 

 

LIBRARY

Michigan State
Unlverslty

 

 

 

PLACE IN RETURN BOX to roman thin checkout from yout record.
TO AVOID FINES mum on or baton duo duo.

DATE DUE DATE DUE DATE DUE

       

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

m1

 

 

THE EFFECT OF WEIGHT CYCLING ON
BLOOD LIPIDS AND BLOOD PRESSURE IN THE MULTIPLE RISK FACTOR
INTERVENTION TRIAL SPECIAL INTERVENTION POPULATION
By

Karen A. Petersmarck

A DISSERTATION
Submitted to
Michigan State University

in partial fulﬁllment Of the requirements
for the degree Of

DOCTOR OF PHILOSOPHY

Department Of FOOd Science and Human Nutrition

1996

ABSTRACT
THE EFFECT OF WEIGHT CYCLING ON BLOOD LIPIDS AND BLOOD
PRESSURE IN THE MULTIPLE RISK FACTOR INTERVENTION TRIAL
(MRFIT) SPECIAL INTERVENTION POPULATION
By

Karen A. Petersmarck

The purpose of this study was to assess whether increases in mortality
associated with weight cycling in a number of populations might be mediated by
the traditional cardiovascular risk factors of total cholesterol, high-density
lipoprotein Cholesterol (HDL), the ratio of total cholesterol to HDL and blood
pressure. The study population consisted of 6,000 men at high risk for heart
disease in the Special Intervention group of the MRFIT, selected because the
data set includes weight, cholesterol, and blood pressure measured every 4
months over a 7-year period, as well as documentation of other cardiovascular
disease risk factors. Three measures of weight cycling were deﬁned: (1) the
number of weight cycles, deﬁned as the number of times an individual lost and
subsequently regained at least 5% of his baseline weight; (2) the standard error
of the estimate (SEE) of the regression of weight on time for each individual, (3)
a combination of number of cycles and SEE, reﬂecting the number and Size of
cycles. A number of analysis of covariance and stepwise regression models
were developed, with the three measures of weight cycling as predictor

variables, changes in the four risk factors as outcome variables and control for

all factors statistically associated with the outcomes. The hypothesis that men
who weight cycled experienced smaller improvements in their blood lipids and
blood pressure than those who did not cycle was not supported. If weight
cycling is causing increased mortality in middle aged men at high risk for heart
disease, it does not appear to be having this effect because of adverse effects

on blood lipids or blood pressure.

Copyright by
Karen A. Petersmarck
1996

ACKNOWLEDGMENTS

The People: I would like to acknowledge some individuals who made the
experience of graduate school at Michigan State University particularly
enriching. Dr. Jenny Bond, Chair of my Guidance Committee, is the most
nurturing person I have ever known. Her solid, consistent support for me in my
role as a graduate student has been no less ﬁrm than her support for me in my
roles as a mother and as a human being. She has come through when I have
needed her, exceeding any expectations for the degree and quality of help an
advisor could provide. It has been a joy and a privilege to work with her.

Dr. Howard Teitelbaum, my Research Director, has been a continual
source of energy, humor, and intelligent direction. His faith in me made possible
the American Heart Association grant which funded the research. He remains a
lifetime role model for making a constructive criticism sound (and feel) like a
commendation!

Dr. Sharon Hoerr has Shared the excitement of my research in a unique
way. Not many people would think it "fun" to Sit in a coffee shop and pore over
computer printouts of weight cycling data analyses. Her creative brainstorming
was greatly appreciated. Dr. Wanda Chenoweth has been challenging in gentle,
helpful ways. Her uncanny ability to spot inconsistencies contributed greatly to
the quality of the dissertation document.

Dr. Leonard Bianchi made statistics seem easy when he was my

V

instructor in CEP 904 and 905. As a consultant for the research, he taught me
how to make SAS obey my commands. Working with him was like taking a
break - only more productive. Dr. MaryFran Sowers gave generously of her time
and expertise. Lending her name and impressive resume to the grant
application was an act of faith in me, and unquestionably was a factor in the
funding agency's decision to approve it. Dr. Randy Fotiu would no doubt say
that he was "just doing his job" as Chief Statistical Consultant at the MSU
Computer Lab, but his quiet competence and absolute mastery of mainframe
operation smoothed out the mountain in the path of my data extraction process.
Dr. Aryeh Stein has been a ﬁrst-rate armchair quarterback - willing to venture an
opinion when one was needed. Dr. Mary Zabik put a face on "The College of
Human Ecology," and the face was smiling at me.

V Dr. Rachel Schemmel was my ﬁrst academic advisor. She got me started
on the research, and opened many doors for me. Dr. David Garner made me
understand clearly the harm that is caused by our ridiculous societal norms for
body size and Shape. His inspiration was the impetus for embarking on the
doctoral program.

Jay McClellan contributed one of the key elements of this research by
writing, just for fun one Saturday afternoon, the brilliant program to count weight
cycles. His long distance help with computer problems and proof reading helped
weave a safety net for my sanity. John Lorenz, my "adopted" brother, spent
hours listening, nodding his head, and throwing out helpful suggestions over the
course of the research. His friendship chased away many headaches over the

vi

past 5 years.

My children, John and Elizabeth Perri, have been patient, and have been
good company to balance the solitude and social isolation that comes with
writing a dissertation. The ﬁnancial sacriﬁce of their father, Giovannino Perri,
offered spontaneously with no strings attached, truly made it possible for me to
work toward this degree.

Financial Support: The major source of funding for this research was
Grant #9468945 from the American Heart Association, Michigan Afﬁliate,
awarded to Dr. Howard Teitelbaum as Principle Investigator and Dr. MaryFran
Sowers as Co-investigator. Additional funding for research expenses came from
the Michigan State University Agricultural Experiment Station. Scholarship
support has come from the American Dietetic Association, the College of Human
Ecology, Drs. Beth and Holly Fryer, the Department of Food Science and Human

Nutrition, and Mid-Michigan Mensa.

vii

TABLE OF CONTENTS

LIST OF TABLES ................................................ ix
LIST OF FIGURES .............................................. xiii
KEY TO ABBREVIATIONS ........................................ xiv
INTRODUCTION ................................................ 1
LITERATURE REVIEW ........................................... 4
Weight and Health ............................................ 4
Weight LOSS and Health ........................................ 6
Weight Gain I Regain and Health ................................ 16
Factors Other than Weight and Weight Change Known to Affect Blood
Lipid Concentrations and Blood Pressure ...................... 23
Weight Cycling and Health ..................................... 31
Limitations of Published Studies of Weight Cycling .................. 33
Blood Lipids and Blood Pressure as Risk Factors for CVD ............ 46
DESCRIPTION OF THE MRFIT DATA BASE ......................... 50
Purpose of the MRF IT ......................................... 50
Subjects .................................................... 50
Data Collection .............................................. 51
Intervention ................................................. 56
Quality Control on Data Collection ............................... 62
METHODS .................................................... 63
Human Subjects Approval ...................................... 63
Obtaining the Data ........................................... 63
Extraction of Data Needed for Analysis ........................... 64
"Cleaning" the Data ........................................... 64
Dealing with Missing Weight Data ............................... 68
Creating Summary Variables Suitable for Statistical Analysis .......... 70
Adequacy of 12-Month Interval Data for Describing Weight Cycling ..... 73
Other Independent Variables Created for Possible Inclusion in Analyses . 74
Statistical Approaches ......................................... 82
Exclusions .................................................. 83
Sample Size I Power .......................................... 84
Deciding Which Possible Variables to Include in Statistical Analysis ..... 85

viii

Preliminary Analysis: ANOVA for Homogeneous Weight Groups ........ 87

Analysis of Covariance Models .................................. 89
Additional Controls for ANCOVA Models .......................... 92
Regression Analysis .......................................... 95
RESULTS ..................................................... 98
Characteristics of the Study Population ........................... 98
ANCOVA Analyses .......................................... 104
Regression Analyses ........................................ 121
Preliminary ANOVA on Homogeneous Groups ..................... 125
Comparisons of Subjects Dropped from Analysis from Those Retained . . 126
DISCUSSION ................................................. 131
Issue: Are the Weight Cyclers in the Present Study the Same Men Who
Were Found to be at Greater Risk of Mortality? .................... 132
Does Weight Cycling Have Different Effects on Heavy Men than on Leaner
Men? ..................................................... 136
Could Weight Cycling be Interpreted as Being Beneﬁcial in Any Cases? 139
Strengths of the Study ........................................ 144
Limitations of the Study ....................................... 147
Low Correlations of Outcomes with Lifestyle Factors ................ 152
Importance of Net Weight Change in Predicting Outcomes ........... 155
What Accounts for the Increased Mortality Associated with Weight
Cycling? ................................................ 1 56
Does Weight Cycling Really Cause Increased Mortality? ............. 158
Future Research Directions ................................... 159
SUMMARY AND CONCLUSION .................................. 162
APPENDIX A : UCRIHS Approval for Research ....................... 163
Appendix B: Response to Freedom of Information Request for MRFIT
Data ...................................................... 164
APPENDIX C: Sample, SAS Program to Extract MRFIT Data from Magnetic
Tapes .................................................... 166
APPENDIX D: SAS Program for Counting Weight Cycles ............... 167
APPENDIX E: Tables Showing Correlation Coefﬁcients for Selected Variables
with Each Outcome .......................................... 169
LIST OF REFERENCES ......................................... 174

LIST OF TABLES

Table 1 - Association of Mild to Moderate Weight Loss with All-Cause Mortality

Reported by Andres et al. ............................................................................. 11
Table 2 - Pattern of Weight Change Associated With Lowest Mortality

Reported Separately for Males and Females by Andres et al. ..................... 17
Table 3 - Selected Characteristics of Published Weight Cycling Studies .......... 41
Table 4 - Summary of Studies of Weight Change in Males Under Extreme

Conditions of Overfeeding and Underfeeding .............................................. 66
Table 5 - Characteristics of Homogeneous Weight Groups .............................. 88

Table 6 - Baseline Characteristics, All Non-excluded Subjects and by Net
Weight Change Groups, MRFIT SI Group ................................................... 99

Table 7 - Weight Cycling Measures in All Non-excluded Subjects and by BMI
Tertile, MRFIT Sl Group ............................................................................. 100

Table 8 - Weight Cycling Measures in All Non-excluded Subjects and by Net
Weight Change Group, MRFIT SI Group ................................................... 100

Table 9 - Outcomes in All Non-excluded Subjects and by Net Weight Change
Groups ....................................................................................................... 101

Table 10 - Mean Values for Selected Characteristics in All Non-excluded
Subjects and by Net Weight Change Groups, MRFIT SI Group ................. 103

Table 11 - Mean Changes (and 95% Conﬁdence Intervals) in Total Cholesterol
by Cycling Status, MRFIT SI Group ........................................................... 105

Table 12 - Mean Changes (and 95% Conﬁdence Intervals) in Total Cholesterol
by Number of Cycles, MRFIT SI Group ..................................................... 106

Table 13 - Mean Changes (and 95% Conﬁdence Intervals) in Total Cholesterol
by Tertile of SEE, MRFIT SI Group ............................................................ 107

Table 14 - Mean Changes (and 95% Conﬁdence Intervals) in Total Cholesterol
by Number and Size of Cycles, MRFIT SI .................................................. 108

Table 15 - Mean Changes (and 95% Conﬁdence Intervals) in HDL by Cycling
Status, MRFIT SI Group ............................................................................. 109

Table 16 - Mean Changes (and 95% Conﬁdence Intervals) in HDL by Number of
Cycles, MRFIT SI Group ............................................................................ 1 10

Table 17 - Mean Changes (and 95% Conﬁdence Intervals) in HDL by Tertile of
SEE, MRFIT SI Group ................................................................................ 111

Table 18 - Mean Changes (and 95% Conﬁdence Intervals) in HDL by Number
and Size of Cycles, MRFIT SI Group .......................................................... 112

Table 19 - Mean Changes (and 95% Conﬁdence Intervals) in the Ratio of Total
Cholesterol to HDL by Cycling Status, MRFIT SI Group ............................ 113

Table 20 - Mean Changes (and 95% Conﬁdence Intervals) in the Ratio of Total
Cholesterol to HDL by Number of Cycles, MRFIT SI Group ....................... 114

Table 21 - Mean Changes (and 95% Conﬁdence Intervals) in the Ratio of Total
Cholesterol to HDL by Tertile of SEE, MRFIT SI Group ............................. 115

Table 22 - Mean Changes (and 95% Conﬁdence Intervals) in the Ratio of Total
Cholesterol to HDL by Number and Size of Cycles, MRFIT Sl Group ........ 116

Table 23 - Mean Changes (and 95% Conﬁdence Intervals) in Diastolic Blood
Pressure by Cycling Status, MRFIT SI Group ............................................ 117

Table 24 - Mean Changes (and 95% Conﬁdence Intervals) in Diastolic Blood
Pressure by Number of Cycles, MRF IT SI Group ....................................... 118

Table 25 - Mean Changes (and 95% Conﬁdence Intervals) in Diastolic Blood
Pressure by Tertile of SEE, MRFIT SI Group ............................................ 119

Table 26 - Mean Changes (and 95% Conﬁdence Intervals) in Diastolic Blood
Pressure by Number and Size of Cycles, MRFIT SI Group ....................... 120

Table 27 - Partial Regression Coefﬁcients (and 95% Conﬁdence Intervals) for
Measures of Weight Cycling from Regression Models for Total Cholesterol,
MRFIT SI Group ......................................................................................... 12

xi

Table 28 - Partial Regression Coefﬁcients (and 95% Conﬁdence Intervals) for
Measures of Weight Cycling from Regression Models for HDL, MRFIT SI
Group ......................................................................................................... 123

Table 29 - Partial Regression Coefﬁcients (and 95% Conﬁdence Intervals) for
Measures of Weight Cycling from Regression Models for the Ratio of Total
Cholesterol to HDL, MRF IT SI Group ........................................................ 124

Table 30 - Partial Regression Coefﬁcients (and 95% Conﬁdence Intervals) for
Measures of Weight Cycling from Regression Models for Blood Pressure,
MRFIT SI Group ......................................................................................... 125

Table 31 - Adjusted Mean Changes (and 95% Conﬁdence Intervals) in Total
Cholesterol and Blood Pressure, Homogeneous Weight Groups, MRFIT SI
Group ......................................................................................................... 126

Table 32 - Comparison of Subjects Present at Year 7 Annual Exam with Those
Absent, Selected Variables, MRFIT SI Group ............................................ 128

Table 33 - Comparison of Subjects Dropped from Analysis Because of Missing
Data with Those Who Had "Enough" Data, MRF IT Population .................. 129

Table 34 - Comparison of Mean Values (with Standard Deviations) for Selected
Baseline Characteristics, MRFIT Usual Care and SI Groups ..................... 130

Table 35 - Correlation Coefﬁcients for Selected Variables with Net Change in
Total Serum Cholesterol for All Non-excluded Subjects and by Net Weight
Change Group, MRFIT SI Group ............................................................... 169

Table 36 - Correlation Coefﬁcients for Selected Variables with Net Change in
HDL for All Non-excluded Subjects and by Net Weight Change Group,
MRFIT SI Group .......................................................................................... 171

Table 37 - Correlation Coefﬁcients for Selected Variables with Net Change in
Ratio of Total Plasma Cholesterol to HDL for All Non-excluded Subjects and
by Net Weight Change Group, MRF IT SI Group ........................................ 172

Table 38 - Correlation Coefﬁcients for Selected Variables with Net Change in

Blood Pressure for All Non-excluded Subjects and by Net Weight Change
Group, MRFIT SI Group ............................................................................. 173

xii

LIST OF FIGURES

Figure 1 - Summary of the Sixteen Primary ANCOVA Models Submitted for Each
of the Four Outcomes of the Research ............................................................ 92

Figure 2 - Summary of the Sixteen Different Regression Models Submitted for
Each of the Four Outcomes of the Research ............................................... 97

Figure 3 - Adjusted Mean Per Cent Change (and 95% Confidence Interval) in
Total Cholesterol from Baseline to Year 6 , by Tertile of Baseline BMI and
Number of Weight Cycles, MRFIT SI Group .............................................. 139

Figure 4 - Adjusted Mean Changes in Total Cholesterol (and 95% Conﬁdence
Interval) by Net Weight Change Group and Number of Weight Cycles, MRFIT
SI Group ..................................................................................................... 142

Figure 5 - Adjusted Mean Changes in HDL (and 95% Conﬁdence Intervals) by
Net Weight Change Group and Number of Weight Cycles,
MRFIT SI Group ......................................................................................... 142

Figure 6 - Adjusted Mean Changes in the Ratio of Total Cholesterol to HDL (and
95% Confidence Interval) by Net Weight Change and Number of Weight
Cycles, MRFIT SI Group ............................................................................ 143

Figure 7 - Adjusted Mean Changes in Diastolic Blood Pressure (and 95%

Conﬁdence Intervals) by Net Weight Change Groups and Number of Weight
Cycles, MRFIT SI Group ............................................................................ 143

xiii

KEY TO ABBREVIATIONS

ANOVA - analysis of variance

ANCOVA - analysis of covariance

ACTH - adrenocorticotropic hormone

BMI - body mass index = kglm’

BT - behavioral therapy

CV - coefﬁcient of variability = s.d./mean

CVD - cardiovascular disease

dI - deciliter

g - grams

HDL - high density lipoproteins

kg - kilograms

lb- pounds

LDL - low density lipoproteins

LTPA - leisure time physical activity

In - meters

mg - milligrams

min - minutes

mm Hg - millimeters of mercury

mgldl - milligrams per deciliter

MRF IT - Multiple Risk Factor Intervention Trial

n - number

NHLBI - National Heart, Lung and Blood Institute

n.s. - not signiﬁcant

P:S Ratio - ratio of polyunsaturated to saturated fats

SEE - standard error of the estimate for the regression of each subject's weight
on time

SI - Special Intervention

s.d. - standard deviation

UC - Usual Care

VLCD - very low calorie diet

VLDL - very low density lipoproteins

wt - weight

xiv

INTRODUCTION

Aims of Research: The aims of the work described here were to
determine the effect of weight cycling on the cardiovascular disease (CVD) risk
factors of blood lipids and blood pressure, using the rich data set of the Multiple
Risk Factor Intervention Trial (MRFIT). It has already been demonstrated in the
MRFIT population‘ and in several other populationsz' 3- ‘ that weight cycling is
associated with Increased risk of mortality. This association does not establish
that weight cycling causes Increased mortality. No credible explanation has
been advanced which could link weight cycling causally to increased
cardiovascular mortality. This study takes a step toward clarifying whether
weight changes are causally related by looking at intermediate outcomes
associated with heart disease - blood lipids and blood pressure.

The speciﬁc hypothesis explored in this research is that men who
experienced weight cycling over the 6-year course of the MRFIT realized smaller
improvements in the CVD risk factors of total serum cholesterol concentration,
high density lipoprotein (HDL) concentrations, the ratio of total plasma
cholesterol to HDL, and diastolic blood pressure, compared with men who did
not weight cycle.

Relevance of the Research to National Health Goals: An important

2

national health goal is to reduce cardiovascular disease, the leading cause of
death in the United States.5 Inasmuch as obesity is a major risk factor for heart
disease and affects 34 million American adults (including 12.5 million severely
overweight individuals),6 long-term health implications for obesity intervention
must be clearly understood.

Until very recently, conventional wisdom and scientiﬁc recommendations’
were to strongly promote weight loss for essentially everyone whose body weight
was above average. The general public seems to be taking this conventional
wisdom to heart: 40% of women and 20% of men are dieting at any one time,8
including many who are not overweight by any objective standard. Recent
research has led many thoughtful health professionals to question whether
weight loss, per se, is an appropriate recommendation for most overweight
individuals. Although weight loss improves cholesterol and blood pressure in
the short run,°"° the vast majority of people who lose weight regain it."'12
Weight gain is associated with worsening of blood lipid proﬁles and blood
pressure,10 and, as cited above, the increased weight variability caused by
weight loss and regain has been Shown in several studies to increase the risk of
mortality. Not only has weight cycling been shown to be associated with
increased CVD mortality, weight loss has also been associated with increased
mortality"2""‘3'“""""""""‘3 in almost every published analysis of weight change
and mortality.

If weight loss (and its almost inevitable weight regain) are, by some

mechanism, increasing health risk rather than decreasing it, national health

3

goals related to decreasing the incidence of obesity may have to be revised.
The National Task Force on the Prevention and Treatment of Obesity recently
published a review acknowledging potential health risks from weight cycling.‘9
Although the Task Force did not recommend against weight loss per se, Its
recommendations were very conservative compared to earlier consensus
documents, calling for "moderate weight loss” for “signiﬁcantly obese patients."
If further research veriﬁes increased health risk from weight loss and regain.
future recommendations for obesity treatment are likely to be even more

conservative.

LITERATURE REVIEW

Weight and Health

Weight and Health Risk Factors: There is universal recognition that
extreme levels of adiposity are associated with increased risk of early mortality
and of various medical condititons. Pi-Sunyer summarized current
understanding of these risks in an extensively documented 1993 review.20
According to Dr. Pi-Sunyer, obesity is associated with insulin resistance,
diabetes mellltus, hypertension, hypertrlglyceridemia, decreased levels of high—
density lipoprotein cholesterol and increased levels of IOw-density lipoprotein
cholesterol. Obesity is also associated with gallbladder disease and some forms
of cancer as well as sleep apnea, chronic hypoxia and hypercapnia, and
degenerative joint disease.

Weight and Mortality: Obesity is an independent risk factor for death
from coronary heart disease, however, there is controversy about the precise
relationship between weight and mortality. Large prospective population studies
in which the relationship between weight at some time and subsequent mortality
have been quantiﬁed produce conﬂicting ﬁndings.

Manson et al."’1 reviewed 25 major prospective studies on weight and

longevity, most of which reported J-shaped or U—shaped weight— mortality curves.

5

They concluded that each of the studies had at least one of three major biases:
(1) failure to control for the effects of smoking, which would Show excess
mortality at lower weights that should be explained by smoking's impact on
health, erroneously yielding a J-shaped mortality curve; (2) inappropriate control
for biologic effects of obesity. This means that researchers eliminated from the
analysis individuals with diabetes or hypertension, conditions which may be
caused by weight. In eliminating such individuals from analysis, negative health
consequences of obesity are not reﬂected in the plot, yielding a ﬂattened curve;
(3) failure to control for weight loss due to subclinical disease. This would also
Show excess mortality at lower weights, actually attributable to undiagnosed
diseases causing weight loss. This bias would also erroneously produce a J-
shaped mortality curve.

The authors believed that the presence of these biases leads to
systematic underestimation of the detrimental impact of obesity on premature
mortality. They suggested the true relationship between weight and mortality is
linear. In an analysis of mortality among 15,195 of the women in the Nurses
Health Study,22 this same group demonstrated that when pertinent confounding
variables were controlled for, a linear relationship was found between BMI at
age 30—55 and subsequent mortality. A Similar linear curve was observed in the
26-year mortality study of Seventh Day Adventists.” The Adventists could be
considered an ideal population for epidemiological study because they eat a
relatively low-fat vegetarian diet, and do not use tobacco, drink alcohol, or take

caffeine—containing beverages.

6

In a more recent review which took into account the concerns expressed
by Manson et al.,"’1 Kushnerz4 reported that, in some studies, weight is found to
have no association with mortality. In a few studies, a linear, dose-response
relationship is reported, such that as weight increases, risk of mortality
increases. In most studies, the weight-mortality curves are found to be J—Shaped
or U-shaped, with increased mortality present at both extremes of weight.

There is still lively debate in the health community about how to deﬁne an
"ideal body weight.“ The issue is made more complex by recent evidence,
reviewed below, that weight loss and weight cycling may increase risk of
mortality.

Weight Loss and Health

Until very recently, it has been believed without'question that, since being
overweight is associated with increased mortality, losing weight would improve
an individual's prospects for a long. healthy life. There is now evidence,
described below, that, despite the fact that weight loss improves known risk
factors for mortality in the short run, weight loss may Ultimately increase the risk
of early mortality.

Weight Loss and CVD Risk Factors: There is no question that weight
loss, in the Short run, is associated with improvements in CVD risk factors. This
was demonstrated vividly in Ashley and Kannel's landmark analysis of
Framingham data.10 Using weight and health data on 5,209 adults, it was
clearly demonstrated that, as weight increased, atherogenic traits (blood

pressure, cholesterol, uric acid, and triglycerides) all worsened. When weight

7

decreased, the same atherogenic traits improved. The rate of improvements in
atherogenic traits with weight loss was greater than the rate of worsening in
traits with weight gain.

More recently, Gloldstein9 conducted a thorough review of the medical
effects of modest weight reduction (loss of approximately 10% or less) in
patients with obesity-related complications. These clinical studies demonstrated
that for obese patients with non—insulin dependent diabetes mellltus (14 studies),
hypertension (13 studies), or hyperlipidemia (6 studies), modest weight
reduction appeared to improve glycemic control, reduce blood pressure, and
reduce cholesterol levels.

One notable exception to the general ﬁnding that weight loss reduces risk
factors was reported by Phinney et al.,”‘who found a consistent
hypercholesterolemia in patients with major weight loss. The condition was
transient, however, resolving when weight stabilized.

Whether weight loss itself causes the improvements in the risk proﬁle is
not altogether clear. While individuals are losing weight, they are also modifying
their diets and often modifying their exercise patterns. It is known that exercise
alone, and dietary change alone, without weight loss, can cause reduction in
blood pressure and cholesterol concentrations. The independent effect of
weight loss is seldom clear. Nevertheless, a large body of research consistently
demonstrates that weight loss is associated with reductions in health risk factors
in the short run.

Weight Loss and Mortality: In contrast to the effect of weight loss on risk

8
factors for mortality, there is evidence to suggest that weight loss may be
associated with increased rather than decreased mortality. This possibility has
been seriously considered only Since 1992, with publication of two critical

reappraisals of the literature available on weight change and mortality?“

plus
the publication of new data analyses.

Critical Reappraisals of Literature of Weight Change and Mortality: In
preparation for the 1992 Technology Assessment Conference on Methods of
Voluntary Weight Loss and Control, sponsored by the National Institutes of
Health, the existing body of research on weight change and mortality was
systematically re-examined and summarized in two papers. The ﬁrst critical
reappraisal was presented by Williamson and Pamuk,26 who reviewed six
observational epidemiologic studies published between 1951 and 1990 in which
weight loss had been found to be predictive of greater longevity.

Each of the studies was found to have serious ﬂaws which weakened or
negated the conclusion that weight loss enhanced longevity. The most credible
studies were those based on actuarial data.“ 23' 2° In these studies, life
insurance companies studied policy holders who were required to pay higher
premiums because they were overweight. Some of these individuals later lost
weight, and were granted lower premiums. The. slenderized policy holders were
compared with those who were never granted premium reductions, and were
found to have lower mortality rates. These studies did not provide information

about the amount and duration of the weight loss or levels of obesity in the

weight loss and comparison groups.

9

Williamson and Pamuk 2‘ considered each of the other four studies to be
seriously ﬂawed. Although each of the studies reported that weight loss was
associated with increased survival, two of the studies found that mortality
actually increased among subgroups of persons who lost weight (the Cancer
Prevention Study I 3° and the British Regional Heart Study 3‘), and the other two
did not provide data to support the ﬁnding (the 1979 Build Study32 and the
Aberdeen Diabetic Study 33). The Aberdeen Diabetic Study also contained a
major mathematical error which weakened its author's conclusions that weight
loss predicted lower mortality. (Williamson and Pamuk pointed out that the
authors had failed to square the slope coefﬁcient in their interpretation of their
regression ﬁndings, so that the effect of weight loss was overestimated. The
.036 years of increased survival that the authors reported was actually only
0.0013 years.)

Taken as a whole, Williamson and Pamuk 2“ concluded that the evidence
from these six studies that weight loss in obese persons increases longevity
was equivocal. None of the six studies provides any information on the type of
weight loss methods used by the subjects. Although cycles of weight loss and
regain could be expected to be common among overweight individuals, none of
the studies included information about history of weight cycling or average
lifetime weight compared to weight at the two points in time measurements were
taken.

The remaining published research on weight change and mortality not

addressed by Williamson and Pamuk was critically reviewed in a second paper

10

at the Technology Assessment Conference by Andres et al.“ who re-examined
published studies to determine which weight changes were associated with the
lowest mortality rate. In each of the twelve studies reviewed, weight change was
computed using weight at two time points during adult life. Participants were
subsequently followed for a period of years to quantify the mortality rate. The
twelve studies included very diverse populations (seven U.S. populations and
four European groups). The segments of adult life during which weight change
was documented were variable, ranging from two points in early adulthood to
two points in later life. The duration of the weight change period ranged from 3.7
years to several decades. The period of follow-up varied from 8-22 years. In
every study the effect of pre—existing illness on weight was addressed in some
way. A variety of techniques was used 'to assess weight change. and methods
of data analysis differed.

Despite the variety of approaches used, the results were largely similar
with respect to the effect of weight loss on mortality. "Mild to moderate weight
loss“ was associated with increased mortality in ten of the studies. Mortality risk
was decreased by weight loss in only one study. “Mild to moderate weight loss”
was deﬁned differently in each study, but typically involved loss of 10% or more,
loss of >1.13 BMI units, or loss of 55-136 kg. This analysis is summarized in

Table 1.

11

Table 1 - Association of Mild to Moderate Weight Loss with All-Cause
Mortality Reported by Andres et al.“

 

 

 

 

 

 

 

 

 

 

 

 

 

High No Effect on Low
Population] Study Mortality Mortality Mortality
Paris Civil Servants 3‘ Males
Dutch Longitudinal Study of Elderly 3“ Females Males
Framingham Heart Study(1988) 36 Males
Females
, Baltimore Longitudinal Study of Ag_ir_1g ‘3 Males
Gothenburg Prospective Study ‘ Males
Females
Framingham Heart Study (1991) 2 Males
Females
Harvard Alumni (1986) 37 Males
Honolulu Heart Pro[am 3" Males
Glostrup (Denmark) Longitudinal Study 39 Males
Females
Kaiser Permanente Medical Care Program ‘° Males Females
Lipid Research Clinics Program (in some Males Females
cases)"1

 

 

 

 

 

 

Five Newer Studies of Weight Loss and Mortality: Additional
information about the effects of weight loss on mortality is found in ﬁve recent
analyses of weight loss and subsequent mortality, four of which report increased
mortality associated with weight loss. Each of the studies focuses on a different
population group. Lee and Paffenbarger‘7 found increased mortality with weight
loss among 11,703 male Harvard alumni. Pamuk et al.15 found increased
mortality with weight loss in a cohort of 2,453 men and 2,739 women, age 45 to
74 years old at the time of their ﬁrst examination in the National Health and

Nutrition Examination Survey. Blair et al.1 found increased mortality with weight

12

loss in 10,529 middle aged males at risk of heart disease in the Multiple Risk
Factor Intervention Trial. Higgins et al."" found increased mortality with weight
loss among 2,500 middle-aged participants in the Framingham study. In
contrast, Williamson et al."3 found decreased mortality with voluntary weight loss
among 43,457 overweight, never-smoking US white women in the Cancer
Prevention Study.

Each of the studies controls for important factors which could cloud
interpretation of the results of the study, including pre-existing illness, smoking,
age and baseline weight. Three of the studies control for physical activity.

It is generally assumed that weight loss is essential for maximizing life
expectancy for overweight individuals, but this assumption was not borne out in
the recent studies, where, in only one instance was weight loss found to improve
risk of mortality for heavier subjects. Speciﬁcally, only overweight women with
obesity—related conditions who voluntarily lost weight saw a decreased risk of
mortality in the Cancer Prevention Study;“3 women who intentionally lost weight
who were free of pre-existing illness did not see a clear improvement in life
expectancy. In none of the other four studies was weight loss associated with
improvements in life expectancy, regardless of initial weight. On the other
hand, the most overweight individuals did not experience the negative impact of
weight loss to the same degree as those who were at lower BMI initially in the
NHANES‘5 and the MRFIT1 populations, where excess mortality was signiﬁcantly
associated with weight loss only in the moderately overweight and lowest-weight

individuals (BMI below 26 in NHANES and below 28.82 in the MRFIT data).

13

A very important limitation to all of the studies considered in the two
critical reviews summarized above, as well as in three of the newer studies, is
lack of control for an absolutely critical variable - lifetime weight pattern. These
analyses are based on weights measured at only two points in time. There is
no reason to assume that a weight loss or gain between two time points is
representative of a lifetime trend in weight. This is well-demonstrated in the
Harvard alumni analysis, where Lee and Paffenbarger had access to
information about lifetime weight changes for a subset of 7,818 of the 11,708
subjects, and were able to calculate a measure of weight variability - mean
lifetime weight loss. They found that the subjects who either gained or lost the
most weight during the study period had also experienced the most weight
ﬂuctuation over their lifetimes. They suggested that total weight loss and weight
gain could be markers for weight cycling, and that weight cycling, rather than
weight loss or gain, may have been the factor actually associated with increased
mortality.

Only two of the ﬁve newer studies discussed here (M RFIT1 and
Framingham“) made use of several weights over a number of years so that it is
certain that the weight loss documented during the observation period was not.
in actuality, a reflection of weight cycling. In both of these studies, weight loss
was associated with increased mortality.

The other glaring limitation of all but one of the above—discussed studies
of weight loss and mortality is lack of information about whether the weight

losses reported by subjects were due to voluntary weight loss efforts or due to

14

illnesses causing weight loss. Most of the researchers who looked at mortality
and weight change made heroic efforts to control for the presence of disease
conditions which could cause weight loss, but in only one study43 were subjects
actually asked whether their weight changes had been intentional. Firm
conclusions about weight loss and mortality cannot be drawn when information
about volition of weight loss has been omitted from so many studies.

AS serious a ﬂaw as this omission is, it may not be a fatal ﬂaw. Even if
the volition of the weight change were known, the picture could still be cloudy.
Whereas it is assumed that involuntary loss would be more closely associated
with mortality, voluntary weight loss does not necessarily imply that healthy
lifestyle was enhanced. For example, voluntary weight loss could be achieved
by health-threatening unsafe diet practices,“4 initiation of smoking for the
purpose of reducing weight, use of recreational drugs or in response to learning
of a diagnosis of a weight-related illness.

In support of the idea that volition of weight loss does not invalidate these
studies, it Should be noted that volition of weight loss has been assessed in four
major weight change studies. In a population of older Iowa women, French at
al.45 found that, whether weight loss in early life had been intentional,
unintentional, or both, women with more weight variability had higher disease
prevalence than stable-weight women. In the Cancer Prevention Study cohort,
Hammond and Garﬁnkle3° found that weight loss had a similar association with
mortality, regardless of intention. In a later analysis of weight and mortality data

in the same Cancer Prevention Study cohort, Williamson et al."3 found that the

15

association between intentional weight loss and longevity in middle-aged
overweight women depended on health status. Intentional weight loss among
women with obesity-related conditions was generally associated with decreased
premature mortality, whereas among women with no Dre—existing illness. the
association was equivocal.

Based on this ﬁnding, Williamson et al. ‘3 suggested that other studies
which eliminated subjects with such illnesses from analysis may have
inadvertently removed those persons for whom weight loss may been most
beneﬁcial. This places researchers who do not know whether weight loss was
intentional in a double bind , because the principle means available for
eliminating the effects of illness on weight is to identify individuals with health
conditions and either eliminate them from analysis or control for the conditions in
data analysis.

Summary - Studies of Weight, Weight Loss and Mortality: No matter
how carefully one scrutinizes the available body of published research on
weight, weight loss and mortality, deﬁnitive conclusions about the effectiveness
of weight loss in increasing longevity cannot be drawn. There is evidence that
being Slim is associated with better life expectancy, but there is deﬁnitely not
compelling evidence that weight loss will produce a comparable life expectancy
as a lifetime of slendemess. The fact that so many of the 23 studies of weight
loss and mortality considered here suggest negative consequences of weight
loss justiﬁes questioning the conventional wisdom that calls for weight loss for

the 50% of the population who are “above average" weight.

16

Weight Gain / Regain and Health

The Effect of Weight Gain on Mortality: Weight gain seems to affect
risk of mortality differently depending on the amount gained. In several
population-based prospective studies, larger amounts of weight gain were
associated with increases in mortality. Speciﬁcally, Lee and Paffenbarger found
that Harvard alumni experienced increased mortality when BMI increased to
greater than 26 in a 1993 study,“3 and also when weight gain over an 11—16 year
period exceeded 5 kg (a 1992 study"). For Paris Civil Servants?4 a gain of
more than 6.5 BMI units was associated with increased mortality. Among
employees of the Western Electric Company,3 the group which gained an
average of 37% had increased mortality compared to employees with stable
weight.

Studies of modest weight gain have yielded conﬂicting results. Willett et
al.47 demonstrated a linear increase in fatal and nonfatal coronary heart disease
with net weight gain between ages 18 and ages 33-55 years among women in
the Nurses Health Study. Even a weight gain as low as 5 to 11 kg. was
associated with a statistically signiﬁcant increase in heart disease deaths
compared to stable weight. Lack of control for physical activity and for weight
cycling are limitations of this study. In contradiction to the ﬁndings of Willett et
al.,‘7 when Andres et al.“ looked closely at 13 major epidemiological studies,
they found that the lowest mortality rates were generally associated with modest
weight gains. The highest mortality rates occurred in adults who either had lost

weight or had gained excessive weight. See Table 2.

17

Table 2 — Pattern of Weight Change Associated With Lowest Mortality
Reported Separately for Males and Females by Andres et al.“

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Moderate No Moderat No
Population Studied Gain Chang a Loss Associatio
e n
Paris Civil Servants“ Males
Dutch Longitudinal Study of Females Males
Elderly35
Western Electric Company Males
Employees3
Framingham Heart Study(1988)"s Males
Females
Harvard Alumni Heart Study Males
D 992) 17
Baltimsre Longitudinal Study of Males
.éSIL'lL
Gothenburg Prospective Study of Males
Men and Women‘ Females
Framingham Heart Study (1991)2 Males
Females
Harvard Alumni (1986) 37 Males
Honolulu Heart Prgram‘i‘3 Males Males
Glostrglgp (Denmark) Longitudinal Males
Study Females
Kaiser Permanente Medical Care Males
Math”
Lipid Research Clinics Program (in Males
some samples)‘1

 

 

 

 

 

 

 

The Effect of Weight Regain on CVO Risk Factors: It is universally
accepted that weight gain is associated with worsening of health risk measures,
although few studies speciﬁcally quantify this phenomenon.” ‘3 A clinical

question of utmost importance is whether it is better to have lost and regained

18

than never to have lost at all.

Dearth of Research: Given the facts that the vast majority of Individuals
who lose weight regain it, and that the rate of weight regain is proportional to the
rate of weight loss,“9 there is an incredible dearth Of published research on the
effects of weight regain on health risk factors.

The surest way to induce weight gain is to induce weight loss. This can
be seen by looking closely at data tables from large, randomized controlled
clinical trials which have demonstrated that weight loss is associated with
reductions in blood pressure.“ 5" 52' ‘3' 5‘ In all cases, the intervention groups,
on average, weighed less at the end of the observation period than the control
groups who had not been prescribed weight loss, and had slightly lower blood
pressure than controls. However, because the weight losers regained lost
weight, the "weight losers“ experienced more weight gain over the observation
period that did the control groups. For example, in both the Hypertension
Control Program50 and the Hypertension Prevention Trial,“ the intervention
group weighed slightly less than the control group at the end of the ﬁve-year
trial, but had gained more than twice as much weight as the control group over
the ﬁve years.

The Difficulty of Interpreting Existing Research: After an individual
regains all the weight lost, are his blood lipids, blood pressure, and blood
glucose going to be the same, better, or worse than their pre-weight-Ioss levels?
Published literature is not very helpful in answering this question. One of the

most serious barriers to answering this question is the common practice of

19

lumping weight losers with weight regainers when research ﬁndings are
summarized. Almost all published studies of weight loss and health risk factors
include some data on subjects who have regained all of their weight, but the
effect of weight regain, per se, is obscured because data are reported in terms of
average weight loss within a cohort at the end of some observation period.
Periods of weight loss and regain are not examined separately.

To give one of dozens of examples, Schotte and Stunkard55 looked at the
effects of weight Changes on blood pressure in 137 hypertensive and 164
normotensive obese adults. There was an average decrease in blood pressure
during the weight loss intervention period and an average increase in blood
pressure during the follow-Up period. However, the follow-up period was not a
time of consistent weight gain. At the time the follow-up measurements were
taken, some of the hypertensive subjects were still losing weight, some had
regained part of the weight, others had regained all, and others had gained more
than they had lost, yet blood pressures for all subjects were averaged together
at the times measurements were taken.

Schotte and Stunkard did attempt to separate out the effects of degree of
weight regain on blood pressure, by dividing subjects into quartiles according to
their "ability to maintain weight loss." One of the quartiles included the
individuals who had regained all the lost weight, but included others as well. All
quartiles showed increases in blood pressure during follow-Up. There was little
statistical relationship between the amount of regain and the amount of increase

in blood pressure.

20

A 1991 study by Wing at al.56 has been widely cited as evidence that
weight loss is beneﬁcial, even if weight is regained.57 Wing and colleagues
looked at long-term glycemic control in obese Type II diabetics who underwent
weight loss and some subsequent regain. Thirty-six patients were randomly
assigned to either a standard behavior therapy (BT) weight loss program, or to a
behavior therapy program which included 8 weeks of very low calorie diet
(VLCD). One year after the end of treatment, there was no difference between
the two groups in weight, because the VLCD group regained more weight than
did the BT group. Both groups showed improvements in glycemic control as
evidenced by reductions in medication requirements. The VLCD group also
maintained several measures of slightly improved glycemic control one year
after treatment (fasting blood glucose and glycosylated hemoglobin), despite the
fact that these individuals experienced a greater weight gain during the post-
treatrnent period than did the individuals in the BT group, and despite similar
exercise patterns, in both groups.

Although this research suggests lingering positive effects on risk factors
after some weight regain, results for subjects who regained all the lost weight
were not reported. In an earlier study of diabetics by the same research group,"’8
improvements in the same risk factors (fasting blood glucose and glycosylated
hemoglobin) were noted only in subjects who whose net weight loss at one year
follow-up was greater than 5% of body weight. In the 41% of subjects who were
not able to sustain that degree of weight loss, glycosylated hemoglobins

returned to pre-weight loss levels. In the other 18% of subjects who regained

21

more weight than they had originally lost, there was an increase in this risk factor
to higher than pre-weight loss values.

In a more recent study of 202 overweight individuals, Wing et al.59 did
differentiate individuals who regained all their weight from those who regained
only part. CVD risk factors of total cholesterol, HDL, LDL, triglycerides, blood
pressure, per cent body fat, and waist to hip ratio were documented at baseline
and 30 months after a 6-month weight intervention. Of greatest interest were
comparisons in CVD risk factors among 3 subgroups of the population - (1) 25
individuals who had remained weight stable, (2) 28 who had regained all of a
moderate weight loss of about 5 kg., and (3) 31 who had regained all of a 12-kg
weight loss. These 3 subgroups were very similar with respect to all risk factors
at 30 months. The group which had regained 12 kg. had slightly better ﬁnal
values for blood pressure and waist to hip ratio than the group regaining all of
the smaller weight loss. The fact that the average values for CVD risk factors
over the 30 months of the study were lower for weight regainers than for weight
stable individuals led the authors to conclude that weight loss followed by regain
could be more beneﬁcial than remaining at a stable high weight. It should be
noted that there were very few individuals in each of the 3 subgroups.

Weight Regain and Visceral Fat Stores: One risk factor for
cardiovascular disease is visceral fat accumulation. It has been suggested that
weight regained after weight loss may be preferentially deposited in visceral fat
stores, as opposed to subcutaneous depots, resulting in a net worsening of

atherogenic risk. There have been several studies quantifying changes in

22

visceral fat with weight loss, °°- 5" 62' “’- 6" 65 but only ﬁve have been found that
describe changes in visceral fat stores or waist-hip ratio with weight gain or
regain. Bouchard et ale“ overfed 12 pairs of identical male twins for 84 days by
1,000 kilocalories per day. Increases in body fat did not correlate with increased
visceral fat, although some sets of twins were found to be more prone to store fat
in the abdominal visceral area than were other sets. Van der Kooy et al.67
followed 32 obese subjects who lost and regained 11.9 kg. Weight regain did not
result in greater body fatness, and there was no indication of preferential
deposition of visceral fat relative to subcutaneous fat. Similarly, Hensrud et al."3
found that weight regain did not result in increased fat mass or increased waist—
hip ratio in 24 obese women who spontaneously regained from 19-216% of lost
weight four years after Signiﬁcant weight reduction. Hainer et al.‘59 found that
waist—hip ratio did not change among women who had regained substantial
amounts of weight one year after a very low calorie diet. Wing et al.59 found that
neither per cent body fat nor waist to hip ratio increased among 31 individuals
who had lost and regained 12 kg.

None of these ﬁve studies support the idea that regained weight is
preferentially deposited in visceral fat depots. The possibility cannot be ruled
out that such a preferential fat deposition does occur, although the bulk of the
animal literature does not support this hypothesis either.70

Summary of Weight Regain and Risk Factors: In summary, the
important clinical question of whether complete weight regain results in a

positive, neutral, or negative effect on cardiovascular risk factors has not been

23

deﬁnitively answered at this time.
Factors Other than Weight and Weight Change Known to Affect Blood Lipid
Concentrations and Blood Pressure

Physical Activity/ Fitness Level: The extreme importance of physical
activity for preventing chronic disease has been well understood only in recent
years. As of1992, the American Heart Association ofﬁcially recognized physical
inactivity as a risk factor for cardiovascular disease, as important as high blood
pressure, smoking, weight and high blood cholesterol.71 Physically inactive
people have almost twice the risk of cardiovascular disease as those who
engage In regular physical activity.72

The powerful protective effect of physical ﬁtness against early mortality
was demonstrated by Blair et al.73 in a population of healthy men and women.
Among individuals with high blood pressure, elevated cholesterol, elevated
blood sugar, high body mass index or who smoked cigarettes, those who were
physically ﬁt had far lower risk of mortality than those who were not ﬁt.
Individuals with a family history of death from coronary heart disease who were
physically ﬁt were 3-6 times less likely to die of coronary heart disease than
individuals with no family history but with low levels of ﬁtness.

Lower mortality rates from cancer of combined Sites have been shown for
both men and women in higher categories of physical ﬁtness." Physical activity
also appears to have other, additional health beneﬁts. Physical activity results

in caloric expenditure which in turn enhances weight loss and control, and is

24

therefore important in the management and prevention of obesity. Physical
activity has been associated with the prevention of osteoporosis in
postmenopausal women”: 7" In addition, regular physical activity can improve
glucose tolerance and insulin sensitivity.77 It has been reported that physical
activity can help prevent and treat depression and anxiety."3 Increased physical
activity, independent of dietary changes, can improve blood lipid proﬁles and
blood pressure.77

The bulk of the exercise and health research indicates that moderate
levels of physical activity exert signiﬁcantly protective effects, however two
recent reports suggest that there is a threshold effect, such that only more
vigorous exercise produces health beneﬁts. Paffenbarger et al.“3 and Marrugat
et al.79 found that vigorous physical activity, but not total activity, was protective.

Since physical activity independently affects weight, blood pressure,
HDL, and total cholesterol, without documentation of exercise practices, the
interpretation of weight and CVD risk factor data is difﬁcult.

Nutritional Factors: There is not universal agreement on the precise
relationships between speciﬁc nutrient intakes and risks of CVD. It is widely
accepted that intakes of total fat, saturated fat, polyunsaturated fat,
monounsaturated fat, dietary cholesterol. and caloric balance have effects on
blood lipids, ‘3‘ however individuals differ substantially in responses to changes
in dietary lipids. Factors that predict response to dietary lipids are poorly
deﬁned but are likely to be largely genetic.ao Other dietary factors thought to

have effects on either blood lipids or CVD risk include fatty acid structure (cis-

25

versus vans-isomers of natural fatty acids), omega—3 fatty acids, soluble ﬁber,
anti-oxidants (including vitamin E, vitamin C, and beta—carotene), magnesium,
chromium, copper, zinc, vitamin Be, nicotinic acid, alcohol, 8° iron 8‘ and folic
acid.168 Nutrition factors reported to be related to high blood pressure include
sodium, alcohol, calcium, potassium, magnesium, chloride and phosphorus. 32'
83. 84

Genetic Contributions to Abnormal Blood Llpld Patterns: Genetic
contributions to CVD can be made by mutations in genes which code for various
lipoproteins. Blood lipid concentrations will be affected by alterations in genes
which code for any of the proteins which control lipoprotein synthesis, lipoprotein
processing, and lipoprotein breakdown.35 Synthesis proteins include the
apolipoproteins (A-l, A-ll, A-IV, B, CI, Cll, Clll, D, E, and apo(a)). Processing
proteins include lipoprotein lipase, hepatic triglyceride lipase, lecithin cholesterol
acyltransferase and cholesteryl ester transfer protein. The breakdown proteins
include the LDL receptor, chylomicron remnant receptor and the scavenger
receptor.

Within-Person Fluctuations of Serum Cholesterol: Cyclical variations
are found in many biochemical measurements in both humans and animals.“ ‘7
It is not possible to quantify the extent to which these variations are responses to
seasonal changes in temperature, exercise, duration of sunlight, diet, or intrinsic
biologic rhythms. Biologic variability in total cholesterol has been recognized for
decades. Variability has been recognized from day to day, week to week, month

to month and across seasons. Hegsted et al.” reviewed data from 11

26

experiments reported in the literature in which repeated cholesterol
determinations were made on the same subject. Whether subjects were
hospitalized, free living, or in metabolic wards, intra-individual variances were
very substantial, ranging from a low of 7 mgldl in subjects consuming a formula
diet, to a high of 48 mgldl when there was very little dietary control.

Although day to day variability is a consistent ﬁnding, some individuals
have more variability than others. Mogadam et al.89 observed variations of more
than 20% in the mean serum cholesterol in 75% of 20 male and female subjects
age 22 to 63 followed for four weeks. The ﬂuctuations followed no predictable
pattern, i.e., they were up and down from week to week at random, and were
unrelated to age, gender or the serum concentrations of lipoproteins. Groover et
al."0 studied 177 ofﬁcers in the Pentagon over 5 years, obtaining at least 6
cholesterol determinations on each individual each year. A variation of less than
20% was observed in 20% of the cohort, while a variation over 50% was
observed in another 21%. During the study, all 15 of the myocardial infarctions
occurred in the men whose cholesterol concentrations had varied by 55% or
more. Bookstein et al.,91 in a study of 51 male and female volunteers age 19-59,
found that the considerable day to day variability in cholesterol concentration
was not dependent on the concentration of cholesterol. In other words,
individuals with higher mean cholesterol concentrations did not experience more
variability than individuals with lower mean cholesterol concentrations.

Seasonal Variations in Blood Cholesterol Levels: Harlap et al.87

reported that nine studies have shown seasonal variations in blood cholesterol

27

levels. Six of the studies were based on populations living in relatively cold
Scandinavian climates, two on North American populations, and one on
residents of Jerusalem, a hot climate. Harlap also notes that three other studies
(in Scandinavia, North America and Australia) found no evidence of seasonal
variations, but these three studies were based on very small samples. In
general, higher levels occur in autumn or winter, with a nadir in late spring and
early summer.” 9"" 93' 9‘ Gordon et al.95 carried out a thorough analysis of
seasonal variations of plasma lipids in 1,446 hypercholesterolemic men who
comprised the placebo group in the Lipid Research Clinics Coronary Primary
Prevention Trial. They found highly signiﬁcant seasonal trends for total
cholesterol, LDL, and HDL, which peaked in the ﬁrst month of winter. There was
no correlation between the seasonal cycles and body mass index or seasonal
intake of total calories, fat or cholesterol. Total cholesterol and LDL variations
were inversely, but Signiﬁcantly correlated with hours of daylight and HDL-
cholesterol was positively and signiﬁcantly associated with ambient temperature.
They concluded that there was no doubt that seasonal plasma lipid and
lipoprotein cycles exist, although etiologic mechanisms remained uncertain. In a
recent review Kritchevsky96 reported that seasonal variations in plasma lipids
have been reported in many animal species. including rats, rhesus monkeys,
vervet monkeys, baboons, woodchucks, badgers and hedgehogs. He suggests
that the seasonal variability in cholesterol may be related to changes in
hormonal levels or in enzyme activity, and points out that not all subjects display

the variation, even when under the same environmental conditions.

28

Rippey” has pointed out that seasonal changes in lipids are not large in
relation to between-person variation or within—person changes from day to day.
Nevertheless, the seasonal differences are as large as those that have been
found in many metabolic experiments. Therefore, season is a possible
confounding variable when comparing lipid levels over time.

Alcohol Consumption: Many epidemiologic studies have shown that
light to moderate alcohol intake is associated with lower mortality from heart
disease, when compared to abstention from alcohol. 98' 99 This effect appears to
be mediated by an increase in plasma HDL levels associated with increased
alcohol consumption. Alcohol intake also increases blood pressure.'°° In the
MRFIT population, Suh et al.‘°1 found a dose-response relationship between the
number of drinks per day and risk of CVD.

Medical Conditions and Medications: A number of medical conditions
and medications can exert effects on blood lipids, blood pressure or weight. For
example, neoplastic disease is well known to decrease cholesterol and in the
case of colon cancer it has been argued that it may do so years before its
clinical presentation.102 Fever, dehydration and recent myocardial infarction
affect cholesterol concentrations. Illnesses such as minor viral infections may
reduce cholesterol in some individuals.‘°" Some diseases produce secondary
hyperlipidemlas, including hypothyroidism, nephrotic syndrome, renal failure and
liver disease. Diabetes is associated with elevated blood pressure and

cholesterol levels, although it cannot be said to be causative for these

29

conditions. Any analysis of weight, cholesterol and blood pressure should take
Into account subjects' medical conditions and medications which could exert
independent effects on cholesterol, blood pressure or weight.

Tobacco Use: It is essential to control any data analyses of weight and
health risk factors for tobacco use. Cigarette smoking is associated with weight
Ioss,‘°" ‘°5 and smoking cessation has been shown to cause weight gain,106 but
stopping smoking is associated with decrease in mortality risk. Smoking has
been associated with higher concentrations of serum cholesterol and blood
pressure in some (not all) studies, and with higher waist to hip ratio.‘°7 Smokers
have increased mortality risk from heart disease and cancer.‘31

Age: Age inﬂuences serum lipids. It is frequently not appreciated that
both cholesterol and triglycerides are at their lowest in adolescence in both boys
and girls. In men it reaches its maximum by the mid 40's but in women a rapid
rise occurs at around the time of the menopause. The increases with age are
more marked in people with the higher levels.92 Blood pressure also tends to
increase with age, perhaps in response to declining renal function and increased
peripheral resistance with aging.108

Mental Health / Stress: The relatively new ﬁeld of psychoneuroim-
munology is producing evidence of what has been known on an intuitive level for
centuries - that mental health affects physical health. The brain and the immune
system are linked via the autonomic nervous system and neuroendocrine outﬂow

from the pituitary.109 Support for the concept that mental health affects heart

3O

disease risk is found in the work of Ornish et al."° in which regression of
coronary artery lesions was observed in patients whose lifestyle changes
included one hour per day of stress reduction plus group social support for one
year.

Bjomtorp has published a thorough review of visceral obesity‘" in which
he suggests that visceral fat deposition may be the result of environmental
stress interacting with neuroendocrine aberrations in genetically-predisposed
individuals. Neuroendocrine aberrations in individuals with visceral fat
deposition include excessive secretion of cortisol in response to environmental
stressors and ACTH, depressed growth hormone concentrations and abnormal
hemodynamic response to stress mimicking the “defeat reaction" observed in
animals faced with ovenrvhelming stress.

The defeat reaction, observed in female primates, has produced the
endocrine proﬁle of visceral obesity in the primates. Female cynomolgus
monkeys in a defeated, uncontrollable situation over time develop enlarged
adrenal glands, decreased sex steroid secretion, insulin resistance. decreased
glucose tolerance, hyperlipidemia, hypertension and coronary atherosclerosis,
with accumulation of triglycerides in the visceral depots.112 Male cynomolgus
monkeys also develop atherosclerotic traits in response to environmental stress,
but in contrast to females, it is the dominant male in socially unstable situations
who develops these traits.113

Bjomtorp "‘ cited epidemiological studies demonstrating that an elevated

waist to hip ratio in men is associated with a poor education, physical types of

31

work, and a low income. Both males and females with elevated waist to hip ratio
use medical facilities more frequently, with frequent incidences of ulcers, gastric
bleeding, depression and anxiety. He suggested that subjects in difﬁcult
socioeconomic situation who have inadequate coping mechanisms may develop
the metabolic characteristics associated with heart disease.

Foreyt et al.” found that weight cyclers had lower general well—being,
less eating self—efﬁcacy, and higher stress than individual of stable weight.
It has been suggested that some of the excess mortality observed in weight
cyclers could be the result of diminution of feelings of self-worth and increased
feelings of helplessness resulting from lack of success in weight loss in the
American culture which glorifies thinness.
Weight Cycling and Health

In recent years, concern has been raised about potential negative health
effects from weight cycling, generally deﬁned as loss and regain of some amount
of weight. Brownell et al.115 showed that rats who lost weight and then regained
it experienced a Slow—down in basal metabolic rate, and experienced increases
in per cent body fat compared to littermates who were not subjected to weight
loss and regain. In a recent thorough review of the animal literature on weight
cycling, Reed and Hill70 concluded that "the published animal literature does not
justify any warnings about the hazards of weight cycling."

The human literature on weight cycling, summarized in Table 3 and
reviewed below, consists of large-scale epidemiological studies of nine

population groups and clinical studies on two small populations. Taken

32

together, this literature suggests an association between weight variability and
cardiovascular disease. In eight of these groups, weight variability was found to
be positively associated with either increased CVD" 2' 4 increased CVD
mortality" 2' 3' "6 or increase in some CVD risk factors." ‘3' "7 In the other three
populations, no statistical association was found between weight variability and
CVD 59. 118. 119

Weight Cycling, Mortality and Cardiovascular Disease: The thrust of
most epidemiological studies of weight cycling has been a search for
associations of weight variability with mortality and CVD. All-cause mortality was
associated with weight variability in six populations - MRFIT,1 Framingham,2
Gothenburg males, Gothenburg females,‘ Zutphen males,119 and Japanese
American men,116 - but not Baltimore males13 or Charleston residents.118
Increased CVD or CVD mortality with weight variability was found among ﬁve
populations — MRFIT,1 Framingham? Western Electric Company employees,3
Gothenburg males,‘ and Japanese American men116 - but not in the Baltimore
men,13 Zutphen males,119 or Gothenburg females.4 Table 3 summarizes and
compares the available studies of human weight cycling.

Weight Cycling and CVD Risk Factors: Since evidence has been found
that weight cycling may be associated with increased deaths from heart disease,
several researchers have looked for associations among measures of weight
cycling and recognized risk factors for CVD. In a well-designed study, Lissner et
al.13 used data for 846 healthy males age 17 to101 participating in the Baltimore

Longitudinal Study on Aging. Three to 16 weights per subject over a 2- to 27—

33

year time span were used. To deﬁne weight variability, weight was regressed on
time for each individual, and the standard error of the estimate (SEE) was used
as the measure of weight variability. Weight variability was not associated with
changes in serum cholesterol, systolic blood pressure, triglycerides or waist to
hip ratio, but was associated with a measure of central fat distribution, the ratio
of subscapular to triceps skinfolds, and was also associated with decreased
glucose tolerance. Increased incidence of diabetes mellitus was also found in
weight cycling Gothenburg females4 but not in Gothenburg males.‘

One clinical study by Wing et al.‘59 identiﬁed 59 overweight individuals
who experienced 1 weight cycle of either 5 or 12 kg. during a 6-month weight
loss intervention and regained all the weight in 30 months. There were no
detrimental effects from weight cycling detectable for total cholesterol, HDL,
LDL, triglycerides. blood pressure, per cent body fat or waist to hip ratio. (In
contrast, a positive association was found between weight variability and waist
to hip ratio in 87 Connecticut women.117
Limitations of Published Studies of Weight Cycling

None of the large epidemiological studies supplying data used for
analyses of weight cycling and health outcomes was designed to study weight
cycling. Each has limitations which preclude establishment of a causal
relationship between weight variability and the endpoints considered. The
limitations of these studies are discussed below.

Limitation: Pauclty of Weight Data: A severe limitation of most

published studies, to date, is that there were too few weights available for each

34

subject, and the intervals between weights were too long to detect the presence
of weight cycles. The study making use of the most weight data was that based
on the Framingham data,2 with 8 measurements over 16 years. Although a great
deal of information was available, the two-year interval between weights allows
the possibility of extreme weight ﬂuctuation between measurements, which could
be completely missed in the data analysis. For the Western Electric Company
employees" ﬁve recalled weights at 5-year intervals were used, which would not
allow documentation of more than one weight cycle.

For the Connecticut women Rodin et al.117 calculated a weight cycling
score based on recalled weights over 5-25 years. This complete documentation
of life weights was a positive aspect of the study, but the absence of measured
weight in both the Connecticut women and Western Electric employees erodes
the validity of the measure of weight cycling. Estimates of the validity of recalled
weights varies from 0.80 to 0.935 38' "8° ‘2° for each measurement. The validity
of multiple recalled weights would be much lower than the validity of one
recalled weight. In the other weight cycling studies, very few measurements
were available. For the Baltimore Longitudinal Study on Aging13 population, four
weight measurements were used for the analysis of weight cycling and mortality.
Analyses of the Gothenburg male data" and the Honolulu Heart Study data "3
were based on only three measured weights. Analyses of the Charleston Heart
Study“9 data and Gothenburg female" data were each based on only two
measured weights and one recalled weight.

One analysis made use of more measurements, but did not have

35

comparable amounts of data for each subject. Analysis of the Zutphen Study119

population included an average of 10.3 measurements over 10 years, with a
minimum of four measurements.

Limitation: Inadequate Mathematical Deﬁnition of Weight Cycling:
The measure of weight cycling used most commonly is the coefﬁcient of
variability (CV), deﬁned as the standard deviation of weights divided by the
mean weight. The standard deviation and CV are inadequate for studies of
weight cycling because individuals who consistently gain or lose weight may
have a high standard deviation of weight and CV even if no weight cycling
occurred at all. ‘2‘

Only three of the published analyses used measures of weight cycling
which separate the linear trend in weight change from the variability around the
trend. Each of these based the deﬁnition of weight cycling on linear regression
of weights on time. In the Zutphen population analysis,"° weight cycling was
quantiﬁed as the residuals of the regression. In the Baltimore ‘3 and the
Honolulu Heart Study"° population analyses, weight cycling was documented as
the SEE. The SEE and residuals do separate out the variability due to
ﬂuctuations in weight from variability attributable to a consistent trend of weight
loss or gain. Neither of these two measures, however, indicate how many
episodes of weight loss and regain occurred.

Limitation: Failure to Specify the Number or Size of Weight Cycles:
To determine whether weight cycling has an effect on health outcomes, it would

be desirable to know how many times weight was lost and regained, and the

36

magnitude of the weight cycle. No published studies of weight cycling include
this information. Only two population studies (Western Electric3 and MRFIT‘)
actually documented that at least one weight cycle had occurred in the
individuals considered to be weight cyclers. Neither of these two analyses
documented whether more than one cycle had occurred.

Limitation: Inadequate lnforrnation about Potentially Confounding
Variables: Many factors affect both weight and CVD risk, including age, physical
activity, tobacco use, existence of other health conditions, intake of dietary
factors, use of drugs that can affect risk factors, alcohol use and mental health
status. None of the previous analyses included all of these factors, and when
any of the factors were included, they were often not speciﬁed completely
enough to adequately control for their potential inﬂuences on the outcome
measures.

I Physical Activity: Physical activity has seldom been documented in
studies of weight and mortality. Without documentation of exercise practices,
the interpretation of weight and mortality data is difﬁcult because physical
activity independently affects weight, blood pressure, cholesterol, insulin
resistance and cardiovascular mortality risk.71 In the weight cycling literature,
physical activity has been considered in analysis only in Model C of the
Framingham population analysis2 and in the mortality analysis of the MRFIT
data.1 For the Framingham analysis, physical activity was documented at only

one time in the trial. A person’s level of physical activity at one point in time is

37

not likely to be representative of his activity level over 16 years. In the MRF IT
analysis by Blair et al.,1 the measure of physical activity used was not
representative of usual physical activity, but rather was the subject's self-
assessment of his activity level compared to his peers at the time of entry into
the study. This was inadequate documentation to eliminate the confounding
effects of exercise on mortality risk because, in the MRFIT population, one of the
interventions in the Special Intervention (SI) group was behavior modiﬁcation to
increase exercise. It can be assumed that exercise habits changed after entry
into the study.

I Cigarette smoking: Smoking can affect weight, blood lipids, and blood
pressure. If smoking behavior is not controlled for in data analyses, the
conclusions drawn are leSS certain. In published weight cycling studies,
smoking was controlled for in Model C of the Framingham analysis,2 the
Western Electric analysis,3 the Gothenburg analyses," the Honolulu Heart Study

"5 and the MRFIT analysis.‘ However, speciﬁcation of smoking

analysis
behavior was incomplete. For example, in both the Framingham and the MRFIT
analyses, smoking status was controlled for by including in the regression model
the number of cigarettes smoked per day measured once, early in the study. If a
person subsequently quit smoking or increased the level of smoking, that was
not taken into account, and could have affected the outcome variable in those
analysis, mortality.

I Causes of Weight Change I Volition of Weight Loss: The most

frequently-mentioned criticism of weight change studies is that they do not

38

differentiate voluntary from involuntary weight loss. As discussed earlier, this
may not be a fatal ﬂaw. Only one of the population studies of weight cycling had
any indication of intentionality of weight change. Women in the Gothenburg
Study " were asked whether they had been on weight reduction diets. Being on
a weight reduction diet was not associated with increased mortality, even though
weight cycling was associated with increased mortality in this study.

In studies of weight cycling and mortality, efforts have been made to
eliminate people whose weight variability was involuntary (due to pre-existing
illness) in several ways. Most commonly, subjects who died or became ill in the
time period immediately after the ﬁnal weight measurement were eliminated from
the analysis, so that weight losses near the end of the observation period due to
illness would not be a factor in the analysis. This is a very reasonable method of
controlling for pre-existing illness. The length of the lag period between ﬁnal
measurement and the beginning of the observation period for deaths in the
studies reviewed here varied from none (Western Electric, Baltimore, and
Zutphen studies 3' 13"‘9), to shorter than one year (Gothenburg and MRF IT
studies": 1), to 5 or 6 years in the Honolulu Heart Study116 and F ramingham
study.2

The other way researchers have attempted to eliminate the effects of
illness on weight is to identify individuals with health conditions and either
eliminate them from analysis or control for the conditions in data analysis. When
Lissner et al." dropped from analysis Gothenburg women with pre—existing

illness, the apparently detrimental effects of weight cycling and weight loss were

39

no longer noted. Similarly, lribarren et al.116 found that weight loss and weight
cycling were associated with reduced life expectancy only in men with pre-
existing health conditions. Volition of weight loss was not directly evaluated by
lribarren, however all deaths occurring in the ﬁve years after the ﬁnal weight
measurement was taken were excluded from analysis, thus eliminating most of
the effects of weight loss due to illness. None of the other epidemiological
studies of weight cycling eliminated individuals with pre-existing weight-related
conditions.

I Dietary factors: Of all the published studies of weight cycling or
weight change, only the Western Electric Study analysis 3 included control for
dietary intake of any nutrient. This would not be considered a serious ﬂaw in
studies which used mortality as an endpoint. However, when blood pressure
and cholesterol are endpoints of interest, such as in the Baltimore Longitudinal
Study on Aging analysis13 or the study by Wing et al., 59 inclusion of dietary
information would be critical.

I Alcohol use: Only two studies of weight cycling have taken alcohol
consumption into account (Western Electric3 and MRFIT‘). In the MRFIT
analysis by Blair et al.,1 the number of alcoholic drinks per week included in the
analysis was based on drinking only at baseline. This is less than optimal
control for the effects of alcohol, because subsequent increases or decreases in
drinking were not taken into account.

I Mental health status: None of the studies of weight cycling include

40

any variables reflective of mental health status.

Summary, Studies of Weight Cycling: The research available suggests
that weight loss with subsequent regain may increase risk of heart disease.
However, serious methodological limitations in all published studies render the
proposed link between weight cycling and negative health outcomes somewhat

tenuous.

41

 

 

 

 

 

 

60.23 22.05396: 2::
8: ”038.0 :3... >532. .o. :o 3:90? n tone .2» Randi:
22.6565: .532. .62 22.25 .0 5.8952 n 565:: 60.9
3. 02 22.25.. Eu :5 233.: ..2m 6. oﬁEamc 2:903 one .:oE a: >035
383: @588 5.? games .2 6:205... .83... 2: .6 8:0 3532: 528:2 tee:
3.63» .oz :2: .8 mo> ”5.28.2 0.85 .8 .mo> 9588.2: 235$ 3...... 30:32. =.:.e:o:
8:222
2.60:3
.2382: coo...
._o.2mo.o:o
E23
533.. .82....
oz ”:2:o>> 53.2.0390
mo> ”:22 2:: .26
”madman. tom... .5 .0 Be: 33..
Sea» 82 22 “$2.22 8> .__2m 53E 6? oEam oEam oEum o .28:
oz 322:0“.
mo> ”no.2... 2:: .26 2925
”033.0 two... .0 ecu: .35.
52:3» .02 mo> H€365. mo> ...2m :aoE 6?. 2:3 2:3 2:5 m .26:
mm on: «NE»:
a £92., No.8
62.32 one 6.2:2
_.2m :35. + £9.25 E: 0.2:
8> + «.55 a $2.... .32 EB.
”mucosa too... .0 5.6.50 .a 2:903 BEES“. < .232
8.3% .oz mo> ”£2.22 oo> 69.. 2355 3528.2 .3533”. «Eng—EEO...
33.5
can: 3.320 too... be 8.33.30 35.2.0
282.“. 3.3.3: 3.8.3: 335.; 2225 new: 83 5.5.2.0..
.3". o>o :. . 03.0.2 :. . 3.3.3260 «c 5.3500 £203 833:» >935

 

 

 

 

 

 

 

 

863m 2:26 29°; 8:23.... .o 8.358.220 832% . n 28»

 

42

 

000..

 

 

 

 

 

 

 

 

 

 

:. 2.00
000 .022:
.0. 0.00.0 3.
.0202 0.00.0
mm 000 mm... .022:
a £0.03 .2 2...;
02.002 80 .022:
02.. 2...; an
=20 :022 08.. 0:0 5.02.2.0
00.020 + 0...).0 m 80.. :. .0 2:00.02 23030
.0: ”00000.0 000... =20 :. 00:20 .0 500.50 0050002 0.0200 :00...
00.030 .oz .8020 .2. ”£222 02 .20 ESE .000 2850 £025 £85: 5.0220
£0.02, 2
0.02 .0 ”20.02
0.50.03 0.0.. .0".
..20.00.2.0 5030 0...
.2502: .0 2:26 031.:
000.0 .02 2:0. :0 50. .0... >030 0:000
28:. .3002... £0.02. .0 a 2.0.0; 5 3..
2200000000 0:005 5.8203. 02:000.: 000 .022: .059.
00:220. .20 00:0..0 .0. 0.02000 50.. 00.00000 :w >030
00000.0 02 ”00000.0 :00... 2.0.03 .0 0.2 2.. .0 .0:m . 3.2.0:: 2.03 $503.30..
”00> 00.020 .0: “£0005. 0z ..s.m :00E .0? 0:00:20 .0“. $5.00... 205.200
0202.
0005 00000.0 :00... .0 00:03:00 u:..0>0
832: 3.53: 2.33: 832.5 20.25 08: 8.0 5.530..
0.03. 0>0 :. . 00:09.2 :. . 0505.0200 .0 5.35.00 2.0.02. £00300 >030

 

.0050. n 0.02.

 

43

 

000a:

 

 

 

 

02.2.0 0.0.
:. 00020:. 0: .00 .00> 000.080 .20 :002 00 0:0 .00 :. 00002.0
”00000.0 :00... 2.0.00.0 + 0.20 0 .00 0000 .0 0.00
0200.0 .0: 00:20 20.03 .0 :000.>00 020000.: 000.00 000.0...
02000.0 .02 H>.._0..02 00> .0: ._20 :002 0.00:20 0.20.02. :. Eon :02 0.00:0500
0.0000 .0:...
.00.0>:0 .0 .0>0. 0:0 .002 .0 000.0
.00.:. .000. .000 20.03 >000 .0..:00
. 2.2:. .02020 .0 .x. 0 .000. 0:. .0.
..0.200.0:0 .0 .0 £002 .00> >.0>0
£0.00 0:0 000.. 0:0 000.0
5020.0 .0 000 0:0 000:2. 000.0 .u:
20000.0 000.0 020.03 .N0 ..0.:_ 0... 05.002.
0200.0 .0: .00000 >000 .0 :000.>00 .0. 05:0... .0 3.00
”00000.0 :00... 0:. 0:00.. 000:... 0.00:0.0 0 >.0>0 000 00000.0
0~v .0:...020 ”0000 00.0002: :00: .0. 0.0..
0200.0 .02 .20 .. .00> H>.._0..02 0~v .20 .00> .20 0:..0000 00.00002 N 20.0.... :0... .0 :02 P0.0.0.2
.20 :00...
._0.0.00_0..0 200......
£0.00 000000 00
2.2:. .0:00.0 .00>.0>.. .00 .00 .00 3001...:
0200.0 0.00.02.20.20 03. 0:...00 .00 0000 00.00 000 0>:0::.00
.0: ”00000.0 000... .200020 *0 F N .0 000. .0 0.20.0; .000>0.0:.0 0.2020
00.030 .02 00> ”20:02 02 08.0 .00.. 0:0 0.00 02.000: 0.0:. 0......) 0.0.00.5
00.02.
0002 00000.0 :00: .0 00:02:00 0:..0>0
0.0.00“. 302.02 5:00.02 0030.05 0.0.02. 000: 0.00 5.5000...
20.0 0>0 :. . 0000042 :. . 0:.0:00.:00 .0 5.05.00 20.02. 0.00300 >030

 

 

 

 

 

 

 

 

.0050. 0 0.00..

 

 

£92;

 

 

 

s 0558:: £9030. Sou:

3:0 £00» or 0:5 .0 0005 8.3 000

00:05:. 3.20096: .0>0 .55 000.2 :0 $903 .0290; 0000:0502

02 ”000005 :00: 9.20:5 .0 0030050. 0050008 .cogasN
00.030 ~02 00:000. .02 ”£8.02 00> 09‘ .0 0.002001 v .000. Z 5 :02 0:00:00.“

mourn:

.wao co .3 .3

E: 223 .918 3a

28» 08.

m 2903 Na? 6 SP

02.228 9.20 02 8:801 .32 .82

0000.: 950.08 95050 K c. .900

.20 z 02 2.0890 in 000.2 +5. 0:0 00500 5

”$05: 9588 09.9.0 Dim m 8.82 Eon 00503
020005 02 ”3000.0 :00... .20 .8 .0050 £903 ~02 00 0030300 c. 02903 _0 0.0Eam «0.0.080...—
H00> 00630 .02 $58.05. 0: 00:3 00> .55 000.2 0.005% 0050005. 50000”. 90000500

0803
000: 000003 :00... .0 . 00:05:00 55.0»0 .

3300“. 3:53: 3.5.0: 00.00....5 2225 won: 800 085000..
0.0.". 0>0 0. _ 3300...? 0. — 9:258:00 .0 005.500 2202. 8003:» >005

 

 

 

 

 

 

 

 

6.2.8. 0 202.

 

45

 

NONI:

 

 

 

 

 

 

 

.meN
000 0.050.
0:0 0.05
.0..0000::..2
05000.0 605. 0:0
000... .9. N. :900850
000.5020... 00.030 .0 h .0....0 0.0305. 50: 0.3.00.
....o._ :5... .0: ”00000.0 :00... ...0.00. .0 5000. 5:050 20.03.05
..0.0.00_0..0 00.030 .0: 0.0: :000 .0 0.0.0500 .0 050.02. 90.05050: :0 .0 052..
.88 62 550.2 00.030 .oz 829. 2:300 05 83 0238.2 55.8... .fgsoigo
L0.0.0
0>0... 00
:0..0:u.000 son:
...00 .0 00.3 000
05.. J; .050:
0:0 .000. .0 .00:0 .:05
0.05 .0 .00. .5500. 0.
or .0 000.. .N 3.05000.
00.030 5052.05 0500... 50:00::00
2.0. .0: 000006 :00... .55 E00... .26 .:0>0I
0.5.0.03 00.030 .0: >50... 20.03 050.02. ...02 50.. . :0502.
H00> 55000.2 00.030 .02 02 :0 0.00m .. 03.000”. :0502. .0». :00
00.02.
0002 00000.0 «.00... :0 00:03:00 05.2.0
0.800.. 0.33: 5.53: 830...; £0.02. 08: 5.0 8.52.0:
0.0... 0>0 :. . 00:00.2 :. . a:.0:=0_.:00 .0 53.5.00 2.0.02. 0.003..» >0...»

 

 

 

 

3.330. n 030...

 

46

Blood Lipids and Blood Pressure as Risk Factors for CVD

Total Cholesterol: Although information on blood lipid subfractions
allows more accurate calculation of CVD risk, total cholesterol concentrations
are still valuable in deﬁning level of risk for CVD. There is a large body of
epidemiologic evidence supporting a direct relationship between the level of
total cholesterol and rates of CVD.‘”' 123° ‘2‘ For both males and females, cohort
studies show an association between total blood cholesterol levels and a CVD
rate. This association is continuous throughout the whole range of cholesterol
levels in the population and is particularly strong at the higher levels of serum
cholesterol. Clinical and epidemiological studies show that a reduction in total
serum cholesterol of 1% is associated with a 2-3% reduction in CVD rates?”5
The 1993 National Cholesterol Education Program recommendations ‘3‘
promote measurement of total cholesterol as the primary screening for CVD risk:
"In individuals free of CVD, total cholesterol levels less than 200 mgldl are
classiﬁed as desirable blood cholesterol, those from 200-239 mgldl as
borderline-high blood cholesterol, and those 240 mgldl or greater as high blood
cholesterol."

Although low density lipoprotein-cholesterol (LDL) is a more sensitive
indicator of risk, most cholesterol in serum is contained in LDL, so the
concentration of total cholesterol in most people is highly correlated with the
concentration of LDL.

HDL Cholesterol: High Density Lipoproteins (HDL) normally carry 20-

47

30% of total cholesterol. HDL concentrations are inversely correlated with CVD
rates over a broad range of HDL levels. For every 1-mg/dl decrease in HDL, the
risk for CVD is increased by 2—3 per cent.126 Low HDL (<35 mgldl) is classiﬁed
as a major risk factor for CVD, and high HDL (60 mgldl and above) is considered
a negative risk factor.131 In elderly Dutch men, HDL was most strongly
associated with the risk of a ﬁrst coronary event. ‘27

A variety of factors contribute to low HDL concentrations. Genetic
inﬂuences are known. "8- “ Inherited influences are accentuated by lifestyle,
including cigarette smoking and physical inactivity and excessive caloric intake
leading to obesity. Certain drugs, including beta-adrenergic blocking agents
(beta blockers), anabolic steroids and progestational agents, also reduce HDL.
Smoking cessation, increasing physical activity, and weight reduction in
overweight individuals can increase HDL concentrations. Most lipid-lowering
drugs have the potential to raise HDL concentrations. No clinical trials have
been reported which speciﬁcally test the efﬁcacy of raising HDL in prevention of
CVD.131

The Ratio of Total Cholesterol to HDL: The ratio of total cholesterol to
HDL has been identiﬁed as the most accurate predictor of coronary heart
disease at all ages in the Framingham population”9 and in the elderly males of
the Zutphen population.127 In the Physicians Health Study cohort, the ratio of
total cholesterol to HDL was a signiﬁcant predictor of risk for myocardial
infarction; after adjusting for other risk factors, a change of one unit in the ratio

of total cholesterol to HDL was associated with a 53% reduction in risk.‘3°

48

Low Density Lipoproteins: LDL typically contain 60-70% of total serum
cholesterol. LDL is a sensitive measure of CVD risk. LDL is the primary target
of cholesterol-lowering therapy because direct clinical trial evidence for the
beneﬁt of lowering LDL is strong. In young men and premenopausal women,
LDL concentrations below 160 mgldl are considered optimal, 160 to 220 mgldl
are considered moderately high, and 220 mgldl is considered very high.131

Triglycerides: Serum triglycerides concentrations are positively
correlated with CVD in most epidemiologic studies, but when risk calculations
take into account total cholesterol and HDL, triglyceride levels are often no
longer statistically signiﬁcant predictors of CVD?“ ‘32

Hypercholesterolemia as a Risk Factor for Mortality from Non-CVD
Causes: It should be noted that the relationship between serum cholesterol and
all-cause mortality is far different than that between serum cholesterol and the
risk for coronary heart disease. AIIred133 described the relationship between
serum cholesterol and all-cause mortality as a U-shaped curve in males, where
low cholesterol concentrations were associated with an equally high risk of all-
cause mortality as elevated concentrations. In the middle range of cholesterol
concentrations (representing about 70% of men), there was no association
between cholesterol concentration and mortality. In women, AIIred found no
change in risk of all-cause mortality as serum cholesterol concentrations
increased above 160 mgldl. It is known that cancer risk, especially lung cancer,

is elevated in both males and females when serum cholesterol concentration is

49

less than 160 mgldl.” There is lack of consensus on whether early-stage
cancers cause a lowering of blood cholesterol levels. The risk of hemorrhagic
stroke in males has been shown to decrease as serum cholesterol increases.135
Blood Pressure as a Measure of CVD Risk: Nonfatal and fatal
cardiovascular disease - including coronary heart disease and stroke - as well
as renal disease, and all-cause mortality, increase progressively with higher
levels of both systolic and diastolic blood pressure.““3 These relationships are
strong, continuous, graded, consistent, independent, predictive, and
etiologically signiﬁcant. In the general population, risks are lower for adults with
an average systolic blood pressure of 120 mm Hg and an average diastolic
blood pressure of less than 80 mm Hg. Higher levels of either systolic or
diastolic blood pressure, or both together are associated with increased risks of
morbidity, disability, and mortality. For blood pressure, meta-analyses of studies
of drug treatment for hypertension show that a reduction of 5-6 mm Hg in mean
diastolic blood pressure was associated with reductions in total mortality of 8-

4296-137, 138

DESCRIPTION OF THE MRFIT DATA BASE

Purpose of the MRFIT

MRFIT was a national, publicly-funded research study extending from
1974 to 1982. Its original objective, as stated in the initial request for proposals,
was "to determine whether or not a preventive program directed at the reduction
of serum lipids, reduction of blood pressure and reduction or elimination of
cigarette smoking among men aged 40-59 who are at high risk of coronary heart
disease from a combination of these risk factors can result in a signiﬁcant
reduction in the incidence of myocardial infarction and death from coronary
disease over a six year primary intervention clinical trial."‘39 In 1972-73, funds to
conduct the study were awarded to 22 clinical centers, a coordinating center, a
central laboratory, standardization laboratory, ECG centers and a drug
distribution center.
Subjects

To be selected for the study, men had to be between ages 35-57, at the
upper end of the risk spectrum for heart disease because of elevated
cholesterol, elevated blood pressure, or smoking, but could not have any
evidence of the existence of heart disease. After screening over 370,000 men,

12,866 meeting the inclusion criteria were randomly assigned to either the

50

51

Special Intervention (SI) group or the Usual Care (UC) group. Excluded from the
study were men who seemed unsuitable for long-term dietary intervention,
including those with blood cholesterol higher than 350 mgldl, those taking
medication for diabetes and those following prescribed diets incompatible with
the MRF IT dietary protocol. Excessive alcohol intake and gross obesity were
also grounds for exclusion. Because of the heavy demands of time and
cooperation required of MRFIT participants, random sampling was not possible.
A convenience sample was used, where men were solicited via advertisements
for free health screening. Because there was not a probability sample and
because subjects were atypical in that they were at high risk for heart disease,
the results of the MRF IT study could not be generalized to the entire adult male
population.
Data Collection

All 12,866 participants underwent an extensive three-phase screening,
during which baseline health and dietary information was documented.
Subsequently, all participants were invited back for extensive annual medical
exams. Additional data were collected on members of the SI group every four
months, including documentation of weight, blood pressure, cholesterol, and
smoking status.

Nutrition Data Collection: The primary method used to document
nutrient intake in the MRFIT was the 24-hour dietary recall. This method was
considered adequate for the research protocol of MRFIT because the unit of

analysis was the mean intake for the entire SI or UC group of each nutrient of

52

interest. Twenty-four-hour dietary recalls were collected at baseline and at
years 1,2,3, 5, and 6 for both SI and UC group members.

Food intake data from 24-hour dietary recalls was converted to nutrient
data in a computerized process, developed collaboratively by MRFIT staff and
staff of the Lipid Research Clinic, with oversight by the National Heart, Lung,
and Blood Institute (NHLBI) and input from the United States Department of
Agriculture. The nutrient data base initially set up in 1973, called the NHLBI
Table of Food Composition, used Agricultural Handbook No. 8“° as the major
reference, with updated ﬁgures for fatty acids and dietary cholesterol found in
the scientiﬁc literature and from food manufacturers. When Agricultural
Handbook 456“1 became available in 1975-76, data were compared and, where
appropriate, updated.

Once the data base was established, NHLBI established the Nutrition
Coding Center (NCC) in the Department of Biometry, School of Public Health,
University of Minnesota. The NCC was responsible for maintaining and
updating the data base, coding dietary records collected in both the Lipid
Research Center Study and the MRFIT, and training ﬁeld nutritionists in data
collection procedures. Nutrient information for margarines, shortenings, and
other processed foods was obtained from food manufacturers and was routinely
updated in the data base.142

The data base was most complete with respect to dietary fats, because it
was developed speciﬁcally for heart disease research which focused on

modiﬁcation of total fat, cholesterol, saturated fat and polyunsaturated fat. Data

53

on other nutrients, such as water-soluble ﬁber and folic acid were most likely not
as reliable as the dietary fat data, because not all foods had been analyzed for
these nutrients at that time, and because laboratory methods for detecting
several nutrients have been reﬁned considerably since that time.

Physical Activity Data: To document physical activity among MRFIT
participants, the Minnesota questionnaire to assess leisure time physical activity
(LTPA) was administered to all participants at baseline, and at annual visits 1, 4,
and 6. This questionnaire lists 18 major activity groups and 62 individual
physical activities. Subjects indicated the number of occasions per month during
the previous 12 months that they performed each activity, and its average
duration in minutes. Trained interviewers helped with the process at all but
baseline assessment. Activities included in the questionnaire were classiﬁed as
light (requiring 8.4 to 16.8 kJ/min.), moderate (requiring 18.9 to 23.0 lemin.), or
heavy (25.2 kJ/min. or more). The instrument had been previously validated
against treadmill exercise performance and energy intake from dietary
records,“3' 1“ and has subsequently been found to predict mortality in the
MRF IT population.145

Another measure of habitual physical activity pattern available in the data
set was the participant's opinion of his own physical activity compared to others
his age (much less active, somewhat less active, about the same, somewhat
more active or much more active). Although this is a subjective measure, it has
predicted mortality in the MRF IT population, has been used as a covariate in

several analysis of the MRF IT data by the MRFIT Research Group " “5 and has

54

been found to correlate with risk factors in ways physical activity levels would be
expected to do. Exercise opinion was asked at each annual visit.

The study protocol for MRFIT included a measure of physical ﬁtness - a
submaximal graded treadmill exercise test, with stepwise increase of slope and
speed until a predetermined target heart rate was reached or until the test was
terminated for other reasons such as heart rhythm irregularities or subject
discomfort. Target heart rates were based on age, and were estimated to
represent 85% of the predicted maximal heart rate.147 Such a test is an
established procedure for evaluation of suspected coronary heart disease and
for assessment of cardiopulmonary ﬁtness. The number of minutes a subject is
able to continue the test ("exercise duration") is considered an accurate
measure of physical ﬁtness.“8 Graded exercise tests were performed as part of
the electrocardiogram at annual visits.

Tobacco Data: Use of tobacco by MRFIT subjects was extensively
documented. For the SI group, the average number of cigarettes per day over
the preceding four months is recorded at each visit, every four months, over the
entire study period. For the UC group, average number of cigarettes over the
past four months is recorded annually. In addition to self-reported tobacco use,
blood levels of the compound thiocyanate were measured on each subject to
independently verify the subject’s self-report on smoking status. Thiocyanate
levels are elevated from cyanide in tobacco smoke. They can also be raised by
consumption of certain foods (the Brassica genus - cabbage, cauliﬂower, kale,

kohlrabi, broccoli, brussels sprouts, turnips, and rutabagas, and also fruit pits

55

and almonds), and by use of diuretics. Inasmuch as the dose-response curve of
number of cigarettes to blood levels of thiocyanate was non-linear, serum
thiocyanate levels were used only to verify smoking cessation, rather than to
verify the reported dose of tobacco.149

Data on Medical Conditions and Medications: For all MRFIT subjects,
extensive physical exams and health histories were performed once each year.
The presence or absence of 55 different conditions was noted by MRF IT
physicians for each subject. Each annual physical exam included a history of
medication use over the previous year. Documentation of medication use was
very speciﬁc for antihypertensive drugs, but most other drugs were identiﬁed
only by broad category. The categories of medications documented in the data
set are as follows:

Digitalis; nitrates including nitroglycerine; propranolol; lipid-lowering

drugs including cloﬁbrate, cholestyramine, sterol-binding resins, beta-

sitosterol (Cytellin), nicotinic acid derivatives, neomycin, dextrothyroxine

(Choloxin), probucol, estrogens, progestins, heparin, haloﬁnate; drugs for

gout including probenicid, allopurinol, or colchicine; insulin or oral

hypoglycemic agents; anticoagulants; antibiotics or anti-infection agents;
steroids; amphetamines or other stimulants; barbiturates or other
sedatives; anti-anxiety drugs including Librium and valium; potassium
supplements.

Mental Health Data: Mental health data were collected on 12,772 of the
MRFIT subjects, to test the hypothesis that men with a Type A Behavior Pattern
would have a higher incidence of ﬁrst major coronary events. 15° Type A
Behavior Pattern was deﬁned to include competitiveness, excessive drive and

enhanced sense of time urgency. Behavior patterns were documented in two

ways. A validated structured interview was administered to a subset of 3,110

56
men by highly-trained technicians and psychologists. All subjects took the

Jenkins Activity Survey, a 54-item self-administered questionnaire.

Blood Lipid Fractions in MRFIT: The MRFIT study used the most
sophisticated criteria for screening for CVD risk available at that time. The blood
lipid fractions recorded throughout the study were total serum cholesterol and
triglycerides. On alternating years, plasma levels of total cholesterol, HDL, LDL
and VLDL were measured. At baseline, serum cholesterol was collected at the
ﬁrst screening, and plasma cholesterol was collected at the second screening.
Lipoprotein subfractions of HDL, LDL, and VLDL were measured from plasma
samples. Because there are established analytic discrepancies between plasma
and serum total cholesterol values, the MRF IT Group always compared serum
values to serum values and plasma values to plasma values.151

Blood Pressure Measurement in MRFIT: Blood pressure was recorded
annually for the UC group and every four months for the SI group. Four
separate blood pressure readings were made at each visit. Systolic blood
pressure and diastolic blood pressure were deﬁned as the average of two
Korotkoff ﬁrst- and ﬁfth-phase readings, respectively, obtained with a random-
zero sphygmomanometer. The posture of the subject was standardized (seated)
and same arm was always used for blood pressure measurement.152
Intervention

Men in the UC group received no health messages or intervention from

MRFIT staff other than the information that they qualiﬁed for the study because

57

of their high risk for heart disease. They received all medical care on their own,
from their usual providers. Many of the members of the UC group received
intervention on their cardiovascular risk factors from their usual providers, but
such intervention was not documented. Men in the SI group participated in an
initial intensive series of group sessions designed to assist in modiﬁcation of
behavior relating to the three risk factors. Subsequently, the SI men were
invited back to their clinics at least three times each year to maintain and
increase risk factor change.

Intervention on Blood Lipids: The goal of the MRFIT nutrition
intervention was to lower the serum cholesterol of all Sl men by 10% or more for
those with baseline levels of 220 mgldl or greater (this included the great
majority of the men). The food pattern initially taught included saturated fats at a
maximum of 10% of calories, polyunsaturated fats at a maximum of 10% of
calories, total fat less than 35% of calories, and dietary cholesterol less than 300
mg per day.151 When initial data collection showed that many men were
consuming lower levels of some of these dietary elements, the pattern was
changed to specify saturated fat no greater than 8% of calories and cholesterol
no more than 250 mg. per day. The dietary intervention was delivered in two
phases. The ﬁrst, intensive phase, consisted of group instruction in the ﬁrst 10-
weeks of participation in the trial. The second phase continued for the
remainder of the trial, and consisted of individual nutrition counseling at least
every four months. If blood lipids did not respond well, more frequent counseling

was done. Consistency of intervention across the 22 centers was accomplished

58

through centralized national training of local staff members and the use of
standardized educational materials at all centers.

The intervention for changing dietary practices used state-of-the-art
behavior change principles, still considered appropriate today. Support for
behavior change was provided continuously throughout the six years of
intervention. Each SI group member was scheduled to meet with a nutrition
interventionist a minimum of every four months for the entire six years of the
trial.

Intervention on Blood Pressure: ‘52 A participant was considered
hypertensive at screening if his diastolic blood pressure was 90 mm Hg or
greater at the third screening visit, conﬁrmed within 4 weeks and/or if he was
taking drugs for hypertension treatment. Men who were not categorized as
hypertensive at baseline were later designated as hypertensive if their diastolic
blood pressure was 90 mm Hg or higher at a regularly-scheduled 4-month visit
(conﬁrmed at an additional visit within 4 weeks). Men whose diastolic blood
- pressure was 115 mm Hg or more at baseline were excluded from the study and
advised to get immediate medical attention.

The primary intervention for hypertensives was drug therapy as deﬁned
by the Stepped Care Program, employed uniformly at all 22 centers. As
described above, overweight men were not started on the Stepped Care
Program until weight loss was attempted. When a man was entered into the
Stepped Care Program, a goal blood pressure was set for the individual.

Antihypertensive drugs were prescribed in a step-wise manner, beginning with a

59

diuretic and adding more potent medications as necessary to achieve a blood
pressure below goal. There were three phases: ( 1) Step-up, where drug
therapy was increased in a step-wise fashion until a satisfactory response was
obtained, (2) Maintenance, where no change in the drug was made, and (3)
Step-down, where men whose diastolic blood pressure had remained below 80
mm Hg for 4 consecutive visits during maintenance could have a step-wise
reduction of medications.

Sodium restriction was also part of the intervention for hypertension. Men
were generally instructed to minimize the intake of foods high in sodium content
and to use salt sparingly, if at all, in food preparation and at the table. More
intensive counseling for sodium restriction was done if a man was taking a
maximum dose of Step 2 medication or taking Step 3 medication and the
diastolic blood pressure was not below goal.

The diuretics drugs used in the trial, chlorthalidone and
hydrochlorothiazide, were found to have a side effect of increasing serum
cholesterol.

Intervention on Smoking: At baseline, 63.8% of the men in the MRFIT
SI group smoked cigarettes, with an average of 21.5 cigarettes smoked per day.
For smokers, the ﬁrst priority for intervention was smoking cessation. Emphasis
was placed on total cessation of smoking, but reduction of dosage by decrease
in the number smoked, change in brand, or. reduced inhalation was encouraged
for those unsuccessful in quitting. Strong anti-smoking messages were given to

all smokers immediately upon their entering the study, at the time of the third

60

screening visit. Intensive encouragement to stop smoking was provided during
the initial 10-week series of group classes. After the ﬁrst 10 weeks, those who
quit smoking were seen on a maintenance basis, and others were placed into an
“extended intervention” phase. A number of state-of-the art behavioral
techniques were used such as stimulus control, positive reinforcement,
contracting, record keeping, and relaxation. Other approaches included
retreats, ex-smoker-Ied groups and special events. Hypnosis and a mild
aversive technique were used, following precise protocols.”3

Intervention on Weight: A basic assumption in protocol development
was that reduction in weight for obese individuals would reduce serum
cholesterol when dietary lipids composition is low in saturated fat and
cholesterol. It was also assumed that weight reduction would reduce blood
pressure for some men. There was a goal often pound weight loss or more for
all men who were overweight. At the beginning of the study, overweight was
deﬁned as weight equal to or greater than 1.2 x Ideal Weight. In late 1976,
when it was noted that cholesterol levels were not coming down as much as had
been anticipated, the deﬁnition was revised downward to weight equal to or
greater than 1.15 x Ideal Weight. Ideal Weight was deﬁned as 0.9 x the average
height-speciﬁc weight of men aged 18-34 in the National Health Survey. Note
that men whose weight was greater than 1.5 x "Standard Weight" were ineligible
for the study.

The "MRFIT diet," which all members of the SI group were advised to

follow, included the provision that calories would be adjusted to achieve 1.15

61

Ideal Weight. In the initial protocol, no emphasis was placed on weight loss, per
se, except for overweight hypertensives were eligible for a weight reduction
program lasting up to 16 weeks in an effort to lower blood pressure before drug
therapy begun. ‘52 It was expected that adherence to the MRFIT eating pattern
would result in weight loss for many participants. When it became clear, a
couple of years into the trial, that the expected reductions in serum cholesterol
were not being achieved, a greater emphasis was placed on weight reduction.
In 1975, a set of detailed nutritional and behavioral guidelines on weight control
was developed. In 1976, the protocol was changed to allow the recommendation
of increased physical activity as a means of controlling body weight.“51

Physical Activity Intervention: The powerful effect of physical activity on
serum cholesterol levels, blood pressure, and weight were not as well-
recognized in the 1970's, when the MRFIT study was designed, as it is today.
Increasing physical activity was not included in the treatment protocols for
hypercholesterolemia or hypertension, but was added to the protocol for weight
loss in 1976. The advice was to be limited to recommending an increase in
duration but not rate of energy expenditure in currently habitual forms of physical
activity. In practice, this meant that most overweight participants were advised
to walk more, inasmuch as most of them were habitually sedentary. More
ambitious increases in physical activity were neither encouraged nor
discouraged, but participants were advised to see their private physicians before

embarking on a program of more vigorous activity.”

62

Quality Control on Data Collection

One of the advantages of using the MRFIT data is the quality of the data
collected. The MRFIT Study was designed in a national spotlight. The protocols
were subjected to intensive review. Because data were collected at 22 different
sites around the country, much attention was given to standardization of
methods. Uniform training on the protocols was provided by the national
coordinating center. Standard forms were used for recording data at all sites. A
central lab using only recognized techniques was used.
Speciﬁc quality control procedures were developed and implemented for the

followingz‘5‘5

Biochemical data at the central lab,

Forms control and detection of errors in reporting at the Coordinating
Center

Resting electrocardiograms

Nutrition modalities

Screening procedures

Clinic operations

Blood pressure measurement

Site visits to clinics.

METHODS

Human Subjects Approval

Approval for the research was obtained from the University Committee on
Research Involving Human Subjects. Appendix A includes a copy of the letter of
approval.
Obtaining the Data

A Freedom of Information request for the MRFIT data set was submitted
to the National Heart Lung and Blood Institute (NHLBI). Appendix B includes
the letter approving the request, which lists the conditions that the MRF IT
Coordinating Center at the University of Minnesota was not available for
extensive assistance and that courtesy copies of any manuscript prepared for
publication be submitted to the MRFIT Coordinating Center. The data set was
received on 19 tapes - 8 tapes with health information from each of the 8 years
of the trial; 4 tapes with physical activity data (collected at baseline and years 1,
4, and 6); and 7 tapes with nutrition data (at baseline and years 1,2,3,5, and 6).
Extensive documentation of each data tape was provided, with every version of
every data collection instrument used over the 7 years of the study, plus brief

descriptions of every variable on each tape.

63

64

Extraction of Data Needed for Analysis

The data set documentation was examined to discern what data elements
were available. Over 1,000 data elements for each subject were identiﬁed as
necessary to answer the research question. Computer programs using SAS
software (SAS Institute, Inc”) were written to allow extraction of the needed
variables from each data tape, using the Michigan State University mainframe
IBM computer. Each program was designed to locate the variables of interest on
the data tape, copy them from the data tape into a SAS data set suitable for
analysis by personal computer and give each variable a new name. The data
sets created in this manner were then sorted by subject identiﬁcation number,
and combined as needed at each stage of data analysis. See Appendix C for a
sample of a successful SAS program for data extraction.
"Cleaning" the Data

Checking for lnaccuracies in the Data Set: The data set was checked
for inaccuracies in several ways. First, all variables for the initial 50 subjects
were printed out and visually Inspected to determine whether values reported for
each variable were feasible. Second, for each variable, descriptive statistics
were generated, including mean, range, minimum and maximum values; these
were evaluated for feasibility.

Identifying Infeasible Weight Data: One of the most critical phases of
data analysis was making decisions about extreme values for weight
measurements. Extreme weight changes which actually did occur were of the

greatest importance for the proposed research. However, extreme weight values

65

that were the result of measurement or coding errors would introduce random
error which would increase the chance of a Type II statistical error (failing to ﬁnd
a difference when one really was present). Criteria for detecting infeasible
weight measurements were developed separately for the four-month weight
changes documented in the SI Group and for the twelve-month weight changes
documented in the UC Group.

Infeasible 4-month Weight Changes: For weight changes in the SI
Group, documented every four months, a review of the literature was done to
determine the maximum biologically-possible weight loss and weight gain for
men over a 4-month interval.

For weight loss, a national multi-center evaluation of the Optifast
program157 was chosen as suitable for estimation of a maximum biologically-
feasible rate of weight loss. The men in the Optifast program were considerably
heavier than the MRFIT subjects and experienced considerably greater caloric
restriction than the MRFIT SI Group were likely to have seen. The mean weight
loss in four months observed in the Optifast subjects could reasonably be
considered a biological maximum for the purpose of this research. For weight
gain, three studies of purposeful weight gain in human males were
found,“ 158"“ in which overfeeding resulted in weight gains of 16-25.5% of
baseline weight in four months. Based on these four studies, it was concluded
that a weight loss or gain of approximately 20% would represent the maximum
biologically-feasible weight change in a four-month interval. Table 4

summarizes the ﬁndings of the relevant studies.

66

Table 4 - Summary of Studies of Weight Change In Males Under Extreme
Conditions of Overfeeding and Underfeeding

 

 

 

 

 

 

 

 

 

 

 

Weight Change in 4
Mean Age, Initial Months, (Per Cent of
Population N Years Mean Wt. Body Weight)
Men from 18 Optifast 110 42.3 128.7 Kg. Loss of 22%
Clinics157
Male Canadian Twins 95 24 21 60.3 Kg. Gain of 16-22%
Vermont Prisoners“ 9 24.8 Not Gain of 19%
Reported
Males, Fattenirstg Ritual in 9 23-25 68.5 Kg. Gain of 25.5%
N. Cameroon1

 

Even with the literature review described above, deﬁning the criteria for
infeasible weights in the data set was still a somewhat arbitrary process. On one
hand, the MRFIT participants were not subjected to the extreme caloric
restriction that Optifast patients had experienced. On the other hand, weight
regain in an ovenrveight person following a period of calorie-restricted dieting for
weight loss could be much more rapid than weight gain in a slender person
caused by overfeeding. The decision was made to deﬁne infeasible weight
changes in the MRF IT as follows: A weight was considered infeasible if that
weight represented a change greater than 15% of body weight in one 4-month
interval and if that weight were also more than 1.96 standard deviations above
or below the subject’s mean weight over the course of the trial.

There were 107 men in the SI group who had weights meeting these
criteria. To verify that the criteria were appropriate, plots were made of weight

on time for 16 of the 107 subjects, selected at random, which supported the

67

criteria. One additional check on the 107 infeasible weights was made. For
each subject with a weight considered infeasible, a printout was obtained of all
weights and all corresponding blood pressures. Since blood pressure is very
responsive to changes in weight, patterns of change in blood pressure were
examined for consistency with patterns of change in weight. This closer
examination revealed that only 95 weights were truly infeasible, and each of
‘ these was re-deﬁned as a missing value.
Infeasible 12-month Weight Changes: To identify infeasible weights for
' the UC group, where weights were recorded only on an annual basis, it was not
possible to ﬁnd in the literature studies reflective of maximum biologically-
feasible weight gain or loss over a twelve-month period. Since a weight gain of
15% in 4 months identiﬁed outliers in the SI group, it seemed obvious that an
infeasible per cent weight change in a year should be deﬁned at higher than
15%. On the other hand, the UC Group did not receive the intense
encouragement to lose weight that the SI Group was given by the MRFIT staff,
so a lower cut off for infeasible weight change may have been appropriate. A
trial and error approach was used to deﬁne exclusion criteria: Several criteria
were applied, and were then veriﬁed by plots of weight on time for a sample of
subjects found to have infeasible weights by each criteria.

Using this method, the criteria for exclusion for the UC Group ﬁnally
adopted were that a weight was considered infeasible if that weight represented
a change of 25% in a one-year interval and that weight was more than 1.96

standard deviations from subject’s mean weight. When these criteria were

68

applied, 46 weights were identiﬁed as infeasible, and were re-deﬁned as missing
data.
Dealing with Missing Weight Data

The next step in preparing the data for initial analyses was deciding how
much missing weight data would be tolerated before a subject was excluded
from analyses. A program to quantify the patterns and numbers of non-missing
weights was developed.

Dropping Year 7 Data: As expected, the longer participants were in the
study, the more missed visits were observed. By the time of the year 7 annual
visit, 7,500 of the initial 12,866 subjects did not have weight data, and most of
these were also missing cholesterol and blood pressure measurements. The
decision was made to consider the data collected at the year 6 annual visit to be
the "ﬁnal" measurements for the purpose of this study. For the outcome
variables HDL and the ratio of total plasma cholesterol to HDL, year 6 was, in
fact, the last year that this information was documented. To assess how results
might have differed if the year 7 data had been retained as the ﬁnal date of
weight measurement, subjects who were present at year 7 were compared with
those absent at year 7 in terms of several characteristics (ANOVA, with post hoc
test Bonferroni Inequality). See Table 32 for a summary of these comparisons.

Identifying Subjects with "Enough" Data: For the 4-month interval
weights measured in the SI Group, 3,510 subjects had a complete set of 19
weights, 841 subjects were missing only one weight, 425 subjects were missing

two weights, and 236 were missing three weights. It was arbitrarily decided that

69

a subject missing more than 3 weights would be excluded from analysis. One ,
additional criterion was added in identifying subjects with enough data for
analysis - the subject needed to have the weight measured at the year 6 annual
visit, which was the weight used to deﬁne the summary variable "net weight
change." When the double criteria of having at least 16 valid weights and also
having the year 6 annual weight, 4,932 SI subjects remained for analysis.

For annual visit weights only, 10,039 subjects had a complete set of 7
weights, 1,101 were missing one annual weight, and 1,726 were missing two or
more weights. It was arbitrarily decided that only subjects with a complete set of
7 annual weights would be retained for analysis, resulting in a data set with
10,039 individuals. To assess possible bias from dropping individuals with
missing data, T tests were run, comparing subjects dropped with those who has
"enough" data. Table 33 summarizes these comparisons.

Replacing Missing Weight Values
The few missing weights left in the SI data set were replaced with

calculated estimates of the missing weight. Two possible computations were

considered:

1. The missing weight could have been replaced by the subject’s mean
weight over the course of the trial. This commonly-accepted method for
replacing missing values was rejected because it could result in
overestimation of a subject’s number of weight cycles and the magnitude

of the residuals for the regressions of weight on time.

70

2. The missing weight could have been replaced by the average of the
weight preceding and the weight following the missing value. Given that
the weight measurements were taken at such short intervals, this method
was considered most appropriate.

Creating Summary Variables Suitable for Statistical Analysis

A number of decisions were made as to how to quantify the data
available in ways which would meaningfully reﬂect factors potentially affecting
changes in blood pressure and changes in blood lipids. The decision-making
process for each category of information is described below.

Outcome Variables: The outcome variables deﬁned for this research
were changes in the cardiovascular risk factors: total serum cholesterol, HDL,
the ratio of total cholesterol to HDL and diastolic blood pressure. Note that the
ratio of total cholesterol to HDL was based on total plasma cholesterol rather
than total serum cholesterol, because the laboratory HDL measurements were
based on samples of blood plasma rather than blood serum. The value for
baseline blood pressure was the value deﬁned by MRF IT staff, which was the
average of four diastolic blood pressure measurements taken with a random-
zero sphygmomanometer - two at the second screening visit, and two at the third
screening visit. The value for the year 6 blood pressure was the average of two
diastolic blood pressure measurements taken with a random-zero
sphygmomanometer at the year 6 annual exam. In each case, the change in
risk factor was deﬁned as the difference between the value at the year 6 annual

exam and the value at baseline.

71

Predictor Variables: For this study, two different deﬁnitions of weight

cycling were used, plus a third which was a combination of the other two. For

each potential deﬁnition of weight cycling, plots of weight on time for a sample of

subjects was generated, to determine whether the weight cycling variables

created by mathematical deﬁnition corresponded to intuitive understanding of

weight cycling. The three measures of weight cycling were as follows.

1.

Number of Cycles: where a weight cycle was deﬁned as a loss and
subsequent regain (or gain and subsequent loss) of at least 5% of
baseline weight. The decision to use 5% of weight as the minimum
change to deﬁne a cycle was based on the ﬁnding by Blair et al.1 that, in
the MRFIT population, men who had undergone at least one 5% weight
cycle had a 55% increase in all-cause mortality compared with stable-
weight subjects. The original SAS program written by Jay McClellan to
count the number of weight cycles is found in Appendix D. In order to
compare the results of this study to the results of the mortality study of
Blair et al.‘, some ANCOVA models were created where the number of
cycles was expressed in terms of "cycling status," where a subject with no
cycles was classiﬁed as a "non-cycler," and an individual with one or
more cycles was classiﬁed as a "cycler."

855: The standard error of the estimate of the regression of each
individual's weight on time was selected over the standard deviation of

weight (used in most other weight cycling studies) because steady weight

72

losses and gains can result in a high standard deviation of weight when

no weight cycles occurred.

3. Number! Size of Cycles: In order to explore the potential impact of small
weight cycles compared to large weight cycles, a third measure of weight
cycling was developed, reﬂective of both the number and size of the
weight cycles, which combined the previous two cycling measures. An
individual with a low SEE (below the median SEE for the entire
population) was assumed to have smaller cycles than a person with a
large SEE. For the third measure of weight cycling, individuals were
classiﬁed into the following groups:

1. Zero cycles

2. 1 cycle, low SEE (<4.5 pounds)

3. 1 cycle, high $55 (24.5 pounds)

4. 2 cycles, low SEE (<4.5 pounds)

5. 2 cycles, high $55 (24.5 pounds)

6. 3 or more cycles.
Individuals with 3 or more weight cycles were not subdivided by magnitude of
SEE because only 34 of the 260 subjects with 3 or more weight cycles had a low

SEE. The third measure of weight cycling was used only in ANCOVA models.

73

Adequacy of 12-Month Interval Data for Describing Weight Cycling

The weight measurements recorded at 4-month intervals over a 6-year
period for the MRFIT SI Group constitute a remarkably complete documentation
of weight changes in a large population over time. The number of weight cycles
and the SEE calculated for the SI population based on 19 four-month interval
weights could be considered reliable estimates of weight cycling.

Before statistical analysis was begun, the question of how well the 7
weights documented at annual visits for all 10,039 subjects would describe the
"true" weight patterns based on 19 weights, known for the 4,931 members of the
SI Group. If weight cycling could be adequately characterized by 7 weights, all
10,039 subjects could be used in data analysis. To answer this question, the
"true" measures of weight cycling (based on 19 weights) were compared with the
measures based on only 7 weights, as follow.

The number of weight cycles and the $55 were calculated for each
member of the SI group, ﬁrst using all 19 validated weights taken at 4-month
intervals, and then using only the 7 validated weights taken at 1-year intervals. It
was found that restricting the weight data to only annual measurements resulted
in underestimation of the number of weight cycles in 54.2% of subjects, and also
resulted in discrepancies in the $55 (Pearson correlation coefﬁcients for SEE
based on 7 weights and SEE based on all 19 weights was 0.90). Because of
these discrepancies, the decision was made to perform statistical tests only on
the SI Group, where conﬁdence in the data was highest. To assess potential

bias from dropping the UC group from analysis, T tests were run comparing the

74

UC and SI groups with respect to the baseline characteristics age, weight,
relative weight, blood pressure, total cholesterol, HDL, and LDL (See Table 34.).
Other Independent Variables Created for Possible Inclusion in Analyses

Nutrition Variables: Nutrition information available for analysis in the
data set were the results of 24-hour dietary recalls recorded by highly trained
dietitians, taken at baseline and at years 1,2,3,5, and 6. Based on review of
literature on nutrients and heart disease, nutrients suspected of affecting blood
lipids and blood pressure were identiﬁed. SAS programs were written to extract
24-hour intakes of the following from the nutrition data tapes: grams of fat,
saturated fatty acids, polyunsaturated fatty acids, alcohol and water-soluble
ﬁber; calories; milligrams of cholesterol, calcium, iron, sodium, and caffeine; and
international units of vitamin 5.

Typical intakes of these nutrients over the trial were deﬁned as follows.
Baseline intake data were not considered representative of intake over the trial,
because the most intensive dietary intervention was done between the screening
visits where baseline data were collected, and the ﬁrst annual visit. Mean
intakes over the three consecutive years for which food intake data were
available (years 1, 2, and 3) were calculated for each subject for each of the
extracted nutrients. Values for calories, fat, saturated fat, and polyunsaturated
fat were used to calculate the summary variables mean per cent of calories from
fat and mean ratio of polyunsaturated to saturated fat (P:S ratio). If subjects did
not have at least two of the three 24-hour dietary recalls, the mean values for all

nutrients were deﬁned as missing and not used in analysis.

75

Pearson correlation coefﬁcients between each nutrient intake variable
and the outcome variables were very low, seldom reaching 0.1, and often failed
to reach statistical signiﬁcance (See Appendix E). For each analytic model, all
nutrition intake measures that were correlated signiﬁcantly with the outcome
were included in analysis.

In addition to the actual nutrient intake data, one additional nutrition
variable was extracted from the data set - the dietitian's rating of dietary
compliance, based on impressions made during dietary counseling. In the early
MRFIT publications,151 the dietitian's rating of dietary compliance was the
nutrition variable found to correlate most highly with changes in blood
cholesterol and blood pressure. This same high correlation was found between
compliance rating and outcomes in these data; however, the variable was
discarded from analysis when it was learned, from personal communication with
former MRFIT nutritionists, that laboratory values for cholesterol and blood
pressure were taken into account by the dietitians when they made their
assessment of dietary compliance.

Smoking-related Variables: The serum thiocyanate level used as a cut-
off to verify smoking status was 100 micromoles per liter, consistent with other
analyses based on this data set and ofﬁcially published by the MRFIT Research
.Group.153

For this analysis, subjects were classiﬁed as non-smokers during the trial
if, at baseline screening and at each annual visit, the subject reported being a

non-smoker and, in addition, his serum thiocyanate level at each visit was less

76

than 100 micromoles per liter. A continuous smoker was deﬁned as one who, at
each annual visit either reported being a smoker, or had a serum thiocyanate
level greater than or equal to 100 micromoles per liter. An intermittent smoker
was deﬁned as one who was veriﬁed to be a non-smoker at least one visit and
who was veriﬁed to be a smoker at least one other visit.

To quantify the amount of smoking, the mean number of cigarettes per
day over the trial was calculated, as was the sum of the number of cigarettes
smoked. Number of cigarettes per day and smoking status at the time of the
ﬁnal visit were also considered for inclusion in the analysis because the effects
of smoking on blood pressure are seen at the time the smoking is occurring and
shortly thereafter."52

Smoking cessation has been shown to be associated with weight gain,“6
and individuals who quit and resumed smoking several times would be likely to
have experienced weight ﬂuctuations. Two measures reﬂective of quitting
smoking were calculated. One was the per cent of annual visits at which the
subject was judged to be a smoker (based on self-report and serum thiocyanate
values). The other variable reﬂective of changes in smoking habits was the
standard deviation of the number of cigarettes reported.

Of the smoking-related variables, the most highly correlated with
outcomes were the mean number of cigarettes/day over the entire trial and
smoking status during the trial. These were highly correlated with one another
(r=0.55), and were not ever included in the same model. Mean number of

cigarettes/day was included in every model submitted.

77

Physical Activity Variables: Several variables reﬂecting physical activity
and physical ﬁtness were chosen. In the MRFIT, physical activity was
documented by means of leisure time physical activity scores (LTPA). It has
been established by the MRFIT Research Group that LTPA scores remained
relatively constant from Annual Visit 1 through Annual Visit 6.‘°° Therefore
LTPA scores for Annual Visit 1 were extracted from the data set for inclusion in
the study. On the MRFIT data tapes, physical activity information is summarized
as average minutes per day of light, moderate, and heavy LTPA, the sum of
which is the total minutes of LTPA. For this research, the primary measure of
physical activity chosen for inclusion was total minutes of LTPA, because that
variable had been found to best predict CVD and all-cause mortality in the
MRFIT population.‘°° Average minutes per day of heavy LTPA was also
included.

For analysis, LTPA was expressed in several ways. Some subjects were
found to have infeasibly high LTPA scores. It was assumed that men whose
total LTPA scores were unrealistically high were likely to be those who
performed a larger number of activities, and may have been among the more
physically active subjects in the trial. To avoid eliminating the most physically-
active subjects from the study, subjects were classiﬁed by quintile for minutes of
LTPA. Additional variables were created where the LTPA scores of subjects
with more than 10 hours per day of LTPA were re-deﬁned as missing values.

Altogether, 8 variables reﬂecting amount of physical activity were created:

total minutes of LTPA , total minutes of heavy LTPA, quintile of total LTPA,

78

quintile of heavy LTPA, plus expurgated versions of each of those 4 variables
where any values greater than 10 hours/day were redeﬁned as missing.
Correlations between each of the 8 physical activity variables and the outcome
variables were very low (always less than 0.1), and seldom reached statistical
signiﬁcance. For each analytic model, if any of the 8 physical activity measures
was correlated signiﬁcantly with outcome, it was included in analysis.

Another measure of habitual physical activity pattern available in the data
set was the participant's opinion of his own physical activity compared to others
his age. Exercise opinion changed from one year to the next in most subjects.
Therefore, three different measures of exercise opinion were created: mean
exercise opinion over years 1-6 of the trial, average of the last two reported
opinions and the exercise opinion at the ﬁnal visit. Of these three measures,
exercise opinion at the ﬁnal year 6 visit correlated most highly with outcomes, so
this was used in analysis.

Physical Fitness Variable: Physical ﬁtness level and physical activity
level are correlated in individuals but not identical. For this analysis, exercise
duration at baseline was used as the only measure of physical ﬁtness. Exercise
duration in subsequent years was not available in the data set. Exercise
duration was signiﬁcantly correlated with changes in blood pressure, but not with
changes in blood lipids, so was included only in models related to changes in
blood pressure.

Medical Conditions and Medications At baseline and at each annual

anniversary, all MRFIT subjects were given thorough medical examinations,

79

including an extensive medical history and inventory of medication use.
Conditions and medications which could affect weight, blood pressure, or
cholesterol were considered in this analysis, as described below.

I Conditions Affecting Weight: Many conditions and medications can
potentially cause increases or decreases in weight. Of the conditions and
medications documented in the MRFIT data set, the following were initially
identiﬁed as having potential effects on weight: diabetes, hyper- or
hypothyroidism, Cushing's disease, primary aldosteronism, nephritis/nephrosis,
congestive heart failure, alcoholism, drug addiction, depression, chronic
obstructive lung disease, tuberculosis, peptic ulcer, gall bladder disease,
cirrhosis and other liver diseases, anemia, Iymphadenopathy, reporting black or
tany stools, stroke, malignant neoplasms, reserpine, prazosin, cholestyramine,
nicotinic acid, insulin or oral hypoglycemic agents, steroids, digitalis,
spironolactone, hydralazine, dextrothyroxine, allopurinol, amphetamines and
other stimulants. Thirty-eight per cent of subjects had one of these conditions or
drugs at some point during the study. This was considered too much of the
sample to exclude from analysis, so the subjects with any of these conditions or
medications were identiﬁed in a variable RXDXWI' (0=no, 1=yes), and this
variable was included in regression models.

In ﬁnal analyses, only conditions judged to have the most dramatic effects
on weight served as the basis for exclusions, including cancer, Cushing's

disease, hyper- or hypothyroidism, and pheochromocytoma.

80

I Conditions Affecting Blood Pressure: Of the conditions and
medications documented in the MRFIT data set, the following were identiﬁed as
having potential effects on blood pressure: angina pectoris, stroke, peripheral
arterial occlusion, pulmonary embolism, thrombophlebitis, atrial ﬁbrillation, other
arrhythmias, pheochromocytoma, primary aldosteronism, alcoholism, and drug
addiction. In the SI group, 831 subjects (17%) had one or more of these
conditions. Subjects with any of these condition were identiﬁed in a variable
BPUPDX (0=no, 1=yes), and this variable was included in regression models.
Only conditions judged as having the most drastic effects on blood pressure
were used to exclude subjects, including angina, renal disease and primary
aldosteronism.

I Medications Affecting Blood Pressure: Eighty-nine per cent of SI
' subjects took antihypertensive drugs at some point during the study. This factor
was taken into account in ANCOVA analyses by blocking, in some models, on
use of antihypertensive drugs.

I Conditions Affecting Cholesterol: Conditions known to signiﬁcantly
affect cholesterol were identiﬁed, to allow control for these conditions in models
predicting change in blood lipids. The two conditions thus identiﬁed were
diabetes (deﬁned either by a diagnosis of diabetes, use of insulin or oral
hypoglycemic agents, or at least 2 fasting serum glucose concentrations greater
than 140 mgldl”1 during the ﬁnal 2 years of the trial) and liver disease.

I Use of Cholesterol-lowering Drugs: One hundred seventy-six

subjects (4% of the SI Group) were using cholesterol-lowering drugs at some

81

time during the trial. A variable was created to identify subjects who had taken
cholesterol-lowering drugs, and these subjects were excluded in some ANCOVA
models.

I Use of Cholesterol-raising Diuretics: Two of the diuretics used in the
MRF IT protocol for lowering blood pressure (chlorthalidone and hydrothiazide)
had a side effect of raising total cholesterol. Almost half of the SI subjects were
taking one or another of these drugs at some point throughout the study, and
were identiﬁed so this factor could be controlled in data analysis.

Mental Health Measures: Review of several published studies of mental
health status and subsequent mortality in the MRFIT cohort revealed that none
of the variables available in the data set were correlated with subsequent
mortality. Therefore, no mental health variables were included in analysis.

Variables Related to Weight Change Three variables were created to
describe the weight changes most likely to affect changes in blood lipids or
blood pressure: (1) net weight change from baseline to year 6, (2) weight change
in the ﬁnal 1-year interval, and (3) weight change in the ﬁnal 4-month interval
before the year 6 visit. When correlations between the outcome measures and
all of the created variables were examined, it was found that the one variable
most highly correlated with outcomes (aside from baseline values for each
outcome measure), was net weight change. (See Tables 35-38, Appendix 5.)
Net weight change had also been shown by Blair et al. to predict mortality in the

MRFIT population‘.

82

Creation of net weight change groups: To facilitate comparisons
between weight cycling's effects on risk factors and its effect on mortality,
subjects were divided into 3 groups by net weight change between baseline and
the year 6 annual visit. The weight loss group included subjects who had lost
more than 5% of baseline weight. The no change group included those whose
weight changed less than 5% from baseline, and the weight gain group gained
more than 5% of baseline weight. Close examination of relationships among
variables were showed that the three net weight change groups were
signiﬁcantly different from one another with respect to all outcome variables
(Table 9), most baseline variables (Table 6 ), and almost all the variables used
in analysis (Table 10 ). Differences in means were compared by Analysis of
Variance (ANOVA), using the Bonferroni Inequality test for post hoc
comparisons.

Because the net weight change groups were different from one another in
so many respects, the decision was made to perform all statistical tests on the
whole population and also separately for each net weight change group.
Statistical Approaches

Statistical analysis was done using the SAS System for Microsoft
Windows, Release 6.10,"56 licensed to Michigan State University. Two primary
statistical approaches were used to test the hypotheses - analysis of covariance
(ANCOVA) and multiple regression. ANCOVA allowed comparisons of mean
changes in blood pressure and blood lipids by each measure of weight cycling,

while controlling for variables thought to affect the outcomes. The output of

83

ANCOVA models allowed an easily-interpretable examination for any potential

dose-response effects from weight cycling.

Stepwise multiple regression was used in addition to ANCOVA to take
advantage of more extensive control for the variables thought to affect the
outcomes. With multiple regression, the effect of each predictor variable is
made more certain because the possibility of distorting influences from other
independent variables is removed.162

In addition to the two primary statistical approaches described above, a
preliminary analysis was done using ANOVA on small homogeneous weight
groups.

The alpha level chosen for determining statistical signiﬁcance for this
research was 0.05.

Exclusions
There were initially 6,428 subjects in the SI Group. Subjects were

excluded from analysis for the following reasons:

1. Inadequate data: As described earlier, 1,496 subjects without at least 16
of 19 possible weight measurements were excluded from analysis. This
left 4,932 subjects.

2. Medical conditions affecting weight: For all analyses, individuals with
conditions affecting weight drastically were excluded, including 235
subjects with cancer or "unexplained weight loss" (assumed to be cancer),
and 31 with hyper-or hypothyroidism. No subjects were found to have the

other two conditions marked for exclusion - Cushing's syndrome or

84

pheochromocytoma. Three of the subjects excluded for these reasons
had more than one condition; 4,669 subjects remained.

3. Medical conditions affecting blood pressure: For analyses related to
changes in blood pressure, subjects with conditions known to drastically
affect blood pressure were excluded, including 9 with renal disease, 284
with angina and 1 with primary aldosteronism. Some of the subjects
excluded for these reasons had more than one condition; 4,393 subjects
remained available for blood pressure-related analyses.

4. Medical conditions affecting blood lipids: For analyses related to changes
in total cholesterol, HDL and the ratio of total cholesterol to HDL, subjects
with conditions likely to drastically affect cholesterol were excluded,
including 346 diabetics, and 49 with cirrhosis and other liver diseases.
Some of the subjects excluded for these conditions had both conditions;
4,302 subjects remained for blood lipid-related analyses.

Sample Size I Power
Minimum differences in outcomes that would have to have been observed

to conclude that weight cyclers were different from non-cyclers were calculated,

using the formula for a 2-sample z-test (below), and solving for the difference

between means:

85

 

Y—Sr')

z: (1 2
2 2
S
4.2
"1 ”2

Based on this calculation, the sample size was large enough to detect at the
alpha = .05 level the following differences in outcomes between cyclers and
non-cyclers: 2.6 mgldl for total cholesterol, 0.73 mgldl for HDL, 0.12 for the ratio
of total cholesterol to HDL, and 0.76 mm of Hg. for blood pressure.
Deciding Which Possible Variables to Include in Statistical Analysis
Correlation with Outcome: Once all the possible covariates were
extracted from the data tapes, summarized, and re-coded as appropriate,
correlation analysis was done to see which of the variables available were
statistically related to each outcome measure. Correlations of each independent
variable with each outcome variable were checked not only in the entire non-
excluded population, but also in each net weight change subgroup. These
checks were done separately for the non-excluded population for blood lipid
analysis and for the non-excluded population for blood pressure analysis. Only
the variables signiﬁcantly correlated with each outcome variable in the
population group or subgroup being examined were included in ANCOVA or
regression models involving that outcome measure. See Tables 35-38 in
Appendix E for a summary of selected correlations. .
Eliminating Multicollinearity: The next stage in selection of variables

was to eliminate any possibility of multicollinearity clouding interpretation of the

86

results. Multicollinearity occurs when two independent variables are highly
correlated with one another and also with the outcome variable. When
multicollinearity is present, parameter estimates become unreliable.

In order to eliminate the possibility of multicollinearity, consideration of
the correlation of each variable with every other variable in the data set was
done, as follows. Once it was determined which variables were statistically
associated with each outcome measure, matrices were generated showing the
correlation of each potential covariate with every other covariate. Eight separate
matrices were generated and checked for possible multicollinearity - four based
on the non-excluded population for blood lipid models (for the entire population
and for each of the 3 net weight change subgroups) and the same four for the
non-excluded population for blood pressure analyses. Each Pearson correlation
coefﬁcient was checked. High correlations were found among three weight
variables - baseline weight, relative weight at baseline, and mean weight over
the course of the trial. Similarly, there were high correlations between total
minutes of leisure time physical activity and total minutes of heavy physical
activity, between baseline HDL and baseline ratio of total cholesterol to HDL,
and between mean number of cigarettes smoked and total number of cigarettes
smoked.

In each of the above instances, decisions had to made regarding which of
the related variables to include in models. The decisions were based on the
degree of correlation with the outcome and on which made the most theoretical

sense.

87

Consistency of Correlation Across Net Weight Change Subgroups:
For ANCOVA models only, there was one additional criterion for selection of
variables. ANCOVA models based on the entire non-excluded population which
blocked on net weight change group included only variables found to be
signiﬁcantly correlated with the outcome in all three net weight change groups.
This limitation was considered necessary because ANCOVA controls for
covariates by holding them at their mean value, with the assumption that the
regression of the outcome variable on the covariate is the same within the
treatment groups.”3

Tables 35-38 in Appendix 5 demonstrate the basis for selection of
variables for inclusion in various models. They show patterns of correlations of
selected variables with outcomes, for the entire non-excluded population as well
as for the each net weight change group. Variables consistently correlated with

outcomes across all net weight change subgroups are highlighted by shading.

88

Preliminary Analysis: ANOVA for Homogeneous Weight Groups

Because weight changes were observed to be so highly predictive of

changes in blood lipids and blood pressure, a preliminary statistical analysis was

done in an attempt to completely eliminate whatever effects baseline weight and

net weight change might be exerting on outcomes.

Three subgroups were identiﬁed, each of which was similar in terms of

baseline weight (within 5 pounds of median baseline weight for that group) and

net weight change (within 5 or 6 pounds of median weight change for that

group). In order to keep the groups homogeneous, only the approximately 50

subjects meeting the criteria for each group were included. Table 5 shows the

characteristics of the homogeneous weight groups.

Table 5 - Characteristics of Homogeneous Weight Groups

 

 

 

 

 

 

 

 

 

Non-excluded Non-excluded
Population for Population for
Net Blood Pressure Cholesterol
Weight Change Baseline Net Weight Analyses Analyses
Group Weight, lb Change, lb n n
Loss 189.5 to 194.5 -13 to -7 51 48
No Change 182.5 to 187.5 -2.5 to 2.5 52 49
Gain 182 to 187 3 to 9 53 51

 

For each of the three homogeneous subgroups, the SAS GLM procedure

was used to perform ANOVA to test whether cycling status, the number of weight

cycles, or the tertile of SEE contributed signiﬁcantly to the variability in changes

in total cholesterol and changes in blood pressure. For these analyses, the

number of weight cycles was re-deﬁned, combining those with 3 or more cycles

 

89

with those with 2 cycles, so that each cell would have at least 6 subjects. The
SAS MEANS procedure was used to calculate mean changes in each outcome
by each measure of weight cycling. To determine whether the mean outcomes
were signiﬁcantly different from one another, the post hoc multiple comparison
procedures used were the Scheffe and Bonferroni Inequality tests. In addition,
95% conﬁdence intervals were calculated for each mean change in outcome,
and are reported in Table 31.
Analysis of Covariance Models

The SAS GLM procedure was used to perform ANCOVA, which tested
whether weight cycling measures contributed signiﬁcantly to the variability in
changes in blood lipids and blood pressure, with adjustments for covariates.
The null hypothesis tested in ANCOVA is that the mean change in outcome by
each measure of weight cycling is the same. Covariates are the continuous
variables which are measures of factors thought to exert independent inﬂuences
on the outcomes. The SAS LSM procedure was used to calculate least square
mean changes in each outcome by each measure of weight cycling. Least
square means are the expected values of means that would be expected for a
balanced design involving class variables with all the covariates at their mean
values.”6 To determine whether the least square mean outcomes were different
from one another, an alpha level of 0.05 was chosen. The 95% conﬁdence
intervals were calculated for each adjusted mean change in outcome, and are

presented in Tables 11-26.

90

For ANCOVA models, the weight cycling measures were class variables,
expressed as four different categorical variables - (1) cycling status (non-cycler
vs. cycler), (2) number of cycles, (3) tertile of 855, and (4) number I size of
cycles.

Models Using Entire Non-excluded Population: Despite known
differences among the net weight change groups, it was desirable to create
models based on the whole population, blocking on net weight change group, so
that the larger number of observations could maximize the power of the
statistical tests. However, because of the differences in characteristics among
the three weight change groups, only the few variables signiﬁcantly correlated
with the outcomes in all 3 subgroups were used in the models (See Tables 35-
38 in Appendix 5). Independent variables meeting the selection criteria for
inclusion as covariates for the models based on the entire non-excluded
populations are listed below.

I Net change in total cholesterol: baseline cholesterol, the mean ratio of
polyunsaturated to saturated fats (P:S), and the mean number of cigarettes/day
reported over the trial.

I Net change in HDL: baseline HDL, mean intake of caffeine per day, and mean
number of cigarettes/day. Mean intake of alcohol per day was also included in
the HDL models because of the known effect of alcohol on HDL.

I Net change in the ratio of total cholesterol to HDL: baseline plasma

cholesterol, baseline HDL, and mean caffeine intake. Note that mean number of

91

cigarettes/day did not meet the criteria for inclusion in the model for cholesterol
ratio, but was included to make the model consistent with other models.

I Net change in blood pressure: baseline blood pressure, and mean number of
cigarettes per day. Note that in the models for change in blood pressure, there
was a statistically signiﬁcant interaction between net weight change group and
each cycling variable, so only models based on weight change subgroups could
be evaluated.

Although race was not signiﬁcantly correlated with the outcome in either
the entire non-excluded population or any subgroup, it was initially included in
ANCOVA models to verify that the ﬁndings did not differ by race. There was no
interaction of race with any measure of weight cycling in any of the models of
blood lipids. In blood pressure models, however, signiﬁcant interactions with
race were noted in two models, necessitating separate analyses by race.

Models Using Net Weight Change Subgroups: In models based on the
net weight change subgroups, all variables signiﬁcantly correlated with change
in outcome within the particular weight subgroup were included. As can be seen
in Tables 35-38 in Appendix 5, each weight group had a somewhat different
group of variables which were signiﬁcantly correlated with the outcomes, so
each model was somewhat different. Tables 11-26 clearly specify which
variables were included in the original models, and which were found to
contribute signiﬁcantly to the model. Only those which contributed signiﬁcantly
to the model were included in the ﬁnal versions where adjusted mean outcomes

were calculated.

92

Summary of Primary ANCOVA Models: In summary, for each of the four
outcomes being examined in this research, there were 4 basic models .
developed - one for each population group. Each of the 4 basic models included
the same variables - the ones correlated with the outcome in that population
group. Each basic model was repeated four times, substituting a different
measure of weight cycling each time. Figure 1 summarizes the 16 resulting
primary ANCOVA models. Each of the 16 models so developed was re-

submitted with additional controls, as explained below.

 

 

 

 

 

 

Cycler vs. Number of Tertile of Number I Size
Population Non-Cycler Cycles 855 of Cycles
All Model 1A Model 1 8 Model 1C Model 1 D
Non-excluded
Weight Loss Model 2A Model 28 Model 20 Model 2D
Group
Weight No Model 3A Model 38 Model 3C Model 30
Change Group
Weight Gain Model 4A Model 48 Model 4C Model 40
Group

 

 

 

 

 

 

Figure 1 - Summary of the Sixteen Primary ANCOVA Models Submitted for
Each of the Four Outcomes of the Research

Additional Controls for ANCOVA Models

Some of the variables reﬂective of factors expected to affect changes in
blood lipids and blood pressure were categorical variables which could not be
included in initial ANCOVA models because inclusion would have resulted in
subgroups with no subjects. For example, if blocking were done on net weight

group (3 levels), number of weight cycles (4 levels), race (2 levels), and smoking

93

status (3 levels), SAS would have computed comparisons among 72 subgroups,
some of which would have no subjects within them. In general, it was not
possible to block on more than 3 class variables in ANCOVA models.

In order to incorporate into the analysis the information contained in
theoretically important categorical variables, some ANCOVA models were
repeated with different categorical variables when race was shown to make
insigniﬁcant contributions to the explanatory power of the models. These
additional controls are described below.

I Additional Controls for Smoking: Number of cigarettes per day was
not found to contribute signiﬁcantly to total cholesterol Models 1A - 1D. To be
sure that smoking was not somehow confounding the results, each of these 4
models was repeated, stratifying on only the cycling variable and on smoking
status during the trial.

I Additional Controls for Exercise: Several models for total
cholesterol, blood pressure, and HDL in which the continuous variables
reﬂective of physical activity or ﬁtness did not contribute signiﬁcantly to the
model were repeated, blocking on exercise opinion at year 6 or quintile of
physical activity (whichever of the two was most highly correlated with the
outcome in the population subgroup being examined).

I Additional Controls for Use of Diuretics Which Raised Cholesterol:
All ANCOVA models for total cholesterol and HDL were repeated blocking on

use of lipid-raising diuretics during the ﬁnal two years of the study.

94

I Additional Controls for Use of Cholesterol-lowering Drugs: One
hundred seventy-six subjects (4%) of the SI Group were using cholesterol-
lowering drugs at some time during the trial. It was not possible to control for
this variable by blocking on it because the small number of subjects resulted in
empty cells. To rule out the possibility that these drugs were confounding the
analysis, all total cholesterol and HDL models were repeated, excluding from
analysis all subjects who had taken a cholesterol-lowering drug at any time
during the trial.

I Additional Controls for Use of Antihypertensive Drugs: The vast
majority of hypertensive subjects (83%) were prescribed antihypertensive drugs.
To rule out the possibility that antihypertensive drug use was obscuring an effect
of weight cycling, all blood pressure models were repeated, blocking on use of
these drugs.

I Additional Controls for Baseline BMI: Blair et al.1 found that weight
cycling was associated with increased mortality only in the MRF IT participants
who were at normal weight or who were moderately overweight. For men who
were the heaviest at baseline, weight cycling was not associated with increased
mortality. To rule out the possibility that weight cycling may be affecting risk
factors differently for heavy men, ANCOVA models for changes in total
cholesterol and blood pressure were repeated, blocking on tertile of baseline

95

Regression Analysis

Stepwise multiple regression models were developed for each outcome to
test the null hypothesis that the partial slope estimate for each measures of
weight cycling was 0. In SAS, stepwise regression begins with no variables in
the model. For each of the independent variables submitted in the model, SAS
calculates an F statistic that reﬂects that variable's contribution to the model if it
were to be included. Variables are added one by one to the model, starting with
the variable that results in the largest F statistic for the model. The selection
criterion for inclusion of a variable in a model is that the F statistic for a variable
have a p value <.50. (Note that variable selection is an exploratory rather than
conﬁrmatory process. The signiﬁcance level required for inclusion in the model
does not have the same connotation as the signiﬁcance level required for
rejecting the null hypothesis about the partial slope estimate. The p value
required for inclusion in the model is higher than the p value required for
rejecting the null hypothesis.) After a variable is added, the stepwise method
looks at the variables already included in the model and deletes any variable
that no longer produces an F statistic signiﬁcant at p<.5 level. Only after this
check is made and the necessary deletions accomplished is another variable
added to the model. The stepwise process ends when none of the variables
outside the model has an F statistic signiﬁcant at the p<.5 level, and every
variable in the model is signiﬁcant at the p<.5 level.156

In testing the hypothesis that the partial slope estimate for each measure

of weight cycling was 0, an alpha level of 0.05 was used. Ninety-ﬁve per cent

96

conﬁdence intervals were calculated for each partial slope estimate, and are
included in Tables 27-30.

Measures of Weight Cycling Used: For the regression analyses, SEE
was used as a continuous variable, and the number of cycles were expressed as
a series of 3 dummy variables:

I CYCDUM1: 0= no cycles, 1:1 or more cycles

I CYCDUM2: 0= 0 or 1 cycles, 1= 2 or more cycles

I CYCDUM3: 0= 0,1, or 2 cycles, 1= 3 or more cycles.

These dummy variables allow comparison of cyclers versus non-cyclers
(CYCDUMI ), and would also make it possible to detect a dose-response effect if
one were present. Only one cycling measure was included in each regression
model submitted.

Selection of Variables for the Models: For each or the four outcomes
being studied in this research, all variables found to be signiﬁcantly correlated
with the outcome in the whole population or any subgroup of the population were
included in the models for that outcome. To eliminate the possibility of
multicollinearity obscuring the interpretation of results, no two variables which
were very highly correlated with one another were both included in the same
model. Each model was forced to include baseline measurements for the
outcome variable being examined, net weight change, and the one cycling
variable being evaluated in that model. Tables 27 - 30 specify which variables

were submitted in each model.

97

Net Weight Change Groups: The regression models developed for

each of the four outcomes of the research (change in cholesterol, HDL, ratio of

total cholesterol to HDL, and blood pressure) were run for the whole non-

excluded population and then repeated for each net weight change group within

that population.

Summary of Regression Models: In summary, a total of 16 different

regression models was submitted for each outcome, as illustrated in Figure 2.

Each of the 16 models for each outcome included exactly the same independent

variables, with the exception of the measure of weight cycling.

 

 

 

 

 

 

 

 

 

 

Weight Cycling All Weight Loss No Weight Weight
Variable Non-excluded Group Change Group Gain Group
0 Cycles vs. Model 1 Model 5 Model 9 Model 13
1,2,3+ Cycles

0-1 Cycle vs. Model 2 Model 6 Model 10 Model 14
2-3+ Cycles

0,1,2 Cycles vs. Model 3 Model 7 Model 11 Model 15
3+ Cycles

855 Model 4 Model 8 Model 12 Model 16

 

Figure 2 - Summary of the Sixteen Different Regression Models Submitted

for Each of the Four Outcomes of the Research

 

RESULTS

Compared with non-cyclers, men with weight cycles did not experience
smaller improvements in either total cholesterol, HDL, the ratio of total
cholesterol to HDL or blood pressure, whether weight cycling was expressed in
terms of number of cycles, SEE or number I size of cycles. The lack of
association of weight cycling measures with CVD risk factors was observed
whether individuals gained weight, lost weight or experienced minimal weight
change.

Characteristics of the Study Population

Baseline Characteristics: Table 6 shows baseline characteristics of the
MRFIT Sl population. The study population was middle-aged (mean age 46.5),
overweight (mean baseline weight 189 pounds), with elevated risk of heart
disease reﬂected by high total cholesterol concentrations (mean 259 mgldl), low
HDL concentrations (mean 42.9 mgldl), high ratios of total cholesterol to HDL
(mean 6.08), and elevated diastolic blood pressure (mean 91.1mm Hg). The net
weight change groups were not uniform with respect to baseline characteristics.
The group which subsequently went on to lose at least 5% of baseline weight
was older, heavier, and at higher risk in terms of total cholesterol concentration,

ratio of total cholesterol to HDL and diastolic blood pressure than other groups.

98

99

Table 6 - Baseline Characteristics, All Non-excluded Subjects and by Net
Weight Change Groups, MRFIT SI Group

 

 

 

 

 

 

 

 

 

 

All Non- Weight No Weight Weight
Excluded Loss Change Gain
(n=4,302) (n=1,056) (n=2,446) (n=800)
Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.)
Variables
Baseline Age 46.5 (5.8) 471 (5,9)1 45.6 (5.8)1 45 3 6 (5.7)1
Baseline Total Cholesterol, 259 (36.6) 263 (34.9)1 258 (36.7)1 254 (37.8)1
mgldl
Baseline HDL, mgldl 42.9 (42.9) 42.7 (11.6) 42.8 (11.7) 43.3 (11.9)
Baseline Ratio of Total 6.08 (1.8) 6.2 (1 .8)1 6.1 (1.8) 5.9 (1 .8)1
Cholesterol to HDL
Baseline Blood Pressure, 91.1 (8.8) 92.4 (8.5)1 91.1 (8.7)1 89.6 (9.1 )1
mm Hg
Baseline Relative Weight 1.26 (.15) 1,3 (15,13 1.25 (.15)1 1.24 (.16)2
Baseline Weight, lb 189 (26.8) 195 (26.7)"2 188 (26.1)1 186 (27.5)2
Baseline BMI, kglmz 27.8 (3.4) 28.5 (3.3)‘-2 27.5 (3.2)1 27.3 (3.7)2

 

 

 

 

 

 

Note. Values in the same row with the same superscripts are signiﬁcantly
different from one another

Weight Cycling Patterns: Patterns of Weight cycling in the SI group are
summarized in Tables 7 and 8. In the entire SI group, 20% of subjects were
non-cyclers, 53% experienced one cycle, 22% experienced two cycles, and 5%
experienced three or more cycles over the 6-year period documented in this
study. Table 7 shows the distribution of weight cycling measures by baseline
BMI. The number of weight cycles was consistent across BMI groups and was
similar to that in the entire population. As expected, more of the heaviest men
and fewer of the leanest men fell into Tertile 3 of 555.

Table 8 shows that the net weight change groups were also very similar to

one another and to the entire population with respect to number of weight cycles,

100

with the exception that there were more non-cyclers in the weight gain group.
The patterns of distribution of subjects by tertile of SEE was similar to that in the
whole population with the exceptions that there were fewer non-cyclers in the
weight loss group and more non-cyclers in the weight gain group.

Table 7 - Weight Cycling Measures in All Non-excluded Subjects and by
BMI Tertile, MRFIT Sl Group

 

 

 

 

 

 

 

 

 

All Non-
Measure of Weight excluded BMI <26.1 26.1<BMI<28.7 BMI>23.7
Cycling n (96) n (96) n (96) n (96)
Number of Cycles
0 848 (20) 293 (21) 310 (22) 245 (17)
1 2298 (53) 731 (53) 775 (54) 792 (54)
2 941 (22) 296 (21) 292 (20) 353 (24)
3+ 215 (5) 63 (5) 61 (4) 91 (6)
Tertile of $55
Tertile 1: 1.4-3.8 lb 1422 (33) 689 (50) 543 (37) 199 (13)
Tertile 2: 3.9-5.3 lb 1413 (33) 453 (33) 507 (35) 453 (31)
Tertile 3: 5.4-20 lb 1467 (34) 241 (17) 397 (28) 829 (56)

 

Table 8 - Weight Cycling Measures in All Non-excluded Subjects and by Net
Weight Change Group, MRFIT SI Group

 

 

 

All Non- Weight Loss No Change Weight Gain

Measure of Weight excluded Group Group Group
Cycling n (96) n (96) n (96) n (96)
Number of Cycle

0 - 848 (20) 226 (21) 411 (17) 21 1 (26)

1 2298 (53) 541 (51) 1368 (56) 389 (49)

2 941 (22) 233 (22) 534 (22) 174 (22)

3+ 215 (5) 56 (5) 133 (5) 26 (3)
Tertile of SEE
Tertile 1: 1.4-3.8 lb 1422 (33) 235 (22) 835 (34) 313 (39)
Tertile 2: 3.9-5.3 lb 1413 (33) 367 (34) 832 (34) 239 (30)
Tertile 3: 5.4-20 lb 1467 (34) 454 (43) 779 (32) 248 (31)

 

 

 

 

 

 

 

101

Outcomes: Most members of the SI group experienced improvements in

CVD risk factors by the year 6 annual visit, presumably because of the intensive

interventions on those risk factors that occurred while they participated in the

MRFIT. Whereas the net weight change groups were similar to one another with

respect to weight cycling patterns, they were very different from one another with

respect to the outcomes of the study. As expected, those who lost weight

showed the greatest improvements in CVD risk factors, those who gained weight

showed the smallest improvements, and those whose weight changed minimally

showed an intermediate improvement. Table 9 shows mean outcomes for each

weight change group. Each net weight change group was found to be

signiﬁcantly different from all others with respect, to all outcome variables.

Table 9 - Outcomes, All Non-excluded Subjects and by Net Weight Change

Groups, MRFIT SI Group

 

 

 

 

 

 

, mm Hg

 

 

 

All Non- Weight No Weight Weight
Excluded Loss Change Gain
Outcomes (n=4,302) (n81.056) (n=2,446) (n=800)
(Year 6 ,, 3339"", Values) Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.)
Change in Total -23.2 (33.8) -32.51 (34.8) -22.11 (32.0) -14.81 (35.1)
Cholesterol, mgldl
Change in HDL, mgldl -1.2 (9.7) +1.51 (11.1 ) -1.61 (8.7) -3.6‘ (9.8)
Change in Ratio of Total -.21 (1.6) -.771 (1.8) -.141 (1.5) +341 (1.7)
Cholesterol to HDL
Change in Blood Pressure, -11 (9.9) -13.6‘ (9.9) -10.8‘ (9.6) -8.2‘ (10.3)

 

 

 

 

Note. Values in the same row with the same superscript are signiﬁcantly

different from one another.

102

Comparisons Among Net Weight Change Groups: Table 10 shows
that the net weight change groups were very different from one another with
respect to most of the independent variables used in ANCOVA and regression

models.

103

Table 10 - Mean Values for Selected Characteristics, All Non-excluded
Subjects and by Net Weight Change Groups, MRFIT Sl Group

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

All Non- Weight No Weight Weight
Variables Excluded Loss Change Gain

(n=4,302) (n=1,056) (n=2,446) (n=8oo)

Mean (s.d.) Mean (s.d.) Mean (s.d.) Mean (s.d.)
Net Weight Change, lb -15 (13.4) -177 (8.9)1 -.64 (52)1 17.4 (8.7)1
:Igleight Change, Final 4 Months, .25 (6.0) -39 (8.1)1 -2.4 (5.4)1 -.7 (8.9)1
X‘Veight Change, Final 12 Months, 0.6 (3.9) -.92 (4.2)1 .69 (3.3)1 2.5 (4.8)‘
SEE 5.1 (2.2) 5.8 (2.7)1 4.7 (2.0)1 5.3 (2.1)1
Alcohol Intake, glday 19 (25) 18 (25)1 19 (25)2 22 (28)"2
Caffeine Intake, mg/day 552 (407) 528 (385)1 548 (400)2 596 (449)13
Calcium Intake, mg/day 628 (307) 644 (299) 618 (307) 635 (318)
Cholesterol Intake, mglday 243 (138) 224 (129.9)1 243 (1:133)‘I 269 (159)1
% Calories from Fat 34 (7) 34 (7.1)1 34 (7.0) 35 (7.0)1
Water-soluble Fiber, grn/day 5 (2) 5 (2)1 5 (2.4)1 4 (2.3)1
Iron Intake, mg/day 14 (5) 14(5) 14(5) 14 (4.5)
st Ratio 1.1 (.6) 12 (.7)1 1.1 (.8)1 .9 (.5)1
Sodium Intake, mg/day 2797 (1138) 2795 (1121) 2780 (1118) 2853 (1223)
Vitamin E Intake, mg/day 10 (5.3) 11 (5.8)‘-2 10 (5.3)2 9 (4.5)1
Exercise Duration, Baseline, min 6.8 (1 .6) 8.7(1.7)1 6.9 (1.7)1 8.9 (1 .6)
Exercise Opinion Year 6 32 (1.0) 3.4 (1 .o)1 32 (1)1 2.9 (1)1
(1 =Iess, 5=more)
Heavy LTPA, min/day 34 (72.1) 33 (84.8) 36 (722) 29 (49.8)
Total LTPA, min/day 105 (120.3) 105 (138.2) 108 (119.1) 97 (99.7)
Cigarettes lday, Mean Over Trial 14 (15) 1o (14)1 13 (15)1 21 (18.4)1

 

 

 

Note. Values in the same row with the same superscripts are signiﬁcantly

different from one another.

 

1 04
ANCOVA Analyses

Tables 11-30 summarize the results of the 16 primary ANCOVA models of
the effect of weight cycling on each of the 4 CVD risk factors addressed in this
research. In ANCOVA models, the measures of weight cycling rarely contributed
signiﬁcantly to the models, and in cases where they did contribute signiﬁcantly,
no dose-response relationship was observed between the degree of weight
cycling and the degree of improvement in risk factors. Additional controls for
smoking status, exercise, use of cholesterol-raising diuretics, use of cholesterol-
lowering drugs and use of anti-hypertensive drugs did not, in any instance, alter

the results of the primary models.

105

Table 11 - Mean Changes (and 95% Conﬁdence Intervals) in Total
Cholesterol by Cycling Status, MRFIT Sl Group

 

 

 

 

 

Adjusted for race, age, baseline cholesterol,
relative wt. at baseline, water-soluble ﬁber intake,
P:S ratio, exercise duration, cigarettes/day

 

 

(-19.3 to 40.3)

 

Population Subgroup! Non-Cycler Cycler
Variables Included in Models '

All Non-excluded (n=4,096) -21.8 -23.5

Adjusted for net wt. change group, race, baseline (-23.9 to -19.7) (-24.7 to -22.3)
cholesterol, P:S ratio, cigarettes/day

Weight Loss (n=961) -32.7 -32.6

Adjusted for. race, baseline cholesterol, net wt. (-36.8 to -28.6) (-34.7 to -30.5)
change, baseline wt., calcium intake, cholesterol

intake, water-soluble ﬁber intake, iron intake, P:S

ratio, vitamin E intake, total minutes of leisure time

physical activity, Cigarettes/day

No Weight Change (n=2,123) -19.5 -22.7

Adjusted for. race, age, baseline cholesterol, net (-20.5 to -18.5) (-24 to -21.4)
wt. change, ﬁnal 4-month wt. change, cholesterol

intake, P:S ratio, vitamin E intake, cigarettes/day

Weight Gain( n=763) -14.8 -14.7

(-17.3 to -12.1)

 

' Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

 

106

Table 12 - Mean Changes (and 95% Conﬁdence Intervals) in Total
Cholesterol by Number of Cycles, MRFIT SI Group

 

 

 

 

 

 

 

 

 

 

Population Subgroup!

Variables Included in Models ' 0 Cycles 1 Cycle 2 Cycles 3+ Cycles
All Non-excluded (n=4,098) -21.8 ‘ -234 2 -23.4 3 -29.4 1.2.:
Adjusted for net wt. change group, (-23.9 to (-245 to (-25.4 to (-33.5 to
race, baseline cholesterol, P:S ratio, 49-7) -21.7) '21-4) '25-3)
cigarettes/day

Weight Loss (n=985) -32.7 -32.5 -33.3 -31.2
Adjusted for. race, baseline (-36.8 to (-35 to (-37.3 to (-39.4 to
cholesterol, net wt. change, baseline 48-6) '30) '29-3) '23)

wt., calcium intake, cholesterol intake,

water-soluble ﬁber intake, iron intake,

P:S ratio, vitamin E intake, total

minutes of leisure time physical

activity, Cigarettes/day

No Weight Change (n=2,184) -19.5 ‘ -22.1 2 -224 3 -31.2 ”'3
Adjusted fon race, age, baseline (-22.5 to (23.7 to (-25 to (-36.4 to
cholesterol, net wt. change, ﬁnal 4- '15-5) '20-5) '19-8) ‘25)
month wt. change, cholesterol intake,

P:S ratio, vitamin E intake,

Cigarettes/day

Weight Gain ( n=763) -14.8 -15.4 -13.2 -13.8
Adjusted for race, age, baseline (-19.3 to (-18.7 to (-18.1 to (-26.4 to
cholesterol, relative wt. at baseline, 403) '12-1) '3-3) 42)
water-soluble ﬁber intake, P:S ratio,

exercise duration, cigarettes/day

 

 

Note. Values in the same row with the same superscript are signiﬁcantly different
from one another.

' Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

107

Table 13 - Mean Changes (and 95% Conﬁdence Intervals) in Total
Cholesterol by Tertile of SEE, MRFIT SI Group

 

 

 

 

 

 

 

 

 

Population Subgroup! $55 $55 $55
Variables Included in Models ' Tertile 1 Tertile 2 Tertile 3
All Non-excluded (n=4,096) -21.3 ‘ -23.5 -244 1
Adjusted for net wt. change group, race, baseline (-23.1 to (-25.1 to (-25 to
cholesterol, P:S ratio, cigarettes/day -19.5) -21.9) -223)
Weight Loss (n=985) -29.8 -34.2 -32.9
Adjusted for". race, baseline cholesterol, net wt. (~34.1 to (-37.6 to (-35.9 to
change, baseline wt., calcium intake, cholesterol -25-5) -30.8) -29.9)
intake, water-soluble ﬁber intake, iron intake, P:S

ratio, vitamin E intake, total minutes of leisure time

physical activity, cigarettes/day

No Weight Change (n=2,184) -205 1 -22.2 -24.4 1
Adjusted for: race, age, baseline cholesterol, net (-225 to (-24.3 to (-26.7 to
wt. change, ﬁnal 4-month wt. change, cholesterol -18.7) '20-1) '22-1)
intake, P:S ratio, vitamin E intake, cigarettes/day

Weight Gain (n=763) -16.7 -14.3 -13.9
Adjusted for race, age, baseline cholesterol, (-21.4 to (-18.1 to (-17.6 to
relative wt. at baseline, water-soluble ﬁber intake, -12) 40.5) -10.2)
P:S ratio, exercise duration, cigarettes/day

 

 

Note: Values in the same row with the same superscripts are signiﬁcantly
different from one another.

' Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

108

Table 14 - Mean Changes (and 95% Conﬁdence Intervals) in Total
Cholesterol by Number and Size‘I of Cycles, MRFIT SI Group

 

 

Population Subgroup!

Variables Included In 1 Small 1 Large ZSmall 2 Large 3+
Models ” 0 Cycle Cycle Cycle Cycles Cycles Cycles
All Non-excluded -21.81'2 -2153" 24.6"” -24.7 -224 6 39;;
(n=4,096) (23.9 to (23.4 to (-26.4 to (-27.6 to (-249 (o "
Adjusted for net wt. -19.7) -19.6) -22.8) -21.8) .193) (3213-2le

change group, race,
baseline cholesterol, P:S
ratio, cigarettes/day

Weight Loss (n=985) -32.6 -29.5 -34.2 -34.3 -32.7 -31.3
Adjusted for". race, (-36.7 to (-34.1 to (-37.6 to (-41 to (-37.7 to (-39.5 to
baseline cholesterol, net -28.5) -24.9) -30.8) -27.6) -27.7) -23.1)
wt. change, baseline wt.,
calcium intake,
cholesterol intake, water-
soluble ﬁber intake, iron
intake, P:S ratio, vitamin
E intake, total minutes of
leisure time physical
activity, cigarettes/day

No Weight Change -19.41 -20.62 .24.o‘-2 -22.9’ -21.8‘ .312 “~34
(n=2,184) (-22.4to (-22.7 to (-26.5 to -26.3 to (-25.8 to (-36.4 to
Adjusted for; race, age, -16.4) -18.5) -21.5) -19.5) -17.8) -26)
baseline cholesterol, net ~
wt. change, ﬁnal 4-month
wt. change, cholesterol
intake, P:S ratio, vitamin
E intake, Cigarettes/day

Weight Gain( n=763) -14.8 -16.4 -14.9 -17.2 -10.5 -13.7
Adjusted for race, age, (-19.3 to {-21.810 (-19.1 to (-24.9 to (-16.9 to (-26.3 to

baseline cholesterol, 403) ‘11) '10-7) '9-5) '4-1) '1-1)
relative wt. at baseline,
water-soluble ﬁber
intake, P:S ratio,
exercise duration,
cigarettes/day

Note: Values in the same row with the same superscript are signiﬁcantly different
from one another.

" Size of Cycle is deﬁned as small if the S55 is < 4.5 pounds, and large if the
$55 is 24.5 pounds.

° Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

 

 

 

 

 

 

 

 

 

 

 

 

109

Table 15 - Mean Changes (and 95% Conﬁdence Intervals) in HDL by
Cycling Status, MRFIT SI Group

 

Population Subgroup! Non-Cyclers Cyclers
Variables Included in Models '

All Non-excluded (n=4,243)
Adjusted for net wt. change group, baseline HDL,
caffeine intake, alcohol intake, Cigarettes/day

Weight Loss (n=1,046) ~ 1.7 1.4
Adjusted for age, baseline HDL, net wt. change, (.3 to 3.1) (0.7 to 2.1)
alcohol intake, caffeine intake, cigarettes/day

No Weight Change (n=2,249) -1 -1.8
Adjusted for. baseline plasma cholesterol, baseline (-1.8 to -.2) (-2.2 to -1.4)
HDL, net wt. change, ﬁnal 4-month wt. change,
alcohol intake, caffeine intake, per cent of calories
from fat, vitamin E intake, Cigarettes/day

Weight Gain( n=787) -4.4 -3.3
Adjusted for. baseline plasma cholesterol, baseline (-5.6 to -3.2) (-4 to -2.6)
HDL, alcohol intake, caffeine intake, cholesterol
intake, per cent of calories from fat, cigarettes/day

 

b b

 

 

 

 

 

 

 

 

Note: Values in the same row with the same superscript are signiﬁcantly different
from one another.

' Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

” Adjusted means were not calculated because the interaction between number
of cycles and net wt. change group was statistically signiﬁcant.

110

Table 16 - Mean Changes (and 95% Conﬁdence Intervals) in HDL by

Number of Cycles, MRFIT SI Group

 

 

 

 

 

intake, caffeine intake, cholesterol
intake, per cent of calories from fat,
cigarettes/day

 

 

 

 

 

Population Subgroup! 0 Cycles 1 Cycle 2 Cycles 3+ Cycles
Variables Included in Models '

All Non-excluded (n=4,096) " b b "
Adjusted for net wt. change group,

baseline HDL, caffeine intake, alcohol

intake, cigarettes/day

Weight Loss (n=1,046) 1.8 0.93 l 1.8 4.6 1
Adjusted for age, baseline HDL, net (0.4 to (0 to (0.5 to (1.9 to
wt. change, alcohol intake, caffeine 3-2) 1-8) 3.1) 7.3)
intake, cigarettes/day

No Weight Change (n=2,249) -1.1 1 -1.7 -2.3 l -1.8
Adjusted for: baseline plasma (-1.9 to (-2.1 to (-3 to (-3.2 to
cholesterol, baseline HDL, net wt. --3) '1-3) '1-6) '-4)
change, ﬁnal 4-month wt. change,

alcohol intake, caffeine intake, per

cent of calories from fat, vitamin E

intake, cigarettes/day

Weight Gain( n=787) -4.4 -3.4 -3.3 -1.0
Adjusted for". baseline plasma (-5.5 to (-4.2 to (-4.6 to (-4.3 to
cholesterol, baseline HDL, alcohol ~32) -2.6) -2) 22)

 

Note: Values in the same row with the same superscript are signiﬁcantly different

from one another.

" Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for

the ﬁnal model in which adjusted means were calculated .

" Adjusted means were not calculated because the interaction between number
of cycles and net wt. Change group was statistically signiﬁcant.

 

111

Table 17 - Mean Changes (and 95% Conﬁdence Intervals) in HDL by Tertile
of $55, MRFIT SI Group

 

 

 

 

 

 

 

 

 

 

Population Subgroup! $55 $55 SEE
Variables Included in Models ' Tertile 1 Tertile 2 Tertile 3
All Non-excluded (n=4,243) -1 -1.4 -1.3
Adjusted for net wt. change group, baseline HDL, (-1.5 to (-1.9 to (-1.8 to
caffeine intake, alcohol intake, cigarettes/day --5) $9) --8)
Weight Loss (n=1,046) 1.9 1.2 1.6
Adjusted for age, baseline HDL, net wt. change, (0.5 to (0.1 to (0.6 to
alcohol intake, caffeine intake, cigarettes/day 33) 2-3) 2-5)
No Weight Change (n=2,249) -1.3 l ' -2.1 l -1.8
Adjusted for“. baseline plasma cholesterol, baseline (-1.8 to (-2.7 to (-2.4 to
HDL, net wt. change, ﬁnal 4-month wt. change, ~03) -1.5) -1.2)
alcohol intake, caffeine intake, per cent of calories

from fat, vitamin E intake, cigarettes/day

Weight Gain( n=787) -4.1 -2.6 l -4.1 ‘
Adjusted to: baseline plasma cholesterol, baseline (-5.3 to (-3.6 to (.5 to
HDL, alcohol intake, caffeine intake, cholesterol -2.9) -1.6) -3.2)
intake, per cent of calories from fat, cigarettes/day

 

Note: Values in the same row with the same superscript are signiﬁcantly different
from one another.

' Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

112

Table 18 - Mean Changes (and 95% Conﬁdence Intervals) in HDL by
Number and Size " of Cycles, MRFIT SI Group

 

Population Subgroup!
Variables Included in
Modelsb

Cycle

Small
Cycle

1
Large
Cycle

Small

Cycles

Large
Cycles

3+
Cycles

 

All Non-excluded (n=4,243)
Adjusted for net wt. change
group, baseline HDL,
caffeine intake, alcohol
intake, cigarettes/day

c

 

Weight Loss (n=1,046)
Adjusted for age, baseline
HDL, net wt. change, alcohol
intake, caffeine intake,
cigarettes/day

1.7 1
(0.3 to
3.1)

1.52
(0 to
3.0)

0.6 3
(-.5 to

1 .9
(0.2 to
3.6)

4.7 ‘33

(2.0 to
7.4)

 

No Weight Change
(n=2,249)

Adjusted for. baseline
plasma cholesterol, baseline
HDL, net wt. change, final 4-
month wt. change, alcohol
intake, caffeine intake, per
cent of calories from fat,
vitamin E intake,
cigarettes/day

-1 .o1
(-1.8 to
.02)

-1.5 3
(-2.1 to
-.09)

-1 .9
(-2.6 to
-1 .2)

-2.6 ‘13
(-3.7 to
-1.5)

-1 .6
(-3 to
-.2)

 

Weight Gain( n=787)
Adjusted for: baseline
plasma cholesterol, baseline

-4.4 3
(-5.6 to
-3.2)

-3.7
(-5.1 to
-2.3)

-3.2
(-4.3 to
-2.1)

-1.8 3
(-3.8 to
.02)

-4.4
(-6.0 to
-2.8)

-1.0
(-4.3 to
2.3)

HDL, alcohol intake, caffeine
intake, cholesterol intake, per
cent of calories from fat,
igarettes/day

 

 

 

 

 

 

 

 

 

Note: Values in the same row with the same superscript are signiﬁcantly different
from one another.

3 Size of cycle is deﬁned as small if the $55 is < 4.5 pounds, and large if the
855 is 24.5 pounds.

'3 Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

° Adjusted means were not calculated because the interaction between number
of cycles and net wt. change group was statistically signiﬁcant.

113

Table 19 - Mean Changes (and 95% Conﬁdence Intervals) in the Ratio of
Total Cholesterol to HDL by Cycling Status, MRFIT SI Group

 

 

 

Population Subgroup! Non-Cyclers Cyclers
Variables Included in Models '

All Non-excluded (n=4,243) -.18 -.19
Adjusted for net wt. change group, baseline HDL, (-.28 to -.08) (-.25 to -.13)
caffeine intake, alcohol intake, cigarettes/day

Weight Loss (n=1,046) -.8 -.8

Adjusted for baseline ratio of total Cholesterol to HDL, (-1 to -.6) (-.9 to -.7)

net wt. change, alcohol intake, caffeine intake, calcium
intake, iron intake, sodium intake, vitamin E intake,
cigarettes/day

No Weight Change (n=2,410) -.20 -0.13
Adjusted for". baseline ratio of total cholesterol to HDL, (-.34 to -.06) (-.19 to -.07)
net wt. change, alcohol intake, caffeine intake,
cigarettes/day

Weight Gain( n=787) .45 .30
Adjusted for: baseline ratio of total cholesterol to HDL, (.23 to .67) (.16 to .44)
caffeine intake, cholesterol intake, cigarettes/day

 

 

 

 

 

 

 

Note: Values in the same row with the same superscript are signiﬁcantly different
from one another.

3 Variables listed were submitted in the original ANCOVA model. These in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

114

Table 20 - Mean Changes (and 95% Conﬁdence Intervals) in the Ratio of
Total Cholesterol to HDL by Number of Cycles, MRFIT SI Group

 

 

 

 

 

 

Population Subgroup!

Variables included in Models 3 0 Cycles 1 Cycle 2 Cycles 3+ Cycles
All Non-excluded (n=4,243) -.18 -.19 -.15 -.37
Adjusted for net wt. change group, (-.28 to (-.25 to (-.25 to (-.57 to
baseline HDL, baseline plasma --03) --13) 7-05) --17)
cholesterol caffeine intake, alcohol

intake, cigarettes/d3 y

Weight Loss (n=1,046) -.8 -.7 -.8 -1.0
Adjusted for baseline ratio of total (-1.0 to (-.82 to (.1.0 to (-1.39 to
cholesterol to HDL, net wt. change, -0-6) '53) '5) '51)
alcohol intake, caffeine intake,

calcium intake, iron intake, sodium

intake, vitamin E intake,

cigarettes/day

No Weight Change (n=2,410) -.20 -.16 -.04 -.22
Adjusted for. baseline ratio of total (-.34 to (-.24 to (-.16 to (-.44 to 0)
cholesterol to HDL, net wt. change, 006) '03) 0-03)

alcohol intake, caffeine intake,

cigarettes/day

Weight Gain( n=787) .45 .30 .37 -.06
Adjusted for. baseline ratio of total (.2310 (.14 to (.13 to (-.71 to
cholesterol to HDL, alcohol intake, .67) .46) .61) .59)
caffeine intake, cholesterol intake,

cigarettes/day

 

 

 

 

 

 

Note: Values in the same row with the same superscript are signiﬁcantly different
from one another.

3 Variables listed were submitted in the original ANCOVA model. These in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

115

Table 21 - Mean Changes (and 95% Conﬁdence Intervals) In the Ratio of
Total Cholesterol to HDL by Tertile of $55, MRFIT SI Group

 

 

 

Population Subgroup! $55 $55 $55
Variables Included in Models ' Tertile 1 Tertile 2 Tertile 3
All Non-excluded (n=4,243) -.17 -.17 -.21
Adjusted for net wt. change group, baseline HDL, (-.25 to (-.25 to (-.29 to
caffeine intake, alcohol intake, cigarettes/day -.09) -.09) -.13)
Weight Loss (n=1,046) -.83 -.76 -.73
Adjusted for baseline ratio of total cholesterol to (-1.03 to (-.92 to (-.87 to
HDL, net wt. change, alcohol intake, caffeine intake, -.63) -.6) -.59)

calcium intake, iron intake, sodium intake, vitamin E
intake, cigarettes/day

 

 

No Weight Change (n=2,410) -.18 -.11 -.12
Adjusted fon baseline ratio of total cholesterol to (-.26 to (-.21 to (-.22 to
HDL, net wt. change, alcohol intake, caffeine intake, -.1) '01) '02)
cigarettes/day

Weight Gain ( n=787) .39 .31 .34
Adjusted for. baseline ratio of total cholesterol to (.17 to (.13 to (.16 to
HDL, alcohol intake, caffeine intake, cholesterol 51) -49) 52)

 

 

 

 

 

intake, cigarettes/day

 

Note: Values in the same row with the same superscript are signiﬁcantly different
from one another.

‘ Variables listed were submitted in the original ANCOVA model. These in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

116

Table 22 - Mean Changes (and 95% Confidence Intervals) in the Ratio of
Total Cholesterol to HDL by Number and Size 3 of Cycles, MRFIT SI Group

 

Population Subgroup!
Variables Included In
Models b

0 Cycle

1Smﬂl
Cycle

1 Large
Cycle

2 Small
Cycles

2 Large
Cycles

3+
Cycles

 

All Non-excluded
(n=4,243)

Adjusted for net wt.
change group, baseline
HDL, caffeine intake,
alcohol intake,
cigarettes/day

-.17
(-.27 to
-.07)

-.16
(-.26 to
-.06)

-.22
(-.3 to
-.14)

-.16
(-.3 to
-.02)

-.14
(-.28 to 0)

-.36
(-.56 to
-.16)

 

Weight Loss (n=1,046)
Adjusted for baseline
ratio of total cholesterol to
HDL, net wt. change,
alcohol intake, caffeine
intake, calcium intake, iron
intake, sodium intake,
vitamin E intake,
cigarettes/day

-.79
(-.99 to
-.59)

-.84
(-1.06
to -.62)

-.63
(-.79 to

-.47)

-.83
(-1.14
to -.52)

-.81

(-1.06 to -.56)

-1.0
(.1 .39 to
-.61)

 

No Weight Change
(n=2,41 0)

Adjusted fon baseline
ratio of total cholesterol to
HDL, net wt. change,
alcohol intake, caffeine
intake, cigarettes/day

-.2o 3
(-.34 to
-.06)

-.14 3
(-.24 to
-.04)

-.18 3
(-.28 to
-.08)

-.14 3
(-.3 to
.02)

.09 13.3.4.5

(-.09 to .27)

-.22 3
(.44 to

 

Weight Gain( n=787)
Adjusted for: baseline
ratio of total cholesterol to
HDL, alcohol intake,
caffeine intake,
cholesterol intake,
cr_'garettes/day

 

 

.45
(23to

.67)

 

.33
(Date
.58)

 

.28
(08to
.48)

 

.40
(03to

.77)

 

35

(.04 to .66)

 

-.06
(-.67 to
.55)

 

Note: Values in the same row with the same superscript are signiﬁcantly different

from one another.

" Size of Cycle is deﬁned as small if the SEE is < 4.5 pounds, and large if the

$55 is 24.5 pounds.

b Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for

the ﬁnal model in which adjusted means were calculated .

 

117

Table 23 - Mean Changes (and 95% Conﬁdence Intervals) in Diastolic
Blood Pressure by Cycling Status, MRFIT SI Group

 

 

 

Adjusted for race, baseline blood pressure, age,
net wt. change, alcohol intake, cigarettes/day

(44.5 to 42.5)

Population Subgroupl Non-Cyclers Cyclers
Variables Included in Models '
6 b
All Non-excluded (n=4,393)
Adjusted for net wt. change group, race, baseline
blood pressure, age, cigarettes/day
Weight Loss (n=1,059) -13.5 -13.6

(-14.1 to -13.1)

 

 

caffeine intake, cholesterol intake, per cent of
calories from fat, exercise duration, cigarettes/day

 

 

 

No Weight Change (n=2,438) -9.6 1 .10,9 1
Adjusted for. race, age, relative wt. at baseline, (-10.3 to-8.9) (-112 to -105)
baseline blood pressure, net wt. change, ﬁnal 12-

month wt. change, caffeine intake calcium intake,

cholesterol intake, iron intake, vitamin E intake,

exercise duration, cigarettes/day

Weight Gain( n=799) -8.5 -8.0

Adjusted for. race, age, baseline blood pressure, (-9.4 to -7.6) (-8.6 to -7.4)

 

Note: Values in the same row with the same superscript are signiﬁcantly different

from one another.

' Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .
” Adjusted means were not calculated because the interaction between number
of cycles and net weight change group was statistically signiﬁcant.

 

118

Table 24 - Mean Changes (and 95% Conﬁdence Intervals) in Diastolic

Blood Pressure by Number of Cycles, MRFIT SI Group

 

Population Subgroupl
Variables Included in Models '

0 Cycles

1 Cycle

2 Cycles

3+ Cycles

 

All Nonexcluded (n=4,393)
Adjusted for net wt. change group,
race, baseline blood pressure, age,
cigarettes/day

b

b

b

b

 

Weight Loss (n=1,059)

Adjusted for race, baseline blood
pressure, age, net wt. change, alcohol
intake, cigarettes/day

43.5
(44.5 to
42.5)

43.6
(44.2 to
43)

-14.1
(-15 to
-13.2)

-12.2
(-14.2 to
-10.2)

 

No Weight Change (n=2,438)
Adjusted for: race, quintile of total
minutes of physical activity, age,
relative wt. at baseline, baseline blood
pressure, net wt. change, ﬁnal 12-
month wt. change, caffeine intake,
calcium intake, cholesterol intake, iron
intake, vitamin E intake, exercise
duration, cigarettes/day

-9.6 ‘3
(-10.3 to
-8.9)

40.3 ‘
(-11.2 to
40.4)

41.4 2
(-12 to
40.3)

40.3
(41.5 to
-9.1)

 

Weight Gain ( n=745 White, 54 Black)
Adjusted for. race,° age, baseline
blood pressure, caffeine intake,
cholesterol intake, per cent of calories
from fat, exercise duration,
i'garetteslday

 

 

White:
-8.7
(-9.7 to
-7.7)

 

White:
~8.1
(-8.8 to
-7.4)

 

White:
-8.2
-9.3 to
-7.1)

 

White:
-6.6
(-9.3 to
-3.9)

 

Note: Values in the same row with the same superscript are signiﬁcantly different

from one another.

" Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute significantly to the model and were retained for

the ﬁnal model in which adjusted means were calculated .

" Adjusted means were not calculated because the interaction between number
of cycles and net weight change group was statistically signiﬁcant.
° Interaction between number of cycles and race necessitated separate analysis
by race; there were too few Black subjects to perform Analysis of Covariance.

 

119

Table 25 - Mean Changes (and 95% Conﬁdence Intervals) in Diastolic
Blood Pressure by Tertile of SEE, MRFIT SI Group

 

 

 

 

 

calories from fat, exercise duration, cigarettes/day

 

 

 

 

Population Subgroupl SEE SEE SEE
Variables Included in Models ‘ Tertile 1 Tertile 2 Tertile 3
All Non-excluded (n=4,393) b b ”
Adjusted for net wt. change group, race, baseline

blood pressure, age, cigarettes/day

Weight Loss (n=1,059) -14.1 -13.0 -13.8
Adjusted for race, baseline blood pressure, age, net (-15.1 to (-13.8 to (~14.5 to
wt. change, alcohol intake, cigarettes/day -‘| 3.1) -12.2) -13.1)
No Weight Change (n=2,500) 40.5 1 41.3 ‘3 -105 2
Adjusted fon race, age, relative wt. at baseline, (40,9 to (-11.8 to (.11 to
baseline blood pressure, net wt. change, ﬁnal 12- 40.1) '10-8) -10)
month wt. change, caffeine intake calcium intake,

cholesterol intake, iron intake, vitamin E intake,

exercise duration, cigarettes/day

Weight Gain( n=815) -8.3 -8.5 -7.8
Adjusted for. race, age, baseline blood pressure, (-9.3 to (-9.3 to (-8.6 to
caffeine intake, cholesterol intake, per cent of -7.3) -7.7) -7)

 

Note: Values in the same row with the same superscript are signiﬁcantly different

from one another .

' Variables listed were submitted in the original ANCOVA model. Those in

italics were found to contribute significantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .
b Adjusted means were not calculated because the interaction between number
of cycles and net weight change group was statistically signiﬁcant.

 

120

Table 26 - Mean Changes (and 95% Conﬁdence Intervals) in Diastolic Blood
Pressure by Number and Size' of Cycles, MRFIT SI Group

 

 

 

 

 

 

Population Subgroupl 1 1

Variables Included in Small Large 2 Small 2 Large 3+
Models b 0 Cycle Cycle Cycle Cycles Cycles Cycles
All Non-excluded (n=4,393) ° ° ° ° ° °
Adjusted for net wt. change

group, race, baseline blood

pressure, age, cigarettes/day

Weight Loss (n=1,059) -13.5 -13.6 -13.6 -1 3.9 -14.1 422
Adjusted for race, baseline (~14.5 to (~14.6 (-14.4 (-1 5.4 to (~15.3 to (-14.2 to
blood pressure, age, net wt. -12.5) to to -12.4) -12.9) -10.2)
change, alcohol intake, -1 2.6) -12.8)

cigarettes/day

No Weight Change White: White: White: White: White: White:
(n=2,248 White, 134 Black) -93 1.2-3 41.0 ‘ 40.5 ‘ 41.7 2" 41.1 ’ 40.4
Adjusted for: race,6 age, (.105 to (~11.5 (-11.1 (~12.5 to (~12.1 to {-11.7 to
relative wt. at baseline, .91) to to -10.9) -10.1) -9.1)
baseline blood pressure, net -10.5) -9.9)

wt. change, ﬁnal 12-month

wt. change, caffeine intake Black: Black: Black: Black: Black: Black:
calcium intake, cholesterol -95 i -13.4 1 -10.6 -12.3 -11 .3 -9.5
intake, iron intake, vitamin E (.113 to (-15.2 (42.8 (-14.8 to (44.2 to (-1 3.4 to
intake, exercise duration, -74) to to -9.8) -8.4) -5.6)
cigarettes/day -1 1 .6) -8.4)

Weight Gain (n-815) -8.5 -8.0 -8.0 -9.1 -7.7 -6.7
Adjusted for: race, age, (-9.4 (-9.1 to (—8.9 to (-10.7 to (~9.1 {-9.2
baseline blood pressure, to -6.9) -7.1) -7.5) to to
caffeine intake, cholesterol -7.6) 6.3) -4.2)
intake, per cent of calories

from fat, exercise duration,

cigarettes/day

 

 

 

 

 

 

 

 

Note: Values in the same row with the same superscript are signiﬁcantly different
from one another.

‘ Size of Cycle is deﬁned as small if the SEE is < 4.5 pounds and large if the
SEE is 24.5 pounds.

” Variables listed were submitted in the original ANCOVA model. Those in
italics were found to contribute signiﬁcantly to the model and were retained for
the ﬁnal model in which adjusted means were calculated .

° Adjusted means were not calculated because the interaction between number
of cycles and net weight change group was statistically signiﬁcant.

‘ Interaction between wt. cycling and race necessitated separate analysis by
race.

121

Regression Analyses

Tables 27-30 summarize the results of regression analyses. In no
regression model was the degree of weight cycling inversely associated with
improvements in CVD risk factors, as was hypothesized. The partial regression
coefﬁcient for the weight cycling measure was not signiﬁcantly different from 0 in
53 of the 64 models. In the 11 instances where the partial regression coefﬁcient
was statistically different from 0, it generally reﬂected small improvements in risk
factors associated with cycling which were not clinically signiﬁcant. For
example, in blood pressure regression Model 1, the partial regression coefficient
for Dummy Variable 1 (where 0 = no cycles and 1 =1, 2, or 3+ cycles) was -.56.
This is interpreted to mean that having 1 or more cycles is associated with a
reduction in blood pressure that is 0.56 mm Hg lower than the reduction

associated with no cycles.

122

Table 27 - Partial Regression Coefﬁcients (and 95% Confidence Intervals)
for Measures of Weight Cycling from Regression Models for Total
Cholesterol, MRFIT SI Group

 

 

 

 

 

 

 

 

 

 

Weight All Weight Loss No Weight Weight Gain

Cycling Non-excluded " Group Change Group Group

Variable (n=3,535) (n=875) (n=2,010) (n=650)

0 Cycles vs. -1.04 -0.5 -3.5 " .32

1,2,3+ Cycles (-3.55 to 1.55) (-5.4 to 4.4) (-6.83 to -.17) (-5.36 to 6)

0-1 Cycle vs. -1.7 -0.7 -2.3 0.5

2-3+ Cycles (-3.86 to .46) (-5.21 to (-5.04 to .44) (-5.18 to 6.18)
3.31)

0-2 Cycles vs. -6.2 b 0.3 -9.8 " 3.1

3+ Cycles (40.3 to 4.79) (-3.52 to (45.3 to -4.31) (42.19 to 13.4)
9.12)

SEE -0.7 ° 01 -0.8 ° -0.7

(4.09 to -.31) (4.03 to .33) (4.39 to -.21) (4.33 to .43)

 

Note. Stepwise multiple regression models submitted included age, race (dummy
variable), baseline cholesterol, net weight change, weight change in the ﬁnal 4-
month interval, relative weight at baseline, use of lipid-raising diuretics (dummy
variable), having a condition or medication affecting weight (dummy variable),
exercise duration at baseline, average total minutes per day of LTPA, mean
number of cigarettes per day and mean intake over years 1-3 of calcium,
cholesterol, water-soluble ﬁber, iron, P:S ratio and vitamin E.
' Excluded: Subjects missing 4 or more weight measurements and subjects with
the following conditions: cancer, unexplained weight loss, thyroid disease,
Cushing's disease, diabetes, cirrhosis or other liver disease
" Signiﬁcantly different from 0

 

123

Table 28 - Partial Regression Coefficients (and 95% Conﬁdence Intervals)

for Measures of Weight Cycling from Regression Models for HDL, MRFIT SI

 

 

 

 

 

Group
Weight All Weight Loss No Weight Weight Gain
Cycling Non-excluded' Group Change Group Group
Variable (n=4,136) (n=1,019) (n=2,344) (n=773)
0 Cycles vs. -0.20 -0.08 -0.53 0.92
1,2,3+ qrcles (-.87 to .47) -1.63 to 1.47) (-1.39 to .33) (-.43 to 2.27)
0-1 Cycle vs. 0.09 1.3 -0.62 0.88
2-3+ Cycles -.52 to .7) (-.11 to 2.71) (-1.35 to .11) (-.49 to 2.25)
0-2 Cycles vs. 0.87 3.4 " -0.52 3.01
3+ Cycles (-.35 to 2.09) (.66 to 6.14) (-1.93 to .89) (-.32 to 6.34)
SEE 0.16 b .09 .05 .26

(.02 to .3) (-.09 to .27) (-.13 to .23) (-.05 to .57)

 

 

 

 

 

 

Note. Stepwise multiple regression models submitted included race (dummy
variable), baseline HDL, net weight change, weight at baseline, use of lipid-
raising diuretics (dummy variable), having a condition or medication affecting
weight (dummy variable), mean number of cigarettes per day and mean intake
over years 1-3 of alcohol, caffeine, cholesterol, vitamin E, and per cent of

calories as fat.

' Excluded: Subjects missing 4 or more weight measurements and subjects with
the following conditions: cancer, unexplained weight loss, thyroid disease,
Cushing's disease, diabetes, cirrhosis or other liver disease
" Signiﬁcantly different from 0

 

124

Table 29 - Partial Regression Coefﬁcients (and 95% Conﬁdence Intervals)
for Measures of Weight Cycling from Regression Models for the Ratio of
Total Cholesterol to HDL, MRFIT SI Group

 

 

 

 

 

Weight All Weight Loss No Weight Weight Gain
Cycling Non-excluded ‘ Group Change Group Group
Variable (n=3,655) (n=918) (n=2,065) (n=672)
0 Cycles vs. 0.03 -0.07 0.05 -0.04
1,2,3+ Cycles (-.09 to .15) (-.32 to .18) (-.11 to .21) (-.31 to .23)
0-1 Cycle vs. 0.02 -0.12 0.09 0.002
2-3+ Cycles (-.1 to .14) (-.36 to .12) (-.05 to .23) (-.27 to .28)
0-2 Cycles vs. -0.15 -0.28 -.03 -0.33
3+ Cycles (-.37 to .07) (-.73 to .17) (-.28 to .22) (-1 to .34)
SEE -0.03 " -0.01 -0.003 , -0.06

(-.05 to -.01) (-.07 to .05) -.04 to .04) (-.12 to 0)

 

 

 

 

 

 

 

Note. Stepwise multiple regression models submitted included age, race (dummy
variable), baseline plasma cholesterol, baseline HDL, net weight change, weight
change in the ﬁnal 4-month interval, weight at baseline, use of lipid-raising
diuretics (dummy variable), having a condition or medication affecting weight
(dummy variable), exercise duration at baseline, average total minutes per day
of LTPA, mean number of cigarettes per day and mean intake over years 1-3 of
caffeine, calcium, cholesterol, sodium, water-soluble ﬁber, iron, P:S ratio and
vitamin E.

' Excluded: Subjects missing 4 or more weight measurements and subjects with
the following conditions: cancer, unexplained weight loss, thyroid disease,
Cushing's disease, diabetes, cirrhosis or other liver disease

” Signiﬁcantly different from 0

125

Table 30 - Partial Regression Coefﬁcients (and 95% Conﬁdence Intervals)
for Measures of Weight Cycling from Regression Models for Blood
Pressure, MRFIT SI Group

 

 

 

 

 

 

 

 

 

 

Weight All Weight Loss No Weight Weight Gain
Cycling Non-excluded ' Group Change Group Group
Measure (n=4,405) (n=985) (n=2,299) (n=761)
0 Cycles vs. -0.6 " -0.4 -1.3 b 0.4
1,2,3+ Cycles (-1.19 to -.01) (-1.58 to 0.78) (-2.6 to -.54) (-.76 to 1.56)
0-1 Cycle vs. -0.3 -0.5 -0.4 0.10
2-3+ Cycles (-.89 to .29) (-1.48 to 0.48) (-1.05 to .25) (-1.12 to 1.32)
0-2 Cycles vs. 0.9 1.8 0.7 1.3
3+ Cycles (-.08 to 1.88) (-.28 to 3.88) (-.55 to 1.95) (-1.44 to 4.04)
SEE -0.09 -0.30 " -0.01 0.12

(-.21 to .03) (-.52 to -.08) (-.17 to 0.15) (-.13 to .37)

 

 

Note. Stepwise multiple regression models submitted included age, race (dummy
variable), baseline blood pressure, net weight change, weight change in the ﬁnal
12-month interval, relative weight at baseline, use of antihypertensive drugs
(dummy variable), having a condition or medication affecting weight (dummy
variable), having a condition or medication affecting blood pressure (dummy
variable), exercise duration at baseline, average total minutes per day of LTPA,
mean number of cigarettes per day and mean intake over years 1-3 of calcium,
cholesterol, water-soluble ﬁber, iron, P:S ratio and vitamin E, alcohol, caffeine
and per cent of calories as fat.

‘ Excluded: Subjects missing 4 or more weight measurements, and subjects with
the following conditions: cancer, unexplained weight loss, thyroid disease, renal
disease, angina, primary aldosteronism and Cushing's disease.

b Signiﬁcantly different from 0.

Preliminary ANOVA on Homogeneous Groups

When the effects of baseline weight and net weight change were
essentially removed by restriction of the population to small groups of men
homogeneous with respect to these two characteristics, weight cycling did not
predict changes in total cholesterol or changes in blood pressure, whether

weight cycling was quantiﬁed by number of cycles, cycling status (non-cycler vs.

126

cycler) or tertile of SEE. Table 31 shows the adjusted mean outcomes by
cycling status only. The differences between cyclers and non-cyclers in mean

changes in cholesterol and blood pressure were not statistically signiﬁcant.

Table 31 - Adjusted‘ Mean Changes (and 95% Conﬁdence Intervals) in Total
Cholesterol and Blood Pressure, Homogeneous Weight Groups, MRFIT SI
Group

 

 

 

Outcome Non-cyclers Cyclers

Change in Total Cholesterol, mgldl -19.2 -25.6

(Year 6 - Baseline) (-31.3 to -7.1) (-31.6 to -19.6)
(n=28) (n=1 15)

Change in Blood Pressure, mm Hg -10.3 -10.6

(Year 6 - Baseline) (-13.8 to -6.8) (-12.3 to -8.9)
(n=29) (n=127)

 

 

 

 

 

' Adjusted for net weight change groups

Effects of Dropping Year 7 Measurements
Table 32 shows that those who were present for their year 7 physical
exam were not signiﬁcantly different from those who were absent in terms of
age, baseline weight, relative weight at baseline, baseline total cholesterol or net
change in total cholesterol as of year 6. However, those who were present at
year 7 had lower baseline blood pressure, showed less improvement in blood
pressure by year 6 and had lost considerably more weight by year 6 than those
who were absent at year 7.
Comparisons of Subjects Dropped from Analysis from Those Retained
Subjects dropped from analysis due to missing weight data were very

similar to those with "enough" data with respect to baseline characteristics.

127

However, as expected with less compliant subjects, they showed poorer weight
loss and smaller improvements in cholesterol and blood pressure than those
with "enough" data (See Table 33).

T-tests comparing the SI and UC Groups showed that the two groups
were not signiﬁcantly different with respect to these baseline characteristics:
age, weight, relative weight, blood pressure, total cholesterol, HDL or low

density lipoproteins (See Table 34 ).

128

Table 32 - Comparison of Subjects Present at Year 7 Annual Exam with
Those Absent, Selected Variables, MRFIT Population

 

 

 

 

 

 

 

 

 

 

 

 

Present Year 7 Absent Year 7
(n=5215) (n=6090)
Variables Mean (s.d.) Mean (s.d.)
Baseline Age, yr 46.3 (5.9) 46.19 (5.9)
Baseline Weight, lb 189.1 (27.1) 188.9 (28.6)
Relative Weight at Baseline 1.26 (1 .6) 1.26 (1 .6)
Baseline Diastolic Blood Pressure, mm 90.3 ‘ (8.8) 91.0 1 (8.6)
H9
Baseline Total Cholesterol, mgldl 257.1 (35.5) 258.1 (38.3)
Change in Blood Pressure -8.4 ‘ (9.9) -8.9 1 (10.3)
(Year 6 - Baseline), mm Hg
Change in Total Cholesterol -19.8 (32.6) -19.8 (33.6)
(Year 6 - Baseline), mgldl
. Weight Change (Year 6 - Baseline), lb —8.0 1 (13.2) +6.9 ‘ (13.7)

 

Note. Values in the same row with the same superscript are signiﬁcantly different

from one another.

 

129

Table 33 - Comparison of Subjects Dropped from Analysis Because of
Missing Weight Data with Those Who Had "Enough" Data, MRFIT SI Group

 

 

 

 

 

 

 

 

 

 

 

 

Dropped " Retained
Parameters Compared ("=1i495) (“34.932 )
Mean (s.d.) Mean (s.d.)
Baseline Age, yr 45 1 (5.3) 47 1 (6.2)
Baseline Weight, lb 190.2 (23.4) 139.0 (27.1 )
Baseline Relative Weight 1.27 (1.7) 1.26 (1.5)
Baseline Total Cholesterol, mgldl 256.9 (37.4) 258.1 (36.7)
Baseline Blood Pressure, mm Hg 39.4 1 (3.9) 91.1 1 (3.7)
Baseline HDL, mgldl 42.2 1 (12.6) 43.1 1 (11.3)
Net Change in Total Cholesterol, 47.3 1 (31.7) -23.3 1 (33.6)
mgldl (Year 6 - Baseline) (n=782) (n=4,698)
Net Change in Blood Pressure, -6.2 1 (10.8) -10.9 ‘ (9.9)
mm Hg (Year 6 - Baseline) (n=829) (n=4,927)
Wt. Change in First 4 Months, lb -.37 1 (3.7) 4.93 1 (9.1)
(n=1224) (n=4350)
Wt. Change in First Year, lb -2.9 1 (10.3) -4.1 1 (10.5)
(n=1,194) (n=4,916)

 

 

 

Note. Values in the same row with the same superscripts are signiﬁcantly
different from one another.
" Dropped for missing more than three 4-month interval weights

130

Table 34 - Comparison of Mean Values for Selected Baseline
Characteristics, MRFIT UC and SI Groups

 

 

 

 

 

 

 

 

 

 

 

Usual Care Special Intervention
(n=6436) (n=6422)
Baseline Characteristic Mean (s.d.) Mean (s.d.)
Age 46.1 (5.9) 46.29 (6.0)
Weight, lb. 139.0 (27) 139.3 (27.4)
Relative Weight 1.26 (.16) 1.26 (.16)
Blood Pressure, mm. Hg 90.7 (8.7) 90.7 (8.8)
Total Cholesterol, mgldl 257.5 (37.5) 257.8 (36.9)
HDL, mgldl 43.0 (11.8) 43.0 (12.0)
Low Density Lipoproteins, mgldl 163.0 (37.0) 162.5 (36.2)

 

Note. There were no signiﬁcant differences between groups.

 

DISCUSSION

This research was undertaken in response to the puzzling ﬁndings among
several population groups,1"3"""3'118 including the MRFIT population‘ studied in
this research, that weight cycling was associated with increased mortality. No
credible mechanism has been identiﬁed to explain this phenomenon. This
research proposes a mechanism and tests whether that mechanism was
operating in the Special Intervention Group of the MRFIT population. It was
hypothesized that individuals who weight cycled experienced smaller
improvements in the cardiovascular risk factors total Cholesterol, HDL, the ratio
of total cholesterol to HDL, and blood pressure - compared with those who did
not cycle.

The results of data analyses, shown in Tables 11—30, provided no support
for this hypothesis. In order for the hypothesis to have been supported, a strong
negative association between measures of weight cycling and improvements in
risk factors would have to have been seen, as well as a dose-response
relationship between the degree of weight cycling and degree of improvement in
risk. Neither of these conditions was met. This lack of association between

weight cycling measures and CVD risk factors is consistent with the few other

131

132

studies published to date addressing the effects of weight cycling on total
cholesterol and blood pressure.‘3'59

Issue: Are the Weight Cyclers in the Present Study the Same Men Who
Were Found to be at Greater Risk of Mortality?

If one is looking for a mechanism to explain increased mortality with
weight cycling, the population studied should be one in which increased
mortality with cycling has been documented. This was the case for the MRFIT
SI Group on which this study was based. The question arises, however, of
whether the exclusions deemed necessary for a study of blood lipids and blood
pressure resulted in purging from the data set the individuals who died during
the 2-5 year follow-up period for which mortality risks were computed by Blair et
al.1 A careful comparison of exclusions in the mortality study and the present
study is warranted.

Comparison of subjects excluded for medical conditions: The only
medical condition which served as a basis for exclusion in the mortality study
was cancer.1 In the present study, in addition to men with cancer, 31 men with
thyroid disease were also excluded because of the drastic effects hypo- and
hyperthyroidism exert on weight. It is certainly possible that the 31 SI men with
thyroid disease who were excluded were among the 98 SI who died during the 2-
5 year follow-up period of the mortality study, but there is no reason to believe
they constituted a disproportionate share of those 98 deaths.

No other men were completely excluded from the present study because

of any medical condition. In models developed for each outcome, men with

133

conditions drastically affecting that outcome were excluded. Leaving in those
individuals would make interpretation of results impossible, in that whatever
effect weight cycling may have had on the outcome would be overshadowed by
the effect of that condition. However, men excluded from blood pressure
analyses were not excluded from blood lipid analyses and men excluded from
blood lipid analyses were not excluded from blood pressure analyses.
Comparison of subjects excluded for missing data: In the mortality
study, approximately 264 men were excluded who were not alive at the seventh
anniversary of their randomization, and 499 men were excluded who were
missing more than 4 of their 8 annual weight measurements or missing the year
6 or year 7 weights. In the present study, 1,498 SI men missing more than 3 of
their 4-month interval weights or missing their year 6 weight were excluded. The
1,498 men thus eliminated would have included most of the 264 men who died
during the study and at least some of the 499 men eliminated in the mortality
study. It is estimated that approximately 800 men were eliminated from the
present study because of missing data who were retained in the mortality study.
The requirement, in the present study, for thorough documentation of
weights was considered essential for creating a reliable and valid measure of
weight cycling; however this stricter criterion may well have eliminated some of
the less-compliant men who could be expected to have a higher risk of mortality.
In the mortality study,1 28 men with any infeasible weight measurements
were dropped. In the current study, infeasible weight measurements were

identiﬁed and replaced with estimated weights. This discrepancy in approach

134

had very minor effect on the composition of the study population, because very
few individuals were found to have infeasible weights, and having an infeasible
weight would not be a reﬂection of either greater or lesser risk of mortality.

Summary of effects of exclusions: In summary, the current study
eliminated two categories of men who were included in the mortality study who
may have been among those who died during the follow-up period in the MRFIT
mortality study - 31 with thyroid disease and approximately 800 subjects who
were missing more than 3 weights. These exclusions were considered
necessary for the scientiﬁc integrity of the present study, but could have
eliminated from the population some of the 98 men who died during the follow-up
period.

Comparison of subjects categorized as weight cyclers: Another
source of discrepancy between the study population for the present research
and that of the mortality study1 was a difference in how subjects were assigned
"cycler" or "non-cyclel" status. For most mortality analyses, only 4-8 weights
taken at 12-month intervals were used when determining whether each subject
had experienced at least one weight cycle. Because they used less than half of
the available weight data to document cycling status, Blair et al. misclassiﬁed
about half of the subjects who experienced at least one weight cycle as "non-
cyclers." In the present study, the same definition of a weight cycle was used
(loss and regain of 5% of weight), but the more complete documentation of
weight in the present study allowed more precise identiﬁcation of weight cycling

status.

135

The continuous measure of weight cycling used in the mortality study was
also different from that used in the present study. Standard deviation of weight
was used in the mortality study, while the current study used standard error or
the estimate of the regression of weight on time (SEE). SEE was used in the
present research because it differentiated between weight change due to steady
loss or gain and weight change from cycling, and was therefore considered a
better measure of weight cycling.

It should be noted that it had been the intention of the researcher to
categorize subjects in exactly the same ways they had been categorized in the
mortality study by Blair et al.‘, so that the results of the present research would
be directly comparable to the mortality ﬁndings. The decision to depart from this
plan was made purposefully, because it was judged that the weight cycling
measures used in the mortality study were too imprecise to allow testing of the
hypothesis related to mechanisms.

Signiﬁcance of the differences in methodology: Because of the
differences in exclusions and in categorization of men by cycling status, there is
no assurance that the subjects classiﬁed as weight cyclers in the present study
were identical with the subjects classiﬁed as cyclers in the mortality study.
Although it would have been ideal if the study populations were identical, the
discrepancy does not invalidate the present study. The question being
addressed in the present study is whether weight cycling had adverse effects on
blood pressure and blood lipids for the MRFIT SI group. No adverse effect of

weight cycling was observed for any outcome, in any subset of the population.

136

In fact, if any effect was found, it was a very slight positive association between
degree of cycling and improvements in outcomes. There is no basis for
believing that weight cycling had a completely opposite effect on the 831
subjects dropped from analysis, or that the magnitude of the opposite effect was
so great that it accounted for most of the deaths observed in the mortality study
by Blair et al.1

Implications for future research: Although the results of the present
study contribute to the understanding of the effects of weight cycling on the CVD
risk factors examined, it would be most worthwhile to obtain the mortality data for
the population, and determine whether the same elevated risk of mortality found
in weight cyclers would still be observed using the more precise deﬁnitions of
weight cycling now available.

Does Weight Cycling Have Different Effects on Heavy Men than on Leaner
Men?

The association Blair and co-workers found between increased mortality
and weight cycling was found only in men in the lower range of the weight
distribution, but was not found in the heaviest men.1 The question of whether
weight cycling could be hamiful for moderately heavy men but not harmful or
somehow beneﬁcial for obese men is an important one, because heavier men
are more likely to diet and are therefore much more likely to be weight cyclers.
This was conﬁrmed in the current study; Table 7 shows that the heaviest men

had similar numbers of episodes of weight cycling as their leaner counterparts,

137

but that the magnitude of the weight variability (SEE) was greatest in the
heaviest men.

In this study, when men were Classiﬁed by tertile of baseline BMI, there
was no signiﬁcant interaction between weight cycling and baseline BMI (data not
shown). In other words, the effects of weight cycling on blood pressure and
cholesterol were no different for the heavier men than for the slimmer men.
Greater magnitude of weight cycling was not associated with either better or
worse Changes in risk factors for men at any weight with one exception. That
one exception, illustrated in Figure 3, was that, among the heaviest men, the 91
men having 3 or more weight cycles had a mean decrease in total cholesterol of
12% compared to decreases of only 8-9% for the 1,390 men with 0-2 weight
cycles. This difference was statistically signiﬁcant.

It is unlikely that this isolated instance of improvement in a risk factor
associated with the most extreme weight cycling constitutes evidence that weight
cycling is beneﬁcial to heavier men for three reasons. First, there were no
signiﬁcant differences between men with no cycles and men with one cycle or
between men with one cycle and men with two cycles. In other words, there was
no dose-response relationship between degree of cycling and improvement in
total Cholesterol. Second, the better improvement in cholesterol for heavy men
with more weight cycling was not observed when weight cycling was deﬁned in
terms of SEE. It should be noted that SEE may be a better reﬂection of
magnitude of weight variability than is number of cycles, and that there was a

greater proportion of heavier men when compared to leaner men in the highest

138

SEE tertile. Third, when other Characteristics of the heaviest men were studied,
it was seen that the mean weight Change during the 4—month interval before the
ﬁnal cholesterol measurement was taken was a loss of 4.4 pounds, signiﬁcantly
greater than the weight Changes among those with 0, 1, or 2 cycles (ANOVA,
post hoc test Bonferroni Inequality). Such a weight loss could be responsible for

small improvements in cholesterol over the ﬁnal 4-month interval of the study.

139

 

BMI 06.1 26.1<BMQ8.7 BM >28.7

 

 

 

 

% Change in Serum Cholesterol

 

 

 

 

 

 

Figure 3 - Adjusted Mean Per Cent Change (and 95% Conﬁdence Interval) in
Total Cholesterol from Baseline to Year 6 , by Tertile of Baseline BMI and
Number of Weight Cycles, MRFIT SI Group

Could Weight Cycling be Interpreted as Being Beneﬁcial in Any Cases?

In the few models where signiﬁcant differences were found in outcomes
between men with differing numbers of cycles, it was occasionally found that
having more cycles was associated with better outcomes than was having fewer
cycles. It is important to raise the question of whether cycling may be beneﬁcial
in some instances. In two instances, regression models suggested that having
three or more weight cycles could be associated with improvements in risk
factors which were of a magnitude approaching clinical signiﬁcance.

I For total cholesterol regression Model 3, based on the entire non-excluded
population, the partial regression coefﬁcient for Dummy Variable 3 was -6.3,
which is interpreted to mean that having 3 or more cycles was associated with

an improvement in cholesterol that was 6.3 mgldl better than the improvement

140

associated with having 0, 1 or 2 cycles. Examination of the total cholesterol
regression results for the weight change subgroups shows that the improvement
in total cholesterol with cycling seen in the whole population is actually
accounted for by the improvement in total cholesterol which occurred in the no
weight change group (Model 3), as there was no relative improvement in either
the weight loss (Model 2) or weight gain (Model 4) subgroups. In the no weight
change subgroup, the magnitude of the partial regression coefﬁcient reﬂected an
even larger improvement in Cholesterol (-9.8 mgldl for those with 3+ cycles
compared to those with 0, 1 or 2 cycles).

I Similarly, in HDL regression Model 7, having 3 or more cycles compared to
having 0, 1 or 2 cycles was associated, only in the weight loss group, with
improvements in HDL of 3.4 mgldl.

The two instances described above where regression analyses showed a
greater improvement in total Cholesterol with 3 or more cycles were conﬁrmed in
Cholesterol ANCOVA Models 1A and 3A and in the HDL ANCOVA Model 28. In
these cases, the magnitude of the improvements associated with having three or
more weight cycles could be considered clinically signiﬁcant.

Figures 4 -7 graphically display the results of ANCOVA models based on
number of cycles to allow examination of overall patterns of changes in risk
factors with progressive degrees of weight cycling. The charts illustrate that
there is not evidence for a beneﬁcial effect of weight cycling. Although there are
instances where higher degrees of weight cycling are associated with larger

improvements in risk factors than lower degrees of cycling, there are also

141

instances where higher numbers of cycles are associated with smaller
improvements. The lack of consistency in patterns of change in risk factors is
even more obvious when weight cycling is expressed in terms of SEE tertile and
as number and size of cycles (data not presented graphically). In other words,
there is no dose-response relationship of progressively greater improvements
with increasing measures of cycling in any risk factor, despite careful structuring
of data analyses to allow detection of dose-response relationships should they
be present. .

Figure 4 shows a signiﬁcant improvement in total cholesterol for men with
3 or more cycles in the no weight Change group. The pattern of changes in that
particular subgroup of the population is consistent with the presence of a
threshold effect. If there were a threshold effect for weight cycling, this would
mean that weight cycling is beneﬁcial but only when the magnitude of the cycling
is great enough to produce the "beneﬁt." Careful examination of all the data,
however, make it Clear that there is no such threshold effect. If there were, it

would be observed in other subgroups and for other risk factors.

142

 

 

I 0 Cycles
I 1 Cycle
[:1 2 Cycles
(:3 3+ Cycles

 

 

 

 

Change In Serum Cholesterol, mgldl

 

 

 

Figure 4 - Adjusted Mean Changes in Total Cholesterol (and 95%
Conﬁdence Interval) by Net Weight Change Group and Number of Weight
Cycles, MRFIT SI Group

 

LOSS m G-tAfGE GAIN

 

I 0 Cycles
I1 O/cle
El 2 Cycles
3+ Cycles

 

 

 

Change In HDL, mgldl

 

 

 

 

Figure 5 - Adjusted Mean Changes in HDL (and 95% Conﬁdence Intervals)
by Net Weight Change Group and Number of Weight Cycles, MRFIT SI
Group

143

 

LOSS m CHArGE GAN

0.5

 

I 0 Cycles
I1 Cycle
CI 2 Wcles
3+ Wcles

 

 

 

 

Change in Ratio Total CholesteroI:HDL

 

-1.5

 

 

 

Figure 6 - Adjusted Mean Changes in the Ratio of Total Cholesterol to HDL
(and 95% Conﬁdence Interval) by Net Weight Change and Number of
Weight Cycles, MRFIT SI Group

 

 

I 0 Wales
.1 Wcle
I] 2 Wcles
a 3+ Cycles

 

 

 

 

Change in Blood Pressure, mm Hg

 

 

 

Figure 7 - Adjusted Mean Changes in Diastolic Blood Pressure (and 95%
Conﬁdence Intervals) by Net Weight Change Groups and Number of Weight
Cycles, MRFIT SI Group

144

Strengths of the Study

The present study has a number of strengths and unique aspects which
lend credibility to its ﬁndings. The data were collected prospectively on a very
large population, with state-of-the-art quality control procedures in place. Other
factors known or thought to affect Changes in cardiovascular risk factors were
documented and were accounted for in data analyses to a greater extent than in
other weight cycling studies.

The attention given to net weight change in every phase of data analysis
is a strong feature of this research. Weight loss and weight gain have been
shown repeatedly to predict Changes in CVD risk factors. In this population the
strongest predictor for change in risk factors other than baseline levels of the
risk factors was net weight change, and each net weight Change group was
different from every other one with respect to most variables used in analysis.
Verifying that the lack of effect of weight cycling on each risk factor examined
was observed in subjects whether they lost weight, gained weight, or remained
at the same weight adds to the generalizability of the ﬁndings. The external
validity of the ﬁnding that weight cycling is not associated with detrimental
effects on blood lipids or blood pressure is enhanced by the consistency of
ﬁndings across these three population subgroups.

The internal validity of the ﬁnding that weight cycling was not associated
with detrimental effects on blood lipids or blood pressure in the MRF IT SI
population is enhanced by the consistency of ﬁndings with two different

statistical approaches (Analysis of Covariance and Multiple Regression) and

145

with ﬁve different ways of measuring weight cycling (number of cycles, cycler vs.
non-cycler, number! size of cycles, SEE and various dummy variables in the
regression analysis).

The documentation of weight is more complete than in any other study of
weight cycling, with weight measured at 4-month intervals over 6 years. Other
studies of weight cycling have used as few as 3 weights to construct an index of
weight cycling, and intervals between weight measurements have ranged from a
minimum of 1 year to more than 5 years.

The measures of weight cycling used in this research are more reliable
and meet more of the criteria for validity than measures used in other studies:

I The reliability of the weight measurements upon which the documentation of
weight cycling was based was veriﬁed via calibration of scales, training of
personnel and testing for consistency between technicians.

I The weight measurements on which documentation of cycling is based are
close enough together and are numerous enough to detect patterns in weight
changes.

I Because the actual number of cycles are counted and the SEE is used, weight
changes due to steady losses or gains are differentiated from changes due to
cycling. This is not the case in other studies which used standard deviation of
weight or coefﬁcient of variability as the measure of weight cycling.

I By combining number of cycles and SEE in one measure of cycling, large
cycles are differentiated from small cycles, maximizing the possibility of

detecting dose-response effects should they be present.

146

The exclusion of subjects without a nearly-complete set of 4-month
interval weights resulted in fewer subjects for analysis, but increased conﬁdence
that the measures of weight change and weight cycling used for analysis reliably
reﬂected actual weight patterns in the study population. Dropping the UC group
should place no limitations on generalizability of the data, because the
randomization procedures employed in the study protocol assured that the SI
group and the UC group were essentially identical. This was veriﬁed by T tests
comparing the UC group with the SI group. The two groups were not found to be
signiﬁcantly different with respect to any baseline characteristic considered.

Another strength of this study of weight cycling is that the weight changes
which were documented are with little doubt the result of voluntary weight loss
efforts rather than illness. Lack of veriﬁcation that weight changes were
volitional has been a limitation of many studies of weight change and
morbidity/mortality. In the MRFIT SI population, intensive intervention on weight
that was part of the study protocol assured that there would be weight losses,
and the lack of long-term effectiveness of any weight loss intervention assured
that there would also be weight regains. Exclusion of individuals with the
conditions likely to cause weight loss, plus control in regression analyses for the
presence of conditions and medications associated with weight changes
effectively eliminated the effects of illness on weight change.

Limitations of the Study
Because of the differences in exclusions and in categorization of men by

cycling status, there is no assurance that the subjects classiﬁed as weight

147

cyclers in the present study were identical with the subjects classiﬁed as cyclers
and found to be at high risk of death in the mortality study by Blair et al.1
Although it would be ideal if the study populations had been identical, the
discrepancy does not invalidate the present study of effects of weight cycling on
risk factors.

The results of this study cannot be generalized to as great a degree as
would be desired. The population studied consisted only of men at high risk of
heart disease, who were probably somewhat healthier than average. There was
little ethnic diversity among the population. In the SI group, there were too few
minorities other than African Americans to allow separate analyses by ethnicity.
The 342 African Americans in the SI group constituted only 7% of the sample.
Although race was used as a blocking variable in all models, and no differences
were found in the effects of weight cycling in African Americans vs. Caucasians,
there is less conﬁdence in the results for African Americans than for Caucasians.
Given the high prevalence of hypertension in the African American community,
inability to generalize to African Americans is an important limitation of the study.
No conclusions can be drawn about the effects of weight cycling on CVD risk
factors for women.

The reliability of the blood lipid outcome measures is somewhat limited by
the fact that baseline and year 6 blood lipid values were each based on only one
measurement. Considering the fact that within-person ﬂuctuations of serum
cholesterol of over 20% have been reported for the majority of individuals in

whom this has been measured,“ 9° and the average change in serum

148

cholesterol found in the MRFIT SI group was only 8.4%, it would have been
highly appropriate to base decisions as to whether Cholesterol concentrations
have changed on the average of several measurement, with the samples taken
on different days?“ ‘35

However, this procedure was not included in the MRFIT protocol.
Baseline and subsequent values for cholesterol concentrations were each based
on one fasting blood sample. To obtain a more representative measure of
baseline cholesterol, averaging of the total cholesterol values at baseline and at
the ﬁrst 4-month visit was considered, but this approach was rejected because
the ﬁrst intensive interventions were initiated at the ﬁnal screening visit, so the
cholesterol concentration at the 4-month visit could not be considered baseline
data. Diurnal variations in cholesterol may not have been an issue, because the
blood samples were taken in the fasting state, presumably collected early in the
day. Month or season of the year may have been somewhat consistent because
all study participants were invited for data visits every 4 months, although strict
observance of anniversary dates was not completely enforceable, given the
many factors involved in scheduling clinic appointments for 12,000 individuals in
22 different centers.

Dropping SI members with more than 3 missing weight measurements
had some potential for introducing bias, since, as shown in Table 33, those
dropped were younger, at lower risk at baseline, and probably less compliant
than those retained. Level of compliance would be a crucial factor mediating the

effects of weight cycling on CVD risk factors. There is no reason to assume,

149

however, that the subjects missing more than 3 weights were the only subjects
who were less than perfect in their compliance with the MRF IT regimen (as
evidenced, for example; by the 800 who gained weight over the course of the
trial). Level of compliance was controlled for, Indirectly, by including nutrition,
smoking, and exercise variables in the models.

Although some mental health information was available in the data set, no
mental health variables were included in the analysis. The rationale for leaving
them out was that none of the variables available was associated with mortality
in this population. In retrospect, it would have been interesting to pursue
whether any mental health factors met the criteria for inclusion in ANCOVA or
regression models. This remains a possibility for future research.

Although the documentation of weight available for deﬁning weight cycles
is superior to that used in any other study of weight cycling, it should be noted
that the number of weight cycles may have been underestimated in some
individuals. Measurements at 4-month intervals made it possible to document
any weight cycles which occurred over a period of 8 months or longer. It is
biologically possible for an individual to lose and regain 5% of body weight in as
short a time interval as two months. “"57'158-‘59 Such rapid weight Changes
would have been possible but unlikely in the MRFIT SI population.

There are three important limitations in the measurement of weight cycling
in this study. One is the exclusion of any indicator of the duration of each weight
cycle. For example, a loss and regain of 20 pounds within a period of 8 months

was indistinguishable from an identical weight loss and gain that occurred over a

150

6- year period. A second limitation in the documentation of weight cycling is lack
of differentiation between cycles characterized by initial weight loss followed by
regain and cycles that begin with a weight gain followed by a loss. A cycle
resulting from loss of weight previously gained could have more positive health
implications than a cycle resulting from loss and regain. A third limitation in the
measurement of weight cycling is restriction of the deﬁnition of a weight cycle to
5% weight changes. A 5% change in weight was chosen to deﬁne a weight
cycle because of the increased mortality noted in this population with weight
cycles of that magnitude.‘ However, It would be interesting to know whether
cycles of 10% would have yielded different results. These limitations could
potentially be overcome using sophisticated pattern analysis techniques to
deﬁne weight cycling patterns; this would be a valuable avenue for future
research.

Control for nutrient intakes which may have affected outcomes was not as
strict as would have been desired. To statistically control for effects of nutrients
on blood Cholesterol changes, it would have been ideal to determine each
subject's change in intake of selected nutrients, because validated prediction
equations have been developed to predict change in blood cholesterol from
changes in dietary cholesterol, saturated fats and polyunsaturated fats.‘°“-“’7

Unfortunately, it was not possible to evaluate individual changes in
nutrient intake over time, because the data tapes received from NHLBI included
only 24-hour dietary recall data at baseline and at year 6; food frequency data

collected on the SI group were not included. A one-day intake of a nutrient is

151
not representative of "usual intake" for an individual.80 Because it was not
possible to calculate changes in nutrients, the decision was made to calculate
"typical" intakes over the trial. It is recognized that the average of three 24-hour
intakes of a nutrient, taken at 1-year intervals is not an ideal marker for food
consumption. This limitation may explain the low Pearson correlation
coefﬁcients observed between all nutrients and changes in blood lipids and
blood pressure.

Another factor that could have contributed to low correlations between
nutrient intakes and outcomes was incompleteness of the nutrient data bases
available at the time nutrient intake data was coded for the MRF IT study. The
University of Minnesota staff who were responsible for the data base placed
greatest emphasis on food composition related to fat intake, since the diet-heart
hypothesis at that time centered on the fat content of the diet. Only in recent
years have food composition data become available to allow the development of
more complete data bases for nutrients such as Vitamin E, folic acid, and dietary
ﬁben

An additional shortcoming in the control for nutrient intake was failure to
include the nutrients folic acid and Vitamin C, which have been recognized
recently as having potential roles in the etiology of heart disease!” ‘58

One measure of CVD risk which has recently been recognized as an
important predictor of CVD morbidity and mortality is waist to hip ratio. There is

very strong epidemiological evidence that body fat distribution is a more

powerful predictor for the morbidity and mortality associated with obesity than is

152

body weight.‘°°- m In studies where body fat distribution is considered along
with body weight, fat distribution correlated more strongly with cardiovascular
risk factors and mortality. Given the fact that individuals with a history of weight
cycling have been shown to have higher waist to hip ratios than those without
such a history,117 it is unfortunate that information about waist to hip ratio was
not included in the data set.

Low Correlations of Outcomes with Lifestyle Factors

A surprising observation was that many of the factors believed to affect
the CVD risk factors studied in this research that were documented in the data
set were not as highly correlated with the outcomes as had been expected. The
extent to which the documentation available reﬂected actual lifestyle practices
was not always as great as would have been desired.

The state of the art for documentation of physical activity patterns leaves
much to be desired. Average minutes of leisure time physical activity over an
entire year, based on one recall at the end of that year, requires complex
thinking and detailed memory retrieval on the part of subjects, and could be
expected to include inaccurate information. Additionally, the researchers
decision to use LTPA reported at year 1, based on the fact that patterns of LTPA
in the MRFIT population remained relatively constant from year 1 through year
6,“"0 could be called into question. Exercise patterns closer to the time of ﬁnal
measurement of each risk factor or changes in exercise patterns between
baseline and year 6 may have had more of an impact on changes in risk factors

than patterns during the ﬁrst year of the study. These factors may account for

153

the low correlations observed between physical activity variables and outcomes,
and may explain why the physical activity variable most correlated with
outcomes was the subjective measure of exercise, exercise opinion at year 6.

Exercise opinion at year 6 was included in several ANCOVA models and
in several regression models. In regression models, because it was a
categorical variable, it had to be converted to dummy variables in order to be
used. Using the dummy variable of greatest interest (0=much less exercise than
others, 1 = much more exercise than others) resulted in dropping from analysis
the majority of subjects with intermediate levels of physical activity, thus
reducing the power of the statistical tests. Even so, adding exercise opinion to
the regression models did not alter the results of the analysis.

It was a disappointment that the measure of actual physical ﬁtness
selected for analysis - the number of minutes each subject was able to continue
the graded exercise test (exercise duration) - was not available in the data set
received from the National Heart, Lung and Blood Institute except at baseline.
Exercise duration at year 6 and change in exercise duration from baseline to
year 6 would have been desirable measure of ﬁtness, and may have been more
highly correlated with outcomes.

It was surprising that the smoking measures used for this analysis
contributed very little to the variability in outcomes. For total cholesterol, mean
number of cigarettes per day was not signiﬁcant in any model. For HDL,
cigarettes per day were signiﬁcant only for the weight gain group, in which

smoking was associated with a small positive effect. For both blood pressure

154

and the ratio of HDL to total cholesterol, smoking was associated with small
improvements in outcomes. The magnitude of the positive effect was small. For
example in the blood pressure regression models, for each additional cigarette
smoked, blood pressure was reduced by .03 mm Hg. In the HDL regression
models, for every additional cigarette, weight gainers saw an average increase
in HDL of 0.04 mgldl.

These counter-intuitive observations are not completely inconsistent with
other studies of risk factor changes in the MRFIT SI population. It has
previously been reported by Caggiula et al.""1 that among MRF IT SI group
members, smokers experienced smaller reductions in cholesterol than did non-
smokers, but analysis of dietary intake patterns revealed that the diminished
Cholesterol response was largely due to poorer dietary compliance by smokers.
Gerace et al.” reported that change in smoking status by MRFIT SI Group
members was not independently associated with change in diastolic blood
pressure, although those who quit smoking did experience an increase in HDL.

Associations of smoking with outcomes may have been higher if
outcomes had been documented over shorter time periods, so that shorter-term
effects of smoking on outcomes could be observed. It should be noted that the
two measures of smoking used in analyses - mean number of cigarettes per day
over the entire trial and smoking status during the trial - were more highly
correlated with changes in blood pressure and cholesterol than either number of

cigarettes per day at the time of the ﬁnal visit or smoking status at the ﬁnal visit.

155

Importance of Net Weight Change in Predicting Outcomes

One of the clearest messages found in this data analysis is that the one
factor most highly correlated with changes in blood pressure and blood lipids
between baseline and year 6 was net weight change over the same period.
Table 9 illustrates this message very clearly - for each outcome, those who lost
weight experienced the greatest improvement, those who gained weight
experienced the smallest improvement (or the greatest deterioration), and the no
change group experienced intermediate changes. For each outcome, each net
weight Change group was signiﬁcantly different from every other net weight
change group. This is not a startling or new ﬁnding, but it adds to the credibility
of the data analysis by conﬁrming what has been found in so many other
studies.

It had been anticipated that weight Change in the ﬁnal year or weight
change in the ﬁnal 4-month interval of the intervention period would have a
higher correlation with outcomes than net weight change. This was found to be
true only in the group with minimal weight change, for whom these later weight
Changes were more predictive of outcome.

What had not been anticipated was the different patterns of correlations
between covariates and outcomes seen in the population as a whole compared
with each weight change group, illustrated in Tables 35-38 in Appendix E. For
example, Table 35 shows that 13 of 24 possible covariates were correlated with
Change in total cholesterol in the entire non-excluded population, 12 of the 24

were signiﬁcantly correlated in the weight loss group, while for the no weight

156

change group and the weight loss group, 8 and 9, respectively, of the 24 were
signiﬁcant. Of the variables signiﬁcantly correlated, only three variables
(highlighted by shading in Table 35) were signiﬁcantly correlated in the entire
population and all 3 subgroups. It could be surmised that the differences in
numbers of signiﬁcant correlations were artifacts due to different numbers of
subjects in each group, but this cannot be the whole explanation because the
group with the most subjects (no weight change group, with 2,446 subjects) had
fewer positive correlations than the weight loss group (1,056 subjects).

A possible interpretation of these differences in patterns of correlation is
that the positive effects of diet and exercise on elevated blood pressure and
blood lipids are enhanced by weight loss, or blunted by weight gain or failure to
lose weight. This is consistent with the ﬁnding of Caggiula et al.151 for the
MRFIT SI Group - that the magnitude of the difference in observed Cholesterol
response between the group losing more than 10 pounds and the group who
gained weight was greater than that predicted by the Keys and Hegsted
equations.‘°5- ‘67 This, again, is not a startling or new ﬁnding, but adds to the
credibility of the data analysis.

What Accounts for the Increased Mortality Associated with Weight
Cycling?

The results of this research do not support the hypothesis that the
increased mortality observed among weight cycling males at high risk of heart
disease is caused by deleterious effects on blood lipids or blood pressure. The

question remains - if weight cycling is causing increased mortality, by what

157

mechanisms could this be occurring? Most speculation as to possible
mechanisms revolves around negative consequences of either the weight loss or
the weight gain phase of the weight cycle.

Damage during weight loss: Perhaps weight cyclers sustain damage to
cardiac tissue during the weight loss phases of each cycle. It was shown in
investigations of deaths from very low calorie liquid protein diets in the 1970's
that rapid, extensive weight loss is associated with atrophy of cardiac tissue."1
Perhaps repeated bouts of weight loss cause cumulative damage to cardiac
muscle, eventually leading to lethal arrhythmias such as those seen in the liquid
protein deaths. Another way weight loss could cause cumulative harm was
suggested by popular nutrition writer of the 1950's, Adelle Davis,172 who
cautioned against extreme weight loss because environmental pollutants that
are potentially cancer-promoting or cardiotoxic such as DDT are stored in fat.
When fat stores are mobilized during weight loss, these toxins become more
concentrated in the remaining adipose tissue, and therefore achieve higher
concentrations in the blood. Data do not exist to conﬁrm this mechanism.

In a similar vein of thought, weight loss may be associated with increase
in some unrecognized risk factor for CVD, such as changes in platelet function,
increases in concentrations of arachidonic acid, or depletion of omega-3 fatty
acid reserves?” ‘7‘ It is known that weight loss is accompanied by some
demineralization of bones, thus increasing the potential for osteoporosis in older
women. Osteoporosis is sometimes called "the silent killer," because it leads to

hip fractures in older women which precipitate a chain of events leading to death

158

within a few months of the fracture.175 Some of the increases in all-cause
mortality observed in weight cycling women could be attributable to increased
fractures and subsequent complications.

A more direct link between the weight loss phase of weight cycling and
increased mortality risk from cycling could be the use of health-threatening
methods of weight loss, such as those documented by Zimmerman et al.“

Damage during weight gain or during periods of elevated body
weight: Others suggest that it is the weight gain phase that could be the culprit.
Keyes et al. speculated that irreversible atherogenesis occurs during periods of
weight gain which is not offset by beneﬁts incurred during weight loss.176 Cutter
conjectured that, if people who are heavier are more likely to weight cycle, the
increased mortality associated with weight cycling could be the result of
increased risk that was incurred while weight was elevated.177
Does Weight Cycling Really Cause Increased Mortality?

One explanation for the lack of association between risk factors for
mortality and measures of weight cycling is that, given the many limitations of
the published studies of weight cycling and mortality, it may be that the
increased mortality found to be associated with weight cycling is an artifact, and
that the association would not be found if some as yet unidentiﬁed factor were
identiﬁed and quantiﬁed. The most vocal proponents of this view point out that
none of the studies of weight cycling and mortality have been able to
differentiate with 100% certainty those whose weight cycles were the result of

voluntary dieting (with subsequent weight regain) from those whose weight

159

varied because of illness. If this were Clariﬁed, it is argued, no ill effects of
weight cycling would be found.

It is tempting to embrace this point of view, given the methodological
problems inherent to the study of weight cycling, and given the absence of a
documented mechanism whereby cycling could increase risk of dying. However,
the fact that increased mortality has been found in so many different populations
by different researchers using different deﬁnitions of weight cycling and different
approaches to controlling for illness makes it impossible to dismiss the
possibility that weight cycling may be detrimental to health.

As long as the situation remains as it is in the United States where 40% of
adult women, 20% of adult men, and ever-increasing numbers of Children are
dieting,8 it is critical that any health effects from weight cycling be clearly
identiﬁed.

Future Research Directions

As mentioned above, it is critical that the question of whether weight
cycling is detrimental to health be clariﬁed. Kuller and Wing178 have suggested
that secondary analysis of existing data sets will not be helpful in gaining
insights to this question. They recommend that future studies need to focus ﬁrst
on the reasons for weight loss and weight cycling, and then on the metabolic and
health consequences. Ongoing clinical and longitudinal trials should collect
data in a way that would permit elucidation of the reasons for weight loss and

weight cycling. Methods used to induce weight loss should also be documented,

160

inasmuch as use of health-threatening weight loss practices could make
independent contributions to mortality risk.

An aspect of weight cycling research which merits much more attention is
the relationship among psychological status, stress, weight cycling, and
mortality. A few studies have demonstrated that individuals with a history of
weight cycling showed more signs of psychological pathology than those without
such a history, independent of weight.179 These studies, along with the studies

112,113 raise

showing accelerated atherogenesis in socially-stressed primates,
interesting possibilities for identifying causal pathways. The relatively new
research ﬁeld of psychoneuroimmunology is building a body of evidence that
emotional states directly affect the immune system. Given cultural norms of
unrealistic slendemess and pervasive discrimination against ovenrreight
individuals, the mental health aspects of weight loss and regain and their
possible contributions to CVD and other causes of mortality should be examined
more closely. The MRFIT data set is not ideal for examining this issue, because
it includes only men and includes a limited range of psychosocial data.

The MRFIT data set, however, holds the potential for addressing several
other questions related to possible effects of weight change on health. It would
be most useful to look at the effects of one complete weight cycle on the CVD
risk factors examined in this study. Speciﬁcally, after loss and complete regain
of at least 5% of body weight, are an individual's blood lipids and blood pressure

the same, higher, or lower than before the weight cycle began? Are the same

patterns of changes in risk factors seen in a second and third weight cycle as in

161

the ﬁrst? Is the rate of change in CVD risk factors the same as the rate of
change in weight? Is the rate the same during weight loss as during weight
gain?

Another crucial weight issue - the question of how best to maintain weight
loss - could also be addressed with this data set. Conventional wisdom holds
that gradual weight loss is more likely to be associated with longer-term weight
maintenance than rapid weight loss. This belief could easily be tested with the
data already extracted from the MRFIT data base.

The ﬁnal area of future research suggested by the results of this study is
validation of the ﬁnding of increased mortality among weight cyclers identiﬁed in
the major studies of weight cycling, including the study by Blair and colleagues‘
of mortality in the MRFIT population. It would be most interesting to learn
whether the increased mortality seen with weight cycling found in other studies
would still be evident if the analyses were repeated using the more reliable and

valid measures of weight cycling developed for this research.

SUMMARY AND CONCLUSION

The hypothesis that men who weight cycled experienced smaller
improvements in blood lipids and blood pressure compared to men who did not
weight cycle was not supported by the ﬁndings of this research, whether weight
cycling was expressed in terms of number of cycles, SEE, or as a combination of
these. The lack of association of weight cycling measures with CVD risk factors
was observed whether individuals gained weight, lost weight, or experienced
minimal weight Change. If weight cycling caused increased mortality risk in the
MRFIT SI Group, it does not appear that the increased mortality was mediated
by effects on total serum cholesterol, HDL, the ratio of total cholesterol to HDL or
diastolic blood pressure. This ﬁnding could be generalized to a population of
Caucasian middle-aged males at high risk for heart disease who are taking
measures to reduce their risk factors. It cannot be generalized to younger men,

women, or minorities, on whom further research is warranted.

162

APPENDICES

 

0neeu=
RESEARCH
AND
GRADUATE
STUUHES

   
 
 
 
 
 

ZQNMMNﬂmn&WMQ
hﬂumWLMUWm
ﬂﬁﬂdOﬁ

MU§5NKI
FAX: 517/432-1171

rhsunmaeuwnh
nushmuuu0muy
ﬂuﬂnxnknn

revenue—aeaen

1€ﬁ3

APPENDIX A : UCRIHS Approval for Research

MICHIGAN STATE

 

U P1 I‘V

September

E I! S I 1' Y

15, 1995

TO: Karen Petersmarck
1557 Rillside
Okemos, Mi 48864

RE: IRBI: 93-C‘9
TITLE: EFFECT OP HEIGHT CYCLING ON CHOLESTEROL AND

BLOOD PRESSURE IN THE MULTIPLE-RISK FACTOR
INTERVENTION TRIAL POPULATION

arvxsrou REQUESTED: u/a
carecoay: 1-3
APPROVAL oars: 09/15/95

The University Committee on Research Involving Human Sub ects‘(UCRIBS)
review of this project is complete. I am pleased to adv se that the
rights and welfare of the human subjects appear to be adequately

protected
herefore,
above.

REVISIONS:

paosLsxs/
censors:

If we can
at (517)35

Sincerel ,

and methods to obtain informed consent are a ropriate.
the UCRIBS approved this project and any rev sions listed

UCRIHS approval is valid for one calendar year, beginning with
the approval date shown above. Investigators planning to
continue a project be and one year must use the green renewal
form (enclosed with t e original a roval letter or when a

pro ect is renewed) to seek u at certification. There is a
max of four such expedit renewals ssible. Investigators
wishing to continue a roject beyond tha time need to submit it
again or complete rev ew.

UCRIHS must review any changes in rocedures involving human
subjects, rior to in tiation of t e change. If this is done at
the time o renewal, please use the green renewal form. To
revise an a roved protocol at an 0 her time during the year
send your wr tten request to the CRIBS Chair, requesting rev se
approval and referencin the project's IRB I and title. Include
in your request a descr ption of the change and any rev

ins ruments, consent forms or advertisements that are applicable

Should either of the following arise during the course of the
work, investigators must noti UCRIHS promptly: ll) problems
(unexpected side effects comp aints, e c.) involv ng uman
subjects or 23 changes in the research environment or new
information n icating greater risk to the human sub ects than
existed when the protocol was previously reviewed an approved.

be of any future helgé lease do not hesitate to contact us

5-2130 or FAX (517)4 - 171.

\ir.——

avid B. Wright, 9 .0.

UCRIHS Chairr
Dzwzkaa/lcp

cc: Jenny Bond
Howard Teitelbaum

164

Appendix B: Response to Freedom of lnfonnation Request for MRFIT Data

1‘
.6 DEPARTMENT or HEALTH a HUMAN SERVICES Pyblic Health Service
gl
”D

d In".
‘4

National Institutes of Health
National Heart. Lung, and
Blood Institute

JUIY 2' 1992 863118568. MOMOM 20892

Karen Petersmarck, M.P.l-l., RD.
1557 Hillside Drive
Okemos, Michigan 48864-2319

Dear Ms. Petersmarck:

I am writing in response to your requ’eSt for seleCIed MRFIT data tapes and
documentation, to be used in your pursuit of a doctoral degree from Michigan
State University. Staff of our Biostatistics Research Branch have determined
that 8 tapes, covering baseline and tri-annual followup visits, will be required to
meet your need for all of the Special Intervention weight data over the course
of the trial. These tapes will also provide age and height: blood lipids, uric acid
and glucose levels; smoking history, leisure time physical activity, use of
Cholesterol lowering drugs, and some dietary data; and most but not all of the
data on blood pressureand antihypertensive drugs. Extensive dietary data are
contained on other tapes.

It will take 1-2 months to ﬁll your request, due to the need to remove certain
identiﬁers and to photocopy a lot of documentation. The cost of the data set
will be $50 per tape for a total of $400. Payment should be in the form of a
check make out to 'DHHSINIH". .

Our understanding with regard to release of these data to you, and the faculty
members you named, are as follows:

I the data will be treated and protected as confidential medical records, no
effort will be made to identify participating individuals, and contacts with
MRFIT centers about their data in this data set will only be made on
terms agreed upon with NHLBI, and

I you will not make the data set or portions of it available outside of your
research group at Michigan State University without our concurrence.

165

Letter to Ms. K. Petersmarck - Page 2
July 2, 1992

I would appreciate a note back from you agreeing to these terms. Please also
indicate the person or office that will be responsible for payment.

Finally, as you know, the publication of MRFIT results on behalf of the original
Research Group is under the direction of an Editorial Committee, chaired by Dr.
Marcus Kjelsberg at the University of Minnesota. On behalf of the Committee, I

would like to request a confidential courtesy copy of any manuscript that is
written as a result of the work you plan to undertake.

If you need any further information, you can contact me at (301) 496-2465, or
Dr. Margaret Wu or Ms. Barbara Geraci, at (301) 496-5905.

Sincerely,

ﬂag—3&3

Je ey utler, M.D., M.P.H.
Chief, Prevention and Demonstration
Research Branch
Division of Epidemiology and
Clinical Applications

  

cc: Dr. M. Wu
Ms. B. Geraci

166

APPENDIX C: Sample, SAS Program to Extract MRFIT Data from Magnetic
Tapes

cms tape rew;

LIBNAME DAT XPORT ’ANN1 TRN A';

cms ﬁledef tapein tap1 sl 2 (lrecl 2100 blksize 23100 recfm fb;
data dat.mrﬁtan1;

inﬁle tapein;

input

ID 3 3-13 WT1 83-86 BMI1 1820-1822 RWT1 1817-1819 BPTX1 399 BPTX21
422 BPG1 1946 BP1 74-76 BPRX1 414 DIUC1 1841—1843 DIUH1 1844-1846
CHOL1 907-911 CHOLA1 1784-1786 LIPRX1 375 TG1 913-917 SMOKE1 119
CIGS1 128-129 CIGD1 1832-1833 EXD1 1417-1420 EXO1 717 ALC1
1876-1878 DIETW1 722 DIETC1 724 MISS1 437 UWL1 559 GLC1 943-947
DATE1 148-153 DAYS1 1869-1871 UA1 919-923 HAP1 708 LE1 1899-1900
KNUT1 1901 -1902 COF1 870-871 TEA1 876-877 COLA1 882-883 WT14
1081-1084 BMI141967-1969 RWT141964-1966 BPG141963 BPTX141068
BPTX214 1080 BP14 1133-1135 BPRX14 1173 CHOL14 965-969 SMOKE14
1136 CIGS14 1144-1145 CIGD14 1974-1975 NUTAD14 1170 DATE14
1070-1075 DAYS14 1982-1984 TG14 971-975 UA14 977-981 GLC14 855-859
WI'18 1224-1227 BMI18 2018-2020 RWT18 2015-2017 BPG18 2014 BPTX18
1223 BP18 1276-1278 BPRX18 1316 CHOL18 1024-1028 SMOKE18 1279
CIGS18 1287-1288 CIGD18 2025-2026 NUTAD18 1313 DATE18 1017-1022
DAYS18 2033-2035 TG18 1030-1034 UA18 1036-1040 GLC18 1060-1064
MHSP1 554 MHTS1 556 MHBS1 557 MHBED1 563 RXA1 372 RXB1 373 RXC1
374 RXD1 375 RXFG1 376 RXH1 377 RX|1 378 RXJ1 379 RXK1 380 RXL1 381
RXM1 382 DXCA1 256 DXCB1 257 DXCC1 258 DXCD1 259 DXCE1 260
DXCF1 261 DXCG1 262 DXCH1 263 DXC|1 264 DXCJ1 265 DXCK1 266
DXCL1 267 DXMNA1 268 DXMNB1 269 DXMNC1 270 DXMNE1 272 DXDM1
273 DXEB1 274 DXEC1 275 DXED1 276 DXEE1 277 DXEF1 278 DXEG1 279
DWEH1 280 DXMDA1 283 DXMDB1 284 DXMDC1 285 DXMDD1 286
DXNDE1 287 DXMDF1 288 DXNA1 289 DXNB1 290 DXMSA1 291 DXNSB1 292
DXRA1 293 DXRB1 294 DXRC1 295 DXRD1 296 DXDA1 297 DXDB1 298
DXDC1 299 DXDD1 300 DXDE1 301 DXGA1 303 DXGB1 304 DXGC1 305
DXGD1 306 DXGE1 307 DXHA1 308 DXHC1 309;

run;

167

APPENDIX D: SAS Program for Counting Weight Cycles
Written by Jay McClellan, M.S.E.E.

* Read in the weight data from the SAS data set siclean5.sd2;
libname karen 'c:\data2';
options nocenter;
data one;
set karen.siclean5;
* Deﬁne the weight array;
array WT(19) WI'B WT04 WT08 WT1 WT14 Wl’18
WT2 WT24 WT28 WT3 WT34 WT38
WT4 WT44 WT48 WT5 WT54 WT58 WT6;
* Deﬁne the number of weights;
count = 19;
* Deﬁne the transition threshold amount;
thresh = WTMEAN * 0.05;
* Set peak 8. valley counts to zero;
peaks = 0; valleys = 0;
* Initial reference weights are the ﬁrst weight in the set;
hirefwt = WT(1);
Iorefwt = WT(1);
* Initial state is WAITING;
state = 0;
* Loop over all weights in the set (after the ﬁrst);
do sample=2 to count;
* Process the sample only if it's not a missing data point (.);
if WT(sample) ne . then do;
select (state);
1' If the state is WAITING, looking for a transition from start weight;
when (0) do;
* Set the high & low reference weights to the max & min so far;
Iorefwt = min(lorefwt, WT(sample));
hirefwt = max(hirefwt, WT(sample));
* Is the weight sufficiently above the low reference weight?;
if WT(sample) >= Iorefwt + thresh then
do;
* Yes - start CLIMBING;
state = 1;
end;
* Is the weight sufﬁciently below the high reference weight?;
if WT(sample) <= hirefwt - thresh then
do;

168

SAS Program to Count Weight Cycles, continued

* Yes - start FALLING;
state = -1;
end;
end;
* If the state is CLIMBING, found a positive excursion - subject is gaining
weight;
when (1) do;
* Set the reference weight to the maximum seen in this excursion;
hirefwt = max(hirefwt, WT(sample));
1' Is the current weight sufﬁciently below the reference weight?;
if WT(sample) <= hirefwt - thresh then
do;
* Yes - got a cycle, so increment the peak count;
peaks = peaks + 1;
* Set this as the new reference weight;
Iorefwt = WT(sample);
* Transition to the FALLING state;
state = -1;
end;
end;
* If the state is FALLING, found a negative excursion - subject is losing
weight;
when (-1) do; ,
* Set the reference weight to the minimum seen in this excursion;
Iorefwt = min(lorefwt, WT(sample));
* Is the current weight sufﬁciently above the reference weight?;
if WT(sample) >= Iorefwt + thresh then
do;
* Yes - got a cycle, so increment the valley count;
valleys = valleys + 1;
* Set this as the new high reference weight;
hirefwt = WT(sample);
* Transition to the CLIMBING state;

state = 1;
end;
end;
end;
end;
end;

* The number of cycles is the larger of peaks or valleys;
cycles = max(peaks, valleys);
run;

169

APPENDIX E: Tables Showing Correlation Coefﬁcients for Selected
Variables with Each Outcome

Table 35 - Correlation Coefﬁcients for Selected Variables with Net Change
in Total Serum Cholesterol for All Non-excluded Subjects and by Net
Weight Change Group, MRFIT SI Group

 

All Non- Weight No Wt. Weight
Variables Excluded ' Loss Change Gain
("4,302, (n=1 ,056) (n=2,446) (n=800)

 

Age -.05 n.s. -.04 -.13

 

 

 

 

 

 

    

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Wt. Change, Final 4 Months.

Net Weight Change .19 .15 .08 n.s.
Relative Weight, Baseline -.05 n.s. n.s. -.12
Weight, Baseline -.03 .06 n.s. -.1 1
Alcohol Intake n.s. n.s. n.s. n.s.
Caffeine Intake n.s. n.s. n.s. n.s.
Calcium Intake n.s. -.14 n.s. n.s.
Cholesterol Intake .06 .06 .06 n.s.
% Calories from Fat n.s. n.s. n.s. n.s.
Water-soluble Fiber Intake -.10 -.21 n.s. -.07
Iron Intake -.04 -.08 n.s. n.s.
Sodium Intake n.s. n.s. n.s. n.s.
Vitamin E Intake -.08 -.12 -.04 n.s.
Exercise Duration n.s. n.s. n.s. .07

Exercise Opinion, Year 6 -.05 n.s. -.04 n.s.
LTPA, Heavy n.s. n.s. n.s. n.s.
LTPA, Total n.s. -.08 n.s. n.s.
Physical Activity Quintile, Heavy n.s. n.s. n.s. n.s.
Physical Activity Quintile, Total n.s. -.08

 

 

 

 

   
 

 

.........................................................
......

e .n-
..........

 

1 70
Table 35 (cont'd)

' Excluded: Subjects missing 4 or more weight measurements, and subjects with
the following conditions: cancer, unexplained weight loss, thyroid disease,
Cushing's disease, diabetes, cirrhosis or other liver disease

171

Table 36 - Correlation Coefﬁcients for Selected Variables with Net Change
in HDL for All Non-excluded Subjects and by Net Weight Change Group,
MRFIT SI Group

 

 

 

All Non- . Weight No Wt. Weight
“mm 33133621 3:19:05) 5.13%) 312300)
Age at Baseline n.s. -.06 n.s. n.s.
Baseline Serum Cholesterol n.s. n.s. n.s. n.s.

 

Baseline Plasma Cholesterol

 

 

 

   
   

-.07

 

 

 

 

 

 

 

 

 

   
 

 

    
   

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

VIII. Change, Final 4 Months n.s. n.s. .05 n.s.
Net Weight Change -.18 -.09 -.06 n.s.
Relative Weight, Baseline n.s. n.s. n.s. n.s.
Alcohol Intake

Calcium Intak;

Cholesterol Intake .04 n.s. n.s. .1 1

% Calories from Fat .04 n.s. .06 .09
Water-soluble Fiber Intake n.s. n.s. n.s. n.s.
Iron Intake n.s. n.s. n.s. n.s.
P:S Ratio n.s. n.s. n.s. n.s.
Sodium Intake n.s. n.s. n.s. n.s.
Vitamin E Intake .04 n.s. .05 n.s.
Exercise Duration at Baseline n.s. n.s. n.s. n.s.
Exercise Opinion, Year 6 .07 .07 n.s. n.s.
LTPA, Heavy n.s. n.s. n.s. n.s.
LTPA, Total n.s. n.s. n.s. n.s.
Physical Activity Quintile, Heavy n.s. n.s. n.s. n.s.
Physical Activity Quintile, Total

 

 

 

 

   
 

 

 

' Excluded: Subjects missing 4 or more weight measurements, and subjects with
the following conditions: cancer, unexplained weight loss, thyroid disease,
Cushing's disease, diabetes, cirrhosis or other liver disease

172

Table 37 - Correlation Coefﬁcients for Selected Variables with Net Change
in Ratio of Total Plasma Cholesterol to HDL for All Non-excluded Subjects
and by Net Weight Change Group, MRFIT SI Group

All Non- . Weight No Wt. Weight
V - b Excluded Loss Change Gain
"a "’5 (n=4,302) (n=1,056) (n=2,446) (n=800)

Age at Baseline n.s. n.s. n.s. n.s.

Baseline Serum Cholesterol -.10 -.13 -.09 n.s.

VIII. Change, Final 4 Months
Net Weight Change
Relative Weight, Baseline
Weight, Baseline

Alcohol Intake

Calcium Intake . . n.s.
Cholesterol Intake . . . . n.s.
% Calories from Fat . . . . n.s. n.s.
Water-soluble Fiber Intake n.s. n.s.
Iron Intake . n.s. n.s.
P:S Ratio n.s. n.s.
Sodium Intake . . n.s. n.s.
Vitamin E Intake . n.s. n.s.
Exercise Duration at Baseline . . n.s. n.s. n.s.
Exercise Opinion, Year 6 -.08 n.s. n.s.
LTPA, Heavy n.s. n.s. n.s. .09
LTPA, Total n.s. n.s. n.s. n.s.
Physical Activity Quintile, Total n.s. n.s. n.s. .09
-.04 n.s. -.11 -.13

 

' Excluded: Subjects missing 4 or more weight measurements, and subjects with
the following conditions: cancer, unexplained weight loss, thyroid disease,
Cushing's disease, diabetes, cirrhosis or other liver disease

173

Table 38 - Correlation Coefﬁcients for Selected Variables with Net Change
in Blood Pressure for All Non-excluded Subjects and by Net Weight

Change Group, MRFIT SI Group

 

Variables

All Non-

Excluded '

Weight
Loss

No Wt.
Change

(n-2 514)

 

(n-4 393)

 

(n=1,060)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Wt. Change, Final 4 Months n.s.

Net Weight Change .19 .10 .10 n 5
Relative Weight, Baseline -.08 n.s. -.07 n s
Alcohol Intake .05 .09 n.s. n.s.
Caffeine Intake .10 n.s. .09 .16
Calcium Intake .04 n.s. .07 n.s.
Cholesterol Intake .07 n.s. .05 .07
% Calories from Fat .04 n.s. n.s. .08
Water-soluble Fiber Intake n.s. n.s. n.s. n.s.
Iron Intake -.04 n.s. .06 n.s.
P:S Ratio -.04 n.s. n.s. n.s.
Sodium Intake n.s. n.s. n.s. n.s.
Vltamin E Intake -.04 n.s. -.06 n.s.
Exercise Duration at Baseline .09 n.s. .12 .08
Exercise Opinion, Year 6 n.s. n.s. n.s. .08
LTPA, Heavy n.s. n.s. n.s. n.s.
LTPA, Total n.s. n.s. n s n.s
Physical Activity Quintile, Heavy n.s. n.s. n s n.s

.04 n.s. 04 .096

 

 

 

 

Physrcal Actrvrty Quintile, Total

 

 

 

‘ Excluded: Subjects missing 4 or more weight measurements, and subjects with
the following conditions: cancer, unexplained weight loss, thyroid disease, renal
disease, angina, primary aldosteronism, Cushing's disease, pheochromocytoma.

10.

IJST OF REFERENCES

Blair SN, Shaten J, Brownell K, Collins G, Lissner L. Body weight change,
all-cause mortality, and cause-speciﬁc mortality in the Multiple Risk
Factor Intervention Trial. Ann Intern Med 119: 749-757, 1993.

Lissner L, Odell PM, D'Agostino RB, Stokes J, Kreger B, Belanger AJ,
Brownell KD. Variability of body weight and health outcomes in the
Framingham population. N Engl J Med 324:1839-44, 1991.

Hamm P, Shekelle R, Stamler J. Large ﬂuctuations in body weight during
young adulthood and twenty-ﬁve-year risk of coronary death in men. Am
J Epidemiol 129: 312-18, 1989.

Lissner L, Bengtsson C, Lapidus L, Larsson B, Bengtsson B, Brownell K.
Body weight variability and mortality in the Gothenburg prospective
studies of men and women. Obesity In Europe 88, eds. P Bjomtorp and
S Rossner. London: Libbey, pp 51 -6, 1989.

American Heart Association. Heart and Stroke Facts. AHA National
Center, Dallas, 1 994.

Kuczmarski FlJ. Prevalence of overweight and weight gain in the United
States. Am J Clln Nutr 55: 4958—5028, 1992

National Institutes of Health Consensus Development Panel on the Health
Implications of Obesity. Health implications of obesity. National Institutes
of Health Consensus Development Conference Statement. Ann Intern
Med 103: 1073-1077, 1985.

Serdula M, Collins ME, Williamson DF, Anda RF, Pamuk ER, Byers T.
Weight control practices of US adolescents and adults: Youth Risk
Behavior Survey and Behavioral Risk Factor Surveillance System. Ann
Intern Med 119: 667—71, 1993.

Goldstein DJ. Beneﬁcial health effects of modest weight loss. Int J Obes
16: 397-415, 1992.

Ashley FW, Kannel WB. Relation of weight change to changes in
atherogenic traits: the Framingham study. J Chron Dis 27: 103-114,
1974. 174

11.

12.

13.

14.

15.

16.

17.

18.

19.

21.

23.

24.
25.

175

Consumers Union: Losing weight: What works. What doesnt. Consumer
Reports June 1993.

National Institutes of Health. Methods for voluntary weight loss and
control: technology assessment conference statement. Nutrition Today
27: 27-33, 1992.

Lissner L, Andres R, Muller D, Shimokata H. Body weight variability in
men: metabolic rate, health and longevity. Int J Obes 14: 373-83, 1990.

Andres R, Muller D, Sorkin J, Long-tenn effects change in body weight on
all—cause mortality. Arch Intern Med 119: 737—43, 1993.

Pamuk ER, Williamson DF, Serdula MK, Madans J, Byers TE. Weight
loss and subsequent death in a cohort of US. adults. Ann Intern Med
119: 744-748, 1993.

Williamson DF, Pamuk ER. The association between weight loss and
increased longevity: a review of the evidence. Ann Intern Med 119: 731—
736, 1 993.

Lee l-M, Paffenbarger RS. Change in body weight and longevity. JAMA
268: 2045-2049, 1 992.

Hammond E, Garﬁnkel L. Coronary heart disease, stroke, and aortic
aneurisms. Arch Environ Health 19: 167-82, 1969.

National Task Force on the Prevention and Treatment of Obesity. Weight
cycling. JAMA 272: 1 196-1202, 1994.

Pi-Sunyer FX. Medical Hazards of Obesity. Ann Intern Med 119: 655-
660, 1993.

Manson, JE, Stampfer MJ, Hennekens CH, Willett WC, Body weight and
longevity: a reassessment. JAMA 257: 353-8, 1987.

Manson JE, Willett WC, Stampfer MJ, Colditz GA, Hunter DJ, Hankinson
SE, Hennekens CH, Speizer FE. Body weight and mortality among
women. N Engl J Med 333: 677—85, 1995.

Lindsted K, Tonstad S, Kuzma JW. Body mass Index and patterns of
mortality among Seventh-day Adventist men. Int J Obes 15: 397-406,
1991.

Kushner RF. Body weight and mortality. Nutr Rev 51: 127-136, 1993.

Phinney SD, Tang AB, Waggoner CR, Tezanos-Pinto RC, Davis PA. The
transient hypercholesterolemia of major weight loss. Am J Clln Nutr 53:

26.

27.

28.

29.
30.

31.

32.

33.

35.

36.

37.

39.

176
1404—10, 1991.

Williamson DF and Pamuk ER. The association between weight loss and
increased longevity. A review of the evidence. Ann Intern Med 119: 731-
736, 1993.

Dublin Ll, Marks HH. Mortality among insured overweights in recent
years. Trans Assn Life Insurance Med Directors of America 35: 235-
63, 1951.

Dublin LJ. Relation of obesity to longevity. New Engl J Med 248: 971 -4,
1953.

Society of Actuaries. Build and Blood Pressure Study Chicago, 1959.

Hammond EC, Garﬁnkle L. Coronary heart disease, stroke and aortic
aneurysm. Arch Environ Health 19: 167-82, 1969.

Wannamethee G, Shaper AG, Weight change in middle—aged British men:
implications for health. Eur J Clin Nutr 44: 133-42, 1990.

Association of Life Insurance Medical Directors of America. Society of
Actuaries. Build Study Chicago, 1979.

Lean ME, Powrie JK, Anderson AS, Gamaite PH. Obesity, weight loss
and prognosis in type 2 diabetes. DIabet Med 7: 228—33, 1990.

Avons P, Dulcimetier P, Rakotovao R. Weight and mortality (Letter).
Lancet 1: 1104, 1983.

Deeg DJ, Miles TP, Van Zonneveld RJ, Curb JD. Weight change, survival
time and cause of death in Dutch elderly. Arch Gerontol Gerlart 10: 97—
1 1 1 . 1 990.

Harris T, Cook EF, Garrison R, Higgins M, Kannel W, Goldman L. Body
mass index and mortality among nonsmoking older persons. The
Framingham Heart Study, JAMA 259: 1520-4, 1988.

Paffenbarger RS, Hyde RT, Wing AL, Hsieh CC. Physical activity, all—
cause mortality, and longevity of college alumni. N Engl J Med 314: 605—
13, 1 986.

Rhoads CG, Kagan A. The relation of coronary disease, stroke, and
mortality to weight in youth and middle age. Lancet 1: 492-5, 1983.

Schroll M. A longitudinal epidemiological survey of relative weight at age
25, 50, and 60 in the Glostrup population of men and women born in
1914. Dan Med Bull 28: 106-16, 1981.

41.

42.

47.

177

Sidney S, Friedman GS, Siegelaub AB. Thinness and mortality. Am J
Public Health 77: 317, 1987.

Wilcosky T, Hyde J, Anderson JJ, Bangdiwala S, Duncan B, Obesity and
mortality in the Lipid Research Clinics Program Follow-up Study. J Clin
Epidemlol 43: 743-52, 1990.

Higgins M, D'Agostino R, Kannel W, Cobb J. Beneﬁts and adverse effects
of weight loss. Observations from the Framingham study. Ann Int Med
119: 758-763, 1993.

Williamson DF, Pamuk E, Thun M, Flanders D, Byers T, Heath C.
Prospective study of intentional weight loss and mortality in never-
smoklng ovenrveight US white women aged 40—64. Am J Epidemiol 141:
1128-41, 1995.

Zimmerman D, Hoerr SL. Use of questionable dieting practices among
young women examined by weight history. J Women's Health 4: 189—96,
1995.

French SA, Jeffery RW, Folsom AR, Williamson DF, Byers T. Relation of
weight variability and intentionality of weight loss to disease history and
health-related variables in a population-based sample of women aged 55-
69 years. Am J Epidemlol 142: 1306—14. 1995.

Paffenbarger RS, Hyde RT, Wing AL, Lee IM, June DL, Kampert JB. The
association of changes in physical activity level and other lifestyle
characteristics with mortality among men. N Engl J Med 328: 538-45,
1993.

Willett WC, Manson JE, Stampfer MJ, Colditz GA, Rosner B, Speizer FE,
Hennekens CH. Weight, weight change, and coronary heart disease. Risk
within the 'normal' weight range. JAMA 273: 461 -465, 1995.

Sedgwick AW, Thomas DW, Davies M, Baghurst K. Relationships
between weight change and blood lipids in men and women: ”The
Adelaide 1000.'Int J Obes 14: 439-50, 1990.

Stunkard A, Penlck S. Behavior modiﬁcation in the treatment of obesity:
The problem of maintaining weight loss. Arch Gen Psychiatry 36: 801 -
806, 1 979.

Stamler R, Stamler J. Gosch F, Civinelli J, Fishman J, McKeever P,
McDonald A, Dyer AR, Primary prevention of hypertension by nutritional-
hygienlc means. Final report of a randomized, controlled trial. JAMA 262:
1801 -07, 1989.

51.

52.

57.

59.

61.

178

Kumanyika SK, Obarzanek E, Stevens VJ, Hebert PR, Whelton PK.
Weight-loss experience of black and white participants In NHLBI-
sponsored clinical trials. Am J Clin Nutr 53: 1631 S—16388, 1991.

Stevens VS, Corrigan SA, Obarzanek E, Bemauer E, Cook NR, Hebert P,
Mattfeldt-Berman M, Oberman A, Sugars C, Dalcin AT, Whelton PK.
Weight loss intervention in Phase 1 of the trials of hypertension
prevention. Arch Intern Med 153: 849-858, 1993.

Langford H, Blaufox D, Oberman A, Hawkins CM, Curb JD, Cutter GR,
WassertheiI-Smoller S, Pressel S, Babcock C, Abernathy JD, Hotchkiss J,
Tyler M. Dietary therapy slows the return of hypertension after stopping
prolonged medication. JAMA 253: 657-664, 1985.

Hypertension Prevention Trial Research Group. Hypertension prevention
trial - three—year effects of dietary change on blood pressure. Arch Intern
Med 150: 153-163, 1990.

Schotte DE, Stunkard AJ. The effects of weight reduction on blood
pressure in 301 obese patients. Arch Intern Med 150: 1701 -1704, 1990.

Wing RR, Marcus MD, Salata R, Epstein LE, Miaskiewicz S, Blair EH.
Effects of very-Iow-calorie diet on long-term glycemic control in obese
Type 2 diabetic subjects. Arch Intern Med 151: 1334-1340, 1991.

Personal communications, George Blackburn, MD. and Fran Peterson,
Ph.D. (Sandoz Pharmaceuticals), 1993.

Wing RR, Koeske R, Epstein LH, Nowalk MP, Gooding W, Becker D.
Long-term effects of modest weight loss In Type II diabetic patients. Arch
Intern Med 147: 1749—1753, 1987.

Wing RR, Jeffery RW, Hellerstedt WL. A prospective study of weight
cycling on cardiovascular risk factors. Arch Intern Med 155: 1416—22,
1 995.

Borkan G Gerzof SG, Robbins AH, Hults DE, Dilbert CK, Sllbert JE.
Assessment of abdominal fat content by computed tomography. Am J
Clln Nutr 36: 172-177, 1982.

Fujioka S, Matsuzawa Y, Tokunaga K, Kawamoto T, Kobatake T, Keno Y,
Kotani K, Yoshida S, Tarui S. Improvement of glucose and lipid
metabolism associated with selective reduction of intra-abdominal visceral
fat in premenopausal women with visceral fat obesity. Int J Obes 15: 853-
859, 1 991 .

62.

67.

70.

71.

72.

73.

179

Fujioka S, Matsuzawa Y, Tokunaga K, Keno Y, Kobatake T, Tarui S,
Treatment of visceral fat obesity. Int J Obes 15: 59-65, 1991 .

Van der Kooy K, Leenen R, Seidell J, Deurenberg P, Droop A, Bakker
CJG. Waist-hip ratio is a poor predictor of changes in visceral fat. Am J
Clln Nutr 57: 327-33, 1993.

Armellini F, Zamboni M, Rigo L, Bergamo—Andrels IA, Robbi R, De Marchi
M, Bosello O. Sonography detection of small intra-abdominal fat
variations. Int J Obes 15: 847-852, 1991.

Despres JP, Poullot M—C, Moorjani S, Nadeau A, Tremblay A, Lupien PJ,
Theiault G, Bouchard C, Loss of abdominal fat and metabolic response to
exercise training in obese women. Am J Physiol 261: E167, 1991.

Bouchard C, Tremblay A, Despres JP, Nadeau A, Lupien PJ, Theriault G,
Dussault J, Moorjani S, Pinault S, Foumier G. The response to long-term
overfeeding of identical twins. N Engl J Med 322: 1477-82, 1990.

Van der Kooy K, Leenen R, Seidell J, Deurenberg P, Hautvast JGAJ.
Effect of a weight cycle on visceral fat accumulation. Am J Clln Nutr 58:
853-7, 1993.

Hensrud DD, Weinser RL, Darnell BE, Hunter GR. A prospective study of
weight maintenance in obese subjects reduced to normal body weight
without weight-loss training. Am J Clln Nutr 60: 688-94, 1994.

Hainer V, Kunesova M, Stich V, Parizkova J, Zak A, Stukavee V, Hrabak

P. Body fat distribution and serum lipids during long-tenn follow-up of
obese patients treated Initially with a very-low-calorie diet. Am J Clln Nutr
56: 2838-58, 1992.

Reed G, Hill J. Weight cycling: a review of the animal literature. Obes
Res 1: 392-402, 1993.

American Heart Association Committee on Exercise and Cardiac
Rehabilitation. Beneﬁts and recommendations for physical activity
programs for all Americans. position statement. Circulation 36: 340-344,
1 992.

Powell KE, Thompson PD, Caspersen CJ, Kendrick JS. Physical activity
and the incidence of coronary heart disease. Annu Rev Public Health 8:
253-287, 1987.

Blair SN, Kohl HW, Paffenbarger RSJR, Clark DG, Cooper K, Gibbons
LW. Physical fitness and all—cause mortality: A prospective study of
healthy men and women. JAMA 262: 2395-401, 1989.

74.

75.

76.

77.

78.

79.

80.

81.

82.

180

Albanes D, Blair A, Taylor PR. Physical activity and risk of cancer In the
NHANES l population. Am J Public Health 79: 744—750, 1989.

Blair SN, Jacobs DR Jr, Powell KE. Relationships between exercise or
physical activity and other health behaviors. Public Health Rep 100:
172-180, 1985.

Epstein LH, Wing RR. Aerobic exercise and weight. Addict Behav 3:
371 -388, 1980.

Lampman RM, Santinga JT, Savage PJ, Bassett DR, Hydrlck CR, Flora
JD, Block WD. Effect of exercise training on glucose tolerance, in vivo
insulin sensitivity, lipid and lipoprotein concentrations in middle-aged men
with mild hypertriglyceridemia. Metabolism 34: 205-11, 1985.

Farmer ME, Locke BZ, Moscicki EK, Dannenberg AL, Larson DB, Radloff
LS. Physical activity and depressive symptoms: The NHANES I
Epidemiologic Follow-up Study. Am J Epldemlol 128: 1340-1351, 1988.

Marrugat J, Elosua R, Covas MI, Molina L, Rubies-Prat J and the
MARATHOM Investigators. Amount and intensity of physical activity,
physical ﬁtness, and serum lipids in men. Am J Epldemlol 143: 562—9,
1996.

Willett W. Nutritional Epidemiology Oxford University Press, New York,
1990.

Sempos CT, Looker AC, Gillum RF. Iron and heart disease: the
epidemiologic data. Nutr Rev 54: 73-84, 1996.

Harlan WR, Hull AL, Schmouder RL, Landis JR. Thompson FE, Larkin
FA. Blood pressure and nutrition In adults. The National Health and
Nutrition Examination Survey. Am J Epldemlol 120: 17-28, 1984.

The INTERSALT Co-operative Research Group. Sodium, potassium,
body mass, alcohol and blood pressure: the INTERSALT Study. J
Hyperten 6: 8584-8586, 1988.

Reusser ME, McCarron DA. Micronutrient Effects on Blood Pressure
Regulation. Nutr Rev 52: 367-75, 1994.

Breslow JL. Genetic basis of lipoprotein disorders. J Clin Invest 84: 373-
80. 1989.

Guyton AC. Textbook of Medical Physiology. W B Saunders Co.,
Harcourt Brace Jovanovich, Inc. Philadelphia, 1991, p.851.

87.

89.

91.

92.

93.

95.

97.

98.

181

Harlap S, Kark JD, Baras M, Eisenberg S, Stein Y. Seasonal changes In
plasma lipid and lipoprotein levels in Jerusalem. Isr J Med Sci 18: 1158-
65, 1982.

Hegsted DM, Nicolosi RJ. Individual variation In serum cholesterol levels.
Proc Natl Acad Sci USA. 84: 6259-6261, 1987.

Mogadam M, Ahmed S, Mensch AH, Godwin ID. Within-person
ﬂuctuations of serum cholesterol and lipoproteins. Arch Intern Med 150:
1645-1648, 1990.

Groover ME Jr, Jernigan JA, Martin CD. Variations in serum lipid
concentration and clinical coronary disease. Am J Med Soc 239: 133—
139, 1 960.

Bookstein L Gidding SS, Fonovan M, Smith FA. Day-to—day variability of
serum cholesterol, triglyceride, and high-density lipoprotein cholesterol
levels. Arch Intern Med 150:1653—1657, 1990.

Durrington PN. Biological variations in serum lipid concentrations. Scand
J Clin Lab Invest 50(Suppl)198: 86-91, 1990.

Boniface DR. Seasonal variation of cholesterol in the cloﬁbrate trial data,
Report on project for practical part of MSc statistics. London School of
Economics, 1972.

Bengtsson C, leblin E, Blohome G, Gustafson A. Serum cholesterol and
serum triglycerides in middle—aged women. The study of women in
Gothenburg 1968-69. Scand J Clin Lab Invest 34: 61 —66, 1974.

Gordon DJ, Hyde J, Trost DC, Whaley FS, Hannan FJ, Jacobs DR,
Eklund L-G. Cyclic seasonal variations in plasma lipid and lipoprotein
levels: the lipid research clinics coronary primary prevention trial placebo
group. J Clln Epldemlol 41: 697-689, 1988.

Kritchevsky D. Variation in plasma cholesterol levels. Nutrition Today 27:
21 -2. 1992.

Rippey RM. Overview: seasonal variations in cholesterol. Prev Med 10:
655-659, 1 981 .

Shaper AG, Wannamethee G, Walker M. Alcohol and mortality In British
men: explaining the U-shaped curve. Lancet 2: 1267-73, 1988.

Klatsky AL, Friedman GD, Siegelaub AB. Alcohol and mortality. A ten-
year Kaiser—Permanente experience. Ann Intern Med 95: 139-45, 1981.

100.

101.

102.

103.

104.

105.

106.

107.

108.

109.

110.

111.

182

MacMahon S. Alcohol consumption and hypertension. Hypertension 9:
111-21,1987.

Suh l, Shaten J, Cutler JA, Kuller LH. Alcohol use and mortality from
coronary heart disease: the role of high-density lipoprotein cholesterol.
Ann Internal Med 116: 881-7, 1992.

Schatzkin A, Hoover RN, Taylor PR, Ziegler RG, Carter CL, Larson DB.
Serum cholesterol and cancer in the NHANES l Epidemiologic Follow-up
Study. Lancet 2: 298-302, 1987.

Groen J, Tijong BK, Kamminga CE, Willebrands AF. The inﬂuence of
nutrition, individuality, and some other factors including various forms of
stress on the serum cholesterol: an experiment of nine months duration in
60 normal human volunteers. Voeding 13: 556-87, 1952.

Albanes DM, Jones Y, Micozzi MS, Mattson ME. Associations between
smoking and body weight in the US population: analysis of NHANES Il.
Am J Public Health. 77: 439-44, 1987.

Lund-Larsen PG, Tretti S. Changes in smoking habits and body weight
after a three year period - the cardiovascular disease study in Finnmark.
J Chronic Dis 35: 773-80, 1982.

Carmelli D, Swan G, Robinette D. Smoking cessation and weight gain In
identical twins. (Letter) N Engl J Med. 325: 517, 1991.

Seidell JC, Cigolini M, Deslypere J-P, Charzewska J, Ellslnger B-M, Cruz
A. Body fat distribution in relation to physical activity and smoking habits
in 38-year-old European men. Am J Epldemlol 133: 257-65, 1991.

Goldman R and Rockstein M, eds. The Physiology and Pathology of
Human Aging Academic Press, Inc., New York, 1975.

Ader R, Cohen N, Felten D. Psychoneuroimmunology: Interaction
between the nervous system and the immune system. Lancet 345: 99-
103, 1 995.

Ornish D, Brown SE, Scherwitz LW. Billings JH, Armstrong WI’, Ports TA,
McLanahan SM, Kirkeeide RL, Brand RJ and Lance Gould K. Can
lifestyle changes reverse coronary heart disease? The Lifestyle Heart
Trial. Lancet 336:129-33, 1990.

Bjomtorp P. Visceral obesity: a 'civilization syndrome.‘ Obes Res 1: 206-
22, 1 993.

112.

113.

114.

115.

116.

117.

118.

119.

120.

121.

122.

123.

124.

183

Shively CA, Clarkson RB, Miller C, Weingard KW. Body fat distribution as
a risk factor for coronary artery atherosclerosis in female, Cynomolgus
monkeys. Atherosclerosis 7: 226-231, 1987.

Kaplan JR, Adams MR, Clarkson TB, Manuck BS, Shively CA. Social
behavior and gender in biomedical investigations using monkeys: studies
in atherogenesis. Lab AnIm SCI 41: 334-43, 1991.

Foreyt JP, Brunner RL, Goodrick GK, Cutter G, Brownell KD, St. Jeor ST.
Psychological correlates of weight ﬂuctuation. Int J Eat DIsord 17: 263-
75, 1 995.

Brownell KD, Greenwood MRC, Stellar E, Shrager EE. The effects of
repeated cycles of weight loss and regain in rats. Physiol Behav 38: 459—
64, 1986.

Iribarren C, Sharp 08, Burchﬁel CM, Petrovitch H. Association of weight
loss and weight ﬂuctuation with mortality among Japanese American men.
N Engl J Med 333: 686-92, 1995.

Rodin J, Radke-Sharpe N, Rebuffe-Scrive M Greenwood, MRC. Weight
cycling and fat distribution. Int J Obes 14: 303-10, 1990.

Stevens J, Lissner L. Body weight variability and mortality in the
Charleston Heart Study. Int J Obes 14: 385-6, 1990.

Hoffman MDAF, Kromhout D. Changes in body mass index in relation to
myocardial infarction (the Zutphen Study). Abstract. Int J Obes 13: A25,
1989.

Casey VA, Dwyer JT, Berkey CS, Coleman KA, Gardner J, Valadian I.
Long-term memory of body weight and past weight satisfaction: a
longitudinal follow-up study. Am J Clln Nutr 53: 1493-8, 1991.

Wing R. Weight Cycling in Humans: a review of the literature: a review of
the literature. Ann Behav Med 14: 113-19, 1992.

Gordon T, Kannel WB, Castelli WP, Dawber TR. Lipoproteins,
cardiovascular disease and death: The F rarningham Study. Arch Intern
Med 141: 1128-31, 1981.

Castelli WP, Garrison RJ, Wilson PWF, Abbott RD, Kalousdian S,
Kannel WB, Incidence of coronary heart disease and lipoprotein
cholesterol levels: The Framingham Study. JAMA 256: 2823-8, 1986.

Stamler J, Wentworth D, Neaton JD. Is the relationship between serum
cholesterol and risk of premature death from coronary heart disease

125.

126.

127.

128.

129.

130.

131.

132.

133.

134.

184

continuous and graded? Findings in 356,222 primary screenees of the
Multiple Risk Factor Intervention Trial. JAMA 256: 2823-8, 1986.

National Cholesterol Education Program. Report on the Expert Panel on
Population Strategies for Blood Cholesterol Reduction. Circulation 83:
2154-232, 1991 .

Gordon DJ, Probstfeld JL. Garrison RJ, Newton JD, Castelli WP, Knoke
JD, Jacobs DR, Bangdiwala S, Tyroler A. High-density lipoprotein
cholesterol and cardiovascular disease: four prospective American
studies Circulation 79: 8-15, 1989.

Weijenberg MP, Feskens EJM, Kromhout D. Total and high density
lipoprotein cholesterol as risk factors for coronary heart disease in elderly
men during 5 years of follow-up. Am J Epldemlol 143: 151-8, 1996.

Genest JJ Jr., Martin—Munley SS, McNamara JR, Ordovas JM, Jenner J,
Myers RH, Silbennan SR, Wilson PWF, Salem DN, Schafer EJ. Familial
lipoprotein disorders in patients with premature coronary artery disease.
Circulation 85: 2025-33, 1992.

Kannel WB, Wilson PWF. Efﬁcacy of lipid proﬁles in prediction of
coronary disease. Am Heart J 124: 768-74, 1992.

Stampfer MJ, Sacks FM, Salvini S, Willett WC, Hennekens CH. A
prospective study of cholesterol, apolipoproteins, and the risk of
myocardial infarction. N Engl J Med 325: 373-81, 1991.

National Cholesterol Education Program. Detection, Evaluation and
Treatment of High Cholesterol In Adults National Institutes of Health,
National Heart Lung and Blood Institute, NIH Publication No. 93-3095,
1993.

Wilson PWF, Anderson KM, Castelli WP. The impact of triglycerides on
coronary heart disease: the Framingham Study, In: Gotto AM Jr., Paoletti
R. eds. Atherosclerosis Reviews Vol 22 Raven Press, New York, 1991,
p 59-63.

AIIred J. Lowering serum cholesterol: who beneﬁts? J Nutr 123: 1453-9,
1993.

Jacobs D, Blackburn H, Higgins M, Reed D, Iso H, McMillan G, Neaton J,
Nelson J, Potter J, Rifkind B, Rossouw J, Shekelle R, Yusuf S. Report of
the conference on low blood cholesterol mortality associations.
Circulation 86: 1046-1060, 1992.

135.

136.

137.

138.

139.

140.

141.

142.

143.

144.

145.

185

lso H, Jacobs DR, Wentworth D, Neaton J, Cohen JD. Serum cholesterol
levels and six—year mortality from stroke in 350,977 men screened for the
Multiple Risk Factor Intervention Trial. N Eng J Med 320: 904-910, 1989.

Joint National Committee on Detection, Evaluation, and Treatment of High
Blood Pressure. The Fifth Report of the Joint National Committee on
Detection, Evaluation, and Treatment of High Blood Pressure
National High Blood Pressure Education Program, National Institutes of
Health, National Heart, Lung, and Blood Institute. NIH Publication No.
93-1088, 1993.

Cutler JA, MacMahon SW, Furberg CD. Controlled clinical trials of drug
treatment for hypertension. A review. Hypertension 13: I-36-l-44, 1989.

Moser M, Hebert P, Hennekens CH. Commentary: An overview of the
meta-analyses of the hypertension treatment trials. Arch Intern Med 151:
1277-9, 1991.

Sherwin R, Kaelber CT, Kezdi P, Kjelsberg M, Emerson Thomas H. The
Multiple Risk Factor Intervention Trial II. The development of the protocol.
Prev Med 10: 402-425, 1981.

Watt BK, Merrill AL. Composition of Foods - Raw, Processed,
Prepared. Rev. USDA Handbook No. 8 United States Department of
Agriculture, 1963.

Adams CE. Nutritive Value of American Foods In Common Units.
USDA Agricultural Handbook No. 456, United States Department of
Agriculture, 1975.

Tlllotson JL. Gorder DD, Kassim N. Nutrition data collection in the
Multiple Risk Factor Intervention Trial (MRFIT). J Am Diet Assoc 78:
235-40. 1 981 .

Taylor HL, Jacobs DR, Schucker B, Knudson J, Leon AS, DeBacker GA.
A questionnaire for the assessment of leisure time physical activities. J
Chronic Dis 31: 741 -55, 1978.

Leon AS, Jacobs DR Jr, DeBacker G, Taylor HL. Relationship of physical
characteristics and life habits to treadmill exercise capacity. Am J
Epldemlol 113: 653-660, 1981 .

Leon AS, Connett J, Jacobs DR, Rauramaa R. Leisure-time physical
activity levels and risk of coronary heart disease and death. The Multiple
Risk Factor Intervention Trial. JAMA 258: 2388-2395, 1987.

146.

147.

148.

149.

150.

151.

152.

153.

154.

155.

156.

186

Gerace TA, Hollis J, Ockene JK, Svendsen K (for the MRFIT Research
Group). Smoking cessation and change In diastolic blood pressure, body
weight, and plasma lipids. Prev Med 20: 602-620, 1991.

Crow RS, Rautaharju PM, Prineas RJ, Connett JE, Furberg C, Broste S,
Stamler J for the MRFIT Research Group. Risk factors, exercise ﬁtness
and electrocardiograph response to exercise in 12,866 men at high risk of
symptomatic coronary heart disease. Am J Cardiol 57: 1075-1082, 1986.

American College of Sports Medicine. Resource Manual for Guidelines
for Exercise Testing and Prescription Lea & Febiger, Philadelphia,
1988.

Neaton JD, Broste S, Cohen L, Fishman EL, IQelsberg MO, Schoenberger
J. (For the MRFIT). The Multiple Risk Factor Intervention Trial (MRFIT).
VII. A comparison of risk factor changes between the two study groups.
Prev Med 10: 519-543, 1981.

Shekelle RB, Hulley SB, Neaton JD, Billings JH, Borhani NO, Gerace TA,
Jacobs DR, Lasser NL, Mittlemark MB, Stamler J, for the Multiple Risk
Factor Intervention Trial Research Group. The MRFIT behavior pattern
study. II. Type A behavior and Incidence of coronary heart disease. Am J
Epldemlol 122: 559-70, 1985.

Caggiula AW, Christakis G, Farrand M Hulley SB, Johnson R, Lassner
NL, Stamler J, Wlddowson G. The Multiple Risk Factor Intervention Trial
IV. Intervention on lipids. Prev Med 10: 443-75, 1981.

Cohen JD. Grimm RH, McFate Smith W (for the MRFIT). The Multiple
Risk Factor Intervention Trial VI. Intervention on blood pressure Prev
Med 10: 501 -518, 1981.

Hughes GH, Hymowitz N , Ockene J, Simon N, Vogt TM. The Multiple
Risk Factor Intervention Trial V. Intervention on smoking. Prev Med 10:
476-500, 1981 .

Sherwin R, Kaelber CT, Kedzi P, Kjelsberg MD, Thomas HE (for the
MRFIT). The Multiple Risk Factor Intervention Trial (MRFIT) II. The
development of the protocol. Prev Med 10: 402-425, 1981.

Morgan J, Kjelsberg MO, eds. The Multiple Risk Factor Intervention Trial:
quality control of technical procedures and data acquisition. Control Clln
Trials 7 (Supplement) 1986.

SAS/STAT User's Guide, Version 6, Fourth Edition SAS Institute, Inc.,
Cary, NC 1989.

157.

158.

159.

160.

161.

162.

163.

164.

165.

166.

167.

168.

187

Wadden TA, Foster GA, Letizia KA, Stunkard AJ. A multi-center
evaluation of a proprietary weight reduction program for the treatment of
marked obesity. Arch Intern Med 152: 961 -6, 1992.

Sims EAH, Goldman RF, Gluck CM, Horton ES, Kelleher PC, Rowe DW.
Experimental obesity in man. Trans Assoc Am Phys 81: 153-170. 1968.

Pasquet P and Apfelbaum M. Recovery of initial body weight and
composition after massive overfeeding in men. Am J Clln Nutr 60: 861 -3,
1994.

Leon AS and Connetti J for the MRFIT Research Group. Physical activity
and 10.5 year mortality In the Multiple Risk Factor Intervention Trial
(MRFIT). Int J Epldemlol 20: 690—695, 1991.

Berkow R, ed. The Merck Manual of Diagnosis and Therapy, Sixteenth
Edition Merck & Co, Inc, Rahway, N.J., 1992.

Lewis—Beck MS. Applied Regression - An Introduction. Series:
Quantitative Applications In the Social Sciences. Sage Publications,
Newberry Park, California, 1990.

Glass GV, Hopkins KD. Statistical Methods In Education and
Psychology, 2nd Edition. Prentice-Hall, Inc. Englewood Cliffs, New
Jersey, 1 984.

Rotterdam EP, Katan MB, Knuiman JL Importance of time interval
between repeated measurements of total or high density lipoprotein
cholesterol when estimating and lndividual‘s baseline concentrations. Clln
Chem 33: 1913-15, 1987.

Keys A. Diet and blood cholesterol In population surveys - lessons from
analysis of the data from a major survey in Israel. Am J Clln Nutr 48:
1161 -1 165, 1988.

Hegsted DM, McGandy RB, Myers ML, Stare FJ. Quantitative effects of
dietary fat on serum cholesterol in man. Am J Clin Nutr 17: 281 -295,
1965.

Keys A, Anderson JT, Grande F. Serurn cholesterol response to changes
in the diet. Metabolism 14: 747-787, 1965.

Verhoef P, Stampfer MJ, Buring JE, Gaziano JM, Allen RH, Stabler SP,
Reynolds RD, Kok FJ, Hennekens CH, Wlllett WC. Homocysteine
metabolism and risk of myocardial infarction: relation with vitamins B6,
8,2, and folate. Am J Epldemlol 143: 845-59, 1996.

169.

170.

171.

172.

173.

174.

175.

176.

177.

178.

179.

188

Bjomtorp P. Classiﬁcation of obese patients and complications related to
the distribution of surplus fat. Am J Clln Nutr 45: 1120-5, 1987.

Hartz AJ, Rupley DC, Rimm AE. The association of girth measurements
with disease in 32,856 women. Am J Epldemlol 119: 71 -80, 1984.

Sours HE, Frattoli VP, Brand CD, Feldman RA, Forbes AL, Swanson RC,
Paris AL. Sudden death associated with very low calorie weight reduction
regimens. Am J Clin Nutr 34: 453-61, 1981.

Davis A. Let's Eat Right to Keep Fit Signet, New York, 1988.

Phinney SD. Weight cycling and cardiovascular risk in obese men and
women (Letter). Am J Clln Nutr 56:781-2, 1992.

Simopoulos AP: Omega-3 fatty acids in health and disease and in growth
and development. Am J Clln Nutr 54:438-463. 1991.

Langlols JA, Harris T, Looker AC, Madans J. Weight change between age
50 years and old age is associated with risk of hip fracture in white
women aged 67 years and older. Arch Intern Med 156: 989-94, 1996.

Keys A, Brozek J, Henschel A, Mickelson P, Taylor HL. The Biology of
Human Starvation University of Minnesota Press, Minneapolis, 1950.

Cutter GR. Obesity and the implications of weight loss (Is there death
after success?) Perspectives In Applied Nutr 1: 3-13, 1993.

Kuller L, Wing R. Weight loss and mortality (letter) Ann Intern Med 119:
630-2. 1993

Brownell KD, Rodin J. Medical, metabolic, and psychological effects of
weight cycling. Arch Intern Med 154: 1325-1330, 1994.