (1.23, 1.89) 2.22*** (1.58, 3.12)

-2 Log L 155.95 2726.93 1222.84
X
2
(df) 23.60 (11) 224.57 (11) 186.62 (11)
a
Overweight or obese BMI>=25.0
b
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA),
US Longitudinal Study on Aging II (LSOA II), Health and Retirement Study 1994 (HRS)
* p<.05 ** p<.01 *** p<.001

higher death rates. Among these variables, stroke and diabetes have greater
effects on death in this sample of the overweight compared to those in Table
6.2 for the total samples. For overweight American older adults, the odds of
dying for people with stroke are 79% higher than the odds for people without
stroke. Also, the odds of dying for people with diabetes are 68% higher than
the odds for people with diabetes. Other than those two variables, the odds of
dying in ADL difficulties, heart disease, currently smoking, and not exercising

101
regularly are not really different from those in the model (see Table 6.2)
including all American older adults. Thus, it may be concluded that the little
reduction in the odds ratio of overweight from 0.80 in Table 6.1 to 0.74 in
Table 6.2 may result from the relation of stroke and diabetes with being
overweight.
Among the U.S. middle-aged, ADL difficulties, cancer, diabetes, heart
disease, currently smoking, and no regular exercise are related to dying
among overweight people. Those variables increase the likelihood of dying by
about 2 times or possibly even higher among overweight people. Particularly,
having cancer increases the odds of dying among the overweight American
middle-aged by a factor of 3.1. Having diabetes has almost the same effect
on death as having cancer. Interestingly, current smokers are more likely to
die than never smokers among overweight people (OR: 1.95; 95% C.I.: 1.24-
3.09), but the odds ratio of current smokers in this model (see Table 6.4) is
smaller than that in the other model (see Table 6.2). Moreover, the odds of
dying for former smokers are not significant among the overweight middle-
aged in the U.S.
Overall, for Japanese older adults, only ADL difficulties and no regular
exercise are associated with dying among the underweight, but there are no
variables affecting death among overweight people. That is why the odds ratio
of underweight remains significant and that of overweight did not change in
Table 6.2 after controlling for health status and behavioral variables. For

102
American older adults, only diabetes and currently smoking have a
relationship with dying among the underweight. Most of the health conditions
and behaviors except for cancer, depression, and being a former smoker are
associated with dying among overweight American older adults. For the U.S.
middle-aged, there are no significant relationships between health and
behaviors and death among the underweight, which is an unexpected result
because the odds ratio on death of being underweight decreased from 7.23 to
3.89 after controlling for health conditions and behaviors. This may be
because there are few deaths among underweight middle-aged people in the
U.S. However, among overweight American older adults, most of the
variables except for stroke, depression, and being a former smoker are
associated with a higher risk of death among the overweight. This may be the
reason that the odds ratio on obesity is reduced from 0.92 in Table 6.1 to 0.62
in Table 6.2 and becomes significant in Table 6.2.
In conclusion, for Japanese older adults, being underweight has a
greater risk of death than being normal weight, which is linked to functioning
difficulties and no regular exercise. For American older adults, being
overweight was associated with a reduced likelihood of dying although most of
the indicators of poor health and behaviors are related to being overweight or
obese. For American middle-aged persons, overweight or obese people are
less likely to die than normal weight although most poor health conditions and
behaviors are common for overweight or obese people. Thus, being

103
overweight may be related to lower mortality even after controlling for health
and behavioral factors in the U.S.
6.2 The Effect of Body Mass Index on Mortality for Only Never Smokers
Some previous research has suggested that when analyzing the
association between BMI and mortality, researchers need to test the sensitivity
of the results by examining the links for never smokers. This is because
smoking is closely related to both underweight and mortality. Accordingly, this
section focuses on examining whether the effect of each BMI level on mortality
is changed when the sample includes only never smokers.
Table 6.5 shows the odds ratio for age, gender, and each BMI level for
never smokers selected from each sample. For Japanese older adults who
are never smokers, only being underweight is significantly associated with
death among BMI levels. Underweight people are more likely to die than
normal weight people (OR: 2.24; 95% C.I.: 1.55-3.26). The odds of dying for
the overweight are 8% lower than the odds of dying for those of normal weight
and the odds of dying for the obese are 12% higher than the odds of dying for
the normal weight among never smokers in Japan. However, those odds
ratios are not significant. For American older adults who have never smoked,
underweight people (OR: 1.54; 95% C.I.: 1.09-2.18) and obese people (OR:
1.35; 95% C.I.: 1.05-1.74) are more likely to die than normal weight people.
Meanwhile, it is very interesting that no category of BMI is significantly related
to death for never smokers who are middle-aged Americans.

104
Table 6.5
Odds Ratios and Confidence Intervals (C.I.) of World Health Organization (WHO) BMI Category
a

on the Likelihood of Dying on Three Samples
b
of Only Never Smokers
NUJLSOA (70+) LSOA II (70+) HRS (51-61)
(N= 1714) (N=4639) (N=3234)
Variables Odds Ratio 95% C. I. Odds Ratio 95% C. I. Odds Ratio 95% C. I.
Age 1.15*** (1.12, 1.18) 1.12*** (1.10, 1.13) 1.12* (1.02, 1.23)
Female 0.60** (0.42, 0.87) 0.74** (0.61, 0.90) 0.60 (0.33, 1.07)
Underweight 2.24*** (1.55, 3.26) 1.54* (1.09, 2.18) 4.84 (0.65, 35.89)
Normal 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight 0.92 (0.58, 1.47) 0.84 (0.69, 1.03) 1.43 (0.68, 3.00)
Obesity 1.12 (0.32, 3.98) 1.35* (1.05, 1.74) 2.11 (0.98, 4.57)

-2 Log L 1115.28 3522.91 482.74
X
2
(df) 154.56 (5) 287.07 (5) 12.64 (5)
a
Underweight: BMI< 18.5; Normal Weight: 18.5<=BMI<=24.9 (the reference category);
Overweight: 25.0<=BMI<=29.9; Obese: BMI >=30.0
b
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA, 70+),
US Longitudinal Study on Aging II (LSOA II, 70+), Health and Retirement Study 1994 (HRS, 51-61)
* p<.05 ** p<.01 *** p<.001

The odds ratios of dying by BMI category after controlling for health
status and behaviors among never smokers in the three samples are
presented in Table 6.6. For Japanese never smokers, being underweight is
still significantly related to a higher death rate even after controlling for the
additional variables. The odds of dying for the underweight are 2.37 times the
odds of dying for the normal weight group. Among health conditions and
behavioral variables, only ADL difficulties and no regular exercise have
significant effects on mortality. Particularly, having one or more ADL
difficulties increases the odds of dying by a factor of 3.61, which is quite a big
effect on mortality.

105
Table 6.6
Odds Ratios and Confidence Intervals (C.I.) of World Health Organization (WHO) BMI Category
a

on the Likelihood of Dying on Three Samples
b
Controlling for Behavior and Health for Only Never
Smokers
NUJLSOA (70+) LSOA (70+) HRS (51-61)
(N= 1381) (N= 3998) (N=3230)
Variables Odds Ratio 95% C. I. Odds Ratio 95% C. I. Odds Ratio 95% C. I.
Age 1.14*** (1.10, 1.19) 1.11*** (1.09, 1.13) 1.08 (0.98, 1.19)
Female 0.68 (0.42, 1.10) 0.68** (0.54, 0.86) 0.46* (0.24, 0.86)
Underweight 2.37*** (1.45, 3.88) 1.05 (0.65, 1.67) 4.87 (0.61, 38.97)
Normal 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight 1.01 (0.57, 1.81) 0.79* (0.62, 1.00) 1.21 (0.56, 2.62)
Obesity 0.85 (0.20, 3.63) 1.19 (0.89, 1.59) 1.07 (0.46, 2.48)

ADL difficulties 3.61*** (2.00, 6.50) 1.50** (1.18, 1.92) 2.02 (0.93, 4.39)
Cancer 1.51 (0.59, 3.87) 1.40** (1.10, 1.79) 2.73* (1.12, 6.66)
Stroke 1.09 (0.54, 2.22) 1.50* (1.09, 2.07) 4.02** (1.59, 10.19)
Diabetes 1.88 (0.99, 3.57) 1.92*** (1.47, 2.51) 3.67*** (1.86, 7.24)
Heart disease 0.74 (0.43, 1.27) 1.70*** (1.37, 2.09) 1.93 (0.93, 4.01)
Depression 0.78 (0.45, 1.33) 1.23 (0.98, 1.54) 1.21 (0.63, 2.31)
No regular exercise 1.94** (1.24, 3.03) 1.54*** (1.24, 1.91) 2.49** (1.32, 4.70)

-2 Log L 703.72 2655.18 417.11
X
2
(df) 134.97 (12) 321.08 (12) 72.40 (12)
a
Underweight: BMI< 18.5; Normal Weight: 18.5<=BMI<=24.9 (the reference category);
Overweight: 25.0<=BMI<=29.9; Obese: BMI >=30.0
b
Nihon University Japanese Longitudinal Study on Aging (NUJLSOA),
US Longitudinal Study on Aging II (LSOA II), Health and Retirement Study 1994 (HRS)
* p<.05 ** p<.01 *** p<.001

For American older adults who never smoked, being overweight results
in a 21% reduction in the odds of dying. However, the odds of dying for both
being underweight and being overweight became insignificant after controlling
for health status and exercise variables. This is because those health and
exercise variables, except for depression, are mediating the relationship

106
between underweight as well as obesity and mortality for never smokers
among American older adults.
Meanwhile, for American middle-aged people, no BMI category is
significantly associated with mortality before or after controlling for health
conditions and exercise variables. However, cancer, stroke, diabetes, and no
regular exercise are associated with higher likelihoods of death and their
effects on mortality are relatively big compared to those for older American
never smokers. The predicted odds of dying for people with cancer are 2.73
times the odds of dying for people without cancer. People with stroke are
more likely to die than people without stroke for middle-aged never smokers in
the U.S. (OR: 4.02; 95% C.I.: 1.59-10.19). Also, the predicted odds of dying
for people with diabetes are 3.67 times the odds of dying for people without
diabetes. No regular exercise increases the odds of dying by a factor of 2.49.
Among Japanese never smokers, being underweight is associated with
a higher probability of death even after controlling for health conditions and
exercise. Therefore, Japanese older adults need to avoid being underweight if
they are never smokers. However, for older American never smokers, being
overweight reduced the odds of dying even after considering all health
conditions and behaviors. Therefore, if American older adults are never
smokers and they are slightly overweight, they do not appear to be at higher
risk of death. However, for middle-aged never smokers in the U.S., body
weight appears to be an unimportant predictor of mortality. Health conditions

107
such as cancer, stroke, diabetes, and not exercising regularly are more
important predictors of mortality.
6.3 Discussion
In this chapter, I attempted to examine whether the association of BMI
with mortality differs by countries and age groups. Also, I would like to
suggest what the optimal BMI is for Japanese older adults, American older
adults, and the U.S. middle-aged.
For all Japanese older adults, underweight people are more likely to die
than normal weight people even after controlling for health conditions and
health behaviors although ADL difficulties and no regular exercise were
revealed to have greater effects on mortality for underweight people. For all
American older adults, overweight people are less likely to die than normal
weight people controlling for the same variables. In fact, the effects of those
health conditions and behaviors on mortality for overweight or obese people
are not much different from the total sample of American older adults.
However, the association of being underweight with a higher likelihood of
death became insignificant probably due to the effect of diabetes and current
smoking. For middle-aged Americans, underweight people are more likely to
die after controlling for health conditions and behaviors in Table 6.2. Moreover,
overweight and obese people are less likely to die when controlling for the
same health conditions and behaviors because most of the health conditions
and behaviors are associated with a higher death rate among overweight or

108
obese people. These results, particularly for the middle-aged Americans,
were not expected.
As Manson et al. (1987) recommended, results shown in Table 6.2
controlled for existing diseases such as cancer, stroke, diabetes, heart
disease, and depression although the analyses did not eliminate early
mortality due to short follow-up period. The model shown in Table 6.2
included all types of smoking: past, current, and never smokers. However, as
suggested by Manson et al. (1987), the smoking variable needs to be
appropriately controlled because it is closely related to both underweight and
mortality (Krueger et al., 2007).
To follow Manson’s suggestion, I ran the same models selecting only
never smokers. The results were almost the same as for the total sample for
Japanese and American older adults. The results for Japanese older adults in
Table 6.6 have supported previous research in which low weight is related to
an increased relative risk of death in Japan (Kuriyama et al., 2004; Miyazaki et
al., 2002; Ohta et al., 2001).
Unlike the results of recent research in which underweight as well as
obesity is associated with increased mortality for older adults in the U.S.
(Allison et al., 1997, Flegal et al., 2005), the results from Table 6.6. showed
that the risk of death for underweight and obese people were not significantly
higher. Instead, overweight older adults are significantly less likely to die
relative to normal weight older adults in the U.S. Therefore, the results for

109
older Japanese and Americans are consistent with some previous research
(Baik et al., 2000; Grabowski & Ellis, 2001) which has shown that higher BMI
is not linearly associated with higher mortality for older people.
For the American middle-aged, no BMI category was significantly
related to mortality after controlling for health conditions and exercise. For
middle-aged never smokers in the U.S., cancer, stroke, diabetes, and no
regular exercise are much more important predictors of death than weight.
These results therefore are somewhat inconsistent with the research by Diehr
et al. (1998) because the relation of higher BMI to mortality was not significant
for the American middle-aged in Table 6.6.
Based on the results from three samples, Japanese and American older
adults, and American middle-aged, it can be said that being overweight is the
optimal BMI only for American older adults. This may be because people with
diseases related to high BMI and earlier death already died before reaching
older age. For Japanese older adults, obese people are less likely to die than
others, but this is not significant. In addition, it cannot be said that normal
weight is the optimal BMI for American middle-aged because it was not
significantly linked to high mortality in this study.
Nevertheless, the results of this chapter give some suggestions for
practitioners including nurses, geriatricians, and directors in facilities for older
adults. In Japan, they need to try to avoid older adults’ being underweight and
loosing body weight because underweight Japanese are significantly more

110
likely to die than normal weight Japanese. In the U.S., practitioners need to
help older adults to not be obese. However, if older adults are slightly
overweight, practitioners may suggest regular exercise rather than pushing
them to loose weight. For the U.S. middle-aged, even if overweight and obese
people are not significantly more likely to die than normal weight people,
middle-aged American people need to maintain normal weight in order to
avoid having diseases related to high BMI such as stoke and diabetes.
Furthermore, practitioners need to encourage all people to exercise regularly
across countries and ages because our results indicate that people who do not
regularly exercise have a greater risk of death in all three samples.
If information covering a longer follow-up period become available in
the future, it would be important to examine the relationship between BMI and
mortality after eliminating early death that might be related to preexisting
diseases. Also, it would suggest more information about the linkage of BMI to
death if specific causes of death such as cancer and heart disease could be
examined.

111
Chapter VII: The Effect of Body Mass Index on Mortality Through
Biomarkers, Nutritional Factors, and Health Behaviors in the NHANES III
In this chapter, I examine whether biological risks, nutritional factors,
and health behaviors differ for thin and overweight people from those of
persons with normal weight in the U.S. In addition, the link between BMI and
all-cause mortality is investigated with controls for other biomarkers which are
risk factors for mortality. The question of how other risks for mortality such as
smoking, drinking, exercise, and diet are linked to the relationship between
BMI and mortality is also addressed. These links are examined among
Americans aged 40 and over, 40 to 64 years old, and 65 years and older. This
chapter uses three BMI categories combining overweight with obesity into one
category of overweight (BMI ≥ 25.0), underweight (BMI<18.5), and normal
weight (18.5 ≤ BMI ≤ 24.9: reference).
7.1 High Risk Level of Biomarkers, Nutrition, and Health Behaviors by Weight
Table 7.1 presents results of tests of the null hypothesis: that the
prevalence of high risk biomarkers among underweight people (or overweight)
is not significantly different from that among normal weight people when age
and gender are controlled.
Interestingly, results show that prevalence of high risk biomarkers
among thin people are not much different from that among normal weight
people except for prevalence of high risk levels of high total cholesterol (Total
cholesterol I ≥ 240 mg/dl) and low total cholesterol (Total cholesterol II ≤ 160

112
Table 7.1
Percentage of Population 40 and Older at High Risk for Biological Markers by Three BMI Categories in the NHANES III
Controlling Age and Gender
% High Risk
Indicator High-Risk Cut Point N Underweight Chi-sq Normal Chi-sq Overweight
Diastolic BP I >= 90 mmHg 9233 2.9 4.8 *** 9.7
Diastolic BP II <= 60 mmHg 9233 11.6 10.7 *** 7.6
Systolic BP >= 140 mmHg 9231 22.1 20.8 *** 26.6
Total cholesterol I >= 240 mg/dl 9233 9.3 *** 23.8 *** 31.6
Total cholesterol II <= 160 mg/dl 9233 16.3 * 8.4 * 6.7
HDL cholesterol < 40 mg/dl 9155 12.0 15.7 *** 29.8
Fibrinogen > 400 mg/dl 9673 19.2 14.4 15.8
C-reactive protein 3.0 < <= 10 mg/dl 8203 12.5 16.0 *** 31.4
Glycated hemoglobin > 6.4% 9153 8.9 3.9 *** 9.8
White blood cell count > 10.8 X10
3
/uL 9673 12.5 8.2 8.4
Fasting LDL cholesterol >= 160 mg/dl 3958 17.7 20.4 *** 27.9
Fasting triglycerides >= 200 mg/dl 4091 12.8 10.7 *** 25.9

113
mg/dl). The prevalence of adverse risk from high total cholesterol for
underweight people is significantly lower than that for normal weight people.
Prevalence of low total cholesterol for underweight people is significantly
higher than that for normal weight people.
Meanwhile, the prevalence of the most high risk biomarkers is
significantly different among the overweight compared to normal weight people
except for high risk levels of fibrinogen and white blood cells. The percentage
of low diastolic blood pressure (Diastolic BP II) and low total cholesterol (Total
cholesterol II) is significantly lower for the overweight than the normal weight;
while the percentages of high risk levels of high diastolic blood pressure
(Diastolic BP I), systolic blood pressure, high total cholesterol (total cholesterol
I), HDL cholesterol, C-reactive protein, glycated hemoglobin, fasting LDL
cholesterol, and fasting triglycerides for overweight people are significantly
higher than those for normal weight people. This means that overweight
people have more risk factors for mortality than normal weight people.
Table 7.2 examines differences in risk from nutritional and behavioral
factors between underweight people (or overweight people) and normal weight
people. The null hypothesis is that there are no differences. The prevalence
of high risk albumin, malnutrition, anemia, and antioxidants for underweight
people are significantly higher than that for normal weight people. This means
that nutrition for underweight people is worse than that for normal weight. In
terms of health behaviors, the percentages of current smokers and those who

114
Table 7.2
Percentage of Population 40 and Older at High Risk for Behavioral Factors by Three BMI Categories in the NHANES III
Controlling for Age and Gender
% High Risk
Indicator High-Risk Cut Point N Underweight Chi- Normal Chi- Overweight
Albumin < 3.8 mg/dl 9673 22.9 * 11.1 * 13.4
Serum Homocysteine >= 15 umol/l 4009 6.7 12.1 * 9.2
Malnutrition Any deficiency from iron, B12,
folate
5751 51.2 *** 31.1 28.7
Anemia
F<130, M<120
9673 16.4 * 8.3 6.7
Antioxidants ≥ 3 low quartiles out of vitamin
A, C, E, selenium, and
lycopene
9223 31.0 *** 14.2 15.5

% daily intake from carbohydrate > 65% 9310 5.4 ** 10.9 * 9.2
% daily intake from fat > 40 % 9310 23.7 22.5 * 25.3
% daily intake from protein < 10% 9310 14.2 8.2 7.2

Current smoker Ref- not current 9693 48.9 *** 28.0 *** 19.3
Heavy drinker Ref- < 3 drinks per day 9366 15.6 14.0 14.1
No regular exercise Ref- exercise 9693 33.2 * 23.3 *** 27.4

115
do not exercise regularly among underweight people are significantly higher
than those among normal weight people. However, the prevalence of people
who consume more than 65 % of daily calories from carbohydrates is
significantly lower among underweight people than that among normal weight
people.
For overweight people, the prevalence of high risk albumin, intake of
more than 40 % of daily calories from fat, and percent of people who do not
exercise regularly are significantly higher than for those of normal weight.
However, the prevalence of high risk serum homocysteine, current smokers,
and people who consume high levels of carbohydrates among overweight
people is significantly lower than normal weight people.
Overall, while underweight people tend to have worse levels than
normal weight people in nutritional markers and adverse health behaviors,
overweight people have more high risks in most biomarkers than normal
weight people.
7.2 Effects of Weight on Mortality When Biomarkers and Health Behaviors Are
Introduced for All-Causes of Death
7.2.1 Effects of Weight on Mortality When Biomarkers and Health Behaviors
Are Introduced for All-Causes of Death Among People Who Are 40 and Older
Table 7.3 presents the results from multivariate analyses using Cox
proportional hazard models examining the effect of low BMI on mortality for
people aged 40 and older limiting the sample to those who are of normal

116

Table 7.3
Multivariate Analyses Using Cox Proportional Hazard Models Indicating Effect of Low BMI for People with Low and Normal Weight
Aged 40 and Older: All Cause Mortality
Model 1 Model 2 Model 3 Model 4
N (# of people who died) 3262 (675) 2626 (474) 1407 (266) 1382 (262)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.
Age 1.10*** (1.09, 1.11) 1.10*** (1.09, 1.11) 1.09*** (1.07, 1.11) 1.09*** (1.07, 1.11)
Female 0.55*** (0.43, 0.69 0.57*** (0.45, 0.74) 0.67* (0.48, 0.93) 0.62* (0.44, 0.89)
Low BMI (<18.5)
1
2.51*** (1.85, 3.41) 2.20*** (1.50, 3.21) 2.22* (1.09, 4.51) 2.32* (1.10, 4.90)

Biomarkers
Diastolic BP I 0.97 (0.59, 1.60) 0.80 (0.36, 1.80) 0.75 (0.33, 1.71)
Diastolic BP II 1.19 (0.86, 1.64) 1.21 (0.74, 1.98) 1.14 (0.69, 1.87)
Systolic BP 1.26 (0.95, 1.68) 1.52* (1.06, 2.18) 1.44* (1.01, 2.05)
Total cholesterol I 1.12 (0.77, 1.64) 1.49 (0.87, 2.54) 1.47 (0.85, 2.55)
Total cholesterol II 1.66** (1.18, 2.34) 1.46 (0.92, 2.33) 1.46 (0.91, 2.33)
HDL cholesterol 1.53* (1.11, 2.11) 1.80** (1.24, 2.61) 1.76** (1.23, 2.53)
Fibrinogen 1.15 (0.86, 1.53) 1.27 (0.82, 1.98) 1.26 (0.80, 1.98)
C-reactive protein 1.16 (0.91, 1.48) 1.40* (1.01, 1.95) 1.44* (1.05, 1.98)
Glycated hemoglobin 1.13 (0.69, 1.83) 0.80 (0.48, 1.33) 0.87 (0.55, 1.40)
White blood cell count 1.45 (0.89, 2.36) 1.92* (1.05, 3.50) 1.97* (1.09, 3.55)

117

Table 7.3, Continued

Model 1 Model 2 Model 3 Model 4
N (# of people who died) 3262 (675) 2626 (474) 1407 (266) 1382 (262)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.

Nutritional Factors
Albumin 1.06 (0.74, 1.53) 1.05 (0.73, 1.52)
Malnutrition 1.22 (0.87, 1.71) 1.20 (0.85, 1.69)
Anemia 1.46 (0.93, 2.29) 1.41 (0.91, 2.19)
Antioxidants 1.21 (0.79, 1.86) 1.16 (0.76, 1.78)
High carbohydrate 0.84 (0.47, 1.49) 0.90 (0.50, 1.63)
High fat 0.91 (0.62, 1.35) 0.92 (0.63, 1.34)
Low protein 0.90 (0.48, 1.68) 0.89 (0.49, 1.63)

Health Behaviors
Current Smoker 1.10 (0.70, 1.72)
Heavy drinker 1.13 (0.54, 2.37)
No regular exercise 1.53* (1.09, 2.16)

Wald Chisq (df) 666.24***(3) 639.73***(13) 633.05***(20) 644.58***(23)
-2 Log L 5686.48 3735.49 1945.68 1895.69
1
Reference category is normal weight (18.5<=BMI<=24.9)

118
weight or less. The analyses used all cause mortality although people who
died accidental or violent deaths were eliminated. Also, to have sufficient
cases for analyses, I excluded two biomarkers and one nutritional variable in
the models because those variables have a smaller number of cases as they
are only collected for the morning sample, or the sample fasted at least 6
hours, or they were not assayed every year. These include fasting LDL
cholesterol, fasting triglycerides and serum homocysteine. There are four
models in Table 7.3, but I focus on Models 1, 2, and 4 because the results of
Model 2 do not much differ from those in Model 3, and X
2
in Model 3 is smaller
than that in Model 2 even after adding the nutritional factors.
The base model shows a hazard ratio for low BMI of 2.51, meaning that
after controlling for age and gender, mortality was 2.45 times higher than
normal BMI. Model 2 shows effects of biomarkers on mortality. The hazard
ratio of low BMI was reduced to 2.20, which means the relationship between
low BMI and high mortality may be partially explained by biomarkers. Low
total cholesterol (Total cholesterol II) and low HDL cholesterol are associated
with a 66% and a 53% increase in the hazard of death, respectively. When we
considered biomarkers, nutritional factors, and health behaviors
simultaneously, the hazard ratio for low BMI is increased to 2.32 in Model 4.
High systolic blood pressure, low HDL cholesterol, high C-reactive protein, and
high white blood cell count are significantly associated with a higher hazard of
death. Particularly, the high white blood cell count increases the hazard of

119
death by about two times. Also, people who do not regularly exercise have a
53% increase in the hazard of death. In fact, it is unusual that the X
2
decreases whenever variables are added into the base model, meaning that
biomarkers, nutritional factors, and health behaviors do not explain much of
the relationship between low BMI and high mortality for people aged 40 and
over. Nevertheless, it can be concluded that people with low BMI are more
likely to die than people with normal weight people even after introducing
biomarkers, nutritional factors, and health behaviors that are likely to be
related to being underweight (HR: 2.32; 95% C.I.: 1.10-4.90).
Table 7.4 presents odds ratios and 95% confidence interval from Cox
proportional hazards models examining the relationship between high BMI and
mortality among people aged 40 and older limiting the sample to those who
are of normal weight or above. As in Table 7.3, I focus on Models 1, 2, and 4
because the effects of high BMI, biomarkers, and nutritional factors in Model 3
are almost the same as those in Model 4.
In the base model, high BMI is not significantly associated with the
hazard of death when controlling for age and gender. Interestingly, after
considering biomarkers in Model 2, high BMI is significantly negatively related
to the likelihood of dying (OR: 0.84; 95% C.I.: 0.71-1.00). High fibrinogen,
high C-reactive protein, high glycated hemoglobin, and high white blood cell
count are linked to an increased risk of death (by 26%, 21%, 48%, and 41%,
respectively). In Model 4, the relative risk for people with high BMI compared

120
Table 7.4
Multivariate Analyses Using Cox Proportional Hazard Models Indicating Effect of High BMI for People With Normal or Overweight
Aged 40 and Older: All Cause Mortality
Model 1 Model 2 Model 3 Model 4
N (# of people who died) 9460 (1468) 7400 (1005) 4153 (571) 4082 (560)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.
Age 1.09*** (1.09, 1.10) 1.10*** (1.09, 1.11) 1.10*** (1.09, 1.11) 1.11*** (1.09, 1.12)
Female 0.63*** (0.54, 0.74) 0.63*** (0.51, 0.77) 0.64*** (0.50, 0.82) 0.59*** (0.46, 0.76)
High BMI (>=25.0)
1
0.98 (0.84, 1.14) 0.84* (0.71, 1.00) 0.76* (0.60, 0.96) 0.77* (0.60, 0.99)

Biomarkers
Diastolic BP I 1.06 (0.73, 1.54) 1.14 (0.70, 1.86) 1.13 (0.69, 1.87)
Diastolic BP II 1.28 (0.99, 1.64) 1.45* (1.09, 1.94) 1.45* (1.08, 1.94)
Systolic BP 1.14 (0.95, 1.37) 0.99 (0.81, 1.22) 0.97 (0.79, 1.19)
Total cholesterol I 1.07 (0.86, 1.32) 1.40* (1.07, 1.83) 1.38* (1.05, 1.82)
Total cholesterol II 1.12 (0.84, 1.48) 1.11 (0.76, 1.63) 1.20 (0.82, 1.75)
HDL cholesterol 1.22 (0.96, 1.55) 1.36* (1.01, 1.82) 1.37* (1.03, 1.83)
Fibrinogen 1.26* (1.01, 1.57) 1.18 (0.82, 1.70) 1.16 (0.80, 1.68)
C-reactive protein 1.21* (1.03, 1.43) 1.31* (1.07, 1.62) 1.33* (1.06, 1.67)
Glycated hemoglobin 1.48* (1.07, 2.04) 1.48 (0.98, 2.24) 1.46 (0.97, 2.20)
White blood cell count 1.41* (1.06, 1.86) 1.52* (1.10, 2.11) 1.39 (0.96, 2.01)

121
Table 7.4, Continued

Model 1 Model 2 Model 3 Model 4
N (# of people who died) 9460 (1468) 7400 (1005) 4153 (571) 4082 (560)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.

Nutritional Factors
Albumin 1.04 (0.79, 1.38) 1.05 (0.79, 1.40)
Malnutrition 1.08 (0.82, 1.41) 1.00 (0.74, 1.34)
Anemia 1.42* (1.04, 1.94) 1.39* (1.02, 1.89)
Antioxidants 1.24 (0.91, 1.68) 1.11 (0.80, 1.54)
High carbohydrate 0.67* (0.46, 0.96) 0.66* (0.47, 0.93)
High fat 0.88 (0.66, 1.19) 0.84 (0.63, 1.13)
Low protein 1.03 (0.72, 1.48) 1.06 (0.72, 1.55)

Health Behaviors
Current Smoker 1.51* (1.02, 2.25)
Heavy drinker 0.73 (0.40, 1.32)
No regular exercise 1.56*** (1.23, 1.97)

Wald Chisq (df) 626.31***(3) 875.20***(13) 599.76***(20) 840.11***(23)
-2 Log L 16756.63 11147.66 5512.93 5285.90
1
Reference category is normal weight (18.5<=BMI<=24.9)

122
to normal weight people is even lower (HR: 0.77; 95% C.I.: 0.60-0.99) after
adjusting for biomarkers, nutritional factors and health behaviors. Low
diastolic blood pressure (Diastolic BP II), high total cholesterol (Total
cholesterol I), low HDL cholesterol, and high C-reactive protein are linked to
an increase of 45%, 38%, 37%, and 33% in the risk of death. Also, people
with anemia have an increased risk of death (HR: 1.39; 95% C.I.: 1.02-1.89).
However, unexpectedly, people who consume more than 60% of their calories
from carbohydrates have a decreased risk of death (HR: 0.66; 95% C. I.: 0.47-
0.93). Current smokers have a higher hazard of death than nonsmokers (HR:
1.51; 95% C.I.: 1.02-2.25). People who do not exercise regularly also have a
higher hazard of death (HR: 1.56; 95% C. I.: 1.23-1.97).
In summary, mortality is higher for people with low BMI than for those
with normal weight, even after adjusting for biomarkers, nutritional factors, and
health behaviors. Underweight people have lower cholesterol than normal
weight people but the indicators of high and low total cholesterol are not linked
to mortality among the underweight relative to those of normal weight. The
effect of differential high systolic blood pressure, low HDL cholesterol, high C-
reactive protein, high white blood cell count, and not exercising regularly only
partially explains the relationship between low BMI and mortality, reducing the
hazard ratio from 2.45 to 2.30. On the other hand, being overweight is not
related to mortality initially. When controls are introduced, low diastolic blood
pressure (Diastolic BP II), high total cholesterol (Total cholesterol I), low HDL

123
cholesterol, high C-reactive protein, anemia, high calorie intake from
carbohydrate, current smoking, and no regular exercise reduced the hazard
ratio linking BMI and mortality so that it is negative and significant (HR: 0.77;
95% C.I.: 0.60-0.99).
7.2.2 Effects of Weight on Mortality When Biomarkers and Health Behaviors
Are Introduced for All-Causes of Death Among People Who Are 40 to 64
Years Old, and 65 and Older
To further clarify the relationship between weight and mortality in
different age groups, I divided subjects into two age groups in this section: 40
to 64 years old and 65 and older and repeated the prior set of analyses. Table
7.5 presents results from multivariate analyses for low BMI among those aged
40 to 64 years old limiting the sample to those who are of normal weight or
less. I focus on Models 1, 2, and 3 due to the lower value of X
2
in Model 4.
When controlling age and gender, low BMI is associated with almost a
4(3.86) times higher hazard of death than for normal weight among people
aged 40 to 64 years old. After considering biomarkers, the effect of BMI is
even stronger, but not significant. Also, none of the biomarkers are
significantly related to mortality. The most interesting thing is that having a
high white blood cell count, being in the bottom quartile for at least three
among the five antioxidants, and having more than 40% of calorie intake from
fat reduced the effect of low BMI on mortality to 0.42, which is not significant
(see Model 3). People with a high white blood cell count have a higher hazard

124

Table 7.5
Multivariate Analyses Using Cox Proportional Hazard Models Indicating Effect of for Low BMI People With Low or Normal Weight
Aged 40 to 64: All Cause Mortality
Model 1 Model 2 Model 3 Model 4
N (# of people who died) 1719 (92) 1440 (63) 805 (41) 795 (41)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.
Age 1.12*** (1.07, 1.18) 1.09** (1.03, 1.16) 1.11*** (1.05, 1.18) 1.12*** (1.06, 1.19)
Female 0.42* (0.21, 0.85) 0.42 (0.17, 1.04) 0.64 (0.22, 1.88) 0.61 (0.22, 1.71)
Low BMI (<18.5)
1
3.86* (1.36, 10.97) 3.91 (0.99, 15.46) 0.42 (0.07, 2.57) 0.45 (0.08, 2.69)

Biomarkers
Diastolic BP I 0.62 (0.19, 2.06) 1.44 (0.27, 7.70) 1.37 (0.27, 6.88)
Diastolic BP II 1.38 (0.45, 4.22) 2.36 (0.52, 10.65) 2.50 (0.56, 11.27)
Systolic BP 1.66 (0.62, 4.49) 1.44 (0.32, 6.45) 1.43 (0.33, 6.24)
Total cholesterol I 1.42 (0.54, 3.72) 1.55 (0.41, 5.92) 1.44 (0.36, 5.73)
Total cholesterol II 1.74 (0.62, 4.90) 0.70 (0.23, 2.09) 0.74 (0.23, 2.36)
HDL cholesterol 1.95 (0.78, 4.85) 1.66 (0.49, 5.65) 1.82 (0.53, 6.27)
Fibrinogen 1.48 (0.70, 3.10) 1.84 (0.85, 3.97) 1.73 (0.75, 4.00)
C-reactive protein 2.02 (0.92, 4.41) 0.65 (0.24, 1.72) 0.59 (0.19, 1.79)
Glycated hemoglobin 2.70 (0.55, 13.38) 0.68 (0.15, 3.06) 0.62 (0.11, 3.48)
White blood cell count 0.90 (0.32, 2.57) 3.55* (1.15, 11.01) 3.88* (1.29, 11.66)

125

Table 7.5, Continued

Model 1 Model 2 Model 3 Model 4
N (# of people who died) 1719 (92) 1440 (63) 805 (41) 795 (41)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.

Nutritional Factors
Albumin 0.55 (0.14, 2.17) 0.48 (0.10, 2.33)
Malnutrition 0.95 (0.17, 5.22) 0.98 (0.15, 6.32)
Anemia 1.19 (0.37, 3.83) 1.08 (0.33, 3.56)
Antioxidants 3.38* (1.12, 10.23) 3.73* (1.37, 10.12)
High carbohydrate 0.22 (0.04, 1.34) 0.22 (0.03, 1.38)
High fat 0.22** (0.08, 0.59) 0.22** (0.08, 0.58)
Low protein 0.81 (0.20, 3.22) 0.81 (0.22, 3.03)

Health Behaviors
Current Smoker 0.74 (0.22, 2.51)
Heavy drinker 1.12 (0.50, 2.52)
No regular exercise 1.44 (0.37, 5.65)

Wald Chisq (df) 46.35***(3) 60.84***(13) 202.71***(20) 196.95***(23)
-2 Log L 749.10 520.04 267.51 264.67
1
Reference category is normal weight (18.5<=BMI<=24.9)

126
of death (HR: 3.55; 95% C.I.: 1.15-11.01). People who have low antioxidants
are significantly more likely to die (HR: 3.38; 95% C.I.: 1.12-10.23). Having
more than 40% of calorie intake from fat decreased the hazard of death by
78%. Based on these results, low BMI is not associated with a higher hazard
of death among those who are 40 to 64 years old when biomarkers and
nutritional factors are controlled.
Table 7.6 shows results from Cox proportional hazard models indicating
the effect of low BMI among people aged 65 and older limiting the sample to
those who are of normal weight or less. For all causes of death, the hazard
ratio of low BMI is 2.38, controlling for age and female. After controlling the
biomarkers, it is reduced to 1.91. Low total cholesterol and low HDL
cholesterol are significantly associated with a higher hazard of death.
However, when variables such as nutritional factors and health behaviors are
included, the hazard ratios indicating the effect of low weight become higher
(HR: 2.91; 95% C.I.: 1.27-6.67). Particularly, low HDL cholesterol, high C-
reactive protein, and not exercising regularly are related to an increased
hazard of death from among those who are 65 and older. Therefore, for
people 40 to 64 years old, low BMI is not significantly related to a higher
hazard of death from all causes when controls for biomarkers and nutritional
factors are included; for those 65 and older, low BMI is significantly associated
with a higher hazard of mortality from all causes when biomarkers, nutritional
factors, and health behaviors are controlled simultaneously.

127

Table 7.6
Multivariate Analyses Using Cox Proportional Hazard Models Indicating Effect of Low BMI People With Low or Normal Weight
Aged 65 and Older: All Cause Mortality
Model 1 Model 2 Model 3 Model 4
N (# of people who died) 1545 (583) 1186 (411) 602 (225) 587 (221)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.
Age 1.08***(1.07, 1.10) 1.10*** (1.08, 1.12) 1.08*** (1.06, 1.11) 1.08*** (1.06, 1.11)
Female 0.58***(0.46, 0.73) 0.60*** (0.47, 0.77) 0.71 (0.50, 1.02) 0.64* (0.43, 0.96)
Low BMI (<18.5)
1
2.38*** (1.75, 3.25) 1.91** (1.19, 3.07) 2.67* (1.21, 5.89) 2.91* (1.27, 6.67)

Biomarkers
Diastolic BP I 1.01 (0.56, 1.83) 0.67 (0.26, 1.76) 0.63 (0.23, 1.72)
Diastolic BP II 1.14 (0.81, 1.61) 1.12 (0.70, 1.79) 1.05 (0.64, 1.70)
Systolic BP 1.24 (0.93, 1.67) 1.51* (1.04, 2.19) 1.41 (0.97, 2.04)
Total cholesterol I 1.05 (0.74, 1.49) 1.45 (0.87, 2.42) 1.44 (0.85, 2.44)
Total cholesterol II 1.78** (1.25, 2.52) 1.62 (0.95, 2.76) 1.56 (0.91, 2.68)
HDL cholesterol 1.43* (1.02, 2.02) 1.77** (1.19, 2.63) 1.69* (1.14, 2.52)
Fibrinogen 1.12 (0.81, 1.54) 1.15 (0.69, 1.93) 1.16 (0.67, 2.01)
C-reactive protein 1.05 (0.78, 1.41) 1.54* (1.10, 2.14) 1.61** (1.16, 2.23)
Glycated hemoglobin 0.94 (0.56, 1.58) 0.79 (0.45, 1.39) 0.87 (0.51, 1.49)
White blood cell count 1.62 (0.95, 2.76) 1.85 (0.92, 3.73) 1.95 (0.98, 3.88)

128

Table 7.6, Continued

Model 1 Model 2 Model 3 Model 4
N (# of people who died) 1545 (583) 1186 (411) 602 (225) 587 (221)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.

Nutritional Factors
Albumin 1.16 (0.80, 1.69) 1.16 (0.80, 1.68)
Malnutrition 1.32 (0.94, 1.86) 1.29 (0.92, 1.81)
Anemia 1.63 (0.94, 2.85) 1.60 (0.95, 2.71)
Antioxidants 1.01 (0.59, 1.72) 0.94 (0.55, 1.61)
High carbohydrate 0.90 (0.48, 1.69) 0.97 (0.52, 1.83)
High fat 1.11 (0.76, 1.62) 1.14 (0.77, 1.69)
Low protein 0.91 (0.43, 1.91) 0.89 (0.43, 1.82)
Health Behaviors
Current Smoker 1.19 (0.77, 1.83)
Heavy drinker 1.09 (0.45, 2.67)
No regular exercise 1.67** (1.19, 2.35)
Wald Chisq (df) 210.75***(3) 428.93***(13) 478.64***(20) 791.30*** (23)
-2 Log L 6298.42 4133.03 2080.82 2015.00
1
Reference category is normal weight (18.5<=BMI<=24.9)

129
Next, the effects of having high BMI on mortality from all causes for
people aged 40 to 64 limiting the sample to those who are of normal weight or
above are presented in Table 7.7. Similar to Table 7.5 for the same age
group, high BMI is related to a higher hazard of death when controlling only
age and gender (HR: 1.49; 1.06-2.10). However, after considering
biomarkers, the hazard ratio on high BMI is reduced to 0.99 and is
insignificant. When nutritional factors and health behaviors are introduced,
low HDL cholesterol, high fibrinogen and high carbohydrate intake reduced the
hazard ratio linking high BMI and mortality, however the hazard ratio of high
BMI is still insignificant (HR: 0.58; 95% C.I.: 0.29-1.17).
Table 7.8 presents the multivariate analyses for high BMI and mortality
for people aged 65 and older limiting the sample to those who are normal or
overweight. Only when biomarkers are controlled is high BMI significantly
associated with a lower hazard of mortality (HR: 0.81; 95% C.I.: 0.68-0.97).
Low HDL cholesterol, high glycated hemoglobin, and high white blood cell
count are all related to higher hazards of death (by 26%, 46%, and 39%).
After controlling for biomarkers, nutritional factors, and health behaviors, the
hazard ratio for high BMI becomes insignificant, and not much different from
that when controlling only for biomarkers, although the overall model fit
increased from X
2
= 362.99 to X
2
=1223.10. Therefore, it can be concluded that
high BMI is not related to a lower hazard of mortality from all causes for those
who are 65 and older.

130

Table 7.7
Multivariate Analyses Using Cox Proportional Hazard Models Indicating Effect of High BMI for People With Normal or Overweight
Aged 40 to 64: All Cause Mortality
Model 1 Model 2 Model 3 Model 4
N (# of people who died) 5527 (295) 4383 (179) 2528 (104) 2492 (99)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.
Age 1.09*** (1.06, 1.11) 1.06*** (1.03, 1.10) 1.09*** (1.04, 1.14) 1.11*** (1.05, 1.16)
Female 0.83 (0.62, 1.12) 0.66 (0.42, 1.04) 0.59 (0.28, 1.23) 0.98 (0.47, 2.03)
High BMI (>=25.0)
1
1.49* (1.06, 2.10) 0.99 (0.61, 1.60) 0.62 (0.36, 1.08) 0.58 (0.29, 1.17)

Biomarkers
Diastolic BP I 0.97 (0.55, 1.74) 2.05 (0.80, 5.26) 2.33 (0.83, 6.54)
Diastolic BP II 1.69 (0.84, 3.41) 2.43 (0.97, 6.07) 2.56 (0.96, 6.79)
Systolic BP 1.27 (0.69, 2.35) 0.76 (0.30, 1.98) 0.75 (0.28, 2.02)
Total cholesterol I 1.12 (0.67, 1.86) 1.80 (0.82, 3.96) 1.66 (0.62, 4.46)
Total cholesterol II 0.95 (0.44, 2.03) 1.09 (0.33, 3.59) 1.22 (0.36, 4.15)
HDL cholesterol 1.11 (0.60, 2.08) 1.51 (0.75, 3.04) 1.96* (1.02, 3.76)
Fibrinogen 2.22* (1.48, 3.33) 2.32* (1.16, 4.62) 2.67* (1.30, 5.48)
C-reactive protein 1.72* (1.14, 2.61) 1.48* (1.00, 2.51) 1.38 (0.84, 2.28)
Glycated hemoglobin 1.54 (0.74, 3.18) 0.97 (0.31, 3.08) 1.13 (0.35, 3.69)
White blood cell count 1.37 (0.72, 2.61) 1.41 (0.57, 3.46) 1.05 (0.35, 3.14)

131

Table 7.7, Continued

Model 1 Model 2 Model 3 Model 4
N (# of people who died) 5527 (295) 4383 (179) 2528 (104) 2492 (99)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.

Nutritional Factors
Albumin 0.81 (0.35, 1.88) 0.85 (0.35, 2.07)
Malnutrition 0.93 (0.43, 2.01) 0.91 (0.35, 2.35)
Anemia 1.13 (0.53, 2.43) 1.19 (0.52, 2.68)
Antioxidants 1.87 (0.75, 4.65) 2.04 (0.83, 5.04)
High carbohydrate 0.26* (0.08, 0.83) 0.28* (0.08, 0.94)
High fat 0.72 (0.34, 1.56) 0.58 (0.26, 1.32)
Low protein 0.68 (0.28, 1.69) 0.68 (0.29, 1.62)

Health Behaviors
Current Smoker 1.66 (0.68, 4.08)
Heavy drinker 1.27 (0.63, 2.57)
No regular exercise 0.69 (0.29, 1.68)

Wald Chisq (df) 67.43***(3) 92.43***(13) 131.71***(20) 290.97***(23)
-2 Log L 3687.21 2093.44 930.59 840.76
1
Reference category is normal weight (18.5<=BMI<=24.9)

132

Table 7.8
Multivariate Analyses Using Cox Proportional Hazard Models Indicating Effect of High BMI for People With Normal or Overweight
Aged 65 and Older: All Cause Mortality
Model 1 Model 2 Model 3 Model 4
N (# of people who died) 3933 (1173) 3017 (826) 1625 (467) 1590 (461)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.
Age 1.10*** (1.09, 1.11) 1.10*** (1.09, 1.12) 1.09*** (1.08, 1.11) 1.09*** (1.08, 1.11)
Female 0.57*** (0.48, 0.68) 0.58*** (0.48, 0.69) 0.61*** (0.49, 0.76) 0.52*** (0.42, 0.66)
High BMI (>=25.0)
1
0.87 (0.74, 1.03) 0.81* (0.68, 0.97) 0.79 (0.62, 1.01) 0.80 (0.62, 1.04)

Biomarkers
Diastolic BP I 1.09 (0.73, 1.64) 0.98 (0.56, 1.73) 1.00 (0.57, 1.77)
Diastolic BP II 1.24 (0.96, 1.60) 1.33 (0.98, 1.81) 1.32 (0.96, 1.79)
Systolic BP 1.12 (0.92, 1.36) 1.00 (0.81, 1.25) 0.96 (0.77, 1.19)
Total cholesterol I 1.06 (0.86, 1.30) 1.35* (1.05, 1.74) 1.37* (1.06, 1.77)
Total cholesterol II 1.15 (0.85, 1.55) 1.14 (0.74, 1.74) 1.25 (0.83, 1.91)
HDL cholesterol 1.26* (1.01, 1.57) 1.33 (0.97, 1.81) 1.31 (0.94, 1.82)
Fibrinogen 1.07 (0.84, 1.36) 0.97 (0.64, 1.46) 0.95 (0.62, 1.45)
C-reactive protein 1.07 (0.89, 1.30) 1.24 (0.98, 1.57) 1.30* (1.03, 1.63)
Glycated hemoglobin 1.46* (1.04, 2.04) 1.60* (1.06, 2.39) 1.51* (1.01, 2.27)
White blood cell count 1.39* (1.04, 1.86) 1.55* (1.11, 2.18) 1.53* (1.02, 2.28)

133
Table 7.8, Continued

Model 1 Model 2 Model 3 Model 4
N (# of people who died) 3933 (1173) 3017 (826) 1625 (467) 1590 (461)
Independent V H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I. H.R. 95% C. I.

Nutritional Factors
Albumin 1.11 (0.83, 1.48) 1.13 (0.82, 1.56)
Malnutrition 1.11 (0.85, 1.45) 1.03 (0.78, 1.36)
Anemia 1.50* (1.07, 2.10) 1.46* (1.07, 1.98)
Antioxidants 1.12 (0.78, 1.59) 0.96 (0.65, 1.42)
High carbohydrate 0.77 (0.53, 1.11) 0.74 (0.52, 1.05)
High fat 0.92 (0.70, 1.20) 0.93 (0.72, 1.21)
Low protein 1.16 (0.78, 1.72) 1.15 (0.75, 1.77)

Health Behaviors
Current Smoker 1.38 (0.87, 2.21)
Heavy drinker 0.53 (0.24, 1.13)
No regular exercise 1.84*** (1.52, 2.22)

Wald Chisq (df) 362.99***(3) 751.29***(13) 793.22***(20) 1223.10***(23)
-2 Log L 15065.60 10129.39 5466.74 5301.43
1
Reference category is normal weight (18.5<=BMI<=24.9)

134
Including a high white blood cell count, being in the bottom quartile of
antioxidants and high fat intake reduced the hazard linking low BMI to mortality
from 3.86 to 0.42 which is insignificance for those 40 to 64 years old. Low BMI
is significantly associated with a higher hazard of death for those who are 65
and older even after adjusting for biomarkers, nutritional factors, and health
behaviors.
7.3 Effects of Weight on Mortality When Biomarkers and Health Behaviors Are
Introduced for Specific Cause of Death: Cancer
Sections 7.3 and 7.4 report the results regarding the effects of weight
on specific causes of death, cancer, and cardiovascular disease (CVD) when
biomarkers and health behaviors are introduced. Figure 7.1 presents the
proportion by causes of death in age group. The multinomial logistic
regressions in sections 7.3 and 7.4 have dependent variables with three
categories, i.e., people who died from cancer, people who died from other
causes, and people who are alive in section 7.3.
According to Figure 7.1, the proportion of deaths from CVD is twice as
high as that from cancer among people who are 40 years and older. For
people who are 40 to 64 years old, the proportion of deaths from cancer is
almost the same as from cardiovascular disease. However, the proportion of
deaths from CVD is twice as high as from cancer among people who are 65
years and older. This means that when I ran the multinomial logistic
regression with a dependent variable which has people who died from cancer,

135
Figure 7.1
Proportion by Causes of Death in Age Group
0
10
20
30
40
50
60
40+ 40-64 65+
age group
%
Cancer
CVD
Others

people who died from other causes, and people who are alive, the category of
people who died from other causes includes people who died from
cardiovascular disease.
7.3.1 Effects of Weight on Mortality When Biomarkers and Health Behaviors
Are Introduced among People Who Are 40 and Older for Specific Cause of
Death: Cancer
This section attempts to examine the relationship between weight and
mortality for those who are 40 and older, but focusing on cancer as a specific
cause of death. Table 7.9 presents results from multivariate analyses using

136
multinomial logistic regressions with a dependent variable which has three
categories: people who died from cancer, people who died from other causes,
and people who are alive (reference category). In the base model controlling
for age and gender, low BMI people have a 3.28 times higher odds of death
from cancer and almost 4 times increase in the odds of death from other
causes than normal weight people. However, being overweight is not
significantly associated with the odds of death from either cancer or other
causes.
After considering biomarkers, the odds of death from cancer are about
3.5 times higher for underweight people as for those of normal weight, which
shows the increased odds of death from cancer compared to the baseline
model. Low diastolic blood pressure (DBP II), low HDL cholesterol, and high
white blood cell count are associated with a 1.86 times, a 1.56 times, a 2.34
times increase in the odds of death from cancer relative to being alive,
respectively. However, after controlling for biomarkers, the odds of death from
other causes decreased to 2.70 for underweight people, but are still
significant. Fibrinogen and glycated hemoglobin are associated with a 49%
and a 63% increase in the odds of death from other causes. Meanwhile, being
overweight is not significantly related to the odds of death from either cancer
or other causes.

137

Table 7.9
Odds Ratios From Multinomial Logistic Models of Specific Cause of Death Limiting the Sample to Those Who are 40 Years and Older: Cancer
Model 1 (N=9642) Model 2 (N=7522)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=367) (n=1176) (n=250) (n=790)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.09*** (1.07, 1.10) 1.12*** (1.11, 1.13) 1.09*** (1.07, 1.11) 1.12*** (1.11, 1.14)
Female 0.60* (0.40, 0.88) 0.58*** (0.48, 0.70) 0.60* (0.38, 0.96) 0.57*** (0.44, 0.73)

Underweight (BMI<18.5) 3.28*** (1.75, 6.14) 3.93*** (2.11, 7.32) 3.49*** (1.80, 6.74) 2.70* (1.22, 5.98)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 1.06 (0.77, 1.47) 0.95 (0.77, 1.18) 0.85 (0.55, 1.30) 0.78 (0.61, 1.01)

Biomarkers
Diastolic BP I 0.91 (0.54, 1.54) 1.19 (0.71, 2.00)
Normal Distolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 1.86* (1.09, 3.17) 1.16 (0.84, 1.60)
Systolic BP 1.18 (0.76, 1.84) 1.14 (0.90, 1.44)
Total cholesterol I 0.78 (0.49, 1.25) 1.15 (0.89, 1.49)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 0.80 (0.40, 1.60) 1.43 (0.96, 2.12)
HDL cholesterol 1.56* (1.04, 2.36) 1.18 (0.83, 1.66)
Fibrinogen 1.15 (0.68, 1.94) 1.49** (1.11, 2.01)
C-reactive protein 1.27 (0.82, 1.97) 1.33 (1.06, 1.66)
Glycated hemoglobin 1.34 (0.73, 2.48) 1.63* (1.06, 2.52)
White blood cell count 2.34* (1.21, 4.53) 1.21 (0.80, 1.82)

138

Table 7.9, Continued

Model 1 (N=9642) Model 2 (N=7522)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=367) (n=1176) (n=250) (n=790)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin
Malnutrition
Anemia
Antioxidants
High carbohydrate
High fat
Low protein

Health behaviors
Current smoker
Heavy drinker
No regular exercise

Wald chi-sq
b
628.21 1012.01
-2 Log L (df) 6313.31(8) 4174.15(28)
*< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

139
Table 7.9, Continued

Model 3 (N=4220) Model 4 (N=4144)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=131) (n=461) (n=130) (n=449)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.11*** (1.08, 1.14) 1.13*** (1.11, 1.15) 1.12*** (1.09, 1.15) 1.13*** (1.11, 1.16)
Female 0.49* (0.27, 0.89) 0.65* (0.46, 0.90) 0.50* (0.28, 0.90) 0.57** (0.39, 0.83)

Underweight (BMI<18.5) 2.13 (0.44, 10.35) 2.43 (0.96, 6.13) 2.09 (0.34, 12.93) 2.25 (0.87, 5.81)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 0.54* (0.33, 0.89) 0.71* (0.52, 0.97) 0.58* (0.33, 1.00) 0.70* (0.49, 0.98)

Biomarkers
Diastolic BP I 1.13 (0.42, 3.06) 1.26 (0.62, 2.58) 1.10 (0.39, 3.11) 1.26 (0.59, 2.71)
Normal Distolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 2.88** (1.47, 5.64) 1.31 (0.85, 2.00) 2.87** (1.50, 5.51) 1.28 (0.82, 2.01)
Systolic BP 0.97 (0.53, 1.78) 0.98 (0.74, 1.29) 1.00 (0.54, 1.83) 0.96 (0.71, 1.28)
Total cholesterol I 1.14 (0.65, 2.01) 1.58** (1.12, 2.21) 1.18 (0.66, 2.12) 1.56* (1.09, 2.25)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 1.00 (0.41, 2.48) 1.23 (0.74, 2.04) 1.00 (0.40, 2.53) 1.36 (0.83, 2.24)
HDL cholesterol 1.78* (1.13, 2.81) 1.30 (0.82, 2.06) 1.76* (1.11, 2.80) 1.33 (0.84, 2.10)
Fibrinogen 1.36 (0.68, 2.74) 1.29 (0.79, 2.12) 1.33 (0.66, 2.69) 1.30 (0.78, 2.14)
C-reactive protein 1.49 (0.83, 2.67) 1.46** (1.10, 1.94) 1.44 (0.78, 2.67) 1.44* (1.05, 1.98)
Glycated hemoglobin 1.52 (0.74, 3.12) 1.55 (0.86, 2.79) 1.51 (0.71, 3.20) 1.54 (0.84, 2.83)
White blood cell count 2.71* (1.05, 6.94) 1.59 (0.85, 2.97) 2.33 (0.90, 6.03) 1.54 (0.81, 2.96)

140
Table 7.9, Continued

Model 3 (N=4220) Model 4 (N=4144)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=131) (n=461) (n=130) (n=449)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin 1.29 (0.80, 2.07) 0.95 (0.68, 1.33) 1.31 (0.79, 2.15) 0.95 (0.66, 1.37)
Malnutrition 0.99 (0.51, 1.89) 1.12 (0.75, 1.67) 0.89 (0.46, 1.72) 1.03 (0.66, 1.63)
Anemia 1.62 (0.81, 3.24) 1.72* (1.15, 2.57) 1.73 (0.84, 3.56) 1.70* (1.12, 2.58)
Antioxidants 1.24 (0.70, 2.20) 1.56 (0.94, 2.58) 1.14 (0.63, 2.09) 1.32 (0.78, 2.24)
High carbohydrate 0.44 (0.18, 1.09) 0.61* (0.39, 0.93) 0.45 (0.18, 1.13) 0.60* (0.39, 0.93)
High fat 0.67 (0.37, 1.22) 0.87 (0.58, 1.29) 0.66 (0.36, 1.20) 0.79 (0.54, 1.17)
Low protein 1.57 (0.84, 2.94) 0.91 (0.54, 1.54) 1.60 (0.83, 3.06) 0.95 (0.54, 1.68)

Health behaviors
Current smoker 2.02 (0.95, 4.27) 1.50 (0.87, 2.59)
Heavy drinker 0.97 (0.38, 2.48) 0.61 (0.33, 1.16)
No regular exercise 1.03 (0.67, 1.60) 2.01*** (1.37, 2.95)

Wald chi-sq
b
4833.69 7579.88
-2 Log L (df) 2194.16(42) 2073.17(48)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

141
In Model 3, adding nutritional factors increased the odds of death from
cancer and other causes for underweight people up to 2.13 and 2.43,
respectively, which became insignificant. Interestingly, overweight people are
significantly less likely to die from both cancer and other causes than normal
weight people (cancer- OR: 0.54; 95% C.I.: 0.33-0.89, other causes- OR: 0.71;
95% C.I. 0.52, 0.97). The effects of low diastolic blood pressure (DBP II), low
HDL cholesterol, and high white blood cell count on death from cancer
became stronger than those in Model 2 (DBP II- OR: 2.88; 95% C.I.: 1.47-
5.64, HDL- OR: 1.78; 95% C.I.: 1.13-2.81, WBC- OR: 2.71; 95% C.I.: 1.05-
6.94). However, none of the nutritional factors are significantly associated with
the odds of death from cancer. Among biomarkers, high total cholesterol and
high C-reactive protein are associated with a 58% and a 46% increase in the
odds of death from other causes. This may be because two-thirds of people
who died from other causes are comprised of people who died from
cardiovascular disease, which is related to high total cholesterol and high C-
reactive protein. Anemia is also linked to a 72% increase in the odds of death
from other causes, but high carbohydrate intake is associated with a 39%
decrease in the odds of death from other causes.
After considering biomarkers, nutritional factors, and health behaviors
simultaneously in Model 4, the effects of being underweight on death from
cancer and other causes became slightly smaller than those in Model 3, but
were still insignificant. Similarly, the odds of death from cancer and other

142
causes for overweight people are not much different from those in Model 3.
Among biomarkers, the effects of low DBP II and low HDL cholesterol on
death from cancer are almost the same as those in Model 3, but the effect of
high white blood cell count on death from cancer became insignificant. None
of the nutritional factors and health behaviors have significant effects on death
from cancer. The odds of death from other causes for high total cholesterol,
high C-reactive protein, having anemia, and high carbohydrate intake are
almost the same as those in Model 3. People who do not regularly exercise
are significantly more likely to die from other causes than those who do
regularly exercise (OR: 2.01; 95% C.I.: 1.37-2.95).
Overall, while low BMI is not significantly related to death from both
cancer and other causes, high BMI is significantly associated with lower odds
of death from both cancer and other causes when controlling for biomarkers,
nutritional factors, and health behaviors, limiting the sample to those who are
aged 40 and older.
7.3.2 Effects of Weight on Mortality When Biomarkers and Health Behaviors
Are Introduced for Specific Cause of Death among People Who Are 40 to 64
Years Old, and 65 and Older: Cancer
Table 7.10 shows the odds ratios and 95% confidence intervals from
multinomial logistic regressions of specific causes of death- cancer and other
causes- limiting the sample to those who are 40 to 64 years old. However, the
odds of death from cancer were not computed in Model 3, which included

143

Table 7.10
Odds Ratios From Multinomial Logistic Models of Specific Cause of Death Limiting the Sample to Those Who are 40 to 64 Years Old: Cancer
Model 1 (N=5603) Model 2 (N=4442)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=96) (n=210) (n=61) (n=123)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.11*** (1.05, 1.17) 1.08*** (1.05, 1.11) 1.12*** (1.05, 1.19) 1.05* (1.01, 1.09)
Female 1.10 (0.58, 2.11) 0.64* (0.45, 0.91) 1.09 (0.51, 2.34) 0.52* (0.28, 0.97)

Underweight (BMI<18.5) 3.25 (0.68, 15.61) 3.70 (0.98, 13.92) 4.56 (0.88, 23.75) 2.45 (0.38, 15.94)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 1.39 (0.72, 2.69) 1.58* (1.01, 2.47) 1.02 (0.39, 2.66) 0.95 (0.50, 1.82)

Biomarkers
Diastolic BP I 0.45 (0.15, 1.36) 1.23 (0.57, 2.63)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 0.84 (0.25, 2.80) 2.14 (0.87, 5.26)
Systolic BP 1.17 (0.40, 3.46) 1.35 (0.67, 2.73)
Total cholesterol I 0.78 (0.36, 1.71) 1.36 (0.69, 2.69)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 0.59 (0.07, 5.17) 1.65 (0.76, 3.58)
HDL cholesterol 2.16 (0.85, 5.51) 0.86 (0.42, 1.78)
Fibrinogen 1.14 (0.44, 2.95) 2.97*** (1.93, 4.59)
C-reactive protein 1.06 (0.42, 2.70) 2.37*** (1.49, 3.77)
Glycated hemoglobin 0.91 (0.21, 4.08) 1.87 (0.67, 5.23)
White blood cell count 1.37 (0.43, 5.73) 1.26 (0.56, 2.82)

144

Table 7.10, Continued

Model 1 (N=5603) Model 2 (N=4442)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=96) (n=210) (n=61) (n=123)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin
Malnutrition
Anemia
Antioxidants
High carbohydrate
High fat
Low protein

Health behaviors
Current smoker
Heavy drinker
No regular exercise

Wald chi-sq
b
81.46 394.60
-2 Log L (df) 2051.4 4(8) 1234.39(28)
*< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

145
Table 7.10, Continued

Model 3 (N=2561) Model 4 (N=2525)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=30) (n=76) (n=30) (n=71)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.19*** (1.10, 1.28) 1.06* (1.01, 1.12) 1.19*** (1.11, 1.29) 1.08* (1.02, 1.14)
Female 0.98 (0.25, 3.88) 0.65 (0.27, 1.58) 1.19 (0.30, 4.67) 0.85 (0.38, 1.94)

Underweight (BMI<18.5) n.a. n.a. 0.70 (0.13, 3.96) n.a. n.a. 0.55 (0.11, 2.77)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 0.25* (0.09, 0.76) 0.91 (0.41, 2.03) 0.29* (0.08, 0.98) 0.79 (0.31, 1.98)

Biomarkers
Diastolic BP I 2.90 (0.79, 10.65) 1.97 (0.61, 6.35) 3.17 (0.84, 12.00) 2.36 (0.65, 8.52)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 0.79 (0.15, 4.25) 3.69* (1.22, 11.16) 0.84 (0.15, 4.79) 3.92* (1.21, 12.63)
Systolic BP 0.76 (0.22, 2.64) 0.79 (0.27, 2.33) 0.70 (0.17, 2.86) 0.79 (0.25, 2.50)
Total cholesterol I 1.74 (0.55, 5.56) 1.80 (0.69, 4.72) 1.82 (0.56, 5.92) 1.59 (0.49, 5.14)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 1.65 (0.10, 28.39) 0.97 (0.38, 2.47) 2.20 (0.13, 37.93) 1.06 (0.39, 2.91)
HDL cholesterol 1.93 (0.58, 6.44) 1.37 (0.53, 3.54) 2.25 (0.73, 7.00) 1.90 (0.76, 4.76)
Fibrinogen 2.28 (0.67, 7.72) 2.70* (1.20, 6.09) 2.56 (0.67, 9.80) 3.07* (1.29, 7.29)
C-reactive protein 1.21 (0.31, 4.76) 1.82* (1.00, 3.32) 1.19 (0.34, 4.08) 1.54 (0.80, 2.98)
Glycated hemoglobin 0.26 (0.05, 1.38) 1.30 (0.33, 5.07) 0.31 (0.05, 1.78) 1.52 (0.37, 6.31)
White blood cell count 1.00 (0.15, 6.72) 1.53 (0.51, 4.59) 0.52 (0.05, 5.83) 1.29 (0.36, 4.60)

146

Table 7.10, Continued

Model 3 (N=2561) Model 4 (N=2525)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=30) (n=76) (n=30) (n=71)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin 1.23 (0.36, 4.15) 0.66 (0.23, 1.91) 1.39 (0.44, 4.42) 0.62 (0.19, 2.02)
Malnutrition 0.55 (0.18, 1.67) 1.15 (0.45, 2.94) 0.44 (0.12, 1.60) 1.20 (0.35, 4.06)
Anemia 1.30 (0.32, 5.20) 1.18 (0.46, 2.98) 1.60 (0.38, 6.76) 1.18 (0.45, 3.14)
Antioxidants 2.82 (0.69, 11.59) 1.69 (0.53, 5.38) 2.62 (0.68, 10.11) 1.92 (0.61, 6.00)
High carbohydrate 0.08 (0.01, 0.81) 0.36 (0.11, 1.18) 0.08* (0.01, 0.92) 0.38 (0.11, 1.26)
High fat 0.55 (0.22, 1.38) 0.81 (0.33, 2.01) 0.57 (0.22, 1.47) 0.58 (0.21, 1.62)
Low protein 0.55 (0.11, 2.87) 0.70 (0.25, 1.95) 0.44 (0.09, 2.14) 0.75 (0.27, 2.03)

Health behaviors
Current smoker 2.63 (0.68, 10.11) 1.38 (0.48, 3.97)
Heavy drinker 1.44 (0.41, 5.03) 1.14 (0.48, 2.70)
No regular exercise 0.33 (0.10, 1.07) 0.91 (0.34, 2.44)

Wald chi-sq
b
116511.59 4750449.66
-2 Log L (df) 574.48(42) 522.62(48)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

147
biomarkers and nutritional factors, and Model 4, which included health
behaviors because the number of people who died from cancer is too small in
this age group.
In the base model, high BMI is significantly associated with a 58%
increase in the odds of death from other causes. When controlling
biomarkers, the odds of death from cancer and other causes are not
significant for being underweight and overweight. High fibrinogen and high C-
reactive protein significantly increased the odds of death from other causes
(Fibrinogen-OR: 2.97; 95% C.I.: 1.93-4.59, CRP-OR: 2.37; 95% C. I.: 1.49-
3.77). Interestingly, when considering biomarkers and nutritional factors,
overweight people are significantly less likely to die from cancer than normal
weight people in Model 3 (OR: 0.25; 95% C. I.: 0.09-0.76). Also, only high
carbohydrate intake is significantly associated with lower odds of death from
cancer (OR: 0.08; 95% C.I: 0.01-0.81). The effects of high fibrinogen and high
CRP still remain significant in Model 3. After adding nutritional factors in
Model 2, the odds of death from other causes for people who have low
diastolic blood pressure (DBP II) became significant in Model 3 (OR: 3.69;
95% C.I.: 1.22-11.16). When controlling for biomarkers, nutritional factors,
and health behaviors in Model 4, the odds of death from cancer for overweight
people still remain significant and the effects of low diastolic blood pressure
(DBP II) and high fibrinogen became stronger on the odds of death from other
causes. However, high C-reactive protein became insignificant in predicting

148
the odds of death from other causes in Model 4. In addition, nothing is
significantly associated with the odds of death from cancer and other causes
among health behaviors.
Table 7.11 is about multivariate analyses for the association between
BMI and death from cancer and other causes, limiting the sample to those who
are 65 years and older.
Low diastolic BP II, low HDL cholesterol and high white blood cell count
reduce the odds ratio linking low BMI and cancer mortality from 3.48 in the
base model to 2.95 in Model 2 controlling for biomarkers, and furthermore, it
becomes insignificant. When controlling for biomarkers, the odds of death
from other causes for those with low BMI reduces to 2.91 which is still
significant. Overweight people are significantly less likely to die from other
causes (OR: 0.75; 95% C. I.: 0.58-0.98), but not from cancer when controlling
for biomarkers. Only glycated hemoglobin is associated with a 67% increase
in the odds of death from other causes.
After adding nutritional factors, the odds of death from cancer are
insignificant although the odds ratio is very high, while the odds of death from
other causes remain significant (OR: 4.24; 95% C.I.: 1.19-15.12). Interesting
thing is that people who consume low protein are more likely to die from
cancer (OR: 2.09; 95% C.I.: 1.07-4.10). The effects of low diastolic BP II, low
HDL cholesterol, and high white blood cell count on cancer mortality become
bigger than those in Model 2. High total cholesterol (total cholesterol I) is

149

Table 7.11
Odds Ratios From Multinomial Logistic Models of Specific Cause of Death Limiting the Sample to Those Who are 65 Years and Older: Cancer
Model 1 (N=4039) Model 2 (N=3080)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=271) (n=966) (n=189) (n=667)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.09*** (1.07, 1.12) 1.14*** (1.12, 1.16) 1.10*** (1.06, 1.13) 1.15*** (1.13, 1.17)
Female 0.41*** (0.27, 0.63) 0.53*** (0.42, 0.66) 0.45** (0.29, 0.72) 0.52*** (0.41, 0.65)

Underweight (BMI<18.5) 3.48* (1.22, 9.96) 4.42*** (2.22, 8.79) 2.95 (0.76, 11.45) 2.91* (1.11, 7.62)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 0.88 (0.61, 1.27) 0.82 (0.64, 1.05) 0.76 (0.48, 1.19) 0.75* (0.58, 0.98)

Biomarkers
Diastolic BP I 1.38 (0.65, 2.94) 1.09 (0.57, 2.08)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 2.30** (1.24, 4.28) 1.09 (0.77, 1.53)
Systolic BP 1.23 (0.73, 2.08) 1.11 (0.86, 1.44)
Total cholesterol I 0.78 (0.47, 1.29) 1.12 (0.89, 1.41)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 0.85 (0.44, 1.64) 1.33 (0.81, 2.17)
HDL cholesterol 1.42* (1.00, 2.02) 1.32 (0.91, 1.92)
Fibrinogen 1.05 (0.61, 1.82) 1.14 (0.79, 1.64)
C-reactive protein 1.25 (0.76, 2.07) 1.10 (0.84, 1.45)
Glycated hemoglobin 1.48 (0.80, 2.75) 1.67* (1.00, 2.77)
White blood cell count 2.83** (1.38, 5.80) 1.16 (0.67, 1.99)

150
Table 7.11, Continued

Model 1 (N=4039) Model 2 (N=3080)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=271) (n=966) (n=189) (n=667)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin
Malnutrition
Anemia
Antioxidants
High carbohydrate
High fat
Low protein

Health behaviors
Current smoker
Heavy drinker
No regular exercise

Wald chi-sq
b
498.82 1894.82
-2 Log L (df) 5147.31(8) 3607.02(28)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

151

Table 7.11, Continued

Model 3 (N=1659) Model 4 (N=1619)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=101) (n=385) (n=100) (n=378)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.10*** (1.06, 1.15) 1.14*** (1.12, 1.17) 1.11*** (1.07, 1.14) 1.15*** (1.12, 1.17)
Female 0.41** (0.24, 0.70) 0.60** (0.43, 0.85) 0.36*** (0.20, 0.65) 0.46*** (0.32, 0.67)

Underweight (BMI<18.5) 4.54 (0.76, 27.09) 4.24* (1.19, 15.12) 5.69 (0.92, 35.12) 4.67* (1.30, 16.81)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 0.65 (0.35, 1.22) 0.71 (0.50, 1.01) 0.69 (0.36, 1.32) 0.71 (0.48, 1.05)

Biomarkers
Diastolic BP I 0.88 (0.26, 2.95) 0.95 (0.39, 2.31) 0.86 (0.24, 3.05) 0.94 (0.36, 2.48)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 3.44* (1.57, 7.53) 1.06 (0.63, 1.80) 3.44** (1.58, 7.53) 1.04 (0.60, 1.78)
Systolic BP 1.02 (0.53, 1.97) 0.99 (0.75, 1.32) 1.03 (0.54, 1.96) 0.95 (0.70, 1.29)
Total cholesterol I 0.95 (0.50, 1.78) 1.56** (1.15, 2.12) 1.02 (0.55, 1.89) 1.62* (1.17, 2.25)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 0.74 (0.32, 1.72) 1.25 (0.66, 2.36) 0.75 (0.32, 1.76) 1.41 (0.74, 2.70)
HDL cholesterol 1.75* (1.10, 2.80) 1.28 (0.80, 2.05) 1.67* (1.06, 2.64) 1.24 (0.74, 2.06)
Fibrinogen 1.05 (0.47, 2.37) 0.94 (0.55, 1.62) 0.99 (0.45, 2.19) 0.92 (0.52, 1.64)
C-reactive protein 1.50 (0.73, 3.09) 1.38 (0.99, 1.93) 1.48 (0.69, 3.14) 1.42* (1.01, 1.99)
Glycated hemoglobin 2.02 (0.90, 4.54) 1.81 (0.89, 3.67) 2.04 (0.86, 4.81) 1.73 (0.84, 3.53)
White blood cell count 3.68* (1.15, 11.73) 1.50 (0.72, 3.16) 3.72* (1.23, 11.23) 1.64 (0.74, 3.61)

152
Table 7.11, Continued

Model 3 (N=1659) Model 4 (N=1619)
Cancer
a
Other Causes
a
Cancer
a
Other Causes
a

(n=101) (n=385) (n=100) (n=378)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin 1.52 (0.85, 2.72) 1.02 (0.70, 1.51) 1.59 (0.85, 2.96) 1.06 (0.67, 1.68)
Malnutrition 1.29 (0.69, 2.38) 1.11 (0.73, 1.68) 1.15 (0.61, 2.19) 0.99 (0.63, 1.57)
Anemia 1.69 (0.81, 3.53) 1.87* (1.14, 3.06) 1.79 (0.85, 3.76) 1.87* (1.16, 3.02)
Antioxidants 0.92 (0.43, 1.99) 1.47 (0.80, 2.71) 0.77 (0.35, 1.68) 1.12 (0.57, 2.20)
High carbohydrate 0.55 (0.23, 1.31) 0.70 (0.44, 1.12) 0.53 (0.22, 1.31) 0.69 (0.43, 1.10)
High fat 0.78 (0.40, 1.51) 0.88 (0.59, 1.30) 0.79 (0.40, 1.55) 0.91 (0.61, 1.36)
Low protein 2.09* (1.07, 4.10) 1.02 (0.56, 1.84) 2.09* (1.02, 4.28) 1.00 (0.51, 1.95)

Health behaviors
Current smoker 2.05 (0.79, 5.34) 1.53 (0.79, 2.99)
Heavy drinker 0.63 (0.21, 1.86) 0.37* (0.14, 0.96)
No regular exercise 1.43 (0.93, 2.19) 2.56* (1.83, 3.58)

Wald chi-sq
b
4121.25 175373.07
-2 Log L (df) 1984.73(42) 1880.19(48)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

153
associated with a 56% increase in the odds of death from other causes.
People who have anemia are more likely to die from other causes, limiting the
sample to those who are 65 years and older.
Even after considering biomarkers, nutritional factors, and health
behaviors simultaneously, the effects of low diastolic BP II, low HDL
cholesterol, high white blood cell count, and low protein on cancer mortality
are almost the same as those in Model 3. Underweight people are more likely
to die from other causes than normal weight when controlling for all other risk
variables (OR: 4.67; 95% C. I.: 1.30-16.81). C-reactive protein is significantly
linked to a 42% increase in the odds of death from other causes. This
becomes significant after adding health behaviors in the model. People who
are heavy drinkers are less likely to die from other causes (OR: 0.37; 95% C.I:
0.14-0.96) and people who do not regularly exercise are more likely to die
from other causes (OR: 2.56; 95% C.I.: 1.83-3.58).
Overall, when controlling for all risk factors limiting the sample to those
who are 40 years and older, being underweight is not significantly related to
the higher odds of death from both cancer and other causes, but being
overweight is significantly associated with the lower odds of death from both
cancer and other causes. Limiting the sample to those who are 65 years and
older, the odds of death from cancer was lowest for underweight people when
controlling for biomarkers in Model 2. In addition, the odds of death from
cancer were insignificant in the model. However, the odds of death from

154
cancer for the overweight remained insignificant in all models for people who
are 65 years and older.
7.4 Effects of Weight on Mortality When Biomarkers and Health Behaviors are
Introduced for Specific Cause of Death: Cardiovascular Disease
7.4.1 Effects of Weight on Mortality When Biomarkers and Health Behaviors
Are Introduced among People Who Are 40 and Older for Specific Cause of
Death: Cardiovascular Disease
Table 7.12 shows results from multinomial logistic regressions of
cardiovascular disease and other causes relative to surviving, limiting the
sample to those who are 40 years and older. For underweight people, high
fibrinogen, high C-reactive protein, and high glycated hemoglobin reduce the
odds of death from cardiovascular disease from 3.20 in the base model to 2.66
in Model 2 and make the odds ratio insignificant. However, unexpectedly,
being overweight is not significantly associated with the odds of death from
cardiovascular disease in any of the four models limited to the sample who are
40 years and older. The odds of death from other causes for overweight
people remain less than 1.00 and are significant even after controlling for
biomarkers, nutritional factors, and health behaviors. This may be because
about one-half of people who died from other causes are people who died
from cancer and thus, they are less likely to die if they are overweight. When
biomarkers and nutritional factors are included in Model 3, the odds of death
from cardiovascular disease for underweight people increase to 3.96 and

155
Table 7.12
Odds Ratios From Multinomial Logistic Models of Specific Cause of Death Limiting the Sample to Those Who are 40 Years and Older:
Cardiovascular Disease
Model 1 (N=9642) Model 2 (N=7522)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=755) (n=788) (n=505) (n=535)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.13*** (1.11, 1.14) 1.09*** (1.08, 1.11) 1.13*** (1.11, 1.15) 1.10*** (1.08, 1.12)
Female 0.56*** (0.44, 0.72) 0.60*** (0.47, 0.76) 0.52*** (0.40, 0.67) 0.62*** (0.45, 0.87)

Underweight (BMI<18.5) 3.20* (1.54, 6.63) 4.13* (2.34, 7.27) 2.66 (0.99, 7.20) 3.06* (1.61, 5.81)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 1.07 (0.87, 1.33) 0.91 (0.72, 1.16) 0.93 (0.73, 1.19) 0.71* (0.52, 0.95)

Biomarkers
Diastolic BP I 1.28 (0.74, 2.22) 0.98 (0.55, 1.73)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 1.03 (0.71, 1.49) 1.62* (1.07, 2.45)
Systolic BP 1.02 (0.74, 1.40) 1.29 (0.93, 1.79)
Total cholesterol I 1.22 (0.89, 1.66) 0.89 (0.66, 1.21)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 0.93 (0.57, 1.53) 1.46 (0.99, 2.15)
HDL cholesterol 1.11 (0.74, 1.67) 1.44 (1.00, 2.07)
Fibrinogen 1.55** (1.14, 2.11) 1.26 (0.84, 1.89)
C-reactive protein 1.39* (1.07, 1.79) 1.26 (0.96, 1.65)
Glycated hemoglobin 1.74* (1.04, 2.90) 1.39 (0.87, 2.23)
White blood cell count 1.25 (0.77, 2.02) 1.74 (0.98, 3.08)

156
Table 7.12, Continued

Model 1 (N=9642) Model 2 (N=7522)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=755) (n=788) (n=505) (n=535)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin
Malnutrition
Anemia
Antioxidants
High carbohydrate
High fat
Low protein

Health behaviors
Current smoker
Heavy drinker
No regular exercise

Wald chi-sq
b
618.33 1431.08
-2 Log L (df) 6553.92(8) 4328.54 (28)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

157
Table 7.12, Continued

Model 3 (N=4220) Model 4 (N=4144)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=298) (n=294) (n=287) (n=292)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.14*** (1.11, 1.17) 1.11*** (1.09, 1.13) 1.15*** (1.12, 1.17) 1.12*** (1.09, 1.14)
Female 0.53*** (0.39, 0.73) 0.68 (0.42, 1.11) 0.48*** (0.33, 0.69) 0.63 (0.38, 1.03)

Underweight (BMI<18.5) 3.96* (1.37, 11.46) 1.39 (0.58, 3.35) 3.66* (1.18, 11.38) 1.39 (0.56, 3.46)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 0.85 (0.62, 1.16) 0.53** (0.35, 0.81) 0.84 (0.59, 1.20) 0.54** (0.34, 0.84)

Biomarkers
Diastolic BP I 1.79 (0.89, 3.59) 0.76 (0.34, 1.69) 1.85 (0.87, 3.91) 0.73 (0.31, 1.72)
Normal Distolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 1.05 (0.66, 1.65) 2.30** (1.36, 3.89) 1.04 (0.65, 1.68) 2.27** (1.34, 3.84)
Systolic BP 0.74 (0.50, 1.10) 1.31 (0.87, 1.97) 0.72 (0.48, 1.08) 1.28 (0.85, 1.93)
Total cholesterol I 1.78** (1.21, 2.61) 1.19 (0.80, 1.79) 1.75** (1.17, 2.62) 1.21 (0.79, 1.87)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 0.99 (0.54, 1.81) 1.35 (0.78, 2.34) 1.08 (0.59, 1.96) 1.45 (0.83, 2.52)
HDL cholesterol 1.17 (0.69, 1.99) 1.68* (1.12, 2.53) 1.22 (0.72, 2.04) 1.66* (1.13, 2.45)
Fibrinogen 1.33 (0.79, 2.23) 1.30 (0.72, 2.33) 1.34 (0.81, 2.21) 1.27 (0.70, 2.29)
C-reactive protein 1.56* (1.13, 2.17) 1.37 (0.96, 1.95) 1.50* (1.01, 2.23) 1.38 (0.98, 1.94)
Glycated hemoglobin 1.76 (0.89, 3.48) 1.34 (0.75, 2.42) 1.79 (0.89, 3.57) 1.34 (0.74, 2.44)
White blood cell count 1.36 (0.70, 2.65) 2.37* (1.06, 5.30) 1.34 (0.66, 2.71) 2.14 (0.96, 4.81)

158
Table 7.12, Continued

Model 3 (N=4220) Model 4 (N=4144)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=298) (n=294) (n=287) (n=292)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin 0.99 (0.69, 1.40) 1.07 (0.69, 1.66) 1.00 (0.68, 1.46) 1.07 (0.69, 1.67)
Malnutrition 1.13 (0.69, 1.85) 1.06 (0.64, 1.74) 1.05 (0.62, 1.78) 0.96 (0.57, 1.62)
Anemia 1.86* (1.14, 3.04) 1.54 (0.93, 2.55) 1.88* (1.14, 3.12) 1.54 (0.92, 2.60)
Antioxidants 1.37 (0.78, 2.38) 1.59 (0.90, 2.80) 1.16 (0.65, 2.10) 1.38 (0.79, 2.43)
High carbohydrate 0.61 (0.35, 1.09) 0.52* (0.28, 0.95) 0.61 (0.34, 1.08) 0.53* (0.29, 0.96)
High fat 1.02 (0.61, 1.71) 0.62** (0.44, 0.89) 0.92 (0.56, 1.51) 0.61** (0.43, 0.88)
Low protein 1.08 (0.58, 2.02) 1.04 (0.63, 1.72) 1.16 (0.59, 2.29) 1.04 (0.62, 1.75)

Health behaviors
Current smoker 1.47 (0.80, 2.70) 1.80* (1.01, 3.20)
Heavy drinker 0.53 (0.22, 1.28) 0.89 (0.43, 1.84)
No regular exercise 1.71* (1.05, 2.78) 1.74** (1.26, 2.41)

Wald chi-sq
b
2642.80 81355.52
-2 Log L (df) 2288.64(42) 2166.50 (48)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

159
become significant. High total cholesterol (total cholesterol I) and high C-
reactive protein are significantly associated with a 78% and a 56% increase in
the odds of death from cardiovascular disease when controlling for biomarkers
and nutritional factors. Also, people who have anemia are significantly more
likely to die than people who do not have anemia (OR: 1.86; 95% C.I.: 1.14-
3.04).
On the contrary, the odds of death from other causes decrease to 1.39
and become insignificant when controlling for biomarkers and nutritional
factors. Low diastolic BP II, low HDL cholesterol, and high white blood cell
count significantly increased 2.30 times, 1.68 times, and 2.37 times in the
odds of death from other causes when controlling for biomarkers and
nutritional factors, which is very similar to results for death from cancer in
Table 7.9. People whose diet includes high carbohydrates or high fat are less
likely to die from other causes than others (high carbohydrate- OR: 0.52; 95%
C.I.: 0.28-0.95, high fat- OR: 0.62; 95% C.I.: 0.44-1.72). Even after adding
health behaviors into the model, the effects of variables on death from
cardiovascular disease and other causes are similar to those in Model 3.
Expectedly, those who do not regularly exercise are more likely to die from
both cardiovascular disease and other causes when controlling for biomarkers,
nutritional factors, and health behaviors (OR: 1.71; 95% C.I.: 1.05-2.78, OR:
1.74; 95% C.I.: 1.26-2.41).

160
7.4.2 Effects of Weight on Mortality When Biomarkers and Health Behaviors
Are Introduced for Specific Cause of Death among People Who Are 40 to 64
Years Old, and 65 and Older: Cardiovascular Disease
Table 7.13 presents results from multinomial logistic regressions of
specific causes of death-cardiovascular disease and other causes- limiting the
sample to those who are 40 to 64 years older. However, the odds of death
from cardiovascular disease for underweight people do not compute in Model
2 through Model 4. This may be because the number of people who died from
cardiovascular disease is too small to calculate the odds of death from
cardiovascular disease. The odds of death from cardiovascular disease for
overweight people decreased from 1.89 in the base model to 1.31, and
furthermore, it becomes insignificant when including biomarkers. Meanwhile,
the odds of death from other causes decrease up to 0.37, which becomes
significant when controlling for biomarkers, nutritional factors, and health
behaviors in Model 4. This may be because people who died from other
causes include people who died from cancer. However, the number of people
who died from both cardiovascular disease and other causes is basically too
small in Models 3 and 4.
Table 7.14 shows interesting results regarding the effects of BMI on
mortality from cardiovascular disease and other causes when biomarkers,
nutritional factors, and health behaviors are introduced into the models limiting
the sample to those who are 65 years and older. High glycated hemoglobin

161

Table 7.13
Odds Ratios From Multinomial Logistic Models of Specific Cause of Death Limiting the Sample to Those Who are 40 to 64 Years Old:
Cardiovascular Disease
Model 1 (N=5603) Model 2 (N=4442)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=116) (n=190) (n=70) (n=114)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.09*** (1.05, 1.13) 1.09*** (1.05, 1.13) 1.05 (1.00, 1.10) 1.08*** (1.04, 1.13)
Female 0.51* (0.29, 0.88) 0.97 (0.63, 1.51) 0.34** (0.16, 0.72) 1.06 (0.60, 1.85)

Underweight (BMI<18.5) 1.34 (0.16, 10.88) 4.29** (1.53, 12.08) n.a. n.a. 4.14*
(1.15,
14.76)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 1.89* (1.04, 3.44) 1.34 (0.90, 2.02) 1.31 (0.62, 2.76) 0.84 (0.45, 1.60)

Biomarkers
Diastolic BP I 1.07 (0.41, 2.75) 0.88 (0.39, 1.97)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 2.85* (1.18, 6.90) 1.05 (0.34, 3.26)
Systolic BP 0.99 (0.41, 2.39) 1.50 (0.76, 2.94)
Total cholesterol I 1.38 (0.64, 2.98) 0.94 (0.54, 1.65)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 0.29* (0.08, 0.98) 1.73 (0.74, 4.00)
HDL cholesterol 0.74 (0.37, 1.49) 1.62 (0.70, 3.75)
Fibrinogen 4.29*** (2.24, 8.23) 1.21 (0.56, 2.63)
C-reactive protein 4.20*** (2.31, 7.64) 1.04 (0.59, 1.83)
Glycated hemoglobin 1.64 (0.56, 4.81) 1.43 (0.60, 3.42)
White blood cell count 1.32 (0.50, 3.50) 1.24 (0.44, 3.50)

162

Table 7.13, Continued

Model 1 (N=5603) Model 2 (N=4442)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=116) (n=190) (n=70) (n=114)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin
Malnutrition
Anemia
Antioxidants
High carbohydrate
High fat
Low protein

Health behaviors
Current smoker
Heavy drinker
No regular exercise

Wald chi-sq
b
105.59 7307.80
-2 Log L (df) 2048.08(8) 1217.46(28)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

163

Table 7.13, Continued

Model 3 (N=2561) Model 4 (N=2525)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=45) (n=61) (n=41) (n=60)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.04 (0.97, 1.11) 1.15*** (1.08, 1.21) 1.06 (1.00, 1.13) 1.15*** (1.08, 1.22)
Female 0.29* (0.10, 0.84) 1.64 (0.52, 5.16) 0.46 (0.15, 1.37) 1.70 (0.55, 5.30)

Underweight (BMI<18.5) n.a. n.a. 0.40 (0.08, 2.02) n.a. n.a. 0.32 (0.07, 1.52)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 1.13 (0.49, 2.62) 0.40* (0.17, 0.97) 1.12 (0.38, 3.28) 0.37* (0.14, 0.99)

Biomarkers
Diastolic BP I 2.27 (0.62, 8.23) 1.71 (0.65, 4.52) 3.06 (0.80, 11.71) 1.82 (0.66, 5.01)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 5.27** (2.00, 13.86) 1.44 (0.33, 6.31) 5.92*** (2.32, 15.12) 1.50 (0.33, 6.70)
Systolic BP 0.70 (0.19, 2.58) 1.06 (0.37, 3.05) 0.64 (0.16, 2.51) 1.07 (0.35, 3.22)
Total cholesterol I 2.53* (1.06, 6.01) 1.22 (0.42, 3.52) 2.18 (0.66, 7.16) 1.31 (0.44, 3.88)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 0.42 (0.08, 2.19) 1.82 (0.48, 6.87) 0.44 (0.08, 2.42) 2.10 (0.51, 8.64)
HDL cholesterol 1.14 (0.44, 2.96) 1.95 (0.81, 4.71) 1.81 (0.69, 4.75) 2.16 (0.93, 5.02)
Fibrinogen 3.11* (1.27, 7.64) 2.36 (0.91, 6.08) 3.75** (1.53, 9.22) 2.47 (0.97, 6.29)
C-reactive protein 3.92*** (2.28, 6.74) 0.63 (0.28, 1.41) 3.16** (1.40, 7.13) 0.65 (0.29, 1.45)
Glycated hemoglobin 1.77 (0.43, 7.26) 0.40 (0.15, 1.07) 2.07 (0.48, 8.97) 0.43 (0.15, 1.19)
White blood cell count 1.08 (0.22, 5.35) 1.78 (0.49, 6.54) 0.67 (0.10, 4.67) 1.44 (0.38, 5.49)

164

Table 7.13, Continued

Model 3 (N=2561) Model 4 (N=2525)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=45) (n=61) (n=41) (n=60)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin 0.35 (0.10, 1.18) 1.21 (0.42, 3.47) 0.30 (0.08, 1.09) 1.29 (0.45, 3.73)
Malnutrition 1.12 (0.46, 2.69) 0.95 (0.31, 2.96) 1.01 (0.33, 3.14) 0.90 (0.25, 3.29)
Anemia 0.81 (0.27, 2.40) 1.47 (0.53, 4.11) 0.72 (0.26, 2.01) 1.57 (0.53, 4.64)
Antioxidants 1.95 (0.41, 9.22) 1.96 (0.74, 5.19) 2.50 (0.52, 11.96) 1.93 (0.72, 5.18)
High carbohydrate 0.43 (0.09, 2.03) 0.15* (0.03, 0.83) 0.42 (0.09, 1.98) 0.18* (0.03, 0.97)
High fat 1.21 (0.42, 3.50) 0.34* (0.15, 0.79) 0.78 (0.24, 2.49) 0.36* (0.15, 0.84)
Low protein 0.45* (0.20, 0.97) 0.87 (0.25, 2.99) 0.45 (0.19, 1.05) 0.83 (0.24, 2.81)

Health behaviors
Current smoker 2.16 (0.56, 8.36) 1.54 (0.54, 4.36)
Heavy drinker 1.17 (0.39, 3.54) 1.32 (0.51, 3.40)
No regular exercise 0.56 (0.10, 3.17) 0.87 (0.37, 2.06)

Wald chi-sq
b
72487.21 n.a.
-2 Log L (df) 561.37(42) 515.39(48)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

165
Table 7.14
Odds Ratios From Multinomial Logistic Models of Specific Cause of Death Limiting the Sample to Those Who are 65 Years and Older:
Cardiovascular Disease
Model 1 (N=4039) Model 2 (N=3080)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=639) (n=598) (n=435) (n=421)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.15*** (1.12, 1.17) 1.11*** (1.09, 1.13) 1.16*** (1.13, 1.19) 1.12*** (1.09, 1.14)
Female 0.54 (0.41, 0.69) 0.46 (0.34, 0.61) 0.52*** (0.41, 0.66) 0.48*** (0.35, 0.67)

Underweight (BMI<18.5) 3.89** (1.68, 8.99) 4.47*** (2.43, 8.23) 3.15 (0.94, 10.50) 2.71* (1.23, 5.96)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 0.92 (0.72, 1.18) 0.75 (0.56, 1.01) 0.87 (0.67, 1.14) 0.66 (0.47, 0.92)

Biomarkers
Diastolic BP I 1.31 (0.64, 2.68) 1.01 (0.54, 1.88)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 0.94 (0.62, 1.44) 1.78* (1.15, 2.78)
Systolic BP 1.03 (0.74, 1.42) 1.27 (0.85, 1.90)
Total cholesterol I 1.15 (0.87, 1.53) 0.91 (0.65, 1.28)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 1.05 (0.60, 1.85) 1.32 (0.82, 2.13)
HDL cholesterol 1.25 (0.78, 2.00) 1.45* (1.06, 1.98)
Fibrinogen 1.06 (0.72, 1.56) 1.17 (0.76, 1.80)
C-reactive protein 1.02 (0.74, 1.41) 1.25 (0.90, 1.75)
Glycated hemoglobin 1.88* (1.04, 3.38) 1.38 (0.84, 2.25)
White blood cell count 1.05 (0.59, 1.89) 2.02* (1.02, 4.00)

166
Table 7.14, Continued

Model 1 (N=4039) Model 2 (N=3080)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=639) (n=598) (n=435) (n=421)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR
(95%
C.I.)

Nutritional factors
Albumin
Malnutrition
Anemia
Antioxidants
High carbohydrate
High fat
Low protein

Health behaviors
Current smoker
Heavy drinker
No regular exercise

Wald chi-sq
b
494.18 1566.78
-2 Log L (df) 5480.14(8) 3825.27(28)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

167

Table 7.14, Continued

Model 3 (N=1659) Model 4 (N=1619)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=253) (n=233) (n=246) (n=232)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)
Age 1.16*** (1.12, 1.19) 1.11*** (1.08, 1.15) 1.16*** (1.13, 1.19) 1.11*** (1.08, 1.15)
Female 0.55* (0.39, 0.78) 0.55* (0.36, 0.83) 0.41*** (0.28, 0.61) 0.46** (0.29, 0.72)

Underweight (BMI<18.5) 7.11** (1.87, 27.02) 2.39 (0.79, 7.28) 7.77** (1.95, 31.04) 2.89 (0.98, 8.54)
Normal weight (18.5<=BMI<=24.9) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Overweight (BMI=>25.0) 0.84 (0.57, 1.22) 0.58* (0.38, 0.89) 0.84 (0.55, 1.29) 0.60* (0.38, 0.95)

Biomarkers
Diastolic BP I 1.33 (0.52, 3.38) 0.58 (0.23, 1.48) 1.37 (0.50, 3.76) 0.54 (0.19, 1.53)
Normal Diastolic BP I 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Diastolic BP II 0.81 (0.45, 1.45) 2.39** (1.37, 4.17) 0.80 (0.43, 1.49) 2.34** (1.35, 4.06)
Systolic BP 0.74 (0.51, 1.09) 1.37 (0.85, 2.20) 0.71 (0.47, 1.06) 1.33 (0.83, 2.14)
Total cholesterol I 1.64** (1.14, 2.37) 1.18 (0.77, 1.83) 1.73** (1.21, 2.49) 1.21 (0.77, 1.90)
Normal total cholesterol 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00) 1.00 (1.00, 1.00)
Total cholesterol II 1.09 (0.53, 2.23) 1.15 (0.60, 2.19) 1.23 (0.60, 2.50) 1.26 (0.65, 2.43)
HDL cholesterol 1.16 (0.61, 2.21) 1.66* (1.12, 2.48) 1.12 (0.57, 2.21) 1.58* (1.06, 2.35)
Fibrinogen 0.92 (0.54, 1.59) 1.01 (0.52, 1.99) 0.90 (0.51, 1.58) 0.98 (0.50, 1.95)
C-reactive protein 1.30 (0.89, 1.90) 1.52 (0.98, 2.37) 1.32 (0.87, 2.00) 1.53 (0.99, 2.36)
Glycated hemoglobin 2.08 (0.91, 4.73) 1.67 (0.86, 3.23) 2.00 (0.88, 4.55) 1.66 (0.83, 3.31)
White blood cell count 1.22 (0.60, 2.48) 2.66 (0.96, 7.36) 1.42 (0.70, 2.90) 2.66 (0.96, 7.37)

168
Table 7.14, Continued

Model 3 (N=1659) Model 4 (N=1619)
Cardiovascular D
a
Other Causes
a
Cardiovascular D
a
Other Causes
a

(n=253) (n=233) (n=246) (n=232)
Independent Variable OR (95% C.I.) OR (95% C.I.) OR (95% C.I.) OR (95% C.I.)

Nutritional factors
Albumin 1.13 (0.75, 1.70) 1.13 (0.70, 1.84) 1.20 (0.74, 1.94) 1.17 (0.68, 1.99)
Malnutrition 1.15 (0.67, 1.97) 1.16 (0.72, 1.85) 1.05 (0.59, 1.87) 1.03 (0.63, 1.69)
Anemia 2.12** (1.23, 3.65) 1.56 (0.85, 2.87) 2.17** (1.28, 3.69) 1.55 (0.84, 2.85)
Antioxidants 1.23 (0.64, 2.37) 1.43 (0.71, 2.88) 0.93 (0.45, 1.94) 1.13 (0.55, 2.32)
High carbohydrate 0.70 (0.39, 1.26) 0.63 (0.33, 1.20) 0.70 (0.38, 1.28) 0.61 (0.33, 1.15)
High fat 0.96 (0.61, 1.51) 0.75 (0.50, 1.12) 1.01 (0.63, 1.62) 0.76 (0.51, 1.13)
Low protein 1.33 (0.65, 2.69) 1.17 (0.73, 1.89) 1.35 (0.61, 2.97) 1.13 (0.67, 1.93)

Health behaviors
Current smoker 1.30 (0.58, 2.94) 2.01* (1.08, 3.71)
Heavy drinker 0.21* (0.07, 0.70) 0.66 (0.28, 1.59)
No regular exercise 2.21*** (1.50, 3.25) 2.30*** (1.73, 3.05)

Wald chi-sq
b
3673.54 22036.50
-2 Log L (df) 2114.47(42) 2004.63(48)

+
< .10 *< .05 **< .01 ***< .001
Notes: OR=Odds Ratio
a
reference= people who are alive
b
Wald chi-sq of model minus intercept

169
reduces the odds of death from cardiovascular disease for underweight people
from 3.89 in the base model to 3.15 in Model 2 and it becomes insignificant.
However, for underweight people, the odds of death from cardiovascular
disease dramatically increase to 7.77, which becomes significant when
controlling for biomarkers, nutritional factors, and health behaviors. However,
for overweight people, the odds of death from cardiovascular disease do not
change very much and all odds ratios are insignificant from Model 1 through
Model 4. In Model 4, only high total cholesterol (total cholesterol I) and having
anemia significantly increase 1.73 times and 2.17 times the odds of death from
cardiovascular disease. People who are heavy drinkers are less likely to die
from cardiovascular disease than non heavy drinkers (OR: 0.21; 95% C.I.:
0.07-0.70). People who do not regularly exercise are more likely to die from
cardiovascular disease than people who do exercise, which is expected
because one of the important risk factors for cardiovascular disease is regular
exercise.
Meanwhile, being underweight significantly increases the odds of death
from other causes by 4.47 times when controlling for age and gender in the
base model. However, for the underweight, the odds of death from other
causes decrease to 2.39 and become insignificant when considering
biomarkers and nutritional factors in Model 3. Overweight people are
significantly less likely to die from other causes than normal weight people
when controlling for biomarkers, nutritional factors, and health behaviors in

170
Model 4 (OR: 0.60; 95% C. I.: 0.38-0.95). Low diastolic blood pressure (DBP
II) and low HDL cholesterol increase the odds of death from other causes,
2.34 times and 1.58 times, respectively. Also, current smokers are
significantly more likely to die from other causes than ever or never smokers
(OR: 2.01; 95% C.I: 1.08-3.71). People who do not regularly exercise are
significantly more likely to die from other causes than people who exercise
regularly (OR: 2.30; 95% C.I.: 1.73-3.05).
Being underweight is significantly related to the higher odds of death
from cardiovascular disease even after controlling for biomarkers, nutritional
factors, and health behaviors in two age groups, 40 years and older and 65
years and older. Also, being overweight is not significantly associated with the
odds of death from cardiovascular diseases in either two groups.
7.5 Discussion
In summary, overweight people have worse health than normal weight
persons because overweight people have significantly more high risk levels in
most biomarkers. Meanwhile, underweight people have significantly worse
nutritional status and health behaviors than normal weight people.
We also found that underweight people 65 years and older have an
increased odds of death from all causes and cardiovascular disease than
normal weight people even after controlling for risk factors such as
biomarkers, diet, and health behaviors. However, overweight people who are
65 years and older do not have significantly lower odds of mortality from all

171
causes, cancer, and cardiovascular disease than normal weight people. For
cancer, overweight people who are 40 to 64 years old have significantly lower
odds of death from cancer than normal weight people after controlling for
biomarkers and nutritional factors. However, overweight people 40 to 64
years old do not have significantly higher odds of death from cardiovascular
disease than normal weight people after controlling for all risk factors.
The result in Table 7.2 regarding the relationship between underweight
and poor nutrition is consistent with previous research. As Fried et al. (1998)
and Reuben et al. (2002) have shown, underweight people have significantly
poorer nutritional status, for example, low albumin and anemia are shown in
this chapter. Also, underweight people have significantly higher risk levels of
antioxidants than normal weight people, which confirms previous research, too
(Michelon et al., 2006; Ray et al., 2006). This may be because underweight
people are more likely to smoke and smoking causes lower levels of
antioxidants (Hanioka et al., 2006; Rock et al., 1996; Northrop-Clewes &
Thurnham, 2007; Wei et al., 2001). Meanwhile, the prevalence of most
indicators of high risk biomarkers for underweight people does not differ
significantly from those of normal weight in Table 7.1, which is not consistent
with previous research (Hammett et al., 2007). On the other hand, low total
cholesterol (Total Cholesterol II) for underweight people is higher than that of
normal weight people as expected (Reuben et al., 2002).

172
Overweight people have significantly higher risk in most biomarkers
than normal weight people, which is not a surprise because the literature has
linked higher weight to cardiovascular risk factors such as high cholesterol and
high blood pressure (Gregg et al., 2005). In addition, the prevalence of current
smokers among overweight people is significantly lower than that for normal
weight people. Overweight people are significantly less likely to regularly
exercise than normal weight people, which was expected based on previous
research (Kruger et al., 2002; Martinez-Gonzalez et al., 1999).
Consistent with previous research, this chapter has found that people
with low weight are more likely to die than normal weight or heavier people
(Allison et al., 1997; Flegal et al., 2005; Grabowski & Ellis, 2001), for both
people aged 40 and older and 65 and older. This is because underweight
people aged 40 and older or 65 and older have higher odds of death from all-
causes and cardiovascular disease than normal weight people even after
controlling for all risk factors including biomarkers, diets, and health behaviors.
However, unexpectedly, underweight people are not significantly associated
with higher odds of death from cancer in all age groups. For all-cause of
death among underweight people aged 65 and over, high C-reactive protein
and low HDL cholesterol are related to high hazard of death as previous
research has shown (Fried et al., 1998; Harris et al., 1999; Reuben et al.,
2002).

173
Regarding death from cancer among people who are 65 years and
older, low diastolic blood pressure (Diastolic blood pressure II), low HDL
cholesterol, and high white blood cell count increased the odds of death as
expected (Fried et al., 1998; Goldman et al., 2006; Hammett et al., 2007).
With respect to death from cardiovascular disease, not exercising regularly is
an important factor to predict high mortality, which is also emphasized in
previous research (Lee et al., 1999).
However, regarding all-causes of death, the association between
underweight and high mortality was not found in the younger -40 to 64 years
old- sample. This means that high white blood cell count, being in the bottom
quartile in antioxidants, and high calorie intake from fat mediated the effects of
low BMI on high mortality and then reduced a hazard ratio to 0.42 which
became insignificant.
Interesting result includes that overweight people who are 40 to 64
years old are significantly associated with lower odds of death from cancer
after controlling for risk factors although the model was somewhat unstable,
which is not consistent with some previous studies because high BMI has
been widely accepted as being related to poor health outcomes including
mortality (Adams et al., 2006; Burke et al., 1999; Colin et al., 2002; Must et al.,
1999). It does not seem to be true for death from cardiovascular disease
because high BMI is not significantly associated with higher odds of death
than normal weight after controlling for risk factors. Also, as a recent study

174
indicated cardiovascular risk factors (Gregg et al., 2005), high total cholesterol
(Total cholesterol I) increased the odds of death from cardiovascular disease
for people who are 65 years and older. Furthermore, antioxidants did not have
an association with death from either cancer or cardiovascular disease in this
analysis, which was unexpected.
Based on the results of this chapter, practitioners working with older
adults need to understand that older people need to avoid being underweight
because being underweight is really an important predictor of mortality from
all-causes of death, particularly cardiovascular disease. Also, they need to
strongly encourage older adults to exercise regularly, particularly if older adults
have experienced cardiovascular disease, because no regular exercise is
highly related to increased odds of death from cardiovascular disease.
On the other hand, relatively young people who are 40 to 64 years old
need to try to keep normal weight to avoid dying from cardiovascular diseases
and if they have cancer, they should try not to loose their weight to avoid
higher mortality from cancer.

175
Chapter VIII: Conclusions and Discussion
The goals of this dissertation were to examine differences in BMI
distribution among Japanese older adults, American older adults, and the U.S.
middle-aged (chapter IV); to investigate which factors predict being
underweight, overweight or obese among Japanese older adults and
American older adults and how weight is associated with chronic diseases and
ADL functioning difficulties (chapter V); to determine how BMI is associated
with all-cause mortality in Japanese older adults and the U.S. middle-aged and
older adults (chapter VI); to investigate how different levels of BMI are related
to all cause mortality and mortality from specific-causes considering the role of
biomarkers indicating physiological risk, nutrition, and health behaviors
(chapter VII). To answer these research questions, I used a number of
datasets and statistical approaches as shown in Table 8.1.
Table 8.1
Summary of Datasets and Statistical Analyses by Chapter
Datasets Statistical Analyses (D.V.)
Chapter IV NUJLSOA (70+), LSOA II (70+), HRS (51-61), %, chi-square test
NHANES III (40+)
Chapter V NUJLSOA (70+), LSOA II (70+) Multinomial logistic regression-
(under-, normal, and overweight)
Logistic regression-
(Six health conditions)
Chapter VI NUJLSOA (70+), LSOA II (70+), HRS (51-61) Logistic regression-
(Death or not)
Chapter VII NHANES III (40+, 40-64, 65+) chi-square test
Cox proportional hazard regression-
(Death or not-all causes of death)
Multinomial logistic regression-
(Cancer, other causes, being alive;
CVD, other causes, being alive)

176
8.1 Findings and Discussion in Chapter IV: Characteristics of Samples
In the fourth chapter, the sample characteristics were described,
forming the basis for the research questions in the following chapters. This
chapter found that more than half of American older men are overweight or
obese in the LSOA II, but only about 14% of Japanese of comparable age
were overweight or obese in the NUJLSOA. This difference in body weight
was almost the same for women. American older men and women have
higher levels of education and income than Japanese older men and women.
A greater proportion of American men reported a number of health conditions
than Japanese men, with the only exception being the higher prevalence of
stroke among the Japanese. Japanese older men and women are more likely
to be heavy drinkers and do more regular exercise than American older men
and women. Japanese older men are more likely to be current smokers than
American older men.
In terms of weight distribution, Japanese older adults (NUJLSOA) have
a lower weight distribution with less dispersion than both American middle-
aged (HRS) and older adults (LSOA II). The highest death rate occurred
among underweight people in all three samples. Interestingly, when I limited
the sample to never smokers, for both Japanese older adults and American
older adults, the lowest death rate occurred among the overweight (25.0 ≤ BMI
≤ 29.9), but for the U.S. middle-aged adults it occurred for those of normal
weight.

177
For Americans who are aged 40 and older (NHANES III), underweight
people have higher levels of low DBP (DBP II), low total cholesterol (Total
cholesterol II), high fibrinogen, high white blood cell count, and high
prevalence of indicators of poor diet. In addition, they are more likely to be
current smokers, heavy drinkers, and less likely to exercise regularly than
people with higher levels of BMI.
The results suggest that American people aged 51 to 61 who maintain
normal body weight are at the lowest risk of death. In addition, those who are
40 years old and older in the U.S. should be aware of the risk of being either
underweight or obese. Both BMI groups are related to higher levels of risk in a
number of biomarkers, poor nutrition, and poor health behaviors.
8.2 Findings and Discussion for Chapter V: Factors Affecting Body Weight and
the Correlates of It with Health Outcomes in Two Different Countries
This chapter used two datasets to examine factors associated with
being underweight, overweight or obese, and how being underweight or
overweight (BMI ≥ 25) is associated with health conditions among Japanese
older adults and American older adults.
This chapter found that older cohorts were actually more represented in
the underweight category and less represented in the overweight category for
men and women in both Japan and the U.S., which is different from previous
research in which higher BMI has been associated with older age (Ogden et
al., 2006). Education and income did not have an effect on BMI for men in

178
both countries. However, as years of education increased, both American and
Japanese women are less likely to be overweight after controlling for
demographic variables, socioeconomic variables, and health behaviors. This
result shows that there is a stronger inverse relationship between SES and
BMI among women than among men, which is consistent with previous
research (Zhang & Wang, 2004a).
There are differences between Japanese older men and American
older men in the way in which health behaviors are associated with being
underweight or being overweight. Current smokers are less likely to be
overweight among both Japanese men and American men, and more likely to
be underweight among American men. This is expected given previous
research (Mizoue et al., 2006), although it was expected that current smoking
would be associated with being underweight for Japanese men, too. Heavy
drinkers are less likely to be underweight among Japanese older men, but
more likely to be overweight among American older men. For men in both
countries, those who do not regularly exercise are less likely to be
underweight. For Japanese older women, health behaviors are not closely
associated with being underweight or overweight, but for American women,
the relationship between current smoking and BMI is stronger than that for
American men. However, American women are the only group for which not
exercising regularly leads to being overweight.

179
With respect to the effects of BMI on health conditions including ADL
difficulties, BMI does not seem to be a significant predictor of more health
problems in Japan except for arthritis for Japanese women. On the other
hand, overweight Japanese men have unexpectedly lower odds of having
diabetes, which is unexpected given results from previous research (National
Heart, Lung, and Blood Institute, 1998). This may be because the proportion
of Japanese men who are overweight is quite low. Also, in contrast to
Americans, even the overweight are not as heavy in Japan.
In the U.S., it is interesting that underweight men and women are more
likely to have ADL difficulties. This may be because American underweight
people are very frail, and low muscle mass leads to having difficulties in
performing ADLs. In addition, it should be pointed out that not only heavier,
but also thin people, may have a higher risk of heart disease compared to
normal weight people in the U.S.
8.3 Findings and Discussion in Chapter VI: The Effect of Body Mass Index on
Mortality in Different Countries and Age Groups
In chapter VI, I attempted to explore whether the relationship between
BMI and mortality differs in Japanese older adults, American older adults, and
the U.S. middle-aged. Before limiting the sample, for Japanese older adults,
being underweight has greater risk of death than being normal weight. For
American older adults, being overweight was associated with a reduced
likelihood of dying although most of the indicators of poor health and

180
behaviors are related to being overweight or obese. For American middle-
aged persons, overweight or obese people are less likely to die than normal
weight, controlling for health status and behaviors.
After limiting the sample to never smokers, I found that the results were
almost the same as for the total sample for Japanese and American older
adults. Limiting the sample to never smokers, Japanese older adults showed
similar results to previous studies in which low weight is related to an
increased relative risk of death (Kuriyama et al., 2004; Miyazaki et al., 2002;
Ohta et al., 2001). When limiting the sample to never smokers, for American
older adults, the results showed that the risk of death for underweight and
obese people were not significantly higher than that for normal weight people,
which is not consistent with the results of recent research in which not only
being obese but also being underweight is associated with increased mortality
(Allison et al., 1997; Flegal et al., 2005). However, I found that overweight
American older adults are significantly less likely to die relative to normal
weight people. Therefore, the results for older Japanese and Americans are
consistent with some previous research (Baik et al., 2000; Grabowski & Ellis,
2001) in which higher BMI was not linearly related with higher mortality for
older people.
For American middle-aged, no BMI category was significantly
associated with mortality when limiting the sample to never smokers. The
results showed that cancer, stroke, diabetes, and no regular exercise are

181
much more important predictors of death than body weight. These results are
thus somewhat inconsistent with previous research (Diehr et al., 1998)
because the link between higher BMI and mortality was not significant for U.S.
middle-aged never smokers.
Being overweight is the optimal BMI only for American older adults.
One reason for this finding may be that people with diseases related to high
BMI had already died before reaching older ages. However, for Japanese
older adults, obese people are not significantly less likely to die than normal
weight people. Instead, underweight people are more likely to die than normal
weight people. Thus, Japanese older adults need to avoid being underweight.
For the U.S. middle-aged, it cannot be concluded that normal weight is the
optimal BMI because normal weight was not linked to significantly less risk of
death in this study.
8.4 Findings and Discussion in Chapter V: The Effect of Body Mass Index on
Mortality Through Biomarkers, Nutritional Factors, and Health Behaviors in the
NHANES III
The next study focused on whether biological risks, nutritional factors,
and health behaviors differ for thin and overweight people from those of
normal weight persons in the U.S. In addition, the link between BMI and
specific causes of mortality, as well as all-cause mortality, was investigated
with controls for other biomarkers, diet, and health behaviors, which are risk

182
factors for mortality. These links were examined among Americans aged 40
and older, 40 to 64 years old, and 65 years and older.
For the first research question, I found that overweight people had
significantly higher levels of risk in many biomarkers than normal weight
people, which is not a surprise because they tend to have high cardiovascular
risk factors, such as high cholesterol and high blood pressure (Gregg et al.,
2005). However, underweight people have significantly worse nutritional
status and health behaviors than normal weight people, which is consistent
with results given in previous research, particularly for low albumin and
anemia (Fried et al., 1998; Reuben et al., 2002) and antioxidants (Michenlon
et al., 2006; Ray et al., 1996). The high risk of low antioxidants for
underweight people may be because underweight people are more likely to
smoke and smoking causes lower levels of antioxidants (Hanioka et al., 2006;
Rock et al., 1996; Northrop-Clewes & Thurnham, 2007; Wei et al., 2001).
Among biomarkers indicating high risk, only low total cholesterol (Total
cholesterol II) in underweight people is higher than that of normal weight
people as expected (Reuben et al., 2002).
It was also found that underweight people who are 65 years and older
have an increased hazard of death from all causes and an increased odds of
death from cardiovascular disease than normal weight people after controlling
for risk factors such as biomarkers, diet, and health behaviors. This result is
somewhat consistent with previous research (Allison et al., 1997; Flegal et al.,

183
2005; Grabowski & Ellis, 2001). Regarding death from cancer among people
who are 65 years and older, low diastolic blood pressure (DBP II), low HDL
cholesterol, and high white blood cell count increased odds of death as
expected (Fried et al., 1998; Goldman et al., 2006; Hammett et al., 2007;
Shankar, Mitchell, Rochtchina, & Wang, 2007). With respect to death from
cardiovascular disease, no regular exercise is an important predictor of high
mortality, which was also emphasized in previous research (Lee et al., 1999).
However, relative to normal weight, being overweight is not significantly
associated with a lower hazard of mortality from all causes or lower odds of
death from cancer and cardiovascular disease when limiting the sample to
those who are 65 years and older.
Overweight people, limiting the sample to those who are 40 to 64 years
old, are significantly associated with lower odds of death from cancer, but not
from cardiovascular disease than normal weight people after controlling for all
risk factors. This is consistent with a recent study which indicated
cardiovascular risk factors (Gregg et al., 2005; Strandberg & Tilvis, 2000), high
total cholesterol (Total cholesterol I) and high C-reactive protein (Jenny et al.,
2007) increased the odds of death from cardiovascular disease for those who
are 40 to 64 years old when controlling for biomarkers and nutritional factors.
The results of this chapter indicate that practitioners working with old
adults need to clarify for older people the need to avoid being underweight
because underweight is an important predictor of mortality from all causes,

184
particularly cardiovascular disease. On the other hand, relatively young
people who are 40 to 64 years old have to maintain normal weight to avoid
dying from cardiovascular diseases and if they have cancer, they should try
not to loose weight to lower risk of mortality from cancer.
8.5 Implications
The results of this study suggest that the level of education is an
important predictor of body weight among Japanese and American older
women. Thus, we would expect in the near future that weight control for the
next cohorts could be better with higher education. This is because the next
cohorts are able to better understand how important it is to maintain proper
body weight for quality of later life.
This study also implies that the relationship between BMI and all- cause
mortality is not the same across countries and age groups. Results of this
study have policy implications for nurses, geriatricians, and directors in
facilities for older adults as well as young adults who need to understand
appropriate weight control in different settings. Specifically, older adults need
to avoid being underweight and loosing body weight because underweight
Japanese are significantly more likely to die than normal weight Japanese. In
the U.S., practitioners need to help older adults not to be obese. However, if
older adults are slightly overweight, practitioners may suggest regular exercise
rather than pushing them to loose body weight. For U.S. middle-aged, even if
overweight and obese people are not significantly more likely to die than

185
normal weight people, middle-aged American people need to keep normal
weight in order to avoid having diseases related to high BMI, such as stroke
and diabetes. Furthermore, practitioners need to encourage all people to
exercise regularly across countries and ages because our results indicate that
people who do not regularly exercise have a greater risk of death in all three
samples.
It is also expected that the findings are clinically relevant for health
practitioners in linking weight to risks of mortality. Since underweight tends to
be related to malnutrition and overweight to high risk in a number of
biomarkers, health practitioners should monitor people who are underweight or
overweight/obese by checking nutritional status and biomarkers regularly.
Through communication between practitioners and their clients, those who are
underweight or overweight/obese will be able to avoid higher risk of death.
8.6 Discussion Regarding the Association Between Body Weight and Frailty
I used BMI as an indicator to predict mortality and investigated how
different levels of BMI are associated with mortality in Japanese older adults,
American middle-aged, and American older adults. Results showed that
underweight is an important predictor of mortality for Japanese older adults,
but not for American older adults. Meanwhile, it is important for American
middle-aged to maintain normal weight to avoid high mortality.
If so, the results bring up some questions about how BMI is associated
with frailty among older adults and how frailty plays a role in the relationship

186
between body weight and mortality. Frailty is defined as a clinical syndrome
characterized by three or more of the following criteria: self-reported
exhaustion, weakness (grip strength), slow walking speed, unintentional
weight loss of 10 lbs in past year, and low physical activity (Fried et al., 2001).
Frailty is not same as either disability or comorbidity, but in a distinct syndrome
(Fried et al., 2001).
Both underweight and overweight, or obese are related to higher risks
of frailty than normal weight (Woods et al., 2005), and particularly obesity is
significantly related to frailty even when controlling for education, presence of
diseases including diabetes, coronary artery disease, knee osteoarthritis, and
so on (Blaum, Xue, Michelon, Semba, & Fried, 2005). In addition, obesity is a
cause of frailty for older adults and intentional weight loss and exercise can
reduce frailty in obese older adults (Villareal, Banks, Sinacore, Siener, & Klein,
2006). These results from previous research thus emphasize that frailty is
associated with not only being underweight but also being obesity among
older adults. This is somewhat unexpected as conventional knowledge is that
frailty more often appears among underweight older adults.
Furthermore, research has shown that frailty results in a greater risk of
mortality among older adults. Frail men were more likely to die than nonfrail
men in all body weight categories (Cawthon et al., 2007). For American older
women, frailty is an independent predictor of hip fracture, ADL disability,
hospitalization, and death (Woods et al., 2005). In addition to evidences of the

187
association between frailty and mortality, some biomarkers related to
inflammation are significantly associated with frailty in older women. Higher
white blood cell and interleukin-6 were independently associated with frailty in
older American women (Leng, Xue, Tian, Walston, & Fried, 2007). Further
research on how BMI is associated with frailty and how it affects the
relationship between BMI and mortality in older adults would be an appropriate
extension of this dissertation.
8.7 Discussion Regarding Other Indicators Related to Body Weight
Body mass index is only one of many indicators of weight and adiposity.
There has been research related to other indicators of anthropometry. Other
markers of visceral adiposity include waist circumferences (WC), waist-to-hip
ratio (WHR), and fat mss (FM) and markers of muscle mass including fat free
mass (FFM) and midarm muscle circumference (MAMC), and so on. Since
this dissertation used the most commonly used measure to indicate body
weight, BMI (Body mass index) in the prediction of mortality, there are still
questions regarding how these other measures of adiposity are related to
mortality.
According to Wannamethee, Shaper, Lennon, and Whincup (2007), the
most effective predictor of mortality is the combination of both WC and MAMC.
The lowest mortality risk occurred among men who have low WC (<=102cm)
and above-median muscle mass. Not only death but also disability is
associated with WC for older adults (Guallar-Castillon et al., 2007). After

188
controlling for age, education, health behaviors including tobacco use, alcohol
consumption, and physical activity, men in the highest waist circumference
quintile had about 2 times more risk of mobility disability than those in the
lowest quintile. BMI reduced a little bit the association between waist
circumference and disability.
In addition to WC and MAMC, fitness and fatness are also important
indicators which are associated with body weight and predict mortality.
Because overweight people are heavier than normal weight people, they
would be also more active and fit than normal weight people. This would lead
to overweight people being less likely to die than normal weight people.
Especially for women, fitness and fatness are important predictors of bone
mineral density because bone mineral density is a risk factor for fracture which
may lead to death in the end (Stewart et al., 2002).
According to recent research (Sui et al., 2007), BMI is strongly
positively correlated with fat mass, but it is only moderately correlated with
waist circumference, percent body fat, and fat-free mass. Interestingly, BMI is
a significant predictor of all-cause mortality risk and fitness is strongly
inversely related to all-cause mortality. However, other markers of adiposity
including percent body fat, fat-free mass, and waist circumference were not
associated with mortality (Sui et al., 2007). Also, higher levels of fitness are
inversely associated with all-cause mortality in both normal- and overweight
people, but all-cause mortality in obese unfit individuals is not significantly

189
higher risk than that in obese fit individuals. Result from recent research
seems to support the results of this dissertation because when overweight
older adults exercise regularly, their mortality is the lowest among all BMI
groups in the U.S. However, more investigation of the association between
markers of adiposity and mortality and the relationship between mortality and
fitness as well as fatness is warranted, as well as research on how BMI is
related to these other indicators in Asian older adults.
8.8 Limitations and Recommendations for Further Research
This dissertation involved multiple analyses of four large datasets with a
focus on, body weight. Several limitations need to be mentioned.
In examining the relationship between BMI and mortality in three
different samples including Japanese older adults, American older adults, and
the U.S. middle-aged, the time period of the analysis was limited to deaths
within four-years of follow-up. In order to fully understand the link between
BMI and mortality, death information covering a longer follow-up period would
be ideal in order to control for early death that might be related to preexisting
diseases.
The lack of a date of death in the NUJLSOA limited the statistical
analyses to logistic regression when I examine the association between BMI
and mortality in the three samples. If information regarding the date of death
were available in the Japanese sample, I could have analyzed all three

190
datasets with Cox proportional hazard regression which is more appropriate
for exploring the link between BMI and mortality over a multi- year time period.
The sizes of the sample within the overweight or obese categories
among Japanese older adults and underweight category among Americans
are too small to observe significant results in linking BMI and mortality. This
particularly limited the analyses by specific age groups and specific causes of
death in NHANES III.
I believe that future research should examine the link between BMI and
mortality in not only Japan, but also other Asian countries, including Korea and
China, for both young and older adults because there is little known about the
association between being underweight and mortality in these countries. In
addition, the proportion of those overweight and obese has grown dramatically
in those countries. Future research should further explore the mechanisms
between being underweight and mortality by including even more
comprehensive biomarkers, nutritional factors, psychological factors, as well
as demographic and socioeconomic variables. Also, future research should
investigate how underweight is associated with frailty, body fat, and fitness
when underweight affects mortality. Such analyses may show additional
mediators in the relationship between underweight and mortality.

191
Bibliography
Adams, K. F., Schatzkin, A., Harris, T. B., Kipnis, V., Mouw, T., Ballard-
Barbash, R., et al. (2006). Overweight, obesity, and mortality in a large
prospective cohort of persons 50 to 71 year old. The New England
Journal of Medicine, 355, 763-778.

Allison, D. B., Gallagher, D., Heo, M., Pi-Sunyer, F. X., & Heymsfield, S. B.
(1997). Body mass index and all-cause mortality among people age 70
and over: The Longitudinal Study of Aging. International Journal of
Obesity, 21, 424-431.

Andersen, R. E., Franckowiak, S., Christmas, C., Walston, J., & Crespo, C.
(2001). Obesity and reports of no leisure time activity among old
Americans: Results from the Third National Health and Nutrition
Examination Survey. Educational Gerontology, 27, 297-306.

Aronson, D., Bartha, P., Zinder, O., Kerner, A., Markiewicz, W., Avizohar, O.,
et al. (2004). Obesity is the major determinant of elevated C-reactive
protein in subjects with the metabolic syndrome. International Journal of
Obesity, 28, 674-679.

Baglietto, L., English, D. R., Hopper, J. L., Powles, J., & Giles, G. G. (2006).
Average volume of alcohol consumed, type of beverage, drinking pattern
and the risk of death from all causes. Alcohol and Alcoholism, 41, 664-
671.

Baik, I., Ascherio, A., Rimm, E. B., Giovannucci, E., Spiegelman, D., Stampfer,
M. J., et al. (2000). Adiposity and mortality in men. American Journal of
Epidemiology, 152, 264-271.

Bender, R., Jockel, K.-H., Trautner, C., Spraul, M., & Berger, M. (1999). Effect
of age on excess mortality in obesity. Journal of the American Medical
Association, 281, 1498-1504.

Blaum, C. S., Xue, Q. L., Michelon, E., Semba, R. D., & Fried, L. P. (2005).
The association between obesity and the frailty syndrome in older
women: The Women’s Health and Aging Studies. Journal of the
American Geriatrics Society, 53, 927-934.

Block, G. (1991). Epidemiologic evidence regarding vitamin C and cancer.
American Journal of Clinical Nutrition, 54, 1310S-1314S.

192
Bostom, A. G., Silbershatz, H., Rosenberg, I. H., Selhub, J., D’Agostino, R. B.,
Wolf, P. A., et al. (1999). Nonfasting plasma total homocysteine levels
and all-cause and cardiovascular disease mortality in elderly
Framingham men and women. Archives of Internal Medicine, 159, 1077-
1080.

Burke, J. P., Williams, K., Gaskill, S. P, Hazuda, H. P., Haffner, S. M., & Stern,
M. P. (1999). Rapid rise in the incidence of type 2 diabetes from 1987 to
1996. Archives of Internal Medicine, 159, 1450-1456.

Calle, E. E., Thun, M. J., Petrelli, J. M., Rodriguez, C., & Health, C. W., Jr.
(1999). Body mass index and mortality in a prospective cohort of U.S.
adults. The New England Journal of Medicine, 341, 1097-1105.

Cawthon, P. M., Marshall, L. M., Michael, Y., Dam, T.-T., Ensrud, K. E.,
Barrett-Connor, E., et al. (2007). Frailty in older men: Prevalence,
progression, and relationship with mortality. Journal of the American
Geriatrics Society, 55, 1216-1223.

Center for Disease Control and Prevention (1994, July). Plan and Operation of
theThird National Health and Nutrition Examination Survey, 1988-1994.
Vital and Health Statistics, 32 (1). Retrieved August 18, 2007, from
http://www.cdc.gov/nchs/data/series/sr_01/sr01_032.pdf

Center for Disease Control and Prevention (1996, December). Third National
Health and Nutrition Examination Survey, 1988-1994 (Data File
Documentation No. 76200). Hyattsville, MD: Centers for Disease Control
and Prevention. Retrieved August, 18, 2007, from
http://www.cdc.gov/nchs/data/nhanes/nhanes3/exam-acc.pdf

Center for Disease Control and Prevention (2007, July). Longitudinal Studies
of Aging LSOAs: (LSOAII). Retrieved August 15, 2007, from
http://www.cdc.gov/nchs/about/otheract/aging/lsoa2.htm

Chei, C. L., Iso, H., Yamagishi, K., Inoue, M., & Tsugane, S. (2008). Body
mass index and weight change since 20 years of age and risk of
coronary heart disease among Japanese: The Japan Public Health
Center-Based Study. International Journal of Obesity, 32, 144-151.

Colin, B. A., Adair, L. S., & Popkin, B. M. (2002). Ethnic differences in the
association between body mass index and hypertension. American
Journal of Epidemiology, 155, 346-353.

193
Comstock, G. W., Alberg, A. J., Huang, H.-Y., Wu, K., Burke, A. E., Hoffman,
S. C., et al. (1997). The risk of developing lung cancer associated with
antioxidants in the blood: Ascorbic acid, carotenoids, α-tocopherol,
selenium, and total peroxyl radical absorbing capacity. Cancer
Epidemiology, Biomarkers & Prevention, 6, 907-916.

Crimmins, E. M., Saito, Y., & Ingegneri, D. G. (1997). Trends in disability-free
life expectancy in the United States, 1970-1990. Population and
Development Review, 23, 555-572.

Davison, K. K., Ford, E. S., Cogswell, M. E., & Dietz, W. H. (2002).
Percentage of body fat and body mass index are associated with mobility
limitations in people aged 70 and older from NHANES III. Journal of the
American Geriatrics Society, 50, 1802-1809.

Diehr, P., Bild, D. E., Harris, T. B., Duxbury, A., Siscovick, D., & Rossi, M.
(1998). Body mass index and mortality in nonsmoking older adults: The
Cardiovascular Health Study. American Journal of Public Health, 88,
623-629.

Escobar-Morreale, H. F., Villuenda, G., Botella-Carretero, J. I., Sancho, J., &
San Millan, J. L. (2003). Obesity, and not insulin resistance, is the major
determinant of serum inflammatory cardiovascular risk markers in pre-
menopausal women. Diabetologia, 45, 625-633.

Farrell, S. W., Braun, L., Barlow, C. E., Cheng, Y. J., & Blair, S. N. (2002). The
relation of body mass index, cardiorespiratory fitness, and all-cause
mortality in women. Obesity Research, 10, 417-423.

Felson, D. T., Zhang, Y., Anthony, J. M., Naimark, A., & Anderson, J. J.
(1992). Weight loss reduces the risk for symptomatic knee osteoarthritis
in women. Annals of Internal Medicine, 116, 535-539.

Flegal, K. M., Carroll, M. D., Kuczmarski, R. J., & Johnson, C. L. (1998).
Overweight and obesity in the United States: prevalence and trends,
1960-1994. International Journal of Obesity, 22, 39-47.

Flegal, K. M., Carroll, M. D., Ogden, C. L., & Johnson, C. L. (2002).
Prevalence and trends in obesity among US adults, 1999-2000. Journal
of the American Medical Association, 288, 1723-1727.

194
Flegal, K. M., Graubard, B. I., Williamson, D. F., & Gail, M. H. (2005). Excess
deaths associated with underweight, overweight, and obesity. Journal of
the American Medical Association, 293, 1861-1867.

Fried, L. P., Kronmal, R. A., Newman, A. B., Bild, D. E., Mittelmark, M. B.,
Polak, J. F., et al. (1998). Risk factors of 5-year mortality in older adults.
Journal of the American Medical Association, 279, 585-592.

Goldman, N., Turra, C. M., Glei, D. A., Seplaki, C. L., Lin, Y. H., & Weinstein,
M. (2006). Predicting mortality from clinical and nonclinical biomarkers.
Journal of Gerontology: Biological Sciences and Medical Sciences, 61A,
1070-1074.

Grabowski, D. C., & Ellis, J. E. (2001). High body mass index does not predict
mortality in older people: Analysis of the Longitudinal Study of Aging.
Journal of the American Geriatrics Society, 49, 968-979.

Gregg, E. W., Cheng, Y. J., Cadwell, B. L., Imperatore, G., Williams, D. E.,
Flegal, K. M., et al. (2005). Secular trends in cardiovascular disease risk
factors according to body mass index in US adults. Journal of American
Medical Association, 293, 1868-1874.

Guallar-Castillon, P., Sagardui-Villamor, J., Banegas, J. R., Graciani, A.,
Fornes, N. S., Garcia, E. L., et al. (2007). Waist circumference as a
predictor of disability among older adults. Obesity, 15, 233-244.

Hammett, C. J., Prapavessis, H., Baldi, J. C., Ameratunga, R., Schoenbeck,
U., Varo, N., et al. (2007). Variation in blood levels of inflammatory
markers related and unrelated to smoking cessation in women.
Preventive Cardiology, 10, 68-75.

Hanioka, T., Ojima, M., Tanaka, K., & Aoyama, H. (2006). Association of total
tooth loss with smoking, drinking alcohol and nutrition in elderly
Japanese: Analysis of national database. Gerodontology, 24, 87-92.

Harris, T. B., Ferrucci, L., Tracy, R. P., Corti, M. C., Wacholder, S., Ettinger,
W. H., Jr., et al. (1999). Associations of elevated interleukin-6 and C-
reactive protein levels with mortality in the elderly. American Journal of
Medicine, 106, 506-512.

Harriss, L. R., English, D. R., Powles, J., Giles, G. G., Tonkin, A. M., Hodge,
A. M., et al. (2007). Dietary patterns and cardiovascular mortality in the

195
Melbourne Collaborative Cohort Study. American Journal of Clinical
Nutrition, 86, 221-229.

Hayashi, R., Iwasaki, M., Otani, T., Wang, N., Miyazaki, H., Yoshiaki, S., et al.
(2005). Body mass index and mortality in a middle-aged Japanese
cohort. Journal of Epidemiology, 15, 70-77.

Heeringa, S. G., & Connor, J. H. (1995). Technical Description of the Health
and Retirement Survey Sample Design. Retrieved August 13, 2007, from
Michigan University, Institute for Social Research Web site:
http://hrsonline.isr.umich.edu/docs/sample/hrs/HRSSAMP.pdf

Himes, C. L. (2000). Obesity, disease, and functional limitation in later life.
Demography, 37, 73-82.

Hu, G., & Cassano, P. A. (2000). Antioxidant nutrients and pulmonary function:
The Third National Health and Nutrition Examination Survey (NHANES
III). American Journal of Epidemiology, 151, 975-981.

Ito, H., Nakasuga, K., Ohshima, A., Maruyama, T., Kaji, Y., Harada, M., et al.
(2003). Detection of cardiovascular risk factors by indices of obesity
obtained from anthropometry and dual-energy X-ray absorptiometry in
Japanese individuals. International Journal of Obesity and Related
Metabolic disorders, 27, 232-237.

Jeffery, R. W., French, S. A., Forster, J. L., & Spry, V. M. (1991).
Socioeconomic status differences in health behaviors related to obesity:
The healthy worker project. International Journal of Obesity, 15, 689-
696.

Jenny, N. S., Yanez, N. D., Psaty, B. M., Kuller, L. H., Hirsch, C. H., & Tracy,
R. P. (2007). Inflammation biomarkers and near-term death in older men.
American Journal of Epidemiology, 165, 684-695.

John, U., Meyer, C., Rumpf, H.-J., Hapke, U., & Schumann, A. (2006).
Predictors of increased body mass index following cessation of smoking.
The American Journal of Additions, 15, 192-197.

Kadowaki, T., Watanabe, M., Okayama, A., Hishida, K., Okamura, T.,
Miyamatsu, N., et al. (2006). Continuation of smoking cessation and
following weight change after intervention in a healthy population with
high smoking prevalence. Journal of Occupational Health, 48, 402-406.

196
Kim, D. M., Ahn, C. W., & Nam, S. Y. (2005). Prevalence of obesity in Korea.
Obesity Review, 6, 117-121.

Krause, N., Liang, J., Shaw, B. A., Sugisawa, H., Kim, H. K., & Sugihara, Y.
(2002). Religion, death of a loved one, and hypertension among older
adults in Japan. Journal of Gerontology: Psychological Sciences and
Social Sciences, 57, S96-S107.

Krueger, P. M., Rogers, R. G., Hummer, R. A., & Boardman, J. D. (2004).
Body mass, smoking, and overall and cause-specific mortality among
older U.S. adults. Research on Aging, 26, 82-107.

Kruger, H. S., Venter, C. S., Vorster, H. H., & Margetts, B. M. (2002). Physical
inactivity is the major determinant of obesity in black women in the North
West Province, South Africa: the THUSA study. Transition and Health
During Urbanisation of South Africa. Nutrition, 18, 422-427.

Kuriyama, S., Ohmori, K., Miura, C., Suzuki, Y., Nakaya, N., Fujita, K., et al.
(2004). Body mass index and mortality in Japan: The Miyagi Cohort
Study. Journal of Epidemiology, 14 (supple1), S33-S38.

Lee, C. D., Blair, S. N., & Jackson, A. S. (1999). Cardiovascular fitness, body
composition, and all-cause and cardiovascular disease mortality in men.
American Journal of Clinical Nutrition, 69, 373-380.

Leng, S. X., Xue, Q.-L., Tian, J., Walston, J. D., & Fried, L. P. (2007).
Inflammation and frailty in older women. Journal of the American
Geriatrics Society, 55, 864-871.

Liu, L., Bopp, M. M., Roberson, P. K., & Sullivan, D. H. (2002). Undernutrition
and risk of mortality in elderly patients within 1 year of hospital
discharge. Journal of Gerontology: Medical Sciences, 57A, M741-M746.

Manson, A. E., Bassuk, S. S., Hu, F. B., Stampfer, M. J., Colditz, G. A., &
Willett, W. C. (2007). Estimating the number of deaths due to obesity:
Can the divergent findings be reconciled? Journal of Women’s Health,
16, 168-176.

Manson, A. E., Stampfer, M. J., Hennekens, C. H., & Willett, W. C. (1987).
Body weight and longevity: A reassessment. Journal of the American
Medical Association, 257, 353-358.

197
Manson, J. E., Willett, W. C., Stampfer, M. J., Colditz, G. A., Hunter, D. J.,
Hankinson, S. E., et al. (1995). Body weight and mortality among
women. The New England Journal of Medicine, 333, 677-685.

Marmot, M. G., & Smith, G. D. (1989). Why are the Japanese living longer?
British Medical Journal, 299, 1547-1551.

Martinez-Gonzalez, M. A., Martinez, J. A., Hu, F. B., Gibney, M. J., & Kearney,
J. (1999). Physical inactivity, sedentary lifestyle and obesity in the
European Union. International Journal of Obesity and Related Metabolic
Disorders, 23, 1192-1201.

McEligot, A. J., Largent, J., Ziogas, A., Peel, D., & Anton-Culver, H. (2006).
Dietary fat, fiber, vegetable, and micronutrients are associated with
overall survival in postmenopausal women diagnosed with breast
cancer. Nutrition and Cancer, 55, 132-140.

Michelon, E., Blaum C., Semba, R. D., Xue, Q.-L., Ricks, M. O., and Fried, L.
P. (2006). Vitamin and carotenoid status in older women: Associations
with the frailty syndrome. Journal of Gerontology: Medical Sciences,
61A, 600-607.

Miyazaki, M., Babazono, A., Ishii, T., Sugie, T., Momose, Y., Iwahashi, M., et
al. (2002). Effects of low body mass index and smoking on all-cause
mortality among middle-aged and elderly Japanese. Journal of
Epidemiology, 12, 40-44.

Mizoue, T., Kasai, H., Kubo, T., & Tokunaga, S. (2006). Leanness, smoking,
and enhanced oxidative DNA damage. Cancer Epidemiology,
Biomarkers & Prevention, 15, 582-585.

Monda, K. L., & Popkin, B. M. (2005). Cluster analysis methods help to clarify
the activity-BMI relationship of Chinese youth. Obesity Research, 13,
1042-1051.

Mora, S., Lee, I. M., Buring, J. E., & Ridker, P. M. (2006). Association of
physical activity and body mass index with novel and traditional
cardiovascular biomarkers in women. Journal of American Medical
Association, 295, 1412-1419.

Must, A., Spadano, J., Coakley, E. H., Field, A. E., Colditz, G., & Dietz, W. H.
(1999). The disease burden associated with overweight and obesity.
Journal of the American Medical Association, 282, 1523-1529.

198
Nakaya, N., Kurashima, K., Yamaguchi, J., Ohkubo, T., Nishino, Y., Tsubono,
Y., et al. (2004). Alcohol consumption and mortality in Japan: The Miyagi
Cohort Study. Journal of Epidemiology, 14(Suppl), S18-S25.

National Heart, Lung, and Blood Institute (1998). Clinical Guidelines on the
Identification, Evaluation, and Treatment of Overweight and Obesity in
Adults. Washington, D.C.: U. S. Public Health Service.

Northrop-Clewes, C. A., & Thurnham, D. I. (2007). Monitoring micronutrients in
cigarette smokers. Clinica Chimica Acta, 377, 14-38.

Ogden, C. L., Carroll, M. D., Curtin, L. R., McDowell, M. A., Tabak, C. J., &
Flegal, K. M. (2006). Prevalence of overweight and obesity in the United
States, 1999-2004. Journal of the American Medical Association, 295,
1549-1555.

Ohta, A., Aoki, S., Takeuchi, K., Yosiaki, S., & Suzuki, S. (2001). Lifestyle and
sociodemographic risk factors for death among middle-aged and elderly
residents in Japan from a five-year follow-up cohort study. Journal of
Epidemiology, 11, 51-60.

Peeters, A., Barendregt, J. S., Willekens, F., Mackenbach, J. P., Mamun, A.
A., & Bonneux, L. (2003). Obesity in adulthood and its consequences for
life expectancy: A life-table analysis. Annals of Internal Medicine, 138,
24-32.

Pierce, J. P., Stefanick, M. L., Flatt, S. W., Natarajan, L., Sternfeld, B.,
Madlensky, L., et al. (2007). Greater survival after breast cancer in
physically active women with high vegetable-fruit intake regardless of
obesity. Journal of Clinical Oncology, 25, 2345-2351.

Ravussin, E., Fontveille, A., Swinburn, B. A., & Bogardus, C. (1993). Risk
factors for the development of obesity. Annals of the New York Academy
of Sciences, 683, 141-150.

Ray, A. L., Semba, R. D., Walston, J., Ferrucci, L., Cappola, A. R., Ricks, M.
O., et al. (2006). Low serum selenium and total carotenoids predict
mortality among older women living in the community: The women’s
health and aging studies. The Journal of Nutrition, 136, 172-176.

Reuben, D. B., Cheh, A. I., Harris, T. B., Ferrucci, L. Rowe, J. W., Tracy, R. P.,
et al. (2002). Peripheral blood markers of inflammation predict mortality

199
and functional decline in high-functioning community-dwelling older
persons. Journal of the American Geriatrics Society, 50, 638-644.

Reynolds, S. L., Saito, Y., & Crimmins, E. M. (2005). The impact of obesity on
active life expectancy in older American men and women. Gerontologist,
45, 438-444.

Rock, C. L., Jacob, R. A., & Bowen, P. E. (1996). Update on the biological
characteristics of the antioxidant micronutrients: vitamin C, vitamin E,
and the carotenoids. Journal of the American Dietetic Association, 96,
693-702.

Sahyoun, N. R., Jacques, P. F., Zhang, Z. L., Juan, W., & McKeown, N. M.
(2006). Whole-grain intake is inversely associated with the metabolic
syndrome and mortality in older adults. American Journal of Clinical
Nutrition, 83, 124-131.

Sahyoun, N. R., Zhang, Z. L., & Serdula, M. K. (2005). Barriers to the
consumption of fruits and vegetables among older adults. Journal of
Nutrition for the Elderly, 24, 5-21.

Sakamoto, M. (2006). The situation of the epidemiology and management of
obesity in Japan. International Journal of Vitamin and Nutrition
Research, 76, 253-256.

Schoenborn, C. A. (2004). Marital status and health: United States, 1999-
2002. Advance Data, 351, 1-32.

Schumacher, A., Peersen, K., Sommervoll, L., Seljeflot, I., Arnesen, H.,
Otterstad, J. E. (2006). Physical performance is associated with markers
of vascular inflammation in patients with coronary heart disease.
European Journal of Cardiovascular Prevention and Rehabilitation, 13,
356-362.

Seidell, J. C., Verschuren, W. M. M., van Leer, E. M., & Kromhout, D. (1996).
Overweight, underweight, and mortality: A prospective study of 48,287
men and women. Archives of the Internal Medicine, 156, 958-963.

Shankar, A., Mitchell, P., Rochtchina, E., & Wang, J. J. (2007). The
association between circulating white blood cell count, triglyceride level
and cardiovascular and all-cause mortality: Population-based cohort
study. Atherosclerosis, 192, 177-183.

200
Sharma, M. (2007). Behavioural interventions for preventing and treating
obesity in adults. Obesity Reviews, 8, 441-449.

Sjodahl, K., Lu, Y., Nilsen, T, I., Ye, W., Hveem, K., Vatten, L., et al. (2007).
Smoking and alcohol drinking in relation to risk of gastric cancer: A
population-based, prospective cohort study. International Journal of
Cancer, 120, 128-132.

Smith, J. P. (1999). Healthy bodies and thick wallets: The dual relation
between health and economic status. Journal of Economic Perspectives,
13, 145-166.

Song, Y.-M., & Sung, J. (2001). Body mass index and mortality: A twelve-year
prospective study in Korea. Epidemiology, 12, 173-179.

Soteriades, E. S., Hauser, R., Kawachi, I., Liarokapis, D., Christiani, D. C, &
Kales, S. N. (2005). Obesity and cardiovascular disease risk factors in
firefighters: A prospective cohort study. Obesity Research, 13, 1756-
1763.

Stevens, J., Cai, J., Pamuk, E. R., Williamson, D. F., Thun, M. J., & Wood, J.
L. (1998). The effect of age on the association between body-mass index
and mortality. The New England Journal of Medicine, 338, 1-7.

Stewart, K. J., Deregis, J. R., Turner, K. L., Bacher, A. C., Sung, J., Hees, P.
S., et al. (2002). Fitness, fatness, and activity as predictors of bone
mineral density in older persons. Journal of Internal Medicine, 252, 381-
388.

Strandberg, T. E., & Tilvis, R. S. (2000). C-reactive protein, cardiovascular risk
factors, and mortality in a prospective study in the elderly.
Arteriosclerosis, Thrombosis, and Vascular Biology, 20, 1057-1060.

Sui, X., LaMonte, M. J., Laditka, J. N., Hardin, J. W., Chase, N., Hooker, S. P.,
et al. (2007). Cardiorespiratory fitness and adiposity as mortality
predictors in older adults. Scandinavian Journal of Medicine and Science
in Sports, 18, 119-120.

Thomas, F., Bean K., Pannier, B., Oppert, J-M, Guize, L., & Benetos, A.
(2005). Cardiovascular mortality in overweight subjects: The key role of
associated risk factors. Hypertension, 46, 654-659.

201
Thorpe, R. J., & Ferraro, K. F. (2004). Aging, obesity, and mortality: Misplaced
concern about obese older people? Research on Aging, 26, 108-129.

Tsugane, S., Sasaki, S., & Tsubono, Y. (2002). Under- and overweight impact
on mortality among middle-aged Japanese men and women: A 10-y
follow up of JPHC Study Cohort I. International Journal of Obesity, 26,
529-537.

USC/UCLA Center of Biodemography and Population Health (2004). Nihon
University Japanese Longitudinal Study of Aging. Retrieved August 13,
2007, from http://www.usc.edu/dept/gero/CBPH/nujlsoa/overview.htm

USC/UCLA Center of Biodemography and Population Health (2005). Individual
biomarkers. Retrieved March 1, 2008, from
http://www.usc.edu/dept/gero/CBPH/biomarker/biomarkers-index3.htm

Villareal, D. T., Banks, M., Sinacore, D. R., Siener, C., & Klein, S. (2006).
Effect of weight loss and exercise on frailty in obese older adults.
Archives of Internal Medicine, 166, 860-866.

Wannamethee, S. G., Shaper, A. G., Lennon, L., & Whincup, P. H. (2007).
Decreased muscle mass and increased central adiposity are
independently related to mortality in older men. The American Journal of
Clinical Nutrition, 86, 1339-1346.

Wei, W., Kim, Y., & Boudreau, N. (2001). Association of smoking with serum
and dietary levels of antioxidants in adults: NHANES III, 1988-1994.
American Journal of Public Health, 91, 258-264.

Weinstein, A. R., Sesso, H. D., Lee, I. M., Cook, N. R., Manson, J. E., Buring,
J. E., et al. (2004). Relationship of physical activity vs body mass index
with type 2 diabetes in women. Journal of American Medical Association,
292, 1188-1194.

Wilkinson, R. G. (1994). The epidemiological transition: from material scarcity
to social disadvantage? Daedalus, 123, 61-77.

Willett, W. C. (1999). Guidelines for health weight. The New England Journal
of Medicine, 341, 427-434.

Woods, N. F., LaCroix, A. Z., Gray, S. L., Aragaki, A., Cochrane, B. B.,
Brunner, R. L., et al., (2005). Frailty: Emergency and consequences in

202
women aged 65 and older in the women’s health initiative observational
study. Journal of the American
Geriatrics Society, 53, 1321-1330.

World Health Organization (2007). World Health Statistics 2007. Retrieved on
May 22, 2007, from
http://www.who.int/whosis/whostat2007_10highlights.pdf

World Health Organization Expert Consultation (2004). Appropriate body-mass
index for Asian populations and its application for policy and intervention
strategies. Lancet, 363, 157-163.

Xiaobin, Z., Li, Z., & Kelvin, S. T. O. (2004). Income inequalities under
economic restructuring in Hong Kong. Asian Survey, 44, 442-473.

Yoshiike, N., Kaneda, F., & Takimoto, H. (2002). Epidemiology of obesity and
public health strategies for its control in Japan. Asia Pacific Journal of
Clinical Nutrition, 11 (suppl), S727-S731.

Yoshiike, N., Seino, F., Tajima, S., Arai, Y., Kawano, M., Furuhata, T., et al.
(2002). Twenty-year changes in the prevalence of overweight in
Japanese adults: The National Nutrition Survey 1976-1995. Obesity
Reviews, 3, 183-190.

Zhang, Q., & Wang, Y. (2004a). Socioeconomic inequality of obesity in the
United States: Do gender, age, and ethnicity matter? Social Science and
Medicine, 58, 1171-1180.

Zhang, Q., & Wang, Y. (2004b). Trends in the association between obesity
and socioeconomic status in U.S. Adults: 1971 to 2000. Obesity
Research, 12, 1622-1632.

Abstract (if available)

Abstract The goals of this dissertation were to examine differences in Body mass index (BMI) distribution among Japanese older adults, American older adults, and the U.S. middle-aged

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Translating two physical activity programs for older adults into home- and community-based settings: "Active Start" and "Healthy Moves for Aging Well"

PDF

Using mixed methods to identify the characteristics of older fraud victims

PDF

Falls among older adults: characteristics of fallers, predictors of falls, and the impact of falls on health care and long-term care utilization

PDF

Childlessness and psychological well-being across life course as manifested in significant life events

PDF

Normative obligations and parental care in social context

PDF

Examination of the long-term psychosocial and functioning effects of a healthy living, life management behavior intervention for older adults

PDF

Investigating the effectiveness of a social work intervention on reducing hospital readmissions among older adults

PDF

Children 's migration and the financial, social, and psychological well-being of older adults in rural China

PDF

Longitudinal change in active life expectancy: the longitudinal studies of aging 1984-2000

PDF

Family caregiver well-being: generational differences and coping strategies

PDF

Translating research into practice: a community-based medication management intervention

PDF

Social determinants of physiological health and mortality in China

PDF

The measurement, life course patterns, and outcomes of intergenerational ambivalence among parent-adult child dyads

PDF

Is stress exposure enough? Race/ethnic differences in the exposure and appraisal of chronic stressors among older adults

PDF

The role of the modern aging network: measuring innovations of Area Agencies on Aging

PDF

Air pollution neurotoxicity throughout the lifespan: studies on the mechanism of toxicity and interactions with effects of sex and genetic background

PDF

Health differences between the elderly in Japan and the United States

PDF

Statistical algorithms for examining gene and environmental influences on human aging

PDF

Aging in a high infection society

PDF

Palliative care: breaking the knowledge barrier among community-based older adults

Asset Metadata

Creator Yeom, Jihye (author)

Core Title The effect of body weight on mortality: different countries and age groups

School Leonard Davis School of Gerontology

Degree Doctor of Philosophy

Degree Program Gerontology

Publication Date 04/10/2009

Defense Date 04/23/2008

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag BMI,body weight,elderly,Japan,mortality,OAI-PMH Harvest,older adults,U.S.

Place Name Japan (countries), USA (countries)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Crimmins, Eileen M. (committee chair), Biblarz, Timothy J. (committee member), Wilber, Kathleen H. (committee member)

Creator Email yeom@usc.edu,yeomwisdom@yahoo.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m2074

Unique identifier UC1228384

Identifier etd-Yeom-2159 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-210420 (legacy record id),usctheses-m2074 (legacy record id)

Legacy Identifier etd-Yeom-2159.pdf

Dmrecord 210420

Document Type Dissertation

Rights Yeom, Jihye

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu