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
/
Methodological approaches to assessing diurnal cortisol rythms in epidemiological studies: how many salivary samples are necessary?
(USC Thesis Other)
Methodological approaches to assessing diurnal cortisol rythms in epidemiological studies: how many salivary samples are necessary?
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
METHODOLOGICAL APPROACHES TO ASSESSING DIURNAL CORTISOL
RHYTHMS IN EPIDEMIOLOGICAL STUDIES: HOW MANY SALIVARY
SAMPLES ARE NECESSARY?
by
Carrie Joy Donoho
A Thesis Presented to the
FACULTY OF THE KECK SCHOOL OF MEDICINE OF USC
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(APPLIED BIOSTATISTICS AND EPIDEMIOLOGY)
August 2012
Copyright 2012 Carrie Joy Donoho
ii
Table of Contents
List of Tables iii
List of Figures iv
Abstract v
Introduction 1
Method 6
Results 11
Discussion 14
References 16
Appendix: Tables and Figures 18
iii
List of Tables
Table
1:
Characteristics
of
the
analytic
sample
for
NSDE
participants
with
cortisol
measures
obtained
on
Wednesday.
18
Table
2:
Growth
curve
models
by
gender
without
covariates
comparing
models
without
and
with
the
before
lunch
cortisol
sample
19
Table
3:
Final
growth
curve
models
of
change
by
gender
with
covariates
using
3
data
points.
20
iv
List of Figures
Figure
1:
Theoretical
diurnal
cortisol
rhythm
21
Figure
2:
Mean
cortisol
levels
and
95%
confidence
intervals
for
each
sample
22
Figure
3:
Distribution
of
person-‐time
for
each
cortisol
sample
23
Figure
4:
Distribution
of
person-‐time
for
the
‘pre-‐lunch’
sample
24
Figure
5:
Distribution
of
clock-‐time
for
the
‘pre-‐lunch’
sample
25
Figure
6:
Growth
models
of
cortisol
using
3
data
points
versus
4
data
points
at
various
time
points
with
cubic
functions
of
time
for
men
and
women
26
v
Abstract
The integration of biomarkers into population-based epidemiological studies has
introduced methodological complexities that have yet to be clearly delineated. Cortisol is
a biomarker that is particularly complicated because it follows a circadian rhythm that is
based on individual wake and sleep times, as well as normal meal times. Thus,
understanding the timing of sampling is important to reduce costs of large studies. This
study used data from the Survey of Midlife in the United States to compare cortisol
rhythms using 3 sample times (waking, +30m, bedtime) and 4 sample times (waking,
+30m, pre-lunch, bedtime), in order to examine the use of a pre-lunch measure for
capturing diurnal cortisol rhythm. Growth curve models for repeated measures were used
to describe the change in cortisol across the day as well as variation across individuals.
Results indicate that including the pre-lunch sample of cortisol in the models does not
capture the expected curvilinear daily rhythm, whereas omitting this sample results in the
expected shape of the daily rhythm. The findings of this study suggest that the use of the
pre-lunch measure is more likely to be deviate from theoretical curve and lead to worse
model fitting Thus, the pre-lunch measure may introduce variability in the time that
samples are taken, as well as variability in levels of cortisol due to other unknown factors
such as normal meal times and food consumption.
1
Introduction
In the past decade there has been a strong movement to understand biomarkers of
aging, with a considerable emphasis placed on biomarkers that may link social,
behavioral, and psychological factors to diseases of aging (Harris, Gruenewald, &
Seeman, 2008). Because of this, many large-scale community and population-based
epidemiological studies have integrated biological measures into their studies, including
the Health and Retirement Study (HRS), the Survey of Health, Ageing, and Retirement in
Europe (SHARE), and the National Longitudinal Study of Adolescent Health (Add
Health). This has offered researchers a wealth of data to examine complex associations
between behavioral and social risk factors and pre-clinical disease. However, the addition
of these measures has also introduced methodological complexities that are only
beginning to be untangled.
One biomarker of particular interest is cortisol – an important hormone regulated
by the hypothalamic-pituitary-adrenal (HPA) axis. Cortisol has regulatory functions for
energy use and metabolism, as well as inflammatory/immune responses. The HPA axis
has been indicated as a major pathway linking social factors to aging (McEwen &
Seeman, 1999; Sapolsky, Krey, & McEwen, 1986). It is thought that psychosocial stress
activates certain regions of the brain (e.g., the hypothalamus), which results in increased
production of corticotropin releasing hormone (CRH). Increases in CRH signal for the
release of adrenocorticotropin hormone (ACTH) from the anterior pituitary, resulting in
increased production of glucocorticoids (cortisol in humans, corticosterone in non-human
animals) from the zona fasciculate of the adrenal cortex. Chronic exposure to
2
glucocorticoids has been shown to cause damage and dysfunction in the central nervous
system, as well as peripheral tissue damage (Sapolsky, Krey, & McEwen, 1986). Thus,
understanding cortisol secretory patterns in human populations has the potential to make
significant contributions to research on stress and aging.
The study of cortisol in human populations is complex because cortisol follows a
circadian rhythm that is based on individual wake and sleep times, as well as normal meal
times. A normal, healthy cortisol rhythm is thought to have a sharp morning increase that
peaks within an hour of waking, followed by a decline throughout the remainder of the
day, with increases around mealtimes, and a nadir in preparation for evening sleep (see
Figure 1; Kirshbaum & Hellhammer, 1989; Pruessner et al., 1997).
There are numerous studies identifying what appear to be pathological aberrations
from normal diurnal cortisol rhythms. Many of these studies suggest that hypoactive and
hyperactive HPA function are both signs of poor health. Blunted responses – indicated by
lower awakening responses and less decline in cortisol over the day – have been observed
among individuals with major depression and post-traumatic stress disorder (Yeheuda,
1996), as well among individuals with metastatic breast cancer (Abercrombie et al.,
2004). A recent meta-analysis found that higher cortisol awakening responses are
observed in individuals experiencing general life stress and that lower cortisol awakening
responses are observed in individuals with post-traumatic stress disorder and among
individuals with positive psychological characteristics such as happiness, optimism, and
self-esteem (Chida & Steptoe, 2009). These findings illustrate the fact that being either
over-reactive or non-reactive (blunted) may be symptomatic of underlying pathology.
3
Previous research findings relating cortisol to health are difficult to interpret
because in the past researchers have tended to focus on overall cortisol production over
the day, levels of cortisol at certain times of the day, or specific parts of the diurnal
rhythm, such as the awakening response or daily decline. Examining average levels over
a day typically requires 24-hour urine collection, which is not feasible for ambulatory
studies. In addition, averaging cortisol values over the day, or examining differences in
levels at one point in the day, is not appropriate for understanding the dynamics of the
cortisol rhythm, which have been shown to have important links to health (Kudielka,
Gierens, Hellhammer, Wüst, & Schlotz, 2012).
An additional complication in studying cortisol is that although cortisol
production follows a diurnal rhythm, it is also sensitive to acute changes in the
environment. Cortisol levels increase with the anticipation of food (Ott et al., 2011), the
experience of a stressful event (Dickerson & Kemeny, 2004; Kirschbaum, Pirke, &
Hellhammer, 1993), or injury (Kushner, 1982; Thompson, 2003). These acute responses
cause problems for researchers interested in studying cortisol in naturalistic settings
because they introduce variability in the diurnal rhythm that is difficult to account for
without creating considerable burden to participants by asking them to provide numerous
samples over the day, or maintain diaries on food intake, social interactions, and pain.
Laboratory studies indicate that hourly measures are desirable in order to fully
capture the complete curve and pulsatility patterns of cortisol (Young, Carlson, & Brown,
2001). However, providing numerous samples across the day is not feasible in
4
ambulatory epidemiological studies because it not only increases the burden on
participants, but also increases the cost of the study.
Theoretically and empirically, the cortisol awakening response (a.k.a. morning
rise) and the daily decline are the two most important aspects of the diurnal rhythm that
have been associated with health, therefore it is important to capture these two aspects of
the curve (Saxbe, 2008). Currently, several large-scale epidemiological studies have
obtained salivary cortisol measures; however, there is not currently a consensus on what
the best research protocol is to assess the diurnal rhythm. Several studies (reviewed in
Adam & Kumari, 2009), such the Chicago Health, Aging, and Social Relationships Study
(CHASRS) and the Cebu Longitudinal Health and Nutrition Survey obtain 3 measures
per day: upon waking, 30-60 minutes after waking, and at bedtime. Other studies, such as
Whitehall II and the Coronary Artery Risk Development in Young Adults Study
(CARDIA) obtained 6 samples per day: upon waking, 30-60 minutes, 2.5 hours, 8 hours,
and 12 hours after waking, as well as at bedtime. A more recent study, the National Study
of Daily Experiences (NSDE), employed a different methodology obtaining saliva
samples at waking, 30-60 minutes after waking, before lunch (not at a specific time), and
at bedtime. Although researchers have used these data for several empirical studies, there
has been little attention paid to differences in the methodologies.
One problem with using mid-day cortisol measures is that they are highly
influenced by routine eating patterns and the anticipation of food - not just food ingestion.
Among individuals that routinely eat lunch, cortisol begins to rise before lunch, in order
to prepare the body for a meal, and continues to rise for several minutes after the meal
5
(Quigley & Yen, 1979). However, when a snack is given before a regular meal, the
cortisol rise before lunch is not as pronounced (Follenius et al., 1982). For individuals
that have abnormal eating habits (i.e., do not eat lunch regularly), then a cortisol rise is
not observed during lunchtime, but is entrained to rise before a normally eaten meal
(Honma, Honma, & Hiroshige, 1984). Moreover, for individuals anticipating food
ingestion (i.e., told they will be given a meal, or presented a meal), but required to refrain
from eating, experience substantial rises in cortisol (Ott et al., 2011). Taken together, this
indicates that a substantial amount of variability in mid-day cortisol levels will be
influenced by regular eating behavior, food availability, dieting, and snacking – factors
that are unlikely to be controlled for by time sampling and are likely to be problematic
among middle-aged and older individuals with highly variable eating behaviors and meal
times. Thus, it is of interest to understand whether a cortisol measure taken before lunch
is meaningful, or if it introduces variability that may mask the true shape of the curve.
The purpose of the current study is to examine whether a pre-lunch sample of
cortisol adds relevant information to the diurnal cortisol rhythm, and if this measure
informs the curve or leads to misspecification of the daily rhythm. To do this, two
separate growth curve models (GCM) were developed and compared to investigate the
necessity of including pre-lunch cortisol measures in the model using empirical data from
the NSDE. The first GCM involved three salivary samples (wake, 30-60 minutes after
waking, and bedtime); while the second considered all four salivary samples (wake, 30-
60 minutes after waking, pre-lunch, and bedtime).
6
Method
The NSDE is a telephone daily diary study was conducted as part of the
MacArthur Foundation National Survey of Midlife in the United States (MIDUS), a
nationally representative sample of adults in the United States. For a detailed description
of the study, see Almeida, McGonagle, and King (2009). For the NSDE, a random
subsample of 3,600 MIDUS respondents were asked to participate, and a total of 2,022
agreed to participate. The age range of participants at the time of the study was 35 to 84.
Compared to MIDUS, the NSDE has a greater percentage of non-Hispanic Whites,
females, and individuals with a college education. The NSDE contains data from two
waves of the study, the first wave (NSDE I) was conducted in 1996, and the second
(NSDE II) was conducted from 2004 to 2006. For the purpose of the current study, we
used only data from the 2004-2006 wave, as this was the only wave in which salivary
cortisol samples were collected.
The NSDE II included short (10-20 minute) telephone interviews about daily
experiences on 8 consecutive days. On four of the days participants provided four
salivary samples (a total of 16 samples). Home collection kits were sent to participants
the week prior to the study start date. The collection kit included directions instructing
participants to take the samples at waking, 30 minutes after getting out of bed, before
lunch, and before bed. Participants were instructed to refrain from eating a large meal 1
hour before sampling, and were told to refrain from dairy products for 20 minutes before
sampling. Less than 3% of the samples were missed or unusable due to contamination.
The exact time of each sample was taken from telephone interviews by staff, and from a
7
paper-pencil log sent with the collection kit. Approximately 25% of the participants
received a “smart box” containing an electronic recording device that recorded the time
of the box opening and closing. Self-reported collection time was highly correlated (r
= .90) with electronic recorded time (Almeida, McGonagle, & King, 2009).
Salivary samples were taken four times per day (waking, 30 minutes after waking,
before lunch, and in the evening before bedtime) for four days. Participants used Starstet
salivettes (Nümbrecht, Germany), which are cotton rolls placed in the cheek until
saturated and subsequently stored in plastic tubes. Samples were returned by mail, in a
pre-addressed paid courier package, to the MIDUS Biological Core center and were
frozen at -60°C for shipping and storage. For analysis, salivettes were thawed and
centrifuged at 3,000 rpm for 5 minutes. Concentrations of cortisol were measured using a
luminescence immunoassay (Hamburg, Germany).
Covariates
Medication Use. Medication use was assessed by self reported use of over-the-
counter and prescription medications. Medications known to alter cortisol production
were recorded and include steroid inhaler, steroid medications, medications containing
cortisone, hormone medications (including birth control pills), anti-depressants and anti-
anxiety medications. A dichotomous indicator was created to indicate whether or not a
person was taking one or more of these medications (0 = current medication use; 1= no
medication use).
8
Smoking Status. A dichotomous indicator was created to indicate whether or not a
person smoked (0 = non-smoker; 1 = current smoker). An individual reporting any
smoking on 1 or more days of the 8-day study period was coded as a smoker.
Analytic Strategy
Cortisol rhythms are primarily driven by time since waking, and not standard
clock-time. Thus, time since waking (i.e., wake time = 0) was used as the measure of
time. Analyses were restricted to Wednesday because previous research indicates there is
a “weekend effect” where cortisol levels differ between weekdays and weekends.
Wednesday was the day of the week with the highest levels of participation in this sample
(N = 1038). Because previous studies indicate that cortisol values over the day are
partially dependent on a person falling asleep (when cortisol should be at its lowest
value), and waking (when cortisol will rise for 30-60 minutes), we eliminated individuals
from the analyses for the following reasons: (1) slept fewer than 4 hours the night before
or stayed awake longer than 20 hours during the day of sampling (n = 38), or (2) did not
take their morning sample within 1-hour of waking (n = 88), as this would limit the
ability to detect the morning cortisol rise. These restrictions result in a final analytic
sample of 912. Distributions of cortisol are typically skewed which was also observed in
this sample; therefore 1 was added to each cortisol value in order to eliminate negative
values, and natural-log transformations were used to normalize these distributions.
Descriptive characteristics of the final analytic sample are presented in Table 1.
Growth curve models for repeated measures were used to describe the change in
cortisol across the day as well as variation across individuals. Models were estimated by
9
gender. Models without covariates were first developed to establish the shape of the
cortisol curve across the day. Strong evidence indicates that diurnal cortisol rhythms are
curvilinear; therefore polynomial functions of time were explored. The growth models
assume that overall growth will follow a quadratic or cubic form, but do not examine
person-specific quadratic growth because this is not expected and this assumption also
introduces a great deal of complexity into an already complex model. Therefore
individual-specific linear trends are explored, and polynomial functions of random effects
are not included in the model. Maximum likelihood was used instead of restricted
maximum likelihood in order to facilitate significance tests between models using
likelihood ratio tests with changes fixed-effects parameters, where maximum likelihood
is appropriate. However, all models were also fit using restricted maximum likelihood to
ensure the standard errors did not change substantially.
A growth curve model, which can be conceptualized as a 2-level model, was
developed to examine variability in cortisol across the day as well as across individuals
(Chou, Bentler & Pentz, 1998). The Level-1, or within-individual model investigates the
variation of repeatedly measured cortisol within an individual with time as the predictor.
Because individuals may vary in the time of their sampling, time in hours since waking is
modeled as the independent variable. The Level-2, or between-individual, model
examines the variation across individuals. Equations of Level-1, Level-2, and Composite
equations are as follows:
Level-1 Model:
Cortisol
ij
= π
0i
+ π
1i
Time
ij
+ π
2i
Time
2
ij
+ π
3i
Time
3
ij
+ε
ij
10
Level-2 Model:
π
0i
= γ
00
+ ζ
0i
π
1i
= γ
10
+ ζ
1i
π
2i
= γ
20
π
3i
= γ
20
Composite Model:
Cortisol
ij
= γ
00
+ γ
10
Time
ij
+ γ
20
Time
2
ij
+ γ
20
Time
3
ij
+ (ε
ij
+ ζ
0i
+ ζ
1i
Time
ij
)
In the Level-1 model, Cortisol
ij
is individual i’s value of cortisol on occasion j and Time
ij
is the amount of time elapsed since waking for individual i on occasion j. The amount of
individual i’s cortisol that is unexplained on occasion j is ε
ij
. In level-2, π
0i
represents
individual i’s true mean cortisol value upon waking and ζ
0i
represents the amount that
individual i differs from the population mean (γ
00
), π
1i
represents individual i’s true
average rate of change in cortisol and ζ
0i
represents the amount that individual i’s rate of
change differs from the population average rate of change (γ
10
), π
2i
represents individual
i’s quadratic rate of change in cortisol, π
3i
represents individual i’s cubic rate of change in
cortisol.
Finally, time invariant covariates were entered into the model. Covariates
included wake time (centered), age (centered), smoking, and medication use. These
covariates were not of substantive interest, but were entered into models to ensure that
the final estimates of time were not substantially changed with the addition of these
covariates.
11
Results
Descriptive characteristics of the sample are presented in Table 1. Men had higher
awakening values of cortisol compared to women, and had higher pre-lunch values as
well. Three quarters of the sample showed evidence of the cortisol awakening response.
A large percentage of men (29.5%) and women (52.9%) were taking medications that are
known to influence cortisol. Chi-square tests indicate that medication use was not
associated with experiencing the awakening response for men (χ
2
(1)
= 2.12, p = .14) or
women (χ
2
(1)
= 0.01, p = .92).
The average level of cortisol for each sample across the day followed the
expected trajectory, increasing between the waking sample and 30-60m after waking,
followed by a decline between the morning samples and lunch, as well as a decline
between lunch and before bed. Mean cortisol levels at each sample time are plotted by
gender in Figure 2. To examine the variability in the time and level of cortisol for all four
samples, cortisol values are plotted by time since waking for each sample. Figure 3 shows
that variability in the timing of the samples was highest in the pre-lunch and bedtime
samples. To examine variability in the time and level of cortisol for the pre-lunch
measure, cortisol values were plotted by time since waking only for the pre-lunch sample
of cortisol. Figure 4 shows that for both men and women, there is considerable variability
in the number of hours since waking that one chooses to eat lunch, ranging from 2.4
hours after waking to 14.4 hours after waking. For men, the median number of hours
since waking for taking the pre-lunch sample was 6 hours, with the 25
th
percentile taking
their pre-lunch sample 5.1 hours past waking and the 75
th
percentile taking their sample 7
12
hours past waking. For women, the median number of hours since waking for taking the
pre-lunch sample was 5.8 hours, with the 25
th
percentile taking their sample 5 hours past
waking and the 75
th
percentile taking their sample 6.9 hours past waking. Figure 5
presents the distribution of cortisol by clock-time for the pre-lunch sample of cortisol.
Clock-time shows a slightly different story, indicating that most individuals eat during
socially normative times. For men, the median lunchtime was 12:00 p.m., with the 25
th
percentile eating lunch at 11:35 a.m., and the 75
th
percentile eating lunch at 1:00 p.m. For
women, the median lunchtime was 12:15 p.m., with the 25
th
percentile eating lunch at
11:45 p.m., and the 75
th
percentile eating lunch at 1:00 p.m. For the cortisol values in
plotted by both clock and person-time (Figures 4 and 5), regression lines and 95%
confidence intervals were plotted, and the slopes of all four lines were not significant.
To test the different modeling strategies with and without the pre-lunch cortisol
sample, the results of the GCMs by gender using Time, Time
2
, and Time
3
are presented
in Table 2. The models using 3 time points are presented in Panel A of Table 2, and the
models using 4 time points are presented in the Panel B of Table 2. In Model 1, the linear
term for time is presented, followed by Model 2, which presents the quadratic term for
time, with Model 3 showing the cubic function for time.
Using three data points, the cubic function for time (Model 3) fit the data best for
both men and women. For men, the likelihood ratio test between Model 1 and Model 3
was significant (χ
2
(2) = 77.34, p < .001), as was the likelihood ratio test between Model 2
and Model 3 (χ
2
(1) = 67.66, p < .001). For women, the results were similar. The
likelihood ratio test between Model 1 and Model 3 was significant (χ
2
(2) = 239.33, p
13
< .001), as was the likelihood ratio test between Model 2 and 3 (χ
2
(1) = 229.13, p < .001),
A graphical depiction of the growth curves from Model 3 for men and women are
presented in Figure 6, which shows the quadratic curve as rising sharply within the first
hours of morning, decreasing throughout the day and coming to a nadir at the end of the
day. It should be noted that there is sparseness of the data between hour 1, and hour 5,
which is why the models indicate cortisol would still be rising during this time.
Comparing the AIC and BIC fit statistics from Panel A to Panel B shows the fit of
the models using 3 data points is consistently higher compared to the models with 4 data
points for both men and women. When using 4 data points, Model 2 fit the data best for
men, whereas Model 3 fit the data best for women. However, examining the coefficients
of the models in Panel B shows that the linear term for time is negative for all models,
indicating that cortisol has an immediate negative slope, and therefore does not model the
morning rise. Graphical depictions of Model 3 for men and women are shown in Figure 6.
These show that cortisol has a consistently negative slope for males, and a negative slope
for women until the evening, when the slope appears to flatten out.
To ensure the quadratic and cubic slopes of cortisol were not a result of
confounding, major confounders were examined in each model. These confounders
included age, wake time, medication use, and smoking. Estimates of time, time
2
or time
3
remained stable with the addition of these covariates. Table 3 presents the final models
for the cubic functions of time with potential confounders.
14
Discussion
The current study examined whether a pre-lunch sample of cortisol adds relevant
information to the diurnal cortisol rhythm, or if it leads to misspecification of the daily
rhythm. Using two separate growth curve models, with and without the pre-lunch
measure, the findings of this study suggest that the use of the pre-lunch measure is more
likely to be deviate from theoretical curve and lead to worse model fitting. The models
without the pre-lunch measure had considerably better model fit, and also revealed the
theoretically expected cortisol rhythm, with a sharp increase in the morning, followed by
a decline throughout the day.
The findings from this study indicate that greater numbers of samples across the
day do not always lead to greater information. Significant costs are incurred for studies
incorporating additional measures of cortisol, and the findings from this study suggest
that the use of a pre-lunch measure may not provide relevant information. Thus, studies
obtaining only 3 cortisol samples over the day, such as CHASRS and Cebu, may not only
reduce costs, but may also result in data that better describes normal diurnal cortisol
rhythms. It would be worthwhile to replicate the methodological comparisons used in this
study on studies that have obtained more than one mid-day measure. Whitehall II or
CARDIA, which have obtained cortisol samples at 2.5, 8, and 12 hours after waking, may
be useful for these comparisons.
There are several limitations to this study. First, the NSDE did not obtain daily
food diaries. Although food and eating behaviors are suspected to result in significant
measurement error in the pre-lunch measure, there is no way of telling whether this is the
15
reason for our findings. It would be interesting to examine diurnal rhythms that include
measures of normal meal times, as well as actual eating behaviors and food availability
on the day that cortisol is being sampled. A second limitation is that this study population
contained a considerable number of individuals on medications, with more than half of
women, and nearly one-third of men taking some type of medication that influences
cortisol. Although these medications did not appear to play a significant role in cortisol
rhythms as indicated by the modest effect size estimates observed in the models, this
could be a result of using a combined category of medications. It may be useful for
further studies to examine different medications separately (i.e., allergy medications,
steroid inhalers, hormones).
16
References
Chou, C, Bentler PM, Pentz MA. (1998). Comparison of two statistical approaches to
study growth curves: The multilevel model and latent curve analysis. Structural
Equation Modeling: A Multidisciplinary Journal, 5:247-266.
Almeida, D. M., McGonagle, K., & King, H. (2009). Assessing daily stress processes in
social surveys by combining stressor exposure and salivary cortisol. Biodemography
and Social Biology, 55(2), 219-237.
Dickerson, S. S., & Kemeny, M. E. (2004). Acute stressors and cortisol responses: A
theoretical integration and synthesis of laboratory research. Psychological Bulletin,
130(3), 355.
Harris, J. R., Gruenewald, T. L., & Seeman, T. E. (2008). An overview of biomarker
research from community and population-based studies on aging. In M. Weinstein, J.
W. Vaupel, K. W. Wachter, J. R. Harris, T. L. Gruenewald & T. E. Seeman (Eds.),
Biosocial surveys (pp. 96-135). Washington, DC: The National Academies Press
(Committee on Population, Division of Behavioral and Social Sciences and
Education): National Academies Press (US).
Honma, K. I., Honma, S., & Hiroshige, T. (1984). Feeding-associated corticosterone peak
in rats under various feeding cycles. The American Journal of Physiology, 246(5 Pt
2), R721-6.
Kirschbaum, C., Pirke, K. M., & Hellhammer, D. H. (1993). The ‘Trier social stress
test’—a tool for investigating psychobiological stress responses in a laboratory
setting. Neuropsychobiology, 28(1-2), 76-81.
17
Kudielka, B. M., Gierens, A., Hellhammer, D. H., Wüst, S., & Schlotz, W. (2012).
Salivary cortisol in ambulatory assessment—some dos, some don’ts, and some open
questions. Psychosomatic Medicine, 74(4), 418-431.
Kushner, I. (1982). The phenomenon of the acute phase response. Annals of the New York
Academy of Sciences, 389(1), 39-48.
McEwen, B. S., & Seeman, T. (1999). Protective and damaging effects of mediators of
stress. elaborating and testing the concepts of allostasis and allostatic load. Annals of
the New York Academy of Sciences, 896, 30-47.
Ott, V., Friedrich, M., Prilop, S., Lehnert, H., Jauch-Chara, K., Born, J., et al. (2011).
Food anticipation and subsequent food withdrawal increase serum cortisol in healthy
men. Physiology & Behavior,
Quigley, M. E., & Yen, S. S. (1979). A mid-day surge in cortisol levels. The Journal of
Clinical Endocrinology and Metabolism, 49(6), 945-947.
Sapolsky, R. M., Krey, L. C., & McEwen, B. S. (1986). The neuroendocrinology of stress
and aging: The glucocorticoid cascade hypothesis. Endocrine Reviews, 7(3), 284-
301.
Saxbe, D. E. (2008). A field (researcher's) guide to cortisol: Tracking HPA axis
functioning in everyday life. Health Psychology Review, 2(2), 163-190.
Thompson, B. T. (2003). Glucocorticoids and acute lung injury. Critical Care Medicine,
31(4), S253.
Young, E. A., Carlson, N. E., & Brown, M. B. (2001). Twenty-four-hour ACTH and
cortisol pulsatility in depressed women.
18
Appendix: Tables and Figures
Table 1
Characteristics of the analytic sample for NSDE participants with cortisol measures
obtained on Wednesday.
Male Female
(n = 392) (n = 520)
Mean (SD) Range Mean (SD) Range
Age (years) 56.6 (11.6) 34 - 83 56.0 (12.1) 33 - 84
Wake Time (hr:min) 6:14 (1:10) 6:32 (1:12)
Cortisol (µg/dL)
a
Awakening
2.79 (.65) .02 – 5.67 2.61 (.57) .09 – 4.37
30 minutes
3.08 (.63) .31 – 5.75 3.02 (.59) .05 – 5.05
Lunch
2.10 (.61) .22 – 5.10 1.92 (.65) .02 – 5.20
Bedtime
1.25 (.74) .02 – 4.74 1.17 (.77) .01 – 6.03
Percent Percent
Awakening Response
> 0 µg/dL
73.0% 78.8%
≤ 0 µg/dL
27.0% 21.2%
Current Smoker
11.4% 12.3 %
Currently on Medication
b
29.5% 52.9%
a
transformed to the natural log plus 1.
b
Medications included steroid inhalers, steroid containing medications, cortisone
containing medications, birth control pills, other hormonal medications, and anti-
depressant and anti-anxiety medications.
Table 2
Growth curve models by gender without covariates comparing models without and with the before lunch cortisol sample.
Panel A: Growth curve models without the before lunch cortisol sample (3 data points)
Male Female
Model 1 Model 2 Model 3 Model 1 Model 2 Model 3
Fixed Effects
Time -.13 (.004)*** -.002 (.04) .58 (.08)*** -.13 (.003)*** -.004 (.03) .76 (.05)***
Time
2
-.008 (.002)*** -.09 (.01)*** -.009 (.002)*** -.12 (.007)***
Time
3
.003 (.000)*** .004 (.000)***
Intercept 3.88 (.03)*** 3.85 (.03)*** 3.70 (.04)*** 3.75 (.03)*** 3.72 (.03)*** 3.53 (.03)***
Random Effects
Intercept .17 (.03) .19 (.04) .23 (.03) .26 (.03) .23 (.03) .28 (.02)
Slope .002 (.000) .002 (.000) .003 (.000) .003 (.000) .004 (.000) .004 (.000)
AIC/BIC 2967/2997 2959/2994 2893/2934 3782/3814 3774/3811 3547/3589
Log Likelihood -1477 -1472 -1439 -1885 -1880 -1765
LRT test from
previous model
- 9.68** 67.66*** - 10.20*** 230.13***
Panel B: Growth curve models with the before lunch cortisol sample (4 data points)
Male Female
Model 1 Model 2 Model 3 Model 1 Model 2 Model 3
Fixed Effects
Time -.13 (.003)*** -.16 (.01)*** -.13 (.03)*** -.14 (.003)*** -.21 (.01)*** -.13 (.02)***
Time
2
.002 (.000)*** -.005 (.005) .004 (.001)*** -.01 (.003)**
Time
3
.000 (.000) .0007 (.0002)***
Intercept
3.85 (.03)*** 3.88 (.03)*** 3.87 (.03)*** 3.69 (.03)*** 3.76 (.03)*** 3.73 (.03)***
Random Effects
Intercept .18 (.03) .18 (.03) .18 (.03) .21 (.02) .21 (.02) .21 (.02)
Slope .003 (.000) .003 (.000) .003 (.000) .003 (.000) .003 (.000) .003 (.000)
AIC/BIC 3366/3397 3357/3394 3357/3400 4312/4345 4250/4289 4235/4279
Log Likelihood -1677 -1671 -1670 -2150 -2118 -2109
LRT test from
previous model
- 10.20*** 2.28 - 64.29*** 16.84***
Note. Model 1: Linear function of time; Model 2: Quadratic function of time; Model 3: Cubic function of time. Each
likelihood ratio is a χ
2
with 1 degree of freedom. LRT = likelihood ratio test.* p ≤ .05 , **p ≤ .01, *** p ≤ .001
19
20
Table 3. Final growth curve models of change by gender with covariates using 3 data
points.
Male Female
Fixed Effects
Time .60 (.08)*** .74 (.06)***
Time
2
-.09 (.01)*** -.12 (.01)***
Time
3
.003 (.000)*** .04 (.000)***
Age .01 (.003)*** .01 (.002)***
Wake time .05 (.03) -.03 (.02)
Medication .10 (.08) -.08 (.06)
Smoker .15 (.12) .05 (.08)
Intercept 3.56 (.12)*** 3.66 (.09)***
Random Effects
Intercept .29 (.04) .22 (.03)
Slope .003 (.000) .004 (.000)
AIC/BIC 2011/2067 2465/2525
Log Likelihood -993 -1220
Note. Medication was coded 0 = medication use, 1 = no medication use; Smoker was
coded 0 = non-smoker, 1 = smoker. Age is centered on 56.26 and wake time is centered
on 6.40. LRT = likelihood ratio test. Each likelihood ratio is a χ
2
with 5 degrees of
freedom.
* p ≤ .05 , **p ≤ .01, *** p ≤ .001
21
Figure 1. Theoretical diurnal cortisol curve
22
Figure 2. Mean cortisol levels and 95% confidence intervals for each sample.
0 1 2 3 4 5
Natural Log Cortisol +1
1 2 3 4
Sample Number
Male Female
95% Confidence Interval
Mean Cortisol Levels for Each Sample
23
Figure 3. Distribution of person-time for each cortisol sample.
-5 0 5 10
Log Cortisol +1 (ug/dL)
0 5 10 15 20
Time since waking (hours)
Wake +30-60 min
Lunch Bed
Distribution of Person-Time for Each Sample
24
Figure 4. Distribution of person-time for the ‘pre-lunch’ sample.
-5 0 5 10
Natural Log Cortisol (ug/dL)
0 5 10 15
Person Time of Lunch Cortisol Sample (hours)
Men
-5 0 5 10
Natural Log Cortsiol (ug/dL)
0 5 10 15
Person Time of Lunch Cortisol Sample (hours)
Women
25
Figure 5. Distribution of clock-time for the ‘pre-lunch’ sample.
-5 0 5 10 15
Natural Log Cortisol (ug/dL)
0 5 10 15 20
Clock Time of Lunch Cortisol Sample (hours)
Men
-5 0 5 10 15
Natural Log Cortisol (ug/dL)
0 5 10 15 20
Clock Time of Lunch Cortisol Sample (hours)
Women
26
Figure 6. Growth models of cortisol using 3 data points (solid lines), versus 4 data points
(dashed lines) at various time points with cubic functions of time for men and women.
1
1.5
2
2.5
3
3.5
4
4.5
5
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Natural
Log
Cortisol
+1
Time
Since
Waking
Female
(3
points)
Male
(3
points)
Female
(4
points)
Male
(4
points)
Abstract (if available)
Abstract
The integration of biomarkers into population-based epidemiological studies has introduced methodological complexities that have yet to be clearly delineated. Cortisol is a biomarker that is particularly complicated because it follows a circadian rhythm that is based on individual wake and sleep times, as well as normal meal times. Thus, understanding the timing of sampling is important to reduce costs of large studies. This study used data from the Survey of Midlife in the United States to compare cortisol rhythms using 3 sample times (waking, +30m, bedtime) and 4 sample times (waking, +30m, pre-lunch, bedtime), in order to examine the use of a pre-lunch measure for capturing diurnal cortisol rhythm. Growth curve models for repeated measures were used to describe the change in cortisol across the day as well as variation across individuals. Results indicate that including the pre-lunch sample of cortisol in the models does not capture the expected curvilinear daily rhythm, whereas omitting this sample results in the expected shape of the daily rhythm. The findings of this study suggest that the use of the pre-lunch measure is more likely to be deviate from theoretical curve and lead to worse model fitting. Thus, the pre-lunch measure may introduce variability in the time that samples are taken, as well as variability in levels of cortisol due to other unknown factors such as normal meal times and food consumption.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Exploring the association of number of cigarettes smoked and confidence to quit smoking in Korean American emerging adults: a multilevel modeling approach
PDF
A multivariate approach to explore nursing home characteristics contributing to high performance on CMS' publicly reported quality measures
PDF
Marital quality, gender, and biomarkers of disease risk in the MIDUS cohort
PDF
Development and validation of survey instrument designed for cervical cancer screening in Malawi, and other low resource settings
PDF
Insulin’s effect on lactate levels in extremely low birth weight neonates. a multi-center, observational study
PDF
Association of maternal and environmental factors with infant feeding behaviors in a birth cohort study
PDF
A pilot study of a global approach to assessing air pollution exposure in port communities: passive air monitoring of nitrogen dioxide concentrations
PDF
Association of traffic-related air pollution and lens opacities in the Los Angeles Latino Eye Study
PDF
Effects of post-menopausal hormone therapy on arterial stiffness in the ELITE trial
PDF
An assessment of necrosis grading in childhood osteosarcoma: the effect of initial treatment on prognostic significance
PDF
Disparities in exposure to traffic-related pollution sources by self-identified and ancestral Hispanic descent in participants of the USC Children’s Health Study
PDF
The risk estimates of pneumoconiosis and its relevant complications: a systematic review and meta-analysis
PDF
Examining exposure to extreme heat and air pollution and its effects on all-cause, cardiovascular, and respiratory mortality in California: effect modification by the social deprivation index
PDF
The evaluation of the long-term effectiveness of zero/low fluoroscopy workflow in ablation procedures for the treatment of paroxysmal and persistent atrial fibrillation
PDF
Long-term effects on vertebral body height growth of dose sculpting intensity modulated radiation therapy for children with neuroblastoma
PDF
Observed and underlying associations in nicotine dependence
Asset Metadata
Creator
Donoho, Carrie Joy
(author)
Core Title
Methodological approaches to assessing diurnal cortisol rythms in epidemiological studies: how many salivary samples are necessary?
School
Keck School of Medicine
Degree
Master of Science
Degree Program
Applied Biostatistics and Epidemiology
Publication Date
01/26/2013
Defense Date
05/26/2012
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
CRP,health,IL-6,Inflammation,marital quality,Marriage,OAI-PMH Harvest,social relationships
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Chou, Chih-Ping (
committee chair
), Azen, Stanley P. (
committee member
), Crimmins, Eileen M. (
committee member
)
Creator Email
carriedonoho@gmail.com,donoho@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-68329
Unique identifier
UC11288893
Identifier
usctheses-c3-68329 (legacy record id)
Legacy Identifier
etd-DonohoCarr-1019.pdf
Dmrecord
68329
Document Type
Thesis
Rights
Donoho, Carrie Joy
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
CRP
IL-6
marital quality
social relationships