Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Association of bone density and bone metabolism markers: the Women’s Isoflavone Soy Health (WISH) trial
(USC Thesis Other)
Association of bone density and bone metabolism markers: the Women’s Isoflavone Soy Health (WISH) trial
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
ASSOCIATION OF BONE DENSITY AND BONE
METABOLISM MARKERS: THE WOMEN’S
ISOFLA VONE SOY HEALTH (WISH) TRIAL
by
Jun Feng
A Thesis Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE (BIOSTATISTICS)
December 2015
Copyright 2015 Jun Feng
I
ACKNOWLEDGEMENTS
I would like to express my sincere gratitude to my committee chair,
Dr. Wendy Mack, for her illuminating and insightful guidance through every
stage of this thesis. Without her invaluable help, this thesis could not be
completed. I also want to express my sincere appreciation to my committee
members, Dr. Stanley Azen and Dr. Howard Hodis, for their kindly support
and inspiring suggestions for my thesis. I also want to show my appreciation
to my family for their love and support.
II
TABLE OF CONTENTS
ACKNOWLEDGEMENTS Ⅱ
LIST OF TABLES Ⅳ
ABSTRACT Ⅴ
INTRODUCTION 1
METHOD 3
Bone Density Sub-Study 5
Study Variables and Measurement 6
Statistical Analysis 7
RESULTS 12
Participants 12
Simple Associations among Bone Density and Bone Markers 14
Change in Bone Density over Visits 15
Unadjusted Associations of Bone Density and Metabolism Markers over Visits 17
Associations of Bone Metabolism Markers with Longitudinal Change in Bone Density 22
Multivariable Models: Associations of Levels of Bone Density 26
and Bone Markers over Visits
Multivariable Models: Associations of Bone Markers with Rate of Change 29
in Bone Density
DISCUSSION 31
REFERENCES 35
APPENDIX 39
III
LIST OF TABLES
Table 1. Baseline Characteristics of Analysis Sample 13
Table 2. Analysis Dataset: Number of Bone Density Outcomes by 14
Trial Follow-up
Table 3. Pearson Correlation Coefficients among Bone Density and Markers 15
Table 4. Comparison of AIC and Number of Parameters for Correlation Structures 16
Table 5. Differences in Log Transformed Bone Density over Visit Times 16
Table 6. Log Transformed Bone density and Metabolism Marker Associations: 17
Measured over the Trial
Table 7. Associations between Log Transformed Lumbar Spine and Bone Markers 19
by Study Visit
Table 8. Associations between Log Transformed Femoral Neck and Bone Markers 20
by Study Visit
Table 9. Associations between Log Transformed Total Hip and Bone Markers 21
by Study Visit
Table 10. Annual Rate of Change in Log Transformed Bone Density 23
Table 11. Associations of Bone Markers with Rate of Log Transformed 24
Bone Density Change
Table 12. Multivariable Model: Associations with Log Transformed 26
Bone Density Levels
Table 13. Multivariable Models: Associations with Rate of Change 29
in Log Transformed Bone Density
IV
ABSTRACT
Objective: The purpose of this study was to assess the association of bone
density and bone metabolism markers among postmenopausal women who
received isoflavone soy protein treatment versus placebo.
Methods: The Women’s Isoflavone Soy Health (WISH) trial was a
randomized, double-blind and placebo-controlled trial. A total of 350
postmenopausal women were randomized to isoflavone soy protein (ISP) or
placebo groups and followed on their randomized intervention for 30 months.
The primary trial outcome was carotid artery intima-media thickness
(CIMT), used to test the effect of isoflavone protein on progression of
subclinical atherosclerosis. Secondary outcomes were change in bone
density and bone metabolism markers. Outcome variables included in this
sub-study were bone density of the lumbar spine, femoral neck and total hip.
Bone markers variables included serum receptor activator of nuclear factor
kB (sRANKL), osteoprotegrin (OPG), bone specific-alkaline phosphatase
(BAP) and n-terminal telopeptide (urine NTX). Measurement of bone
density occurred at baseline, 12 months and 30 months; measurement of
urine and serum bone metabolism markers occurred at baseline, 3 months, 6
months, 12 months, 24 months and 30 months. Associations of bone
metabolism markers were evaluated both with bone density level (cross-
V
sectional associations over visits) and with bone density change
(longitudinal models) using time-specific mixed effects linear models.
Evaluation of collinearity, correlation structure and model assumptions were
performed to ensure model validity.
Results: The association of urine NTX with absolute levels of bone density
at all sites differed over the trial follow-up (all interaction p-value<0.0001).
Urine NTX was significantly inversely associated with density of lumbar
spine (p=0.0021 and p<0.0001), femoral neck (p=0.04 and p<0.0001) and
total hip (p<0.0001 and p<0.0001) in 12 months and 30 months. Higher
levels of urine NTX were significantly associated with greater rates of
decline at all measured bone sites (all interaction p-value<0.0001). The
association of osteoprotegrin with absolute levels of lumbar spine differed
over the trial (interaction p-value=0.02). Osteoprotegrin was significantly
positively associated with density of lumbar spine only in 30 months
(p=0.0005). The association of BAP with absolute levels of total hip differed
over the trial (interaction p-value=0.035). Higher levels of BAP were
associated with greater rates of decline in density of total hip (p=0.02).
SRANKL was not associated with bone density at any sites. Higher levels of
sRANKL were marginally associated with greater rates of decline in density
of lumbar spine (p=0.06).
VI
Conclusion: Osteoprotegrin was significantly positively associated with
bone density of the lumbar spine over the trial follow-up. BAP and urine
NTX were significantly negatively associated with bone density of the
lumbar spine, femoral neck and total hip over the trial follow-up. SRANKL
was not associated with bone density at any sites.
VII
INTRODUCTION
Osteoporosis and bone fracture related to loss of bone mineral
density are a serious issue among older people, especially among
postmenopausal women [1][2][3]. Thus research on bone mineral density
aims to improve the health of postmenopausal women, as bone density is
strongly related to both osteoporosis and bone fracture.
Bone density is commonly measured by dual energy X-ray
absorption scan (DXA scan) [4]. Grading of bone density ranges from
normal bone mass to osteoporosis. The World Health Organization has
applied T scores to define osteoporosis by comparing the bone density of the
patient to the bone density expected of a healthy, young female. T scores
less than -2.5 (2.5 standard deviations below the average of healthy, young
females) indicate osteoporosis [5][6]. T scores between -1.0 and -2.5 are
graded as osteopenia. T scores greater than -1.0 are considered as normal
bone density [7]. Research related to bone density commonly evaluates bone
density at the lumbar spine, femoral neck and total hip [1][8][9][10].
Numerous intervention studies have evaluated the effect of different
interventions on bone density. Some studies, primarily randomized trials,
have tested the effect of soy and soy isoflavones on bone density loss. While
1
some of these studies claimed isoflavones prevented bone loss [11][12],
other studies failed to demonstrate a benefit [13][14].
The level of bone mass is influenced by the two processes of bone
resorption and bone formation. Bone formation occurs after bone injury and
during routine bone remodeling. Bone resorption is the process of breaking
down bone and releasing the minerals into the bloodstream [15]. The bone
density level is determined by the balance of bone formation and resorption,
which are under the coordination of bone-forming cells (osteoblasts) and
bone-resorbing cells (osteoclasts) [16]. Interventions to influence bone mass
either decrease bone resorption or increase bone formation. For example one
study concluded a protective effect of isoflavone supplementation was due
to reduction of bone resorption [10]. Since bone metabolism markers are
related to bone formation and resorption, understanding the associations
between bone density and bone metabolism markers is important for the
interpretation of intervention effects on bone density and bone metabolism
markers.
Serum receptor activator of nuclear factor kB levels (sRANKL) and
n-terminal telopeptide (urine NTX) are markers of bone resorption [17]. We
therefore expect an inverse association between these markers and bone
density [7][18]. Bone specific-alkaline phosphatase (BAP) is a bone
2
formation marker [19][20]. Some studies indicated osteoprotegrin as a
critical bone formation regulator, blocking osteoclast differentiation to result
in reduction of bone resorption and increase in bone mass [16][21][22].
Positive associations are therefore expected between bone density and BAP
and osteoprotegrin.
In the Women‘s Isoflavone Soy Health (WISH) Trial, the primary
outcome and objective was to evaluate the effects of soy isoflavones on
progression of subclinical atherosclerosis. Secondary outcomes tested the
intervention effects on bone density and bone metabolism markers. Our
objective was to use the longitudinally-collected data on bone density and
bone metabolism markers to evaluate the associations between levels of
bone metabolism markers and both absolute levels and change in bone
density, measured over 2.5 years.
METHOD
WISH was a randomized, double-blinded, placebo-controlled trial
conducted to test whether isoflavone soy protein (ISP) reduces progression
of subclinical atherosclerosis [23]. The trial included a recruitment period of
2 years and randomized treatment period of 2.5 years. Participants were
3
recruited from the greater Los Angeles area through media advertisement,
primarily by phone screening followed by a clinical screening visit.
Eligible participants were postmenopausal women without vaginal
bleeding for at least one year and serum estradiol <20pg/ml. Exclusion
criteria were: serum creatinine >2 mg/dL, fasting plasma triglycerides >500
mg/dL, diabetes mellitus, CVD, untreated thyroid disease, uncontrolled
hypertension, serious illness with prognosis <5 years, alcohol abuse >5
drinks daily, estrogen-based hormone therapy, and allergies of soy, nut or
related materials.
350 eligible participants were randomly assigned evenly into the ISP
group and placebo group within 2 strata of carotid artery intima-media
thickness (CIMT; <0.75 mm, >0.75 mm). Women randomized to ISP
received 25 g soy protein daily containing 45 mg genistein, 35 mg daidzein
and 5 mg glycitein, all in aglycone weight, with total isoflavones weight
respectively 80 mg, 60 mg and 10mg. Women randomized to placebo
received milk protein without isoflavones.
Participants were followed by clinic visits every month in the first 6
months and every other month after that for a 2.5-year period. For every
clinical visit, data collection included product compliance, dietary intake,
prescription and non-prescription medications not related to the trial
4
intervention, nutritional supplementation, vital signs and adverse events.
Plasma sample collection for isoflavone measurement, lipid measurement
and sample storage, and lifestyle surveys (smoking and alcohol use, physical
activity) were completed every 6 months. Carotid ultrasound to measure
CIMT was performed every 6 months. Bone density scans were conducted
at baseline, 12 and 30 months. Blood and urine samples were collected for
bone metabolism markers at baseline, 3, 6, 12, 24 and 30 months.
Bone Density Sub-Study
While the primary WISH trial outcome was carotid artery intima-
media thickness (CIMT) to test the effect of ISP on progression of
subclinical atherosclerosis, the outcomes of BMD of the lumbar spine (L1-
L4), femoral neck and total hip for bone density and measurements of blood
and urine for bone metabolism markers were also collected as secondary
outcomes in the WISH trial. To be included in this sub-study, participants
should have bone density measurement by DXA and bone metabolism
marker measurements from blood and urine collection for the same visit
month. A total of 349 subjects were included in this sub-study; 1 subject was
excluded due to missing bone density baseline data.
5
Study Variables and Measurement
Variables included in this sub-study were bone density of the lumbar
spine (L1-L4), femoral neck and total hip, measured by dual-energy x-ray
absorptiometry (DXA) bone scans using a Hologic Bone Densitometer
conducted at baseline, 12 and 30 months.
Blood and urine samples were collected to measure bone metabolism
markers at baseline, 3, 6, 12, 24 and 30 months. Serum levels of bone
metabolism markers included serum receptor activator of nuclear factor kB
levels (sRANKL), osteoprotegrin (OPG) and bone specific-alkaline
phosphatase (BAP). Urine measures included n-terminal telopeptide (urine
NTX). The assessment of sRANKL was conducted by measuring the
generation of formazan. Plasma was added to a microtiter plate with OPG
then biotinylated anti-sRANKL. An enzyme cycling system was used for the
signal to enhance the color level generated by alkaline phosphatase (AP)
linked to the antibody; AP converted NADPH to NADH. Two enzymes,
diaphorase and alcohol dehydrogenase, were then added to convert NADH
to NAD to generate formazan. Osteoprotegrin was measured by a
sandwiched ELISA kit; the capture antibody is a monoclonal IgG antibody
and the detection antibody was a biotin-labeled polyclonal antihuman OPG.
BAP was measured by an immunoassay kit; the monoclonal anti-BAP
6
antibody coated strip detected BAP in the sample. For the purpose of
binding sites of a monoclonal antibody labeled with horseradish peroxidase,
n-terminal telopeptide in the specimen competed with the solid phase n-
terminal telopeptide. Urinary creatinine analysis was used for urinary
dilution and urine NTX was expressed in nMol BCE (bone collagen
equivalents per liter)/mMol creatinine per liter.
Statistical Analysis
The primary purpose of this sub-study was to assess the longitudinal
association between bone density and bone metabolism markers.
Collinearity among bone metabolism markers was assessed by Pearson
correlations. Lumbar spine, femoral neck and total hip bone density were
used as dependent variables in separate linear regression models. Bone
metabolism markers were included as independent variables to test
associations with bone density. Due to the repeated measurements of both
dependent and independent variables over the trial and the fact that the
independent variables changed over time, time-specific mixed effects linear
models were used. Mixed effects linear models used a combination of fixed
and random effects to assess the association between bone density and bone
7
metabolism markers. All models specified the model intercept as a random
effect. Visit time was included as a categorical variable.
For the time-specific mixed effects linear model, comparison of four
commonly used correlation structures was conducted to determine the most
appropriate correlation structure for the repeated bone density measures.
Correlation structures that were evaluated included: unstructured (UN),
compound symmetry (CS), simple and first-order autoregressive (AR(1)).
An unstructured correlation structure has no constraint for the covariance
matrix, uniquely estimating each element of the covariance matrix. Thus, an
unstructured correlation structure can obtain the best fit for the correlations
among repeated measurements, at a cost of higher degrees of freedom (more
parameter estimates than more restrictive correlation structures). A
compound symmetry correlation structure assumes the correlation among
repeated measures is constant regardless of how far apart the measurements
are. A simple correlation structure assumes repeated measures are not
correlated, even on the same subject. A first-order autoregressive correlation
structure assumes homogeneous variances and correlations among repeated
measurement that decline exponentially with distance (i.e. time between
measurements). Criteria for the selection of the appropriate correlation
structure to be used in the mixed linear models were Akaike information
8
criterion (AIC), Bayesian information criterion (BIC), and adjusted Akaike
information criterion (AICC) scores and number of parameters; these are
goodness of fit measures that indicate the comparative fit of the model. The
correlation structure with the smaller AIC, BIC and AICC values and
number of parameters indicated smaller residual error in the model. After
implementing this comparison, the compound symmetry correlation
structure (CS) was used in the mixed linear models.
A random intercept model was used to specify the subject-specific
(within-subject) variability in bone density. The three bone density variables
(lumbar spine, femoral neck and total hip) were modeled separately as
dependent variables. Metabolism markers sRANKL, OPG, BAP and urine
NTX were included as independent variables; these were first modeled
separately and later in joint models to assess the independent associations of
bone markers with bone density. Visit month (at baseline, 12 months and 30
months) was included as a categorical variable. Age, race and marital status
were included in the models and evaluated as possible confounders of the
bone density-bone marker associations. A coefficient change (comparing
unadjusted and adjusted estimates) of the regression coefficients for
metabolism markers that was larger than 10% indicated confounding.
9
Associations of bone metabolism markers with absolute levels of
bone density (repeated cross-sectional associations evaluated over study
visits) and change in bone density (longitudinal models of bone density rate
of change) were analyzed. Association models with absolute BMD (bone
mass density) level tested the association between levels of bone density and
bone metabolism markers measured at the same study visit. Metabolism
markers used as independent variables included only those measured at the
same visit as BMD was measured (i.e., at baseline, 12 months, and 30
months); metabolism markers measured at 3 months, 6 months and 24
months were excluded from analysis. Trial visit was included as a class
variable, so that separate indicator variables were modeled for each visit.
Tests in these cross-sectional association models included a test for
difference in bone density over visits (main effect of visit) as well as the
overall association of bone markers with bone density. The simple model
assumed that these associations were equivalent across study visit. An
added interaction term of marker by visit tested whether the marker-bone
density association differed over visit.
For the longitudinal model, a continuous variable of years was
generated using the first bone measurement date of the trial and the bone
density visit date. The regression coefficient for years estimated the annual
10
rate of change in the BMD variable, measured over the 2.5-year trial period.
Longitudinal models tested: (1) whether the BMD change rate was different
from zero; (2) whether the rate of change in bone density was associated
with the bone metabolism markers. This second hypothesis was tested with
the addition of a years-by-bone marker interaction term.
Four model assumptions were evaluated to test the validity of the
model. The assumptions are linearity, independence, normality and equal
variance. The linearity assumption was graphically evaluated using a
scatterplot of model residuals versus model-predicted values; a flat scatter of
model-predicted values around the zero residual line indicated that the
linearity assumption was satisfied. Independence was evaluated by
researching the data source and data collection scheme. At the individual
level if there were independent subjects (e.g. no between-subject correlations,
such as family members), then independence could be assumed to hold.
Normality was measured by a histogram of model residuals and the Shapiro-
Wilk test for normality of model residuals; a p-value larger than 0.05
indicated normality. Because of non-normality of residuals using
untransformed dependent variables, log transformation of the bone density
variables was applied to all model results to satisfy the normality assumption.
11
The equal variance assumption was measured by the scatter plot of
independent variables against model residuals.
For the 349 subjects included in the sub-study, some missing data
existed. Importantly, 69 subjects were missing the measurement of
sRANKL at the baseline. Based on a previous study, mixed linear models
can accommodate missing data in longitudinal data [24]. SAS 9.4 was used
for all analyses; mixed effects models used the proc mixed procedure. All
hypothesis testing was conducted at a two-sided 0.05 level.
RESULTS
Participants
350 postmenopausal women were randomized to (ISP) isoflavone
soy protein or placebo groups in the Women’s Isoflavone Soy Health (WISH)
trial. One participant was excluded from this analysis due to missing
baseline bone density data. Table 1 summarizes the characteristics of the
study subjects at baseline. The average age of the 349 subjects was 60.6
years.
Table 2 shows the distribution of observations in the analysis sample
by study visit. The analysis dataset used the joint datasets of serum markers,
12
urine markers and bone density. While per-protocol bone density
measurements occurred at baseline, 12 months and 30 months, some
subjects did not have their bone density assessments at the protocol-
specified visit month. Thus some bone density assessments occurred in
months other than 0, 12 or 30; these were included in the nearest protocol-
specified months in the bone density dataset.
Table 1. Baseline Characteristics of Analysis Sample (n=349)
Characteristics
1
Age, mean(range),(year) 60.6(44-91)
Race, n (%)
White non-hispanic 223(63.9)
Black non-hispanic 20(5.7)
Hispanic 55(15.8)
Asian or Pacific Islander 38(10.9)
Other 13(3.7)
Marital Status, n (%)
Single, never married 34(9.7)
Married, marital situation 197(56.5)
Separated 7(2.0)
Divorced 80(22.9)
Widowed 31(8.9)
Lumbar spine density, mean(SD), (g/cm
2
) 0.9(0.1)
Femoral neck density, mean(SD), (g/cm
2
) 0.7(0.1)
Total hip density, mean(SD), (g/cm
2
) 0.9(0.1)
Osteoprotegrin, mean(SD), (pmol/L) 3.6(1.3)
BAP, mean(SD), (µg/L) 20.4(7.5)
SRANKL, mean(SD), (pmol/L) 0.2(0.3)
Urine NTX, mean(SD), (nM BCE/mM creatinine) 44.2(23.1)
1 subject was excluded due to missing baseline bone density data. 70
1
missing
baseline sRANKL, 2 missing baseline femoral neck and total hip bone density, 3 missing
baseline lumbar spine bone density, 3 missing baseline urine NTX
13
Table 2. Analysis Dataset: Number of Bone
Density Outcomes by Trial Follow-up
Visit Frequency(%)
Analysis dataset
Baseline 349(38.44)
12 months 287(31.61)
30 months 272(29.96)
Urine and serum dataset
Baseline 350(19.27)
3 months 311(17.13)
6 months 301(16.57)
12 months 293(16.13)
24 months 281(15.47)
30 months 280(15.42)
Bone density dataset
Baseline 349(38.35)
12 months 287(31.54)
30 months 273(30)
42 months 1(0.11)
Simple Associations among Bone Density and Bone Markers
Table 3 summarizes the simple associations among bone density and
markers using Pearson correlations. This is a descriptive exercise only since
we have not accounted for repeated measures. The baseline of the analysis
dataset was used for the analysis. The bone markers were relatively
uncorrelated. Although urine NTX was positively correlated with BAP, the
correlation was not of a magnitude indicative of collinearity (>0.8 indicating
14
a potential risk of collinearity). Thus, the metabolism marker variables could
be modeled simultaneously.
Table 3. Pearson Correlation Coefficients among Bone Density
and Markers (n=349)
Osteoprotegrin BAP SRANKL Urine NTX
Osteoprotegrin
1 0.04
-0.08 -0.09
BAP
0.04 1 -0.02 0.4
SRANKL
-0.08 -0.02 1 0.1
Urine NTX
-0.09 0.4
0.1 1
Lumbar spine
0.08 0.07 -0.06 0.02
Femoral neck
-0.01 0.06 -0.08 0.03
Total hip
-0.03 0.1 -0.05 -0.01
BAP = bone specific − alkaline phosphatase
1
SRANKL= serum receptor activator of
nuclear factor kB levels Urine NTX= n-terminal telopeptide.
Change in Bone Density over Visits
Time-specific mixed effects linear models were applied to analyze
the association between bone density and bone metabolism markers.
Indicator variables for visit time were included in the model. Lumbar spine
density and all four metabolism markers were included as a test model to
evaluate correlation structures; these results are shown in Table 4. The
compound symmetry correlation structure had the smallest number of
parameters and small AIC, BIC and AICC values, relative to the other
correlation structures considered. The compound symmetry correlation
structure was therefore used in subsequent models.
15
Table 4. Comparison of AIC and Number of Parameters for Correlation Structures
Unstructured
Compound
Symmetry
Simple
First-order
Autoregressive
Akaike Information Criterion(AIC) -2513.8 -2462.2 -946 -2516.6
Adjusted AIC(AICC) -2513.8 -2462.2 -946 -2516.6
Bayesian Information Criterion(BIC) -2525.8 -2466.2 -948 -2520.6
Number of Parameters 6 2 1 4
Mixed linear model was used for the comparison.
1
Bone density lumbar spine was
included as dependent variable and bone markers osteoprotegrin, BAP, sRANKL and
urine NTX were included as independent variables. Model included intercept as a random
effect, and visit time as a fixed effect. Visit time was included as a categorical class
variable.
Table 5. Differences in Log Transformed Bone Density over Visit Times
Effect
Visit
Time
Estimate
Standard
Error
P-value
Lower
Limit
Upper
Limit
Type 3
Test P -
value
Dependent Variable: Lumbar Spine Bone Density
Intercept
-0.08 0.011 . . .
Visit Time
V-12 0.005 0.002 0.008 0.001 0.009
0.020
V-30 0.004 0.002 0.061 0.0002 0.008
Dependent Variable: Femoral Neck Bone Density
Intercept
-0.33 0.006 . . .
Visit Time
V-12 -0.003 0.001 0.018 -0.006 -0.001
<0.0001
V-30 -0.010 0.001 <.0001 -0.013 -0.007
Dependent Variable: Total Hip Bone Density
Intercept
-0.12 0.006 . . .
Visit Time
V-12 -0.002 0.001 0.15 -0.004 0.001
<0.0001
V-30 -0.008 0.001 <.0001 -0.01 -0.005
Mixed linear effects model was used, with a compound symmetry correlation structure.
1
Model included intercept as a random effect, and visit time as a fixed effect. Visit time
was included as a categorical class variable. The model estimates for visit time are the
mean differences in bone density from V-0.
Log transformations were applied to bone densities of lumbar spine, femoral neck and
2
total hip.
16
Visit times were included in the model to test for differences in bone
density over visits (Table 5). Density of the log transformed lumbar spine
(p=0.02), femoral neck (p<0.0001) and total hip (p<0.0001) differed among
the visits. Bone density of the log transformed lumbar spine increased while
bone density of the log transformed femoral neck and total hip decreased
over the trial.
Unadjusted Associations of Bone Density and Metabolism Markers over
Visits
Table 6. Log Transformed Bone density and Metabolism Marker
Associations: Measured over the Trial
Model
Coefficients
P-value
95% Confidence
Interval for β
β
Standard
Error
Lower
Bound
Upper
Bound
Dependent variable: Lumbar Spine Bone Density
SRANKL -0.0005 0.0025 0.83 -0.005 0.004
Osteoprotegrin 0.0046 0.0018 0.01 0.001 0.008
BAP -0.007 0.0016 0.0001 -0.01 -0.0045
Urine NTX -0.007 0.0017 <.0001 -0.01 -0.004
Dependent variable: Femoral Neck Bone Density
SRANKL -0.0004 0.0016 0.8 -0.0037 0.0028
Osteoprotegrin 0.0002 0.0013 0.89 -0.0023 0.0027
BAP -0.0059 0.0011 0.0001 -0.0080 -0.0038
Urine NTX -0.0032 0.0012 0.0065 -0.0055 -0.0009
Dependent variable: Total Hip Bone Density
SRANKL -0.0019 0.0015 0.2 -0.0048 0.0010
Osteoprotegrin 0.0016 0.0012 0.17 -0.0007 0.0038
BAP -0.0052 0.001 0.0001 -0.0071 -0.0033
Urine NTX -0.0046 0.001 0.0001 -0.0066 -0.0025
17
Mixed effects linear models were used, with a compound symmetry correlation
1
structure. Model included intercept as a random effect, and visit time as a fixed effect.
Visit time was included as a categorical variable. β coefficients are interpreted as change
in log bone density (in g/cm
2
) per unit change in independent variable bone markers.
Log transformations were applied to bone densities of lumbar spine, femoral neck and
2
total hip.
Table 6 shows the results of unadjusted associations between log
transformed bone density levels and bone metabolism markers. Each of the
bone markers was modeled separately. Osteoprotegrin was significantly
positively associated with log transformed lumbar spine bone density only
(p=0.01). BAP and urine NTX were significantly inversely correlated with
log transformed bone density at all sites. SRANKL was not associated with
log transformed bone density at any site.
18
Table 7. Associations between Log Transformed Lumbar Spine and Bone Markers
by Study Visit
Visit Time
Coefficients
P-value
95% Confidence
Interval for β
P-value for
Interaction
Term
2
β
1
Standard
Error
Lower
Bound
Upper
Bound
Metabolism Marker: SRANKL
0 months -0.002 0.0028 0.45 -0.0075 0.0034
0.15 12 months
-0.0007
0.0027 0.8 -0.006 0.0046
30 months
0.0026 0.0029
0.38
-0.0032 0.0085
Metabolism Marker: Osteoprotegrin
0 months 0.0042 0.0024 0.07 -0.0004 0.0089
0.022 12 months 0.0009 0.002 0.68 -0.003 0.005
30 months 0.0076 0.002 0.0005 0.0033 0.012
Metabolism Marker: BAP
0 months -0.0045 0.0023 0.051 -0.009 0.00001
0.12 12 months -0.0088 0.0022 <0.0001 -0.013 -0.0045
30 months -0.008 0.0017 <0.0001 -0.012 -0.005
Metabolism Marker: Urine NTX
0 months -0.0025 0.0021 0.22 -0.0066 0.0016
<0.0001 12 months -0.0071 0.0023 0.0021 -0.012 -0.0026
30 months -0.016 0.0026 <0.0001 -0.02 -0.01
Mixed effects linear models were used, with a compound symmetry correlation
1
structure. Model included intercept as a random effect, and visit time as a fixed effect.
Visit time was included as a categorical variable. Associations between bone density
lumbar spine and bone markers at each visit were tested. β coefficients are interpreted as
change in log bone density (in g/cm
2
) per unit change in independent variable bone
markers at each visit(baseline, 12 months, 30 months).
Interaction term between the metabolism marker and visit time, testing whether
2
association with bone density differ over trial.
Log transformation was applied to lumbar spine density.
3
Table 7 summarizes the results for models testing whether
associations between log transformed lumbar spine and bone metabolism
markers differ over visit time. Associations with absolute levels of lumbar
spine differed over the trial for osteoprotegrin (p-value for marker by time
19
interaction = 0.022) and for urine NTX (p-value for marker by time
interaction = 0.0001).
Table 8. Associations between Log Transformed Femoral Neck and Bone Markers
by Study Visit
Visit Time
Coefficients
P-value
95% Confidence
Interval for β
P-value for
Interaction
Term
2
β
1
Standard
Error
Lower
Bound
Upper
Bound
Metabolism Marker: SRANKL
0 months -0.0014 0.0025 0.58 -0.0064 0.0
12 months
0.0006
0.0024 0.79 -0.004 0.0055 0.49
30 months
-0.0016 0.0027
0.55 -0.007 0.0037
Metabolism Marker: Osteoprotegrin
0 months 0.0027 0.002 0.22 -0.002 0.007
12 months -0.0009 0.002 0.68 -0.005 0.003 0.22
30 months 0.0007 0.002 0.73 -0.003 0.005
Metabolism Marker: BAP
0 months -0.005 0.0021 0.009 -0.009 -0.001
12 months -0.006 0.002 0.0006 -0.01 -0.003 0.18
30 months -0.009 0.002 <0.0001 -0.012 -0.006
Metabolism Marker: Urine NTX
0 months -0.0005 0.002 0.79 -0.008 0.0005
0.0001 12 months -0.004 0.02 0.04 -0.009 -0.0003
30 months -0.015 0.002 <0.0001 -0.019 -0.01
Mixed effects linear models were used, with a compound symmetry correlation
1
structure. Model included intercept as a random effect, and visit time as a fixed effect.
Visit time was included as a categorical variable. Associations between bone density
femoral neck and bone markers at each visit were tested. β coefficients are interpreted as
change in log bone density (in g/cm
2
) per unit change in independent variable bone
markers at each visit(baseline, 12 months, 30 months).
Interaction term between the metabolism marker and visit time, testing whether
2
association with bone density differ over trial.
Log transformation was applied to femoral neck density.
3
20
Table 8 summarizes the results for models testing whether
associations between log transformed femoral neck and bone metabolism
markers differ over visit time. Associations with absolute levels of femoral
neck differed over the trial only for urine NTX (p-value for marker by time
interaction = 0.0001).
Table 9. Associations between Log Transformed Total Hip Density and Bone
Markers by Study Visit
Visit Time
Coefficients
P-value
95% Confidence
Interval for β
P-value for
Interaction
Term
2
β
1
Standard
Error
Lower
Bound
Upper
Bound
Metabolism Marker: SRANKL
0 months -0.003 0.002 0.16 -0.006 0.0011
0.69 12 months
-0.002
0.002 0.28 -0.006 0.002
30 months
-0.001 0.002
0.55
-0.005 0.003
Metabolism Marker: Osteoprotegrin
0 months 0.0034 0.0016 0.047 0.00005 0.0067
0.09 12 months -0.000003 0.0016 0.99 -0.003 0.003
30 months 0.0025 0.0015 0.1 -0.0005 0.0055
Metabolism Marker: BAP
0 months -0.0034 0.0016 0.03 -0.006 -0.0003
0.035 12 months -0.0065 0.0015 <0.0001 -0.009 -0.0035
30 months -0.007 0.0012 <0.0001 -0.009 -0.005
Metabolism Marker: Urine NTX
0 months -0.0003 0.001 <0.85 -0.003 0.002
0.0001 12 months -0.006 0.002 <0.0001 -0.009 -0.003
30 months -0.0014 0.0017 <0.0001 -0.017 -0.01
Mixed effects linear models were used, with a compound symmetry correlation
1
structure. Model included intercept as a random effect, and visit time as a fixed effect.
Visit time was included as a categorical variable. Associations between bone density total
hip and bone markers at each visit were tested. β coefficients are interpreted as change in
log transformed bone density (in g/cm
2
) per unit change in independent variable bone
markers at each visit(baseline, 12 months, 30 months).
21
Interaction term between the metabolism marker and visit time, testing whether
2
association with bone density differ over trial.
Log transformation was applied to total hip density.
3
Table 9 summarizes the results for models testing whether
associations between log transformed total hip and bone metabolism markers
differ over visit time. Associations with absolute levels of total hip differed
over the trial for BAP (p-value for marker by time interaction = 0.035) and
for urine NTX (p-value for marker by time interaction = 0.0001).
Associations of Bone Metabolism Markers with Longitudinal Change in Bone
Density
Dates of enrollment into the trial and visit dates were applied to
generate the follow-up time variable “year”, which indicated the time period
measured in unit years from the date of first bone density measurement to
the specific visit date for each subject. Table 10 displays the estimates of
annual change in log transformed bone density as dependent variables in the
mixed effects linear models. Log transformed lumbar spine density
increased at a rate of 0.0014 g/cm
2
annually (p=0.06), with 95% CI (-0.0001,
0.003). Log transformed femoral neck density decreased at a rate of 0.0038
g/cm
2
annually (p<0.0001), with 95% CI (-0.0049, -0.0027). Log
22
transformed total hip density also decreased at an annual rate of 0.003 g/cm
2
annually (p<0.0001), with 95% CI (-0.004, -0.002).
Table 10. Annual Rate of Change in Log Transformed Bone Density
Model
Coefficients
P-value
95% Confidence
Interval for β
β
1
Standard
Error
Lower
Bound
Upper
Bound
Dependent Variable: Lumbar Spine Bone Density
Intercept -0.075 0.0077 . . .
Years on study
2
0.0014 0.0008 0.06 -0.0001 0.0030
Dependent Variable: Femoral Neck Bone Density
Intercept -0.325 0.008 . . .
Years on study -0.0038 0.0006 <.0001 -0.006 -0.003
Dependent Variable: Total Hip Bone Density
Intercept -0.115 0.0072 . . .
Years on study -0.0030 0.0005 <.0001 -0.004 -0.002
Mixed effects linear models were used, with a compound symmetry correlation
1
structure. Model included intercept as a random effect. β coefficients for year on study
are interpreted as annual change in bone density (g/cm
2
per year).
Year length measured as the time period from baseline bone mass measurement to
2
specific visit time for each subject.
Log transformation was applied to bone densities of lumbar spine, femoral neck and
3
total hip.
23
Table 11. Associations of Bone Markers with Rate of Log Transformed Bone
Density Change
Model
Coefficients
P-value
95% Confidence
Interval for β
β
Standard
Error
Lower
Bound
Upper
Bound
Dependent Variable: Lumbar Spine Bone Density
SRANKL*Year
2
0.0018 0.0009 0.06 -0.0001 0.0033
Osteoprotegrin*Year 0.0014 0.0008 0.11 -0.0002 0.0031
BAP*Year -0.001 0.0008 0.15 -0.0026 0.0007
Urine NTX*Year -0.004 0.0009 <0.0001 -0.006 -0.0022
Dependent Variable: Femoral Neck Bone Density
SRANKL*Year -0.0001 0.0006 0.83 -0.0014 0.0011
Osteoprotegrin*Year -0.0006 0.0006 0.43 -0.002 0.0009
BAP*Year -0.0011 0.0006 0.07 -0.003 0.0001
Urine NTX*Year -0.004 0.001 <.0001 -0.006 -0.003
Dependent Variable: Total Hip Bone Density
SRANKL*Year 0.0005 0.0006 0.42 -0.0007 0.0017
Osteoprotegrin*Year -0.0003 0.0006 0.63 -0.0015 0.0009
BAP*Year -0.0013 0.0005 0.02 -0.0023 -0.0002
Urine NTX*Year -0.005 0.0006 <.0001 -0.006 -0.0033
Mixed effects linear models were used, with a compound symmetry correlation
1
structure. Model included intercept as a random effect. β coefficients are interpreted as
rate of change in bone density per unit change in bone marker variables.
Interaction term between sRANKL and year length on study.
2
Log transformation was applied to bone densities of lumbar spine, femoral neck and
3
total hip.
Table 11 shows the associations of bone metabolism markers with rate
of change in log transformed bone density variables; these associations were
tested with the interaction term of marker with year on-study. The
interaction term for urine NTX with follow-up time indicated significant
inverse associations with rates of change in log transformed density of the
lumbar spine (p<0.0001), femoral neck (p<0.0001) and total hip (p<0.0001);
higher levels of urine NTX were associated with greater decline rates at all
24
measured bone sites. In addition, BAP was inversely associated with rate of
change in the total hip (interaction p-value = 0.02). The interaction term for
sRANKL with follow-up time indicated marginally significant positive
associations with rates of change in density of the lumbar spine (p=0.06).
Osteoprotegrin was not associated with rates of change in bone density at
any site.
25
Multivariable Models: Associations of Log Transformed Bone Density Levels
with Bone Markers over Visits
Table 12. Multivariable Model: Associations with Bone Density Levels
Model Visit
Coefficients
P-value
95%
Confidence
Interval for β
P-value
for Type
3 Test
β
Standard
Error
Lower
Bound
Upper
Bound
Dependent Variable: Lumbar Spine Bone Density
Intercept
-0.0595 0.0095
BAP
-0.0049 0.0017 0.0033 -0.0077 -0.0082 0.0033
Visit Time V-12 0.0113 0.005 0.025 0.0014 0.021
0.0002
Visit Time V-30 0.021 0.005 0.0001 0.011 0.031
Urine NTX* Visit Time V-00 -0.0015 0.002 0.0001 -0.006 0.003
<0.0001
Urine NTX* Visit Time V-12 -0.0055
0.002 0.018 -0.01 -0.001
Urine NTX* Visit Time V-30 -0.014 0.003 <0.0001 -0.02 -0.008
Dependent Variable: Femoral Neck Bone Density
Intercept
-0.3118 0.01
BAP
-0.006 0.0015 0.0001 -0.009 -0.003 0.0001
Visit Time V-12 -0.0008 0.005 0.99 -0.009 0.009
0.34
Visit Time V-30 0.0056 0.0046 0.23 -0.003 0.015
Urine NTX* Visit Time V-00 0.0007 0.0019 0.0074 -0.003 0.0044
<0.0001
Urine NTX* Visit Time V-12 -0.003
0.002 0.24 -0.007 0.0017
Urine NTX* Visit Time V-30 -0.012 0.002 <0.0001 -0.016 -0.007
Dependent Variable: Total Hip Bone Density
Intercept
-0.1065 0.008
BAP
-0.004 0.001 0.0004 -0.006 -0.0018 0.0004
Visit Time V-12 0.008 0.003 0.022 0.001 0.014
0.0107
Visit Time V-30 0.0098 0.003 0.0036 0.003 0.016
Urine NTX* Visit Time V-00 0.0006 0.0014 <0.0001 -0.002 0.003
Urine NTX* Visit Time V-12 -0.0055 0.0015 0.0005 -0.009 -0.002
<0.0001
Urine NTX* Visit Time V-30 -0.012 0.002 <0.0001 -0.015 -0.008
Mixed effects linear models were used, with a compound symmetry correlation
1
structure. Model included intercept as a random effect, and visit time as a fixed effect.
Visit time was included as a categorical variable. Associations between bone density total
hip and bone markers at each visit were tested. Log transformations for dependent
26
variables bone density were applied. β was the model coefficient interpreting as change in
log bone density per unit change in independent variables bone markers or visit times.
Estimate of association between bone density and markers at each visit.
2
Based on the prior results for log transformed bone density and
individual bone metabolism markers, multivariable models evaluating
associations of multiple bone metabolism markers with absolute levels of
log transformed bone density were evaluated. Table 12 summarizes the
results of the multivariable modeling. Inclusion and exclusion criteria for
variable selection were p-value less than 0.10. The association of urine
NTX remained significant on multivariate modeling at all three sites. The
interaction term of urine NTX with visit time also remained significant in all
three multivariable models (interaction p-value<0.0001), indicating that the
association of urine NTX with absolute levels of density of lumbar spine,
femoral neck and total hip differed over visits. The inverse association of
BAP remained significant on multivariate modeling at all three sites:
density of lumbar spine (p=0.0033), femoral neck (p=0.0001) and total hip
(p=0.0004).
Age, race and marital status were included in the three multivariate
models to consider as potential confounders. None of these variables
changed regression coefficient estimates of the marker variables by more
27
than 10%, indicating that age, race and marital status were not confounders
of these associations.
Linearity, independence, normality and equal variance model
assumptions for the three multivariable models were also tested to check the
validity of the models. As the subjects from the WISH trial are not related
and they received treatments as independent individuals, the independence
assumption was considered satisfied.
Figures 1-3 (Appendix) show the scatterplot of model residuals and
model-predicted values of bone density aimed to test the linearity
assumption of the three multivariable models. The residuals remained flat
around the zero residual line across predicted levels of bone density,
indicating satisfaction of the linearity assumption.
Figures 4-7 (Appendix) show the distribution of the residuals. Figure
4 shows the residual distribution of the lumbar spine model, indicating
violation of normality. Thus a log transformation of lumbar spine density
was performed to satisfy the normality assumption in figure 5. Figures 5-7
indicate satisfaction of the normality assumption for density of lumbar spine,
femoral neck and total hip after log transformation. Log transformation was
applied to all model results to satisfy the normality assumption.
28
Figures 8-10 (Appendix) show the scatterplot of standardized urine
NTX and residuals of the multivariable models. The dots remained in the
same trend indicating satisfaction of the equal variance assumption.
Multivariable Models: Associations of Bone Markers with Rate of
Change in Log Transformed Bone Density
Table 13. Multivariable Models: Associations with Rate of Change in
Bone Density
Model
Coefficients
P-value
95% Confidence
Interval for β
β
Standard
Error
Lower
Bound
Upper
Bound
Dependent Variable: Lumbar Spine Bone Density
Intercept -0.0672 0.011
Years On-study 0.0073 0.0018 <0.0001 0.0004 0.01
BAP -0.006 0.0016 <.0001 -0.009 -0.003
Osteoprotegrin 0.003 0.0018 0.053 -0.00005 0.0069
Urine NTX -0.0006 0.0021 0.77 -0.0046 0.0035
Urine NTX*year -0.0038 0.001 0.0002 -0.0058 -0.0018
Dependent Variable: Femoral Neck Bone Density
Intercept -0.3122 0.009
Years On-study 0.002 0.0017 0.24 -0.0013 0.0053
BAP -0.006 0.0015 <.0001 -0.009 -0.003
Urine NTX 0.0021 0.0019 0.27 -0.0016 0.0058
Urine NTX*year -0.004 0.0009 <0.0001 -0.0062 -0.0025
Dependent Variable: Total Hip Bone Density
Intercept -0.1042 0.008
Years On-study 0.0032 0.0012 0.0096 0.0008 0.006
Urine NTX 0.001 0.0014 0.48 -0.002 0.0037
Urine NTX*year -0.0044 0.0007 <0.0001 -0.006 -0.003
BAP -0.0046 0.001 <0.0001 -0.0067 -0.0024
Mixed effects linear models were used, with a compound symmetry correlation
1
structure. Model included intercept as a random effect. Visit time was included as a
categorical variable. Log transformations were performed for the bone density dependent
29
variables. β was the model coefficient interpreting as change in log bone density per unit
change in independent variables bone markers or year in study.
Based on the prior results for log transformed bone density and
individual bone metabolism marker, longitudinal multivariable models
including multiple bone metabolism markers were evaluated. Table 13
summarizes the results of the longitudinal multivariable modeling. BAP
remained significant on multivariate modeling in the inverse association
with level of log transformed density of lumbar spine (p<0.0001), femoral
neck (p<0.0001) and total hip (p<0.0001). In addition, osteoprotegrin was
marginally significantly positively associated with absolute log transformed
density of the lumbar spine (p=0.053). The interaction term of urine NTX
with year in study remained significant in all three multivariable models,
indicating significant inverse associations with rates of change in log
transformed density of the lumbar spine (p=0.0002), femoral neck (p<0.0001)
and total hip (p<0.0001).
Model assumptions were also tested for the validity of the
longitudinal multivariable model.
Figures 11-13 (Appendix) show the scatterplot of model residuals
and model-predicted values of bone density aimed to test the linearity
assumption of the three multivariable models. The residuals remained flat
30
around the zero residual line over values of predicted bone density,
indicating satisfaction of the linearity assumption.
Figure 14-16 (Appendix) show the distribution of the residuals,
indicating approximate normality after log transformation.
Figure 17-19 (Appendix) show the scatterplot of standardized urine
NTX and residual of the multivariable models. The dots remained in the
same trend indicated satisfaction of the equal variance assumption.
DISCUSSION
This analysis from the Women’s Isoflavone Soy Health (WISH) trial
indicated associations between bone density dependent variables and
individual bone metabolism markers. Specifically, analysis of association of
bone markers with absolute levels of bone density indicated that
osteoprotegrin was significantly positively associated with lumbar spine
density. BAP and urine NTX were significantly inversely correlated with
bone density of the lumbar spine, femoral neck and total hip. Consistent
directions of association were found in all bone density models, showing
that osteoprotegrin was positively associated with bone density, while
sRANKL, BAP and urine NTX were inversely associated with bone density.
31
However, sRANKL was not significantly associated with levels of bone
density at any site.
Positive associations were expected between BAP, osteoprotegrin
and bone density variables [18][16][19][20]. Based on previous study one
very small cohort study included 27 patients (13 males and 14 females; age
20-69 years) for a 12-month assessment in bone density sites of lumbar
spine and femoral neck, indicated a positive association between BAP and
bone density [18] Negative associations were expected in sRANKL and
urine NTX [25]. All results for the study above were consistent with
expected associations except BAP. Inconsistency in the direction of the BAP
association with bone density in our study compared to previous studies may
due to confounding effects of unmeasured bone resorption factors. One
study stated that bone mass of the lumbar spine and femoral neck decreased
because bone resorption factors deoxypyridinoline (DPD) and pyridinoline
(PYD) increased more than the bone formation factor BAP, resulting in a
decrease in bone mass despite in an increase in BAP [18]. Further research is
needed to resolve these inconsistencies in the direction of the associations.
Bone density of the lumbar spine (p=0.02), femoral neck (p<0.0001)
and total hip (p<0.0001) significantly differed over visit time. Bone density
of the lumbar spine increased over visit while femoral neck and total hip
32
bone densities decreased over visit. This was consistent with our estimates
of longitudinal rates of change of bone density at each of these sites (Table
10). Osteoprotegrin was significantly positively associated with only lumbar
spine bone density (p=0.01). BAP and urine NTX were significantly
inversely correlated with bone density at all sites. SRANKL was not
significantly associated with bone density at any site. The inverse
association of urine NTX with absolute levels of bone density at all sites
differed over the trial follow-up, indicating a decreasing trend (i.e., the
inverse association between urine NTX and bone density became stronger
over the 30-month trial follow-up). In addition, the association of
osteoprotegrin with absolute levels of lumbar spine differed over the trial
(interaction p-value=0.02); the association of BAP with absolute levels of
total hip differed over the trial (interaction p-value=0.035).
For the longitudinal model results, lumbar spine increased annually
at a non-significant level (p=0.06) over the trial, while femoral neck and
total hip significantly decreased annually. Consistent with our analyses
associating bone markers with absolute levels of bone density, urine NTX
was significantly inversely associated with rates of change in densities of the
lumbar spine (p=0.0003), femoral neck (p<0.0001) and total hip (p<0.0001),
33
and BAP was inversely associated with rate of change in the total hip
(p=0.02).
Although missing data existed in these analyses, time-specific mixed
effects linear model can handle the missing data issue and deal with the
repeated measures correlation [24]. The validity of the model was assured by
selection of the correlation structures in mixed linear model, a check for the
collinearity of the metabolism markers and model assumptions for linearity,
independence, normality and equal variance for the multivariable models.
Associations for cross-sectional and longitudinal models, including
associations of individual bone markers with bone density and difference of
associations over visits or year in study were assessed to ensure the
completeness of the analysis. The limitation for this study is that some
subjects of the WISH trial could not make the visit on time, thus some visits
were scattered around the scheduled visit time for bone density (baseline, 12
months and 30 months). All the scattered visits were combined in the nearest
scheduled visit time (mostly within 6 months), which may lead to some
issues. There was also inconsistent direction of the association for the
metabolism marker BAP compared to that reported in previous studies.
Further research is required for these issues.
34
REFERENCES
[1]. Ettinger B, Black DM, Mitlak BH, et al. Reduction of Vertebral Fracture Risk in
Postmenopausal Women With Osteoporosis Treated With Raloxifene: Results
From a 3-Year Randomized Clinical Trial. Obstet Gynecol Surv. 2000;55:39-39.
[2]. Chesnut CH, Silverman S, Andriano K, et al. A Randomized Trial of Nasal Spray
Salmon Calcitonin in Postmenopausal Women with Established Osteoporosis:
The Prevent Recurrence of Osteoporotic Fractures Study. Am J Med.
2000;109:267-276.
[3]. Neer RM, Arnaud CD, Zanchetta JR, et al. Effect of Parathyroid Hormone (1-34)
on Fractures and Bone Mineral Density in Postmenopausal Women with
Osteoporosis. N Engl J Med. 2001;344:1434-1441.
[4]. Tenenhouse A, Joseph L, Kreiger N, et al. Estimation of the Prevalence of Low
Bone Density in Canadian Women and Men Using a Population-Specific DXA
Reference Standard: The Canadian Multicentre Osteoporosis Study (CaMos).
Osteoporosis Int. 2000;11:897-904.
[5]. Black DM. Fracture Risk Reduction with Alendronate in Women with
Osteoporosis: The Fracture Intervention Trial. Journal of Clinical Endocrinology
& Metabolism. 2000;85:4118-4124.
[6]. Kanis JA. Assessment of Fracture Risk and its Application to Screening for
Postmenopausal Osteoporosis: Synopsis of a WHO Report. Osteoporosis Int.
1994;4:368-381.
[7]. Oelzner P, Franke S, Lehmann G, et al. Soluble Receptor Activator of NFkappa
B-ligand and Osteoprotegerin in Rheumatoid Arthritis - Relationship with Bone
35
Mineral Density, Disease Activity and Bone Turnover. Clin Rheumatol.
2007;26:2127-2135.
[8]. Kenny AM, Kulldorff M, Kleppinger A, Prestwood KM. Ultralow-Dose
Micronized 17β-Estradiol and Bone Density and Bone Metabolism in Older
Women: A Randomized Controlled Trial. JAMA. 2003;290:1042-1048.
[9]. Vis M, Havaardsholm EA, Haugeberg G, et al. Evaluation of Bone Mineral
Density, Bone Metabolism, Osteoprotegerin and Receptor Activator of the NFκB
Ligand Serum Levels During Treatment with Infliximab in Patients with
Rheumatoid Arthritis. Ann Rheum Dis. 2006;65:1495-1499.
[10]. Kritz-Silverstein D, Goodman-Gruen DL. Usual Dietary Isoflavone Intake, Bone
Mineral Density, and Bone Metabolism in Postmenopausal Women. J Womens
Health Gend Based. 2002;11:69-78.
[11]. Potter SM, Baum JA, Teng H, Stillman RJ, Shay NF, Erdman J,J W. Soy Protein
and Isoflavones: Their Effects on Blood Lipids and Bone Density in
Postmenopausal Women. Am J Clin Nutr. 1998;68:1375S.
[12]. Atkinson C, Compston JE, Day NE, Dowsett M, Bingham SA. The Effects of
Phytoestrogen Isoflavones on Bone Density in Women: a Double-blind,
Randomized, Placebo-controlled Trial. Am J Clin Nutr. 2004;79:326.
[13]. Brink E, Coxam V, Robins S, et al. Long-term Consumption of Isoflavone-
Enriched Foods does not Affect Bone Mineral Density, Bone Metabolism, or
Hormonal Status in Early Postmenopausal Women: a Randomized, Double-blind,
Placebo controlled study. Am J Clin Nutr. 2008;87:761.
36
[14]. Hsu CS, Shen WW, Hsueh YM, Yeh SL. Soy Isoflavone Supplementation in
Postmenopausal Women. Effects on Plasma Lipids, Antioxidant Enzyme
Activities and Bone Density. J Reprod Med. 2001;46:221.
[15]. Teitelbaum SL. Bone Resorption by Osteoclasts. Science. 2000;289:1504-1508.
[16]. Simonet WS, Lacey DL, Dunstan CR, et al. Osteoprotegerin: A Novel Secreted
Protein Involved in the Regulation of Bone Density. Cell. 1997;89:309-319.
[17]. Garnero P, Hausherr E, Chapuy MC, et al. Markers of Bone Resorption Predict
Hip Fracture in Elderly Women: the EPIDOS Prospective Study. Journal of Bone
and Mineral Research : the Official Journal of the American Society for Bone and
Mineral Research. 1996;11:1531-1538.
[18]. Crosbie OM, Freaney R, McKenna MJ, Hegarty JE. Bone Density, Vitamin D
Status, and Disordered Bone Remodeling in End-Stage Chronic Liver Disease.
Calcif Tissue Int. 1999;64:295-300.
[19]. Eastell R, Delmas PD, Hodgson SF, Eriksen EF, Mann KG, Riggs BL. Bone
Formation Rate in Older Normal Women: Concurrent Assessment with Bone
Histomorphometry, Calcium Kinetics, and Biochemical Markers. J Clin
Endocrinol Metab. 1988;67:741-748.
[20]. Pepmueller P, Cassidy J, Allen S, Hillman L. Bone Mineralization and Bone
Mineral Metabolism in Children with Juvenile Rheumatoid Arthritis. Arthritis
Rheum. 1996;39:746-757.
[21]. Bucay N, Sarosi I, Dunstan CR, et al. Osteoprotegerin-deficient Mice Develop
Early Onset Osteoporosis and Rrterial Calcification. Genes Dev. 1998;12:1260-
1268.
37
[22]. Nakamura M, Udagawa N, Matsuura S, et al. Osteoprotegerin Regulates Bone
Formation through a Coupling Mechanism with Bone Resorption. Endocrinology.
2003;144:5441-5449.
[23]. Hodis HN, Mack WJ, Kono N, et al. Isoflavone Soy Protein Supplementation and
Atherosclerosis Progression in Healthy Postmenopausal Women: a Randomized
Controlled Trial. Stroke. 2011;42:3168-3175.
[24]. Krueger C, Tian L. A Comparison of the General Linear Mixed Model and
Repeated Measures ANOVA Using a Dataset with Multiple Missing Data Points.
Biol Res Nurs. 2004;6:151-157.
[25]. Doumouchtsis KK, Kostakis AI, Doumouchtsis SK, et al. Associations between
Osteoprotegerin and Femoral Neck BMD in Hemodialysis Patients. J Bone Miner
Metab. 2008;26:66-72.
38
APPENDIX
39
40
41
42
43
44
45
46
47
48
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Effect of soy isoflavones on anthropometric and metabolic measurements in postmenopausal women
PDF
Association between endogenous sex hormone levels and cognition: the Women’s Isoflavone Soy Health (WISH) trial
PDF
Associations between isoflavone soy protein (ISP) supplementation and breast cancer in postmenopausal women in the Women’s Isoflavone Soy Health (WISH) clinical trial
PDF
Associations between physical activities with bone mineral density in postmenopausal women
PDF
The relationship of resistin, leptin, adiponectin, and ghrelin with bone mineral density in healthy postmenopausal women: longitudinal analysis
PDF
Soy isoflavone supplements for the treatment of menopausal hot flashes: the Women’s Isoflavone Soy Health (WISH) trial
PDF
Associations between inflammatory markers and change in cognitive endpoints
PDF
Bone mineral density is associated with carotid atherosclerosis in healthy postmenopausal women: a longitudinal analysis of randomized clinical trial data
PDF
The associations of inflammatory markers with progression of subclinical atherosclerosis in early and late postmenopausal women
PDF
Effect of estradiol on circulating levels of inflammatory cytokines in postmenopausal women
PDF
Physical activity and sex hormone levels in postmenopausal women
PDF
Association of subclinical atherosclerosis with plasma B-vitamin, cysteine, homocysteine, and cysteinyl glycine in a cardiovascular disease-free population
PDF
The effects of hormone therapy on carotid artery intima-media thickness and serum lipids by ApoE4 genotype
PDF
Relationship of blood pressure and antihypertensive medications to cognitive change in the BVAIT, WISH, and ELITE clinical trials
PDF
A randomized controlled trial of hormone therapy on postmenopausal women’s quality of life
PDF
The effect of vitamin D supplementation on the progression of carotid intima-media thickness and arterial stiffness in elderly African American women: Results of a randomized placebo-controlled trial
PDF
The association between self-reported physical activity and cognition in elderly clinical trial participants
PDF
Relationship between progression of atherosclerosis and coagulation measures in a randomized-controlled trial
PDF
Effects of post-menopausal hormone therapy on arterial stiffness in the ELITE trial
PDF
Comparative study of laparoscopy vs. laparotomy for ovarian mass removal
Asset Metadata
Creator
Feng, Jun
(author)
Core Title
Association of bone density and bone metabolism markers: the Women’s Isoflavone Soy Health (WISH) trial
School
Keck School of Medicine
Degree
Master of Science
Degree Program
Biostatistics
Publication Date
09/17/2015
Defense Date
09/05/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
association,bone density,bone metabolism marker,isoflavone,OAI-PMH Harvest
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Mack, Wendy (
committee chair
), Azen, Stanley (
committee member
), Hodis, Howard (
committee member
)
Creator Email
fengjun798@gmail.com,junf@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c40-183821
Unique identifier
UC11273474
Identifier
etd-FengJun-3931.pdf (filename),usctheses-c40-183821 (legacy record id)
Legacy Identifier
etd-FengJun-3931.pdf
Dmrecord
183821
Document Type
Thesis
Format
application/pdf (imt)
Rights
Feng, Jun
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
bone density
bone metabolism marker
isoflavone