University of Southern California Dissertations and Theses

Pain and multi-morbidity among veterans: theory-guided, data-driven, and narrative approaches

(USC Thesis Other)

Pain and multi-morbidity among veterans: theory-guided, data-driven, and narrative approaches

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content Copyright 2024 Shaddy Saba
Pain and Multi-Morbidity among Veterans:
Theory-Guided, Data-Driven, and Narrative Approaches
by
Shaddy K. Saba, MA, LCSW
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirement of the Degree
DOCTOR OF PHILOSOPHY
(SOCIAL WORK)
August 2024

ii
Dedication
To my grandfather, Amal, who saw the future.

iii
Acknowledgements
This dissertation would not be possible without the effort of my doctoral committee and
mentorship team, who have dedicated countless hours to my training. To Dr. Jordan Davis, you
are a research unicorn, and this has given me the opportunity to learn so much: how to think and
write more clearly, how to select and learn appropriate research methods, and how important
strong interpersonal connections are for doing this work. You teach your mentees that the world
of research is their oyster; thank you for this lesson. To Dr. Carl Castro, thank you for sharing the
wisdom you have accumulated over an incredible career and helping remind me that there is no
substitute for lived experience; thank you for your confidence in my work and for stewarding me
across the finish line. To Dr. Eric Pedersen, thank you for sharing in your success and providing
countless opportunities to learn on your research projects, and for helping me feel like a valued
contributor from day one; you make both research productivity and kindness look effortless. To
Dr. John Prindle, thank you for sharing your quantitative knowledge with me over many, many
hours and working meetings; I have been lucky to learn from someone who can teach virtually
any quantitative method I have expressed interest in. To Dr. Kathryn Bouskill, thank you for your
curiosity and the wisdom you bring about how to dig deep with people to understand what is
truly important to them and how to share that with others. To Dr. Benjamin Henwood, thank you
for your willingness to join this project during the home stretch; it is impressive how much my
work improved in just a few months with your input.
I would like to acknowledge others who have been critical in directly supporting my
work on this dissertation. To Angeles Sedano, thank you for your effort contacting and recruiting
research participants and coding qualitative interviews. To the National Institute on Drug Abuse,

iv
thank you for seeing promise in my work and providing funding for me to focus on my training
as a behavioral health scientist and to carry out this research.
I would also like to thank the faculty and staff at the USC Suzanne-Dworak Peck School
of Social Work more broadly. I appreciate the financial and intellectual investment the School
and its faculty have made in me as a doctoral trainee. I have especially benefitted from strong
PhD program leadership in Dr. Michael Hurlburt, Dr. Olivia Lee, and Dr. Eric Rice. Thank you
each for your dedication to creating opportunities for students and to fostering an inclusive,
supportive, and dynamic intellectual environment, which has been crystal clear. To Malinda
Sampson, PhD Program Manager, thank you for your support, guidance, and knowledge on
navigating the program and University. Thank you also to leadership at the USC Center for AI in
Society, namely Dr. Eric Rice, Ms. Hailey Wintrobe, Dr. Bistra Dilkina, and Dr. Phebe Vayanos,
for your support and for building what became such an exciting academic home for me at USC.
Outside my immediate committee and program leadership, my scientific thinking has
benefitted tremendously from the influence of many other strong mentors. Thank you to Dr.
Elizabeth D’Amico, Dr. Daniel Hackman, Dr. Eric Rice, Dr. Olivia Lee, Dr. David Black, Dr.
Todd Herrenkohl, Dr. Anthony Rodriguez, Ms. Kayla Williams, Dr. Justin Hummer, Dr. Rajeev
Ramchand, Dr. Carrie Farmer, Dr. Meredith Kleycamp, Dr. Daniel Dickerson, Dr. Lauren
Broyles, Dr. Adam Gordon, Dr. Monica DiNardo, Dr. Susan Jarquin, Dr. Ajay Wasan, Ms. Billie
Jo Smith, and Dr. Daniel Leightley for the lessons and opportunities you have provided.
It has also been a pleasure to do this work alongside the many student colleagues and
friends I have made at USC. Thank you especially to Dr. Graham DiGuiseppi, Dr. Sara
Semoborski, Dr. Rory O’Brien, Dr. Daniel Lee, Liv Canning, John Bunyi, Aubrey Sutherland,
Olga Koumoundourous, Dr. Amy Lee, Dr. Nina Christie, Angeles Sedano, Reagan Fitzke, Dr.

v
Sheila Pakdaman, Dr. Denise Tran, Keegan Buch, Emma Carpenter, and Dr. Caroline Johnston.
You have made substantive contributions to my work, provided invaluable support, and made my
time as a doctoral student more memorable.
I would not have embarked on this academic journey without the inspiration of my family
members and close friends who did so before me, including Nehme Ayoub, Raja Ghanem,
Robert Sanders, Sorabh Tomar, Shruthi Venkatesh, Sarah Najjar, Max Rubinstein, Corinne
Beaugard, Pete Faggen, and Jonathan Keim. I also appreciate the support of Nathan Healy,
Delaney Gold-Diamond, Neil Bakshi, Nnaemeka Alozie, Peter Grimm, Laura Walsh, Michael
Ruhl, Jessica Williams, Daniel Menges, Wyatt Fertig, Dan Olivieri, Odette Zero, Josh Standiford,
Jim Lynch, and Lee Foster, as well as my grandparents, aunts, uncles, cousins, and in-laws.
To my fiancée Andrea, I have said my work improved after I met you a couple of years
into my PhD program and I was not joking. Your endless support and love make everything
better. I am grateful for you, and excited for the life we are building together and with Babybel.
To my parents Mary and Khalil, my sister Dani, and my brother-in-law Amir, I would not be
where I am today without your unconditional love, belief in my abilities, and support in pursuing
my goals. To my nieces Tala and Leya, you are shining stars, and I am always here for you.
Finally, I would like to thank the veterans who have generously shared their time so I
could carry out this research, especially those 20 individuals who were willing to participate in a
one-on-one interview with me. I am inspired your desire to help others by contributing to
science. I hope I have done justice to telling your stories.

vi
Table of Contents
Dedication ....................................................................................................................................... ii
Acknowledgements........................................................................................................................ iii
List of Tables................................................................................................................................ viii
List of Figures................................................................................................................................ ix
Abbreviations.................................................................................................................................. x
Abstract.......................................................................................................................................... xi
Chapter 1: Introduction ................................................................................................................. 12
Existing theory of multi-morbidity ........................................................................................... 14
Limitations of existing research on multi-morbidity among veterans...................................... 16
The present dissertation study: Advancing the science of multi-morbidity among veterans ... 17
Paper 1: Daily and dynamic associations between PTSD symptoms, pain, and cannabis use . 17
Paper 2: Biopsychosocial predictors of pain among veterans with PTSD................................ 18
Paper 3: Towards a bio-cultural vulnerability model of veteran multi-morbidity .................... 20
Conceptual model ..................................................................................................................... 21
Chapter 2: Daily and dynamic associations between PTSD symptoms, pain, and cannabis use
among veterans (Paper 1).............................................................................................................. 23
Abstract..................................................................................................................................... 23
Introduction............................................................................................................................... 25
Methods..................................................................................................................................... 29
Results....................................................................................................................................... 36
Discussion................................................................................................................................. 38
Supplemental analyses: Weekly symptom dynamics................................................................ 52
Chapter 3: Biopsychosocial predictors of pain among veterans with PTSD ................................ 61

vii
Abstract..................................................................................................................................... 61
Introduction............................................................................................................................... 63
Methods..................................................................................................................................... 66
Results....................................................................................................................................... 72
Discussion................................................................................................................................. 77
Chapter 4: “My body will remember what my mind wants to forget”: Towards a bio-cultural
vulnerability model of veteran multi-morbidity (Paper 3)............................................................ 93
Abstract..................................................................................................................................... 93
Introduction............................................................................................................................... 95
Methods................................................................................................................................... 100
Results..................................................................................................................................... 103
Discussion................................................................................................................................116
Chapter 5: Conclusion................................................................................................................. 124
Review of major findings and integration with existing literature ......................................... 124
Summary of theoretical implications...................................................................................... 127
Summary of clinical implications........................................................................................... 128
Future research directions....................................................................................................... 129
References................................................................................................................................... 131

viii
List of Tables
Table 2.1 Participant characteristics (n = 74)............................................................................... 44
Table 2.2 Study variable descriptive statistics and bivariate correlations.................................... 45
Table 2.3 DSEM model output depicting daily, dynamic symptom associations ........................ 46
Table 2.4 DSEM model output depicting the moderating role of anxiety sensitivity on daily,
dynamic symptom associations..................................................................................................... 48
Table 2.5 ALT-SR model output depicting weekly, dynamic symptom associations................... 57
Table 2.6 Model fit procedure and comparisons for bivariate ALT-SR model ............................ 59
Table 2.7 Model fit procedure and comparisons for trivariate ALT-SR model ............................ 60
Table 3.1 Predictor variables descriptions, descriptive statistics, and missingness ..................... 84
Table 4.1 Qualitative participant characteristics (n = 20) .......................................................... 122

ix
List of Figures
Figure 1.1 Dissertation conceptual model.................................................................................... 22
Figure 2.1 Conceptual model of a bivariate DSEM model.......................................................... 50
Figure 2.2 Conceptual model of a bivariate DSEM model with a random effects moderator..... 51
Figure 2.3 Conceptual model of a bivariate ALT-SR model........................................................ 56
Figure 3.1 Variable importance metrics from random forest classifier predicting frequent pain 87
Figure 3.2 Odds ratios from regularized logistic regression predicting frequent pain................. 88
Figure 3.3 Survival plot depicting probability of pain intensity increases over time for the
full sample..................................................................................................................................... 89
Figure 3.4 Variable importance metrics from random survival forest predicting pain intensity
increases........................................................................................................................................ 90
Figure 3.5 Hazard ratios from regularized Cox regression predicting pain intensity increases .. 91
Figure 3.6 Survival plots depicting probability of pain intensity increases over time for
subsamples based on top two predictors from random survival forest model .............................. 92
Figure 4.1 Preliminary, bio-cultural vulnerability model of veteran multi-morbidity............... 123

x
Abbreviations
ASI-3: Anxiety Sensitivity Index-3
AUDIT: Alcohol Use Disorders Identification Test
CI: Confidence interval
CUD: Cannabis use disorder
CUDIT-R: Cannabis Use Disorder Identification Test - Revised
DSEM: Dynamic structural equation modeling
HR: Hazard ratio
LASSO: Least Absolute Shrinkage and Selection Operator
mMBRP: Mobile Mindfulness-Based Relapse Prevention
NRS: Numerical Rating Scale
OIF/OEF/OND: Operation Iraqi Freedom/Operation Enduring Freedom/Operation New Dawn
OR: Odds ratio
PCL-5: PTSD Checklist for DSM-5
PROMIS: Patient Reported Outcomes Measurement Information System
PSR: Potential scale reduction
PTSD: Posttraumatic stress disorder
SD: Standard deviation
TBI: Traumatic brain injury
THC: Tetrahydrocannabinol
U.S.: United States
VHA: Veterans Health Administration

xi
Abstract
Military veterans experience a high burden of multi-morbid physical and behavioral
health problems, including physical pain, posttraumatic stress disorder, and problematic
substance use. Researchers have developed several theoretical models to explain multimorbidity, including the self-medication model, the mutual maintenance model, and the shared
vulnerability model. There are varying degrees of evidence for these models and they generally
consider bivariate associations, which limits what is known about associations between all three
problems. Empirical research with veterans has also yet to explore how existing theories might
integrate with one another, which could lead to a more nuanced understanding of symptom
processes and move from bivariate symptom associations to associations among three or more
problems. There are also many additional biopsychosocial factors ranging from traumatic
injuries and military culture to trouble sleeping and physical inactivity that are known to
influence individual physical and behavioral health problems but have been less considered in
studies of multi-morbidity. This three-paper dissertation seeks to extend knowledge of multimorbid problems and their etiology among veterans. Paper 1 uses dynamic structural equation
modeling with daily diary data to explore trivariate associations between pain, PTSD symptoms,
and problematic cannabis use among a sample of veterans (n = 74); this paper seeks to test and
extend existing theories of multi-morbidity. Given the breadth of additional biopsychosocial
factors that are known to influence pain, paper 2 uses machine learning with the daily diary
sample to clarify the predictors of pain – particularly of pain intensity increases and frequent
pain – among veterans with multi-morbid PTSD. Finally, paper 3 engages veterans (n = 20) in
qualitative and narrative life history interviews to understand their subjective experiences of
multi-morbid problems and how military experiences and culture influence such problems.

12
Chapter 1: Introduction
Military veterans experience a high burden of physical and behavioral health problems
relative to their civilian counterparts (Boden & Hoggatt, 2018; Hoerster et al., 2012). Two
especially common behavioral health problems among United States (U.S.) veterans are
posttraumatic stress disorder (PTSD; Fulton et al., 2015) and the problematic use of substances
including alcohol and cannabis (Boden & Hoggatt, 2018). Among post-9/11 veterans, prior
works estimates the prevalence of PTSD to be 23% (Fulton et al., 2015), a 36% lifetime
prevalence of alcohol use disorder, and a 7% lifetime prevalence of cannabis use disorder (Boden
& Hoggatt, 2018). Veterans also have high physical health challenges (Hoerster et al., 2012),
with one especially prevalent problem being physical pain (Nahin, 2017; Stecker et al., 2010). A
recent systematic review reported that approximately 30% of veterans have chronic pain (i.e.,
pain lasting three or more months; Raja et al., 2020) (Qureshi et al., 2023). In a recent study
involving a U.S. nationally representative sample veterans had 1.5 higher odds of reporting
severe pain compared to non-veterans (Nahin, 2017).
When seeking to understand the factors that explain veterans’ high prevalence of physical
and behavioral health problems such as pain, PTSD, and problematic substance use, one
potentially important clue is the high degree of multi-morbidity between these problems. That is,
large numbers of veterans have not only one, but multiple problems concurrently, and having one
problem appears to be associated with heightened risk for having another (Stecker et al., 2010).
There is a particularly robust literature linking pain and PTSD among veteran samples, as largescale chart review studies in the Veterans Health Administration (VHA) indicate an especially
high prevalence of multi-morbid pain and PTSD (18% of patients have both conditions; Stecker
et al., 2010). Studies of post-9/11 veterans engaged in treatment for pain or PTSD also suggest a

13
high degree of overlap between these conditions, with 66% of those in treatment for PTSD
having pain (Shipherd et al., 2007) and 49% of those seeking treatment for pain meeting criteria
for PTSD (Otis et al., 2010). Similarly, both PTSD and pain are associated with heightened rates
of problematic substance use. In nationally representative samples, veterans with PTSD have
2.13 greater odds of having alcohol use disorder, and 3.19 greater odds of having a drug use
disorder, relative to veterans without PTSD (Smith et al., 2016). Large-scale studies of VHA
patients have demonstrated 1.8 greater odds of alcohol use disorder (Tiet & Moos, 2021) and
over two times greater prevalence of cannabis use disorder (Mannes et al., 2022) among veterans
with pain compared to those without pain.
Advancing knowledge on the etiology of multi-morbidity among veterans holds great
potential for relieving the physical and behavioral health burden among them. This is true not
only because of the high prevalence of multi-morbid problems in this population, but also
because having more than one problem is associated with especially poor outcomes. A systematic
review of the impact of multi-morbid PTSD on veterans with pain reported PTSD symptoms
were associated with a litany of adverse outcomes including greater pain, poorer mental health,
and worse physical functioning (Benedict et al., 2020a). Similarly, problematic alcohol use has
been linked with poorer physical and psychological functioning among veterans with PTSD
(Norman et al., 2018), as well as worse pain interference (i.e., the extent to which pain interferes
with daily living activities) among VHA patients (Caniglia et al., 2020). Thus, a better
understanding of the factors that drive multi-morbidity could inform advances in practice and
prevention efforts and improve veteran health and well-being.

14
Existing theory of multi-morbidity
Looking to the existing scientific literature, researchers have developed several
theoretical models seeking to explain multi-morbidity (Asmundson et al., 2002; Khantzian,
1997). These models generally consider bivariate associations between pairs of problems and
have varying degrees of empirical evidence (i.e., there is evidence both in favor of and against
several theorized symptom relationships, and others have yet to be tested) particularly when it
comes to their applicability for explaining multi-morbid pain, PTSD, and substance use among
veterans. Regarding links between either pain or PTSD and substance use, the self-medication
model has a particularly robust evidence base (Khantzian, 1997). This model states that some
individuals with physical or behavioral health problems (e.g., pain or PTSD) use substances to
cope with such problems, which can lead to increasing use and a greater risk for substancerelated problems (Khantzian, 1997). Some of the earliest empirical evidence for self-medication
came from studies of veterans with PTSD (Lacoursiere et al., 1980), and since then evidence in
support of the model has proliferated among veterans with many problems, including PTSD and
pain (Saba et al., 2021). One open question is whether substance use temporarily relieves these
problems and might exacerbate them long term, as some have considered that self-medicating
with substances may interfere with effective coping behaviors (Hawn et al., 2020) or make
individuals more sensitive to behavioral health symptomology or pain (Egli et al., 2012; Metrik
et al., 2022).
Theoretical models have also been developed to help explain multi-morbidity between
pain and PTSD. The mutual maintenance model posits that the overlap between pain and PTSD
is due in part to how each condition influences the other (Sharp & Harvey, 2001). Specifically,
the mutual maintenance model suggests a bidirectional association where PTSD symptoms make

15
pain worse and vice versa via several mechanisms, increasing both the severity and persistence
of each problem (Sharp & Harvey, 2001). For example, theoretical literature on the mutual
maintenance model suggests PTSD can result in attentional biases where someone pays more
attention to their pain, and pain can serve as a trauma reminder if someone suffered a traumatic
injury (Sharp & Harvey, 2001). Results from existing empirical studies testing the mutual
maintenance model among veterans have been mixed: two studies have indeed reported a
bidirectional association between pain and PTSD symptoms (Lee et al., 2019; Stratton et al.,
2014), however, another study reported greater evidence for a unidirectional association (where
pain more robustly predicted heightened PTSD symptoms rather than the other way around;
McAndrew et al., 2019).
Couched as an alternative to the mutual maintenance model, the shared vulnerability
model states that pain and PTSD tend to occur together because they share predisposing factors
(i.e., veterans may be susceptible to risk factors for PTSD that are also risk factors for pain;
Asmundson et al., 2002). Whereas the mutual maintenance model suggests the overlap between
pain and PTSD can be explained by each problem’s direct association on the other problem (a
bidirectional association), the shared vulnerability model implicates “third variables” that
heighten risk for both pain and PTSD independently. To date, much of the theoretical literature
on shared vulnerability for PTSD and pain has focused on one predisposing factor: anxiety
sensitivity (i.e., worry about physiological symptoms of anxiety; Asmundson et al., 2002). This
literature details how PTSD hyperarousal symptomology such as a rapid heart rate or chest
tightening could trigger worry that one’s health is in real danger. While separate empirical studies
have demonstrated heightened anxiety sensitivity among individuals with PTSD (Raines et al.,

16
2017) and pain (Ocañez et al., 2010) the role of anxiety sensitivity in vulnerability for multimorbid pain and PTSD among veterans remains far less explored.
Limitations of existing research on multi-morbidity among veterans
There are several additional limitations in the existing theoretical and empirical literature
on multi-morbidity. Namely, empirical research has yet to explore how existing theories might
integrate with one another, which could lead to a more nuanced understanding of symptom
processes and move from bivariate symptom associations to associations among three or more
variables. To illustrate, while the mutual maintenance and shared vulnerability models are often
presented as alternatives to one another (i.e., either pain and PTSD exacerbate one another or
anxiety sensitivity influences both), it may be the case that there are shared vulnerability factors
that catalyze mutual maintenance. For example, perhaps anxiety sensitivity not only directly
influences PTSD and pain, but moderates associations between them. There also may be
trivariate associations between pain, PTSD, and problematic substance use that are largely
unexplored. For example, if pain and PTSD symptomology exacerbate one another, does this
later catalyze increases in substance use? Or if substance use can help individuals temporarily
cope with pain or PTSD (or makes it more challenging to cope over time), does this influence the
other condition? Complicating manners further, there are many additional factors ranging from
traumatic injuries and military culture to trouble sleeping and physical inactivity that are known
to influence pain (Nampiaparampil, 2008; Wei et al., 2018), PTSD (Neilson et al., 2020;
Whitworth et al., 2022), and/or substance use (Meadows et al., 2022) individually but have been
less considered in existing theory of multi-morbidity. How might these additional factors
influence multi-morbid symptomology among veterans?

17
The present dissertation study: Advancing the science of multi-morbidity among veterans
This three-paper dissertation seeks to explore such questions and extend knowledge of
multi-morbid problems and their etiology among veterans. This research is grounded in existing
theories of multi-morbidity, including the self-medication, mutual maintenance, and shared
vulnerability models, and seeks not only to test these models, but to explore how they may
integrate with one another to explain links between pain, PTSD, and problematic substance use
(Paper 1). Given the breadth of additional biopsychosocial factors that are known to influence
pain, PTSD, and problematic substance use individually, this work also begins to broaden
knowledge of the role of these factors in multi-morbid symptomology – namely, how they
influence pain among veterans who also have PTSD symptoms (Paper 2). Finally, the present
study engages veterans in qualitative interviews to understand their subjective experiences of
multi-morbid problems and how military experiences and culture influence such problems (Paper
3). Further detail on each paper and its methodology is outlined below.
Paper 1: Daily and dynamic associations between PTSD symptoms, pain, and cannabis use
PTSD, pain, and problematic cannabis use are three problems known to be highly
prevalent among veterans (Stecker et al., 2010; Stratton et al., 2014). Existing theory of multimorbidity generally explain bivariate associations between pairs of problems; empirical tests of
these theories either have mixed results (in the case of the mutual maintenance model) or have
not been conducted at all (in the case of the shared vulnerability model) among veterans. This
paper seeks to advance knowledge of how PTSD symptoms, pain, and problematic cannabis use
influence one another, and thus advance and integrate existing theory of multi-morbidity. One
key element of this study is that it involves data on symptomology that has been collected on a

18
daily basis (i.e., daily diary data; Gunthert & Wenze, 2012). This is crucial, because while most
prior work on multi-morbid symptomology involves data that has been collected in 3–6-month
intervals, pain and behavioral health symptomology are known to fluctuate daily (Eather et al.,
2019; Mun et al., 2019; Schuler et al., 2021), perhaps explaining inconsistencies in prior results.
Thus, paper 1 utilizes secondary data from a daily diary study of post-9/11 U.S. veterans
who have screened positive for PTSD and problematic cannabis use (n = 74). This paper
involves dynamic structural equation modeling (Asparouhov et al., 2018), a statistical modeling
technique that was designed to leverage data with many repeated measurements (intensive
longitudinal data) such as daily diary data and can model multivariate associations between
variables. Aim 1 is to test the bivariate associations between pain and PTSD at the daily level, in
line with the mutual maintenance model. For Aim 2, in line with the mutual maintenance and
self-medication models, analyses also include cannabis use to test the trivariate associations
between all three problems. Aim 3 seeks to integrate the shared vulnerability model and explore
whether anxiety sensitivity moderates associations between PTSD symptoms, pain, and cannabis
use. Thus, this paper exemplifies how advanced methodology may push forward theoretical
knowledge and could potentially inform more targeted interventions for veterans with multimorbidity.
Paper 2: Biopsychosocial predictors of pain among veterans with PTSD
Paper 2 focuses on understanding predictors of pain, which is the most commonly
endorsed health problem among veterans and is highly prevalent among veterans with multimorbid behavioral health problems such as PTSD (Stecker et al., 2010). The development and
maintenance of pain is known to be influenced by many biological, psychological, and social

19
(biopsychosocial) factors not limited to physical injury, socio-demographics, and mental health
symptomology (Gatchel et al., 2007). Prior work among veterans with PTSD implicates factors
like these as determinants of pain, but such work typically uses traditional statistical methods
common in the social sciences that can only model a relatively small subset of predictors at a
time (Giordano et al., 2018; Lew et al., 2009; Tan et al., 2009). Thus, researchers have been
unable to disentangle the most important factors contributing to pain and how they interact. And
while most existing research on factors influencing pain among veterans involves retrospective
measures of average pain during multiple weeks (Suzuki et al., 2020), far less is known about
what predicts important components of pain, such as increases in pain intensity (i.e., how long
until someone experiences a clinically meaningful increase in pain) or the frequency with which
one experiences pain (i.e., whether one experiences pain relatively often).
Thus, paper 2 begins to clarify the predictors of pain – particularly of pain intensity
increases and frequent pain – among veterans with multi-morbid PTSD. This paper is grounded
in the contemporary biopsychosocial model of pain, which states pain is influenced by a complex
set of biological, psychological, and social factors and their interactions (Engel, 1981). It
leverages data from the aforementioned daily diary study of veterans (n = 74) including sociodemographic and military characteristics at baseline, as well as clinical symptomology and
changes in clinical symptomology measured daily. Aim 1 is to explore which of these factors
predict whether one experiences moderate pain relatively frequently over a two-month period,
and Aim 2 is to explore which factors predict the time until one experiences a clinically
significant increases in pain intensity over a two month-period among veterans with multimorbid PTSD. This study utilizes machine learning methods, which are well suited for including
many possible predictor variables and determining which variables are most important in

20
predicting an outcome (Lötsch & Ultsch, 2018). Analyses such as this one could catalyze the
development of interventions that are delivered in a time sensitive manner.
Paper 3: Towards a bio-cultural vulnerability model of veteran multi-morbidity
Finally, paper 3 turns again to the question of vulnerability, by qualitatively exploring
among veterans themselves why they may be at particular risk for having concurrent, multimorbid physical and behavioral health problems. While existing theoretical work on shared
vulnerability among veterans has focused narrowly on psychological factors such as anxiety
sensitivity (Asmundson et al., 2002), many additional factors have been linked with physical and
behavioral health problems and thus may be implicated in multi-morbidity between them. Likely
salient among veterans are exposures to physically and psychologically challenging experiences
that previously occurred during their military service. These include traumatic brain injuries,
combat trauma, and military sexual trauma, all of which have been empirically linked with multimorbid problems including pain, PTSD, and/or problematic substance use (Chui et al., 2022;
Cifu et al., 2013; Raines et al., 2017). Additionally military cultural beliefs and practices such as
masculinity and a heavy drinking culture have been linked with multi-morbid problems
(Meadows et al., 2022; Neilson et al., 2020), while certain factors such as camaraderie could
confer protection (Nevarez et al., 2017).
Thus, paper 3 endeavors to document not only how military experiences may influence
multi-morbidity among veterans, but how these experiences may interact with military culture to
determine risk. This idea is in line with bio-cultural models of health (McElroy, 1990;
Zuckerman, 2018) that suggest some cultural practices that develop to help individuals cope with
stressors can ultimately make them worse (e.g., perhaps masculinity helps service members

21
complete combat missions but then makes it more challenging to seek mental health treatment).
To explore these ideas, we completed qualitative interviews with veterans who had PTSD
symptoms, pain, and problematic substance use to learn about salient experiences they had
during their time in the military, their perception of how their multi-morbid problems developed,
and how they cope with such problems. Results can move us towards a richer understanding of
vulnerability for multi-morbidity, grounded in the bio-cultural perspective.
Dissertation conceptual model
The below conceptual model (Figure 1.1) displays at a high level various theoretical
models, constructs, and proposed symptom processes that are central to this dissertation and its
various aims. In the center are pain, PTSD, and substance use, which are the multi-morbid
problems of primary interest in this dissertation. The bivariate associations between pain and
PTSD proposed by the mutual maintenance model, and between substance use and both pain and
PTSD proposed by the self-medication model, are depicted by the double-sided arrows linking
these constructs. The circle surrounding the multi-morbid problems and connecting biological,
psychological, social, and cultural factors (e.g., aspects of military culture) indicate that such
factors have known associations with pain, PTSD, and/or substance use, and could therefore help
explain veterans’ shared vulnerability for these problems.

22
Figure 1.1 Dissertation conceptual model

23
Chapter 2: Daily and dynamic associations between PTSD symptoms, pain, and cannabis
use among veterans (Paper 1)
Abstract
Introduction: Military veterans experience high rates of multi-morbid problems, including
posttraumatic stress disorder (PTSD), physical pain, and problematic cannabis use. Several
theoretical models, including the mutual maintenance, self-medication, and shared vulnerability
models, have been developed to explain the associations between these multi-morbid problems.
However, theoretical models and empirical tests of these models have generally focused on
bivariate associations, limiting our understanding of the dynamic interplay between all three
problems.
Methods: 74 U.S. veterans with PTSD symptoms and problematic cannabis use completed daily
diary surveys on their PTSD symptoms, pain intensity, and cannabis use over an 88 day period.
Dynamic structural equation modeling was used to explore the day-to-day, dynamic associations
between PTSD symptoms, pain, and cannabis use, and test whether anxiety sensitivity moderated
these associations.
Results: There were no significant within-person lagged associations between pain and PTSD
symptoms. However, there was a significant same-day correlation between pain and PTSD
symptoms, and a between-person correlation between pain and PTSD symptoms. There were
significant negative lagged associations between pain and subsequent cannabis use and between
cannabis use and subsequent pain. Anxiety sensitivity did not significantly moderate any daily
associations.

24
Discussion: Though there were not lagged associations between pain and PTSD symptoms, the
same-day correlation between these variables suggests there may be other time-varying factors
influencing both problems concurrently. Given lagged associations between pain and cannabis
use, pain providers should be prepared to discuss the benefits and consequences of cannabis with
veterans. Results also suggest there may be other between-person factors besides anxiety
sensitivity that confer shared vulnerability for multi-morbid problems.

25
Introduction
Among military veterans, empirical links between posttraumatic stress disorder (PTSD),
physical pain, and problematic cannabis use are well-established. Most (66%) veterans in
treatment for PTSD experience pain (Shipherd et al., 2007), and nearly half (49%) of those
seeking treatment for pain meet criteria for PTSD (Otis et al., 2010).Both PTSD and pain are
associated with cannabis use, as 72% of veterans with cannabis use disorder (CUD) are also
diagnosed with PTSD (J. L. Bryan et al., 2021), and pain is among the most common reasons
veterans cite for using cannabis (Metrik et al., 2018). Several theoretical models have been
developed to explain links between multi-morbid problems such as pain, PTSD, and problematic
cannabis use (Asmundson et al., 2002; Khantzian, 1997). However, these theoretical models, and
empirical tests of these theoretical models, generally describe bivariate associations between
pairs of problems, such as associations between pain and PTSD, and associations between either
pain and PTSD and cannabis use (McAndrew et al., 2019; Ravn et al., 2018). This limits what is
known about how multi-morbid pain, PTSD, and cannabis use might influence one another
(trivariate associations) among individuals with all three problems, a likely symptom
presentation given the prevalence of multi-morbidity and recent increases in cannabis use among
veterans (Wolfgang & Hoge, 2023).
One promising method for extending theory of multi-morbidity and exploring trivariate
associations are intensive longitudinal studies where participants provide many frequent
measurements of their symptoms over time, potentially offering increased clarity on the nature of
symptom associations (Bolger & Laurenceau, 2013). In daily diary studies, for example,
participants self-report their symptoms one or multiple times on a daily basis for several days or
weeks, which can minimize retrospective recall bias and allow for precise longitudinal modeling

26
of associations (Gunthert & Wenze, 2012). Since pain and behavioral health symptoms are
known to fluctuate daily, daily diary studies can provide important information on within-person
dynamic associations that traditional studies using quarterly (3 to 6 month) intervals cannot
accurately capture (Bolger & Laurenceau, 2013).
In the existing literature on multi-morbidity, the primary theoretical model that explains
bivariate association between PTSD and pain is the mutual maintenance model, and it might also
have implications for understanding problematic cannabis use among veterans. The mutual
maintenance model posits that symptoms of PTSD and pain exacerbate one another via several
cognitive, affective, and behavioral pathways (Sharp & Harvey, 2001). It presumes a
bidirectional association where PTSD symptoms lead to a worsening of pain and vice versa via
several symptom pathways. For example, those with PTSD are said to have attentional biases
that could result in them paying more attention to their pain, while pain from a traumatic injury
could serve as a reminder of trauma, and both conditions could result in avoidant coping
behavior that ultimately could contribute to a worsening of symptomology (Sharp & Harvey,
2001). Given self-medication models of substance use suggest that individuals use substances to
cope with physical and psychological distress (Khantzian, 1997), pain and PTSD symptom
escalation due to mutual maintenance may subsequently lead to increases in cannabis use. Still,
tests of either the mutual maintenance or self-medication models have been siloed (i.e., testing
one model and bivariate rather than trivariate associations), and results have been mixed. Two
studies testing the mutual maintenance model reported a bidirectional association between pain
and PTSD symptoms (Lee et al., 2019; Stratton et al., 2014), however, another study reported
stronger support for a unidirectional association as pain was associated with subsequent PTSD
but PTSD was not always associated with subsequent pain (McAndrew et al., 2019). And while

27
the self-mediation model implies those with PTSD and/or pain use cannabis for symptom
improvement, one recent longitudinal study with veterans revealed baseline cannabis use was
associated with worse self-reported PTSD symptoms 6-months later (Metrik et al., 2020). With
respect to pain, there is some evidence from intervention studies that cannabis may have
analgesic (i.e., pain relieving) qualities (Metrik et al., 2018; Nugent et al., 2017), but results from
a large-scale survey study (non-veteran sample) suggested frequent cannabis use may induce
hyperalgesia (i.e., increased sensitivity to pain; Boehnke et al., 2020).
One potential explanation for these mixed results is prior study designs often have not
utilized time intervals that can capture frequent fluctuation in symptomology, and this could
impede the validity of results. Existing longitudinal studies of PTSD and pain among veterans
have assessed changes measured at quarterly (i.e., 3-, 6-, or 9-month) intervals. Thus, prior work
may be subject to retrospective recall bias and may not effectively capture changes in pain and
PTSD symptoms, which have both been shown to fluctuate daily (Eather et al., 2019; Mun et al.,
2019; Schuler et al., 2021). Daily diary studies are therefore promising for clarifying associations
between pain and PTSD, as well as both variables’ associations with cannabis use. To date, daily
diary studies have not been used to explore multi-morbid pain, PTSD symptoms, and cannabis
use among veterans, though a small number of these studies have been conducted in non-veteran
samples with pain and PTSD. In a daily diary study among individuals who experienced
whiplash, for example, there was evidence that pain was associated with subsequent PTSD
symptoms, however PTSD was not associated with subsequent pain (Eather et al., 2019). In
another daily diary study with college students, pain interfered with daily life more so among
those with probable PTSD compared with those without (Berghoff et al., 2018).

28
It may also be the case that there are between-person differences in how multi-morbid
symptoms are associated at the daily level and predisposing psychological factors may predict
these individual differences in symptom associations (i.e., act as moderators). The shared
vulnerability model was developed to explain how these predisposing factors can contribute to
the development and maintenance of two or more multi-morbid problems concurrently
(Asmundson et al., 2002). Much of the literature on shared vulnerability for pain and PTSD has
focused on a predisposing factor known as anxiety sensitivity, which is a tendency to respond
fearfully to symptoms of anxiety (Ocañez et al., 2010; S. Taylor, 2003). Because pain and PTSD
involve symptoms that mirror anxiety such as muscle tightness and shortness of breath,
theoretical literature suggests individuals high in anxiety respond fearfully to both conditions,
cope relatively poorly, and experience worse symptomology (Asmundson et al., 2002). Empirical
literature has also implicated anxiety sensitivity as a determinant of problematic cannabis use
(Zvolensky et al., 2018), thus veterans who are high in anxiety sensitivity may be especially
likely to engage in problematic cannabis use in response to the physiological symptoms of pain
and PTSD symptoms thinking that cannabis could help manage such symptoms (Khantzian,
1997). As such, anxiety sensitivity may actually moderate multi-morbid, trivariate symptom
associations (i.e., make it more likely that pain or PTSD predict cannabis use, or that pain and
PTSD are bidirectionally associated) rather than simply having a direct association with each
individual problem. However, the moderating role of anxiety sensitivity in the context of the
day-to-day trivariate associations between pain, PTSD, and cannabis use has not been explored.
The present study uses intensive longitudinal methods (i.e., dynamic structural equation
models) (Asparouhov et al., 2018) with daily diary data from a sample of post-9/11 U.S. veterans
to explore the dynamic associations between multi-morbid PTSD symptoms, pain, and cannabis

29
use, and the moderating role of anxiety sensitivity. The first Aim, in line with the mutual
maintenance model, is to test the dynamic, bidirectional associations between pain and PTSD at
the daily level. The second Aim, in line with the mutual maintenance and self-medication
models, is to include cannabis use frequency to understand the dynamic associations between
PTSD symptoms, pain, and cannabis use (whether pain and/or PTSD are associated with
cannabis use and vice versa) at the daily level. The third Aim, in line with the shared
vulnerability model, is to assess the moderating role of anxiety sensitivity and whether daily
associations between PTSD symptoms, pain, and cannabis use differ between those high vs. not
high in anxiety sensitivity. Exploring multi-morbid symptom associations at the daily level could
improve theoretical knowledge and lead to improvements in interventions and systems of care
for veterans.
Methods
Participants and Procedures
Data are from a larger study (1R21DA051802) of PTSD symptoms and cannabis use
among U.S. veterans. Participants (n = 74) were recruited via social media platforms using
previously developed measures to screen out those misrepresenting themselves as veterans
(Pedersen et al., 2017). Criteria for the parent study were: (a) U.S. veteran aged 18 or older
separated or discharged from military service from the Air Force, Army, Marine Corps, or Navy
within the past three years; (b) not currently on active duty service or serving in the active
reserves or national guard; (c) served in post-9/11 conflicts (i.e., in Iraq/Afghanistan); (d) able to
read English; (e) own a smartphone released since 2012 with Internet access and have interest in
using apps on that phone; (f) not receiving treatment for cannabis, alcohol, or other drug use or

30
PTSD at the VA or other health care provider; (g) Cannabis Use Disorder Identification Test
score of 8 or higher, indicating at least minimal levels of problematic cannabis use; Thake &
Davis, 2011) (h) use of cannabis in the prior month, (i) experienced a traumatic event during
military service, and (j) had a Primary Care PTSD screen score of 1 or more associated with
military trauma exposure, representing at least one PTSD symptom cluster; Cameron & Gusman,
2003).
Data on participants’ symptomology over the subsequent three months were collected via
the smartphone application MAVERICK, which was developed by investigators at King’s
College London along with United Kingdom military personnel. Data were collected at baseline,
monthly, and twice daily: during the morning (regarding symptoms experienced during the prior
evening/night, between 6pm and 6am) and during the evening (regarding symptoms experienced
during that day, between 6am and 6pm). Participants were asked to provide daily survey data
during 86 days, as they did not receive daily surveys on the same day in which they received
longer, baseline or monthly surveys. To encourage users to complete daily surveys, in-app
notifications were sent via smartphone push notification and text messaging. Participants were
paid $20 for completing the baseline and each monthly survey, $1 per day they completed both
daily surveys, and a $6 bonus if they completed daily surveys at least 90% of days.
See Table 2.1 for participant demographics and clinical characteristics at baseline. Study
participants were 79% male. The sample was relatively diverse in terms of race and ethnicity,
with the racial make-up of participants being 60% White, 9% Black, and the remining (30%)
mixed race or other, and 37% of participants were of Hispanic/Latinx ethnicity. On average,
participants had experienced 2.78 deployments. At baseline, participants reported using cannabis
an average of 19.5 out of the past 30 days. The mean pain intensity score on the PROMIS Pain

31
Intensity (short form) measure was 7.53 representing mild-moderate pain on average (Stone et
al., 2016). The mean PTSD symptom score on the PTSD Checklist for DSM-5 (PCL-5) was
38.7, with 39 (53%) participants meeting criteria for likely PTSD (Bovin et al., 2016). The mean
anxiety sensitivity score on the Anxiety Sensitivity Index-3 (ASI-3) was 29.7, with 46 (62%)
exhibiting a high level of anxiety sensitivity (Allan et al., 2014).
Individual participants provided a mean of 51.75 days of daily data (SD = 32.3 days; min
= 1 days; max = 86 days; median = 64 days). In total, participants provided data a combined
3881 days. See Table 2.2 for descriptive statistics and correlations between daily study variables
and baseline anxiety sensitivity. Correlations between study variables were mostly significant
and in the expected directions.
Measures
Baseline characteristics. Participants’ self-reported sex at birth, race and ethnicity, and
military characteristics (such as branch, number of combat deployments, time since last
deployment) were collected at baseline. Cannabis use was reported as the total number of days
used out of the prior thirty days participants used cannabis products that contain Delta-9-
Tetrahydrocannabinol (THC, the ingredient in some cannabis products that gets users high).
PTSD symptomology over the past month was assessed with the 20-item PCL-5 (Bovin et al.,
2016). Pain intensity over the past seven days was assessed using the three-item PROMIS Pain
Intensity (short form) measure (Stone et al., 2016).
Daily PTSD symptoms. The 4-item version of the PCL-5 (PCL4-5; Geier et al., 2020)
was used to measure PTSD symptoms twice per day, in the evening (regarding symptoms during
that day) and in the morning (regarding symptoms during the prior night). Participants were
asked to rate, on a scale from 0 (not at all) to 4 (extremely) the severity of four specific

32
symptoms: intrusive thoughts, avoidance of trauma reminders, negative expectations of self/the
world, and being easily startled. Items were summed for a total PCL4-5 score. The PCL4-5 has
been shown to correlate well with the full PCL-5 (Geier et al., 2020) and has been used as a
continuous scale of PTSD symptoms in daily diary studies (Huang et al., 2022).
Daily pain intensity. The numerical rating scale (NRS) of pain intensity was used to
measure daily pain intensity twice per day, in the evening (regarding pain during that day) and in
the morning (regarding pain during the prior night). Participants were asked rate their pain 0 (no
pain) to 10 (worst pain ever). The NRS has been validated widely in clinical and research
contexts (Williamson & Hoggart, 2005).
Daily cannabis use. Participants were asked about their cannabis use twice per day, in
the evening (regarding cannabis use during that day) and in the morning (regarding cannabis use
during the prior night). Participants responded to a single item inquiring how many times they
used cannabis products containing THC.
Anxiety sensitivity. The 18-item ASI-3 (Taylor et al., 2007) was administered at baseline
to measure anxiety sensitivity. Example items include “It scares me when my heart beats
rapidly,” “When I feel pain in my chest, I worry that I’m going to have a heart attack,” and
“When my thoughts seem to speed up, I worry that I might be going crazy.” Participants were
asked to rate each item from 0 (very little) to 4 (very much). In line with prior psychometric
work with the ASI-3 among adults (Allan et al., 2014), a score of ≥ 23 indicates high anxiety
sensitivity and was used as a cutoff for moderation analyses.
Analytic Plan
The present study utilizes dynamic structural equation modeling (DSEM), which was
developed for intensive longitudinal data and merges time series analysis, multilevel modeling,

33
structural equation modeling, and time-varying effects modeling (Asparouhov et al., 2018).
Specifically, DSEM models the autocorrelation of intensive longitudinal data through lagged
regressions between observed and latent variables, allows for observations to be nested within
persons, and for the modeling of individual differences in parameters all in one framework.
DSEM can test between- and within-person associations in a single model and has several
advantages over other intensive longitudinal methods, which are relevant for the proposed
analyses: unlike traditional time series models, DSEM allows for modeling data from more than
one individual concurrently, and unlike traditional multilevel models, DSEM allows for
structural associations between multiple outcomes (i.e., bidirectional effects) (McNeish &
Hamaker, 2020). DSEM also overcomes statistical biases commonly affecting multilevel models,
including Nickell’s bias (the estimated autoregressive parameter is often smaller than the true
effect, which would mean that cross-lagged associations between pain, PTSD, and cannabis use
would not properly account for prior symptomology) and Lüdtke’s Bias (occurs because
observed person-level means for pain, PTSD, and cannabis use could be susceptible to
measurement error). DSEM uses Bayesian Markov Chain Monte Carlo (MCMC) for estimation;
as Bayesian MCMC leverages conditional distributions that enable greater computational
efficiency compared with traditional maximum likelihood methods, DSEM can model a large
number of random effects to test moderation (such as by anxiety sensitivity) in a complex
multivariate model. Mplus version 8 was used to carry out study aims, as it has built-in DSEM
functionality (Muthén & Muthén, 2017).
Daily diary data was aggregated at the daily level by taking the maximum of each day’s
two scores on a variable if an individual responded to both surveys that day, or using the single
available score if an individual responded to one survey. Days with no responses were marked as

34
missing. Aggregating data in this way maximized the number of non-missing datapoints and was
conceptually in line with our study aims of studying day-to-day symptom associations. DSEM
treats remaining missing data similarly to random effects, where an individual’s missing values
on a given day are estimated conditional on the neighboring days and the individual’s
autoregressive parameter (Hamaker et al., 2018). Prior to analyses, stationarity of all three
constructs was assessed as DSEM assumes stationarity (i.e., the means, variance, and
autocorrelations of the PTSD symptoms, cannabis use, and pain do not systematically change
over time; Falkenström et al., 2017). While slight time trends were detected for PTSD symptoms
and cannabis use, residualizing these variables to control for time trends (non-stationarity) did
not change the substantive results (parameter estimates were similar in models that did and did
not control for time trends), and at times led to worse DSEM model fit (indicated by the
Deviance Information Criterion), thus the raw (non-residualized) variables were used in
analyses.
To model the daily, dynamic associations between PTSD symptoms and pain, we fit a
bivariate DSEM (See Figure 2.1 for a conceptual model). This model includes within-person,
autoregressive parameters for both constructs (an individual’s current day pain predicted by their
level of pain the prior day [𝑏𝑃𝐴𝑃𝐴]; current day PTSD symptoms predicted by their level of
PTSD symptoms the prior day [𝑏𝑃𝑇𝑃𝑇]), and within-person cross-lagged parameters (an
individual’s current day pain predicted by their level of PTSD symptoms the prior day [𝑏𝑃𝑇𝑃𝐴];
current day PTSD symptoms predicted their level of pain the prior day [𝑏𝑃𝐴𝑃𝑇]). Models also
included within-person correlations (correlations between an individual’s pain on a specific day
their PTSD symptoms on the same day [φ𝐶𝑜𝑣]) and between-person correlations (correlations

35
between an individual’s average pain level, relative to the rest of the sample, and their average
PTSD symptoms, relative to the rest of the sample).
To understand the daily, dynamic associations between symptoms of PTSD, physical
pain, and cannabis use, we fit a trivariate DSEM that also included a within-person,
autoregressive parameter for cannabis use, within-person cross-lagged parameters between
cannabis use and both pain and PTSD, and within- and between-person correlations between all
three constructs. To test whether anxiety sensitivity moderated the associations between PTSD,
physical pain, and cannabis use, we refit both the bivariate and trivariate models, estimating
random effects for within-person parameters and regressing the random effects on the
dichotomous anxiety sensitivity variable. This allowed us to test the effect of high anxiety
sensitivity on individual model parameters. See Figure 2.2 for a conceptual model of a bivariate
DSEM testing the role of a moderator (anxiety sensitivity), where the arrows from anxiety
sensitivity to model parameters [indicated by 𝑏 and φ for within-person parameters] at the
between level indicate random effects.
Participants’ self-reported biological sex, race/ethnicity, and number of deployments were
included as covariates on the between-person parameters for pain, PTSD, and cannabis use in all
models. Mplus uses the Potential Scale Reduction (PSR) value to check model convergence. The
PSR is calculated by comparing the variance within Bayesian MCMC chains to the variance
between chains. A PSR close to 1.00 indicates that the chains are converging well, suggesting
parameter estimates are reliable, thus Mplus provides DSEM model output after the potential
scale reduction (PSR) drops close below 1.10 during estimation. In line with recommendations,
all models were rerun with double the number of iterations to check convergence and that the
PSR value did not increase.

36
Results
Tables 2.3 and 2.4 provide model results with unstandardized parameter estimates and
standard errors for final DSEM models. In the text below, unstandardized posterior median
estimates are reported for lagged effects (b, the median value of the distribution of a model
parameter based on the posterior distribution in Bayesian analysis, a measure of the typical
association between variables across time points and individuals), and correlations are
represented by φstandardized.
Daily dynamics for pain and PTSD symptoms (bivariate DSEM model)
The bivariate DSEM model with daily pain and PTSD symptoms converged after 200
iterations with the PSR dropping below 1.03, and the PSR remained below 1.03 after the
iterations were doubled to 400 providing additional evidence for convergence. Cross-lagged
associations between pain and PTSD symptoms were not significant at the within-person level.
However, there was a significant within-person correlation between same-day pain and PTSD
symptoms (φstandardized = 0.16; 95% CI = [0.13, 0.19]), meaning, on days in which a veteran had
heightened pain they also tended to have heightened PTSD symptoms. At the between-person
level, veterans who tended to report high daily pain levels relative to other veterans in the sample
also tended to report relatively high levels of PTSD symptoms (φstandardized = 0.24; 95% CI =
[0.01, 0.45]) relative to other veterans in the sample. See Table 2.3 (Model 1) for unstandardized
model results.
Daily dynamics for pain, PTSD symptoms, and cannabis use (trivariate DSEM model)
The trivariate DSEM model with daily pain, PTSD symptoms, and cannabis use
converged after 400 iterations with the PSR dropping below 1.03, and the PSR dropped below
1.01 after the iterations were doubled to 800 providing additional evidence for convergence.

37
There were significant, negative cross-lagged associations between pain and cannabis use at the
within-person level. That is, when a veteran had heightened pain on one day, they tended to have
reduced cannabis use the next day (b = -0.05; 95% CI = [-0.10, -0.002]), and when a veteran had
heightened cannabis use on one day, they tended to have reduced pain the next day (b = -0.02;
95% CI = [-0.05, -0.001]). There was also a significant within-person correlation between sameday pain and PTSD symptoms (φstandardized = 0.16; 95% CI = [0.13, 0.19]). Pain, PTSD symptoms,
and cannabis use did not significantly covary at the between-person level. See Table 2.3 (Model
2) for unstandardized model results.
The moderating role of anxiety sensitivity on daily dynamics (random effects DSEM
models)
The bivariate DSEM model (of pain and PTSD symptoms) with anxiety sensitivity as a
between-person predictor of random effects converged after 500 iterations with the PSR
dropping below 1.05, and the PSR dropped to 1.02 after the iterations were doubled to 1000
providing additional evidence for convergence. High anxiety sensitivity did not predict any
random effects in the model, suggesting anxiety sensitivity in this sample does not play a
moderating role in bivariate symptom dynamics between pain and PTSD at the daily level. See
Table 2.4 (Model 3) for unstandardized model results.
The trivariate DSEM model (of pain, PTSD symptoms, and cannabis use) with anxiety
sensitivity as a between-person predictor of random effects converged after 1600 iterations with
the PSR dropping to 1.08, and the PSR dropped below 1.06 after the iterations were doubled to
3200 providing additional evidence for convergence. High anxiety sensitivity did not predict any
random effects in the model, suggesting anxiety sensitivity in this sample does not play a

38
moderating role in trivariate symptom dynamics between pain, PTSD symptoms, and cannabis
use at the daily level. See Table 2.4 (Model 4) for unstandardized model results.
Discussion
Several theoretical models have been developed to explain multi-morbidity, including the
mutual maintenance, self-medication, and shared vulnerability models (Asmundson et al., 2002;
Khantzian, 1997). These models are relevant for understanding physical and behavioral health
problems among veterans, who have high rates of multi-morbid pain, PTSD, and problematic
cannabis use (J. L. Bryan et al., 2021; Metrik et al., 2020; Otis et al., 2010). However, these
models and empirical tests of these models have focused on associations between pairs of
problems, limiting what is known about symptom dynamics among individuals with all three
problems. This study exemplifies how recent advances in data collection and quantitative
methods can enhance theoretical and empirical knowledge of such symptom dynamics (Bolger &
Laurenceau, 2013; Giordano et al., 2018). By leveraging survey data on symptom levels
measured daily (i.e., daily diary data) and a modeling technique that can leverage such data to
capture multivariate and dynamic change (i.e., DSEM; Asparouhov et al., 2018), our analyses
provide insight into the day-to-day dynamic associations between pain, PTSD symptoms, and
cannabis use among a sample of veterans. This represents a critical advance given that pain and
behavioral health symptoms are known to fluctuate daily (Eather et al., 2019; Mun et al., 2019;
Schuler et al., 2021) but most prior research with veterans explores associations between
symptoms assessed over several month intervals (Lee et al., 2019; McAndrew et al., 2019;
Metrik et al., 2022; Stratton et al., 2014; Turna & MacKillop, 2021). Below, we discuss our

39
pattern of results in the context of existing theory and empirical work, implications for
intervening more effectively, and avenues for additional research.
While the mutual maintenance model suggests there are bidirectional associations
between pain and PTSD, DSEM results do not provide evidence for within-person lagged
associations between these variables at the daily level. That is, if a veteran’s pain was heightened
on one day relative to their own average, this did not predict heightened PTSD symptoms the
next day or vice versa. Since prior work with veterans has reported bidirectional associations
between pain and PTSD symptoms assessed in 3-6-month intervals (Lee et al., 2019; Stratton et
al., 2014), this again suggests the timing of measurement matters when assessing symptom
associations and mutual maintenance may occur over time periods besides one day to the next. In
fact, in all our DSEM models, same-day correlations between pain and PTSD were significant at
the within-person level, such that on days in which a veteran’s pain was heightened, that
veteran’s PTSD symptoms were also heightened. Mutual maintenance may in fact exist at the
day level but require multiple assessments per day to capture such as with ecological momentary
assessment data (Shiffman et al., 2008). Such methodology could capture more proximal risk
factors that could affect one or both conditions throughout the day, such as PTSD triggers (e.g.,
challenging social interactions, experiencing loud noises) or injuring or otherwise overexerting
oneself physically. It may also be that there are other variables that vary day to day that can
influence daily pain and PTSD symptoms concurrently; in other words, daily pain and PTSD
may not necessarily maintain one another so much as be maintained by a time-varying third
variable. For example, poor sleep and daily stressors have been linked with both heightened
PTSD symptoms (Messman et al., 2022; Short et al., 2017) and pain (Ivey et al., 2018; Wei et al.,
2018) at the daily level among non-veteran samples. Continuing to explore the influence of

40
sleep, stress, and other time-varying psychosocial variables on daily pain and PTSD among
veterans could provide salient prevention targets to reduce the number of days with high levels of
multi-morbid symptomology.
Including cannabis use in DSEM models revealed significant within-person lagged
associations between cannabis use and pain. That is, if a veteran experienced higher cannabis use
than their own average on one day, they were predicted to experience lower pain than their own
average the next day. While prior studies with veterans had demonstrated cannabis use is
associated with reduced pain (Metrik et al., 2018; Nugent et al., 2017), this is the first study to
demonstrate that cannabis use on one day might have an analgesic (i.e., pain relieving) effect that
persists through the next day. While the self-medication model suggests that a veteran
experiencing heightened pain would be at risk of increased cannabis use (Khantzian, 1997), our
results run contrary to the self-medication model and as a veteran experiencing higher pain than
their own average on one day was predicted to use lower amounts of cannabis than their own
average the next day. Given that this is a sample with relatively high cannabis use, it is possible
that veterans regulate (i.e., reduce) their use of cannabis after pain goes up it is because they
recognize they are building tolerance to cannabis’s analgesic (i.e., pain reducing) effects
(Romero-Sandoval et al., 2018). That is, heavy cannabis users who begin to experience pain may
believe that by reducing their use for a period, they will be able to again derive benefit from
cannabis in terms of reducing their pain. This could be assessed using a time-varying covariate
model and exploring whether the effect of cannabis on pain differs the longer one has used
cannabis, or if the effect of pain on cannabis differs at different levels of cannabis use.
Our analyses revealed no differences in associations between pain, PTSD symptoms, and
cannabis use among those with differing levels of anxiety sensitivity. While prior work with

41
veterans revealed that anxiety sensitivity may mediate links between pain and PTSD symptoms
(Raines et al., 2022), ours was the first to test anxiety sensitivity as a moderator of pain and
PTSD symptom associations. This could be interpreted as a lack of evidence of the shared
vulnerability model (i.e., the idea that some individuals have predisposing factors such as high
anxiety sensitivity that increase vulnerability for multi-morbid problems at the between-person
level; Asmundson et al., 2002). However, the significant and positive between-person
associations between pain and PTSD in most models provides evidence that there may be other
predisposing factors at the between-person level, as this parameter suggests those with high pain
relative to the entire sample had high PTSD relative to the entire sample. Factors that confer
vulnerability for both problems could be as wide ranging as physically and psychologically
strenuous deployment experiences (Chui et al., 2022), physical disability (Duckworth & Iezzi,
2010), neurological changes (Cifu et al., 2013), insomnia (Babson & Feldner, 2010; Saconi et al.,
2021), and lack of access to healthcare (McDermott et al., 2017). Researchers should continue to
disentangle the complex links and interactions between factors like these and multi-morbid
problems, as this could inform strategies to address and prevent such problems from developing.
Results should be interpreted in light of methodological limitations. First, analyses
involved a single sample of post-9/11 veterans. While intensive longitudinal methods employed
in this study allow for precise day-to-day dynamic associations among this sample, it is possible
that results would not generalize to other veterans, in particular those with less severe behavioral
health symptomology, more severe pain (such as veterans recruited from a pain clinic), those
who are in traditional behavioral health treatment, or those who are less likely to participate in a
daily diary mobile app study. Considering this was a sample that was relatively high in cannabis
use at baseline, it would be germane to explore whether symptom associations differ for alcohol,

42
for other substances, or for co-use in this sample as well as in a low cannabis using sample.
Considering the relevance of opioids in particular for pain, it is a limitation that we do not assess
or control for opioid use or initiation. Additionally, the lagged associations reported are specific
to the day level, and it is possible that symptom associations might differ hour-to-hour or
moment-to-moment. Also, while within-person analytic methods are powerful because they hold
all between-person differences constant (each individual serves as their own control), we are still
unable to draw definitive causal conclusions and are unable to say with certainty, for example,
that heightened pain on one day causes reduced cannabis use the next day. Future studies
combining intensive longitudinal methods with an intervention (such as one targeting pain) could
help clarify the causal influence of pain on cannabis use at the daily level. Our pain variable was
selected as a validated measure with a low participant burden (a single item measuring pain
intensity), but there are additional dimensions of pain (i.e., disability, pain-related negative
affect; Clark et al., 2003) that are not captured by this item. Researchers may develop pain
measures that are both multidimensional and low burden (with a minimal number of items) to
capture the complexity of pain in studies that also leverage novel intensive longitudinal
methodology. As we did not survey participants about the type or cause of their pain (e.g.,
musculoskeletal vs. neuropathic pain, whether pain was due to an injury), future research should
consider whether type or cause of pain influence multi-morbid symptom associations.
Still, this work provides a novel picture of the daily, dynamic associations between multimorbid problems among veterans, and can therefore prompt us to think about how we can
monitor, predict, and address symptomology in a more effective manner. As veterans who had
relatively high pain were likely to have relatively high PTSD at the between-person level, and
days with heightened pain were also likely to be days with heightened PTSD at the within-person

43
level, clinical interventions and coping strategies that target transdiagnostic processes implicated
in both conditions (i.e., mindfulness; Black, 2009; Garland et al., 2014) should be prioritized.
The lagged association between cannabis use and pain suggests veterans may be deriving
benefits from cannabis when it comes to managing pain. Providers should stay informed on the
changing scientific and legal landscape when it comes to cannabis use and be prepared to discuss
the benefits and consequences their veteran patients may be experiencing. Finally, nascent efforts
to integrate novel symptom reporting tools such as those used in this study (i.e., daily diary
mobile apps) directly into healthcare systems could improve providers’ ability to monitor and
address multi-morbid symptomology (Daniëls et al., 2021).

44
Table 2.1 Participant characteristics (n = 74)
Mean or # SD or %
Age 34.18 8.01
Sex
Male 59 79%
Female 15 20%
Race
White 45 60%
Black 7 9%
Asian 3 4%
American Indian or Alaska Native 1 1%
Mixed Race or Other 19 25%
Ethnicity
Non-Hispanic/Latnix 46 61%
Hispanic/Latinx 28 37%
Branch of service
Air Force 10 14%
Army 34 48%
Marine Corps 16 23%
Navy 11 15%
Number of deployments 2.78 3.26
Time since last discharge (Years) 3.62 1.93
Pain intensity 7.53 2.8
PTSD symptoms 38.7 17
Cannabis using days (during past 30 days) 19.5 11.4
Anxiety sensitivity 29.7 17
Note: Participant characteristics collected during baseline and screening surveys.
Sex is biological sex at birth reported by the participant. PTSD symptoms were
measured by the PTSD Checklist for DSM-5 (range: 0-80). Pain was measured by
the PROMIS Pain Intensity scale (range: 3-15). Cannabis using days are number of
days participant drank or used cannabis during the past 30 days. Anxiety sensitivity
was measured by the Anxiety Sensitivity Index (range: 0-72). Total n for branch of
service was 71 due to missingness on the screening survey.

45
Table 2.2 Study variable descriptive statistics and bivariate correlations
Variable M SD N 1 2 3 4
1. Anxiety sensitivity (baseline) 29.73 17.06 75 -
2. Pain intensity (daily) 3.45 2.71 3871 .13 -
3. PTSD symptoms (daily) 7.84 3.58 3785 .30 .27 -
4. Cannabis use (daily) 2.45 4.82 3871 .02 .09 .14 -
Note: Correlations were calculated using Pearson's R. All correlations were significant at p
< .05, except baseline anxiety sensitivity with daily cannabis use. Anxiety sensitivity was
measured at baseline with the Anxiety Sensitivity Index (range = 0-72); Pain intensity was
measured daily using the Numerical Rating Scale (range = 0-10); PTSD symptoms were
measured daily using the PCL4-5 (range = 0-16); Cannabis use was measured daily with a
single item asking the participant the number of times they used that day (range = 0-30).

46
Table 2.3 DSEM model output depicting daily, dynamic symptom associations
Model 1: PTSD and Pain
Model 2: PTSD, Pain, and
Cannabis
Estimate 95% CI Estimate 95% CI
Within-person day cross-lags
Paint-1 → PTSDt -0.003 [-0.05, 0.04] -0.006 [-0.06, 0.05]
PTSDt-1 → Paint 0.01 [-0.01, 0.03] 0.01 [-0.01, 0.03]
Paint-1 → Cannabist -0.05 [-0.10, -0.002]
Cannabist-1 → Paint -0.02 [-0.05, -0.001]
PTSDt-1 → Cannabist -0.02 [-0.05, 0.01]
Cannabist-1 → PTSDt 0.003 [-0.03, 0.04]
Within-person day autoregressions
PTSDt-1 → PTSDt 0.4 [0.37, 0.44] 0.4 [0.37, 0.43]
Paint-1 → Paint 0.38 [0.34, 0.41] 0.38 [0.35, 0.41]
Cannabist-1 → Cannabist 0.37 [0.34, 0.40]
Within-person day (co)variances
Pain variance 1.38 [1.32, 1.45] 1.38 [1.31, 1.44]
PTSD variance 3.61 [3.44, 3.77] 3.63 [3.47, 3.79]
Cannabis variance 3.28 [3.14, 3.42]
Pain with PTSD covariance 0.35 [0.28, 0.43] 0.35 [0.28, 0.43]
Pain with Cannabis covariance 0.06 [-0.01, 0.13]
PTSD with Cannabis covariance -0.07 [-0.18, 0.04]
Between-person means and (co)variances
Pain mean 2.98 [1.92, 4.13] 2.34 [1.88, 3.94]
PTSD mean 7.09 [5.73, 8.52] 7.11 [5.69, 8.54]
Cannabis mean 3.15 [1.15, 5.07]
Pain variance 6.29 [4.39, 9.04] 6.31 [4.51, 9.19]
PTSD variance 9.6 [6.85, 14.18] 9.83 [6.98, 14.33]
Cannabis variance 21.05 [15.07, 30.72]
Pain with PTSD covariance 1.86 [0.08, 3.86] 1.83 [-0.12, 4.18]
Pain with Cannabis covariance 0.06 [-3.00, 3.05]
PTSD with Cannabis covariance 2.28 [-1.24, 6.40]
Fit statistics
Free parameters 24 39
DIC 36833.32 56772.93
Number of iterations 400 800
PSR value 1.027 1.005

47
Note: Unstandardized posterior median estimates and 95% credible intervals provided from bivariate and trivariate
DSEM models capturing associations between pain, PTSD symptoms, and cannabis use at the daily level. Estimates
for control variables (sex, race/ethnicity, and number of deployments) on between-person mean parameters not
shown for readability. Subscripts (t) identify time of measurement.
Bold = significant, credible interval does not cross 0; DIC = Deviance Information Criterion; PSR = Potential Scale
Reduction.

48
Table 2.4 DSEM model output depicting the moderating role of anxiety sensitivity on daily,
dynamic symptom associations
Model 3: PTSD and Pain with
ASI Random Effects
Model 4: PTSD, Pain, and
Cannabis with ASI Random
Effects
Estimate 95% CI Estimate 95% CI
Within-person day cross-lags
Paint-1 → PTSDt 0.01 [-0.05, 0.06] 0.01 [-0.05, 0.07]
PTSDt-1 → Paint 0.03 [-0.01, 0.07] 0.03 [-0.01, 0.07]
Paint-1 → Cannabist -0.01 [-0.04, 0.03]
Cannabist-1 → Paint -0.01 [-0.06, 0.04]
PTSDt-1 → Cannabist 0.01 [-0.02, 0.03]
Cannabist-1 → PTSDt 0.001 [-0.05, 0.04]
Within-person day autoregressions
PTSDt-1 → PTSDt 0.31 [0.19, 0.42] 0.32 [0.21, 0.42]
Paint-1 → Paint 0.29 [0.19, 0.39] 0.29 [0.19, 0.39]
Cannabist-1 → Cannabist 0.23 [0.11, 0.35]
Within-person day (co)variances
Pain variance -0.09 [-0.47, 0.27] -0.04 [-0.43, 0.34]
PTSD variance 0.45 [ -0.11, 1.01] 0.46 [-0.09, 0.99]
Cannabis variance -0.17 [-0.95, 0.69]
Pain with PTSD covariance 0.11 [0.03, 0.20] 0.12 [0.02, 0.21]
Pain with Cannabis covariance -0.02 [-0.08, 0.03]
PTSD with Cannabis covariance -0.04 [-0.11, 0.02]
Between-person means and (co)variances
Pain mean 2.86 [1.41, 4.30] 2.87 [1.36, 4.37]
PTSD mean 6.31 [4.49, 7.98] 6.25 [4.41, 8.17]
Cannabis mean 3.67 [1.07, 6.20]
Pain variance 6.07 [4.26, 8.98] 6.23 [4.39, 8.99]
PTSD variance 9.5 [6.69, 13.65] 10.04 [7.02, 15.21]
Cannabis variance 19.3 [13.71, 27.98]
Pain with PTSD covariance 2.1 [0.26, 4.44] 2.16 [0.15, 4.62]
Pain with Cannabis covariance 0.34 [-2.51, 3.37]
PTSD with Cannabis covariance 2.48 [-1.26, 6.59]
ASIHIGH moderating within-person random effects
( Paint-1 → PTSDt ) mod ASIHIGH = 1 -0.02 [-.10, 0.06] -0.02 [-0.11, 0.06]
( PTSDt-1 → Paint ) mod ASIHIGH = 1 -0.04 [-0.08, 0.01] -0.03 [-0.08, 0.01]

49
( Paint-1 → Cannabist ) mod ASIHIGH = 1 0.01 [-0.04, 0.05]
( Cannabist-1 → Paint ) mod ASIHIGH = 1 -0.01 [-0.07, 0.06]
( PTSDt-1 → Cannabist ) mod ASIHIGH = 1 -0.01 [-0.04, 0.02]
( Cannabist-1 → PTSDt ) mod ASIHIGH = 1 -0.01 [-0.07, 0.06]
( PTSDt-1 → PTSDt ) mod ASIHIGH = 1 0.06 [-0.09, 0.21] 0.03 [-0.10, 0.17]
( Paint-1 → Paint ) mod ASIHIGH = 1 0.07 [-0.07, 0.21] 0.08 [-0.05, 0.20]
( Cannabist-1 → Cannabist ) mod ASIHIGH = 1 0.04 [-0.11, 0.20]
( Pain variance ) mod ASIHIGH = 1 0.09 [-0.38, 0.55] 0.04 [-0.45, 0.51]
( PTSD variance ) mod ASIHIGH = 1 0.61 [-0.05, 1.34] 0.63 [-0.04, 1.34]
( Cannabis variance ) mod ASIHIGH = 1 0.09 [-0.98, 1.10]
( Pain w/ PTSD covariance ) mod ASIHIGH = 1 0.05 [-0.08, 0.16] 0.05 [-.07, .17]
( Pain with Cannabis covariance ) mod ASIHIGH = 1 0.06 [-0.02, 0.13]
( PTSD with Cannabis covariance ) mod ASIHIGH = 1 0.03 [-0.6, 0.11]
Fit statistics
Free parameters 40 74
DIC 33228.15 46331.34
Number of iterations 1400 3200
PSR value 1.074 1.059
Note: Unstandardized posterior median estimates and 95% credible intervals provided from bivariate and trivariate DSEM models
capturing associations between pain, PTSD symptoms, and cannabis use at the daily level. Estimates for control variables (sex,
race/ethnicity, and number of deployments) on between-person mean parameters not shown for readability. Subscripts (t) identify
time of measurement. To test moderation by anxiety sensitivity, random effects were estimated for model parameters and high
anxiety sensitivity was included as a predictor of random effects. Models also include high anxiety sensitivity as a predictor of
between-person means, and these estimates were non-significant.
Bold = significant, credible interval does not cross 0; DIC = Deviance Information Criterion; PSR = Potential Scale Reduction.

50
Figure 2.1 Conceptual model of a bivariate DSEM model
Note: Conceptual model of bivariate DSEM model estimating dynamic associations between
pain and PTSD symptoms at the daily level. The above conceptual model is simplified and does
not represent every aspect of a DSEM model. The DSEM model decomposes the data into a
within-person and between-person component. At the within-person level, a time-series model is
specified with pain (PA) and PTSD (PT) autoregressions, cross-lags, and within-day correlations.
At the between-person level, person-level means are estimated and correlated. Subscripts
indicate time of measurement. Superscripts indicate whether estimate is at the within (w) or
between (b) level.

51
Figure 2.2 Conceptual model of a bivariate DSEM model with a random effects moderator
Note: Conceptual model of bivariate DSEM model estimating dynamic associations between
pain and PTSD symptoms at the daily level and the moderating role of anxiety sensitivity. The
above conceptual model is simplified and does not represent every aspect of a DSEM model. In
this model, within-person model parameters (autoregressions, cross-lags, and within-day
correlations) are estimated as random effects (indicated by filled black circles), as are betweenperson means, meaning each individual has their own values for them. These parameters are
modeled as latent variables at the between-person level. Anxiety sensitivity is included as a
predictor of each of these parameters at the between-person level.

Supplemental analyses: Weekly symptom dynamics
Because a focus of this study was the significance of the timing of assessments in
understanding associations between multi-morbid symptoms, we ran additional analyses of
symptomology at the week level. To do so, daily diary data were first aggregated at the week
level by taking the average (mean) of available daily data for each week. Participants who had
less than two daily diary responses in a given week were marked as missing for that week.
We used an autoregressive latent-trajectory with structured residuals (ALT-SR) model to
understand the within-person associations between pain, PTSD symptoms, and cannabis use at
the weekly level (Curran et al., 2014). Like DSEM, the ALT-SR model offers an analysis of both
between- and within-person effects, but the ALT-SR is more appropriate for data with fewer time
points such as in weekly analyses. See Figure 2.S1 for a conceptual model of the ALT-SR model
with two variables, pain and PTSD symptoms. At the between-person level, effects are captured
by correlating random latent intercept and random latent growth factors (𝑃𝐴𝑖𝑃𝑇𝑖 and 𝑃𝐴𝑠𝑃𝑇𝑠
).
The remaining variance is “pushed” into the residual auto-regressive and cross-lagged portion of
the model. This allows the ALT-SR model to isolate bidirectional associations at the withinperson level (e.g., 𝑏𝑃𝐴𝑃𝑇1 and 𝑏𝑃𝑇𝑃𝐴1
).
In line with the daily analyses, we fit a bivariate ALT-SR to the weekly symptom
associations between pain and PTSD symptoms, as well as a trivariate ALT-SR model including
pain, PTSD symptoms, and cannabis use. In both cases, we employed a model fitting procedure
using model constraints and likelihood ratio tests (Curran et al., 2014) to identify the best fitting
ALT-SR model. Specifically, we tested whether slope variances and co-variances for study
constructs (pain, PTSD symptoms, cannabis use) should be estimated or constrained to zero, as
well as whether within-person lagged parameters (auto-regressive, within-time correlations, and

53
cross-lagged associations) should be constrained to be equal or freely estimated over time. We
also tested whether residual variances for pain, PTSD symptoms, and cannabis use should be
constrained to be equal or allowed to vary over time. We used a comparative fit index (CFI) of
0.90 of greater, root mean square error of approximation (RMSEA) of 0.08 or less, and
standardized root mean square error (SRMR) of less than 0.10 to indicate adequate final model
fit (Wang & Wang, 2019). Participants’ self-reported biological sex, race/ethnicity, and number
of deployments were included as covariates of between-person parameters (i.e., intercepts and
slopes). All analyses were conducted in Mplus version 8, which uses maximum likelihood to
handle missing data.
We fit our ALT-SR models using the first four weeks of data, as attempting to include all
twelve weeks resulted in convergence issues, likely due to there being a larger number of model
parameters than participants. We were also unable to test the moderating role of anxiety
sensitivity in weekly analyses: moderation with ALT-SR models typically involves dividing the
sample using a grouping variable and fitting a multi-group model, but dividing our sample based
on high/not high anxiety sensitivity resulted in relatively small sample sizes that would be
inappropriate for an ALT-SR model.
Results: Weekly dynamics for pain and PTSD symptoms (bivariate ALT-SR model)
For the bivariate model with pain and PTSD symptoms aggregated at the week level, our
ALT-SR model building process indicated residual variances for pain and PTSD symptoms
should be freely estimated over time, and slopes should be constrained to be equal. The final
model also included constrained cross-lags, autoregressions, and within-time correlations, as a
model with these parameters freely estimated would not converge. See Table 2.S2 for a summary

54
of all models and model fit comparisons for the bivariate ALT-SR model. All fit indices indicated
good model fit (CFI = 1.00; SRMR = .05; RMSEA = 0) for the final model.
Cross-lagged associations between pain and PTSD symptoms were not significant at the
within-person level. At the between-person level, veterans who reported high levels (intercept
value) of pain also reported relatively high (intercept value) PTSD symptoms (φstandardized = 0.43).
This between-person association between pain and PTSD symptoms was in line with DSEM
result (a between-person association between pain and PTSD when using a daily measurement
interval). See Table 2.S1 (Model 1) for unstandardized model results.
Results: Weekly dynamics for pain, PTSD symptoms, and cannabis use (trivariate ALT-SR
model)
For the trivariate model with pain, PTSD symptoms, and cannabis use aggregated at the
week level, our ALT-SR model building process indicated residual variances for pain, PTSD
symptoms, and cannabis use should be freely estimated over time. The final model also included
constrained cross-lags, autoregressions, and within-time correlations, as a model with these
parameters freely estimated would not converge. Attempting to also include random slopes for
pain and cannabis use resulted in a negative variance estimate, indicating problems with
convergence and overfitting. Because random slope variances were relatively small and
estimating these did not change our substantive results, we ultimately fit our final model without
random slopes. See Table 2.S3 for a summary of all models and model fit comparisons. Fit
indices generally indicated adequate model fit (CFI = .92; SRMR = .08; RMSEA = .12) for the
final model.

55
Cross-lagged associations between pain, PTSD symptoms, and cannabis use were not
significant at the within-person level. At the between-person level, veterans who reported high
levels (intercept value) of pain also reported relatively high (intercept value) PTSD symptoms
(φstandardized = 0.44), though cannabis use was not associated with pain or PTSD symptoms at the
between-person level. See Table 2.S1 (Model 2) for unstandardized model results.
Thus, ALT-SR results at the weekly level varied somewhat from the DSEM daily-level
results. Both analyses revealed between-person associations between pain and PTSD symptoms,
and a lack of lagged associations between these variables. However, while we report lagged
(negative) associations between pain and cannabis use at the daily level, the ALT-SR model did
not reveal evidence of such associations at the weekly level.

56
Figure 2.3 Conceptual model of a bivariate ALT-SR model
Note: Conceptual model of bivariate ALT-SR model estimating dynamic associations between
pain and PTSD symptoms at the week level. The above conceptual model is simplified and does
not represent every aspect of an ALT-SR models. Correlations between random intercepts and
slopes (𝑃𝐴𝑖𝑃𝑇𝑖 and 𝑃𝐴𝑠𝑃𝑇𝑠
) represent between-person effects. While we only show intercept-tointercept and slope-to-slope correlations, any estimated random intercepts are also correlated
with any estimated random slopes in final models. Residuals (e.g., error) of measured variables
(e.g., Pain1 or PTSD1) are identified as latent residuals as observed means, and variances are
constrained to zero. Residual variances are estimated as newly created latent variables (e.g., 𝜖𝑃𝐴1
and 𝜖𝑃𝑇1
) and “residuals of the residuals” are estimated as newly created within-person latent
constructs. These structured residuals are used to estimate the remaining variance left over after
random intercept and random slope variances are accounted for. Finally, the ALT-SR model
includes the within-time correlations of structured residuals, auto-regressive components of
structured residuals (e.g., 𝑏𝑃𝐴𝑃𝐴1 and 𝑏𝑃𝑇𝑃𝑇1
), and within-person cross-lagged effects (e.g.,
𝑏𝑃𝐴𝑃𝑇1 and 𝑏𝑃𝑇𝑃𝐴1
), allowing us to determine the within-person bidirectional effects of pain and
PTSD symptoms.

57
Table 2.5 ALT-SR model output depicting weekly, dynamic symptom associations
Model 1: PTSD and
Pain
Model 2: PTSD,
Pain, and Cannabis
Estimate SE Estimate SE
Within-person week cross-lags
Paint-1 → PTSDt
0.02 0.12 0.01 0.12
PTSDt-1 → Paint
-0.07 0.05 -0.07 0.05
Paint-1 → Cannabist 0.05 0.07
Cannabist-1 → Paint -0.14 0.08
PTSDt-1 → Cannabist 0.02 0.06
Cannabist-1 → PTSDt -0.02 0.18
Within-person week autoregressions
PTSDt-1 → PTSDt
0.67 0.12 0.67 0.12
Paint-1 → Paint
0.81 0.07 0.81 0.06
Cannabist-1 → Cannabist -0.01 0.09
Between-person intercepts and slopes
Pain intercept 2.66 0.5 2.6 0.49
PTSD intercept 6.99 0.68 6.96 0.65
Cannabis intercept 3.05 0.93
Pain slope 0.02 0.1 0.003 0.1
PTSD slope 0.03 0.16 0.03 0.16
Cannabis slope 3.05 0.08
Between-person (co)variances
Pain intercept variance 3.35 1.02 3.42 1.02
PTSD intercept varance 7.72 1.65 7.74 1.64
Cannabis intercept variance 17.44 2.97
Pain intercept with PTSD intercept covariance 2.22 0.97 2.25 0.97
Pain intercept with Cannabis intercept covariance 0.23 1.12
Cannabis intercept with PTSD intercept covariance 1.78 1.49
Fit statistics
Free parameters 33 57
-2LL 1791.84 2697.98
AIC 1857.84 2811.99
BIC 1932.51 2940.96
RMSEA 0 0.12
SRMR 0.05 0.08
CFI 1 0.92

58
Note: Unstandardized estimates and standard errors provided from bivariate and trivariate ALTSR models capturing associations between pain, PTSD symptoms, and cannabis use at the
weekly level. Estimates for control variables (sex, race/ethnicity, and number of deployments) on
latent intercept and slope parameters not shown for readability. Subscripts (t) identify time of
measurement. A single t indicates paths were constrained to be equal over time. Because latent
slope variances were not estimated in final model, we only include intercept-to-intercept
covariance to estimate between-person effects.
Bold = significant at p < .05; -2LL = negative two log-likelihood; AIC = Akaike Information
Criteria; BIC = Bayesian Information Criteria; RMSEA = Root Mean Square Error of
Approximation; SRMR = Standardized Root Mean Square Residual; CFI = Comparative Fit
Index; SE = standard error.

Table 2.6

60
Table 2.7

61
Chapter 3: Biopsychosocial predictors of pain among veterans with PTSD
Abstract
Introduction: Pain is highly prevalent among veterans with PTSD. Especially understudied
among veterans are determinants of pain intensity increases and the frequency with which one
experiences pain. The biopsychosocial model of pain implicates determinants, but most research
relies on traditional statistical methods that incorporate a relatively small number of predictors.
This exploratory study utilizes machine learning to examine which of many biopsychosocial
factors predict pain intensity increases and frequent pain among veterans with PTSD.
Methods: Data were from a larger daily diary study of PTSD symptoms and cannabis use among
U.S. veterans (N=74). Predictors included baseline socio-demographics, military characteristics,
and clinical traits, as well as self-reported symptoms and behaviors during days 1-30 and
proximal changes in symptoms and behaviors during days 24-30. Pain outcomes included
clinically significant increase in pain intensity and frequent pain assessed during days 31-88.
Machine learning methods were used to analyze the data, where random forest models provided
the rank order of importance of predictors and regularized regression models provided the
magnitude of associations between predictors and pain outcomes.
Results: Most machine learning models demonstrated adequate performance. In analyses of
frequent pain, pain interference emerged as a top predictor and had a relatively large magnitude
positive association with frequent pain. Variables capturing proximal changes in mental health
also emerged as top predictors of frequent pain; several of such variables (proximal changes in
anxiety, stress, depression, and negative affect) had relatively large positive associations with
frequent pain, and proximal changes in positive affect had a relatively large negative association

62
with frequent pain. Top predictors of pain intensity increases were posttraumatic cognitions,
which demonstrated a large positive association with pain intensity increases, and proximal
changes in positive affect, which demonstrated a large negative association with pain intensity
increases. Female sex also demonstrated a large magnitude positive association with pain
intensity increases. Across outcomes, substance use and social interaction variables also emerged
in several models, sometimes with differing directions of effects.
Conclusion: Results of this exploratory study exemplify how machine learning with novel data
can explore complex health phenomena like pain through a biopsychosocial lens. Findings point
to the clinical significance of many variables and several potentially important treatment and
prevention targets, including individual differences and changes in mental health. Future
research should validate these findings in larger samples and consider incorporating objective
measures to complement self-reported data.

63
Introduction
Given the high prevalence of physical pain among military veterans with multi-morbid
problems, it is vital to understand factors that contribute to pain among such veterans (Benedict
et al., 2020b). It is also challenging to do so given the complex etiology of pain and the litany of
related challenges that veterans commonly experience (Bourn et al., 2016; Bowe & Rosenheck,
2015; Khazaie et al., 2016; Lew et al., 2009; Tan et al., 2011). Pain is influenced by many of the
physical and psychological problems common among veterans, including traumatic exposure and
injury, sleep disturbances, heightened stress, and problematic substance use (Giordano et al.,
2018; Lew et al., 2009; Tan et al., 2009), which makes understanding the precise determinants of
pain among veterans a challenge. Additionally, while pain is known to vary within individuals
over short periods of time (Mun et al., 2019), particularly little research has examined the
determinants of time-related components of pain, such as the time until someone experiences an
increase in pain intensity or whether someone experiences pain relatively frequently (Salamon et
al., 2014). As pain is especially prevalent and impactful among veterans with multi-morbid
problems such as PTSD (Benedict et al., 2020b), clarifying the factors that most contribute to
pain intensity increases and frequent pain among veterans with PTSD is crucial.
As numerous factors are associated with pain, the biopsychosocial model of pain
provides a framework for organizing the potential determinants of pain. While earlier
conceptions of pain focused narrowly on biological factors such as injury and tissue damage, the
biopsychosocial model posits the development and maintenance of pain can also be influenced
by psychological factors such as mental health symptoms, sleep disturbances, and problematic
substance use and social factors such as lack of social interaction and financial challenges
(Gatchel et al., 2007). While the biopsychosocial model does not make specific predictions about

64
the direction and magnitude of effects of different factors on pain, it provides a useful organizing
framework that suggests many factors should be considered in tandem. A growing body of
empirical work provides evidence for associations between many biopsychosocial factors and
pain, albeit when pain is measured on average over relatively long intervals. For example,
longitudinal studies using three-to-six-month follow-ups have reported that PTSD symptoms are
associated with subsequent pain among veterans (Lee et al., 2019; Stratton et al., 2014).
Similarly, psychological stress is associated with pain in veteran samples (Ang et al., 2006;
Davis, Prindle, et al., 2022; Tan et al., 2009). Sleep quality, often poor among those with PTSD
(Babson & Feldner, 2010), is an especially robust predictor of pain, with a recent systematic
review identifying 26 studies linking poor sleep quality with greater pain among veterans
(Saconi et al., 2021). Additionally, while some veterans use substances such as alcohol and
cannabis to cope with pain and/or PTSD (Goebel et al., 2011; Metrik et al., 2018), there is
evidence substance use might actually worsen their pain by diminishing their capacity to cope or
leading to negative consequences even if it improves pain in the short term (i.e., see Paper 1
results; Boehnke et al., 2020; Mahdavi et al., 2021; Zale et al., 2015). Finally, combat
experiences common among veterans with PTSD such as traumatic exposure (Jakob et al., 2017)
also precipitate pain (Giordano et al., 2018) and PTSD and pain are especially highly co-morbid
among veterans with traumatic brain injury (Lew et al., 2009; Nampiaparampil, 2008).
Still, existing knowledge of pain among veterans with PTSD is limited when fully
considering the biopsychosocial perspective and its implications. Existing research has mostly
relied on traditional statistical modeling methods common in the social and behavioral sciences,
which enable researchers to test one or a small number of predictors (see, for example: Ranney et
al., 2022; Saba et al., 2021; Saconi et al., 2021). Since the biopsychosocial model suggests pain

65
is multi-factorial (i.e., relative risk for pain is likely determined by the presence or absence of
several factors and their interactions; Engel, 1981), traditional statistical modeling methods are
limited in their ability to study pain from the biopsychosocial perspective and elucidate which
factors convey the highest risk for pain. Methodological advances in machine learning, which
have recently been applied in studies of pain among non-veterans (Matsangidou et al., 2021),
may be better suited to this task. Machine learning is a class of flexible algorithmic and statistical
techniques used to study a specific outcome by generating a predictive model based on many
predictors and their interactions (Matsangidou et al., 2021). Because machine learning
approaches can learn from and model many predictors concurrently, they are well suited for
studying phenomena such as pain from the complex biopsychosocial perspective (Lötsch &
Ultsch, 2018). While machine learning has been underused in studies of pain among veterans, it
has been applied to predict pain among several non-veteran populations using rich clinical and
socio-demographic data, including individuals with arthritis (Loetsch et al., 2020), lower back
pain (Goldstein et al., 2020), and cancer pain (Juwara et al., 2020). To date, no study has applied
machine learning methods to study the determinants of pain among veterans with PTSD.
Additionally, while pain is understood to be varying and changeable (i.e., pain intensity is
known to change within an individual over short periods of time such as weekly or daily, and
individuals vary in how often they experience pain; Mun et al., 2019) prior work including the
above studies on pain among veterans primarily rely on simple retrospective reports of average
pain intensity experienced during multiple weeks rather than on pain measured more frequently
(see, for example: Davis et al., 2022; Lee et al., 2019; Saconi et al., 2021). Thus, existing studies
of pain among veterans provide far less information on what influences time-related variations in
pain, such as the time until one experiences an increase in pain intensity or whether one

66
experiences pain relatively often (i.e., frequent pain) (Salamon et al., 2014). This is an important
gap in the literature given the clinical significance of both pain intensity increases and frequent
pain (Kamper et al., 2019; Suzuki et al., 2020). That is, in non-veteran samples increases in pain
intensity predict worsening behavioral health symptoms and predict higher pain interference (the
extent to which pain interferes with day-to-day life) (Suzuki et al., 2020). Similarly, those who
experience pain relatively frequently have greater psychological distress and higher quantity of
substance use (Kamper et al., 2019).
The present study is an exploratory investigation of the use of machine learning methods
to explore the biopsychosocial determinants of pain intensity increases and frequent pain among
veterans with PTSD. It leverages daily diary data from a sample of such veterans (n = 74). The
first Aim is to explore the factors that predict whether veterans with PTSD experience pain
relatively frequently, and the second Aim is to explore the factors that predict clinically
significant increases in pain intensity among such veterans. Results could shed light on the
predictors of both types of pain outcomes, thereby providing potentially useful information for
clinicians seeking to target pain in an increasingly targeted and time sensitive manner. Given the
biopsychosocial model of pain, models include many candidate predictors including biological
(e.g., biological sex at birth, traumatic brain injury), psychological (e.g., mental health, substance
use), and social (e.g., social interaction, income) variables.
Methods
Participants and procedures
The present study utilizes data from a larger study (1R21DA051802) of PTSD symptoms
and cannabis use among U.S. veterans (n = 74; see dissertation paper 1 for information on

67
recruitment and inclusion/exclusion criteria). This study utilizes self-report measures collected
via the MAVERICK daily diary app at baseline and twice daily: during the morning (regarding
symptoms experienced during the prior evening/night, between 6pm and 6am) and during the
evening (regarding symptoms experienced during that day, between 6am and 6pm. Participants
provided up to 86 days of daily diary data. See paper 1 (Table 1) for participant demographics
and clinical characteristics at baseline. Briefly, study participants were 79% male, 60% were
White, and 63% non-Hispanic. The mean pain intensity score at baseline was 7.53 (SD = 2.80)
on the Patient Reported Outcomes Measurement Information System (PROMIS) Pain Intensity
measure, representing mild-moderate pain on average (Stone et al., 2016).
Measures
Physical pain outcomes. Pain outcomes (pain intensity increases and frequent pain) were
assessed during the final two months of data collection (days 31-88). Pain intensity was
measured twice per day using the single item Numerical Rating Scale (scored from 0 “no pain”
to 10 “worst pain ever”), which has shown a high degree of validity in clinical and research
contexts (Suzuki et al., 2020). As pain is self-reported twice per day, pain scores for each day
were first averaged for one daily pain score during this time period. For models predicting
whether one experiences frequent pain, pain frequency first was calculated as the percent of days
during days 31-88 on which a participant had data and reported pain of at least moderate severity
(greater than 4 on the Numerical Rating Scale; Woo et al., 2015), and participants that
experienced moderate pain on a majority of days (greater than 50% of days) were classified as
experiencing frequent pain.
For models predicting increases in pain intensity, increases were operationalized as the
number of days (starting on day 31) until they experienced a clinically important increase in pain

68
intensity (an increase in 2 points or greater on the Numerical Rating Scale, in line with a
systematic review of prior studies; Olsen et al., 2017) relative to their average pain intensity
score on the Numerical Rating Scale during days 1-30. Those with an average pain intensity
score above 8 were considered to have a pain intensity increase if they reported a pain intensity
score of 10, even if this did not represent an increase of 2 points. A time to event variable was
created indicating the day each participant reported a clinically important pain intensity increase.
Thus, lower values indicated an earlier clinically important pain intensity increase. Those who
did not experience a pain increase throughout the study period were censored.
Predictors. Predictors were from the MAVERICK baseline assessment and the first 30
days of the daily diary data. Predictors were selected to capture a breadth of biological,
psychological, and social factors in line with the biopsychosocial model of pain (Gatchel et al.,
2007). Predictors from the baseline assessment included socio-demographics, military
characteristics, and clinical traits. From daily diary surveys, predictors included self-reported
clinical symptomology and behavior during days 1-30 (aggregated), as well as changes in these
constructs proximal to assessed pain outcomes (the prior week, during days 24-30). See Table 3.1
for a full list of predictor variables, definitions, descriptive statistics, and missingness.
Analytic Plan
Study aims were achieved using machine learning methods, which are appropriate for
testing a large set of predictors simultaneously and identifying the most important predictors and
their relative influence on an outcome (Lötsch & Ultsch, 2018). Specifically, this study utilized
random forest and regularized regression (Breiman, 2001; Zou & Hastie, 2005) models, as prior
research has shown how these machine learning approaches can complement one another (Davis
et al., 2021; Davis, Rao, et al., 2022). Random forest models are an ensemble decision tree

69
approach (Breiman, 2001). That is, they combine many decision trees to determine the rank order
importance of predictors in terms of how much each predictor improves the ability of the model
to make predictions. Each decision tree within a random forest is assigned a random subset of
participants from the full sample, and each split within each tree examines a random subset of
predictor variables to determine which predictor variables best partition the data with respect to
the outcome. Random forest models provide a variable importance metric to identify impactful
and relevant predictors (i.e., how important each predictor was to the model’s decision making).
Specifically, random forest models use permutation based variable importance where they
calculate importance of a variable by permuting (shuffling) its values in the out-of-bag data (the
observations that were not used to construct a particular tree) and measuring the resulting
increase in prediction error. Regularized regression allows for the modeling of basic linear
associations between many possible predictors and an outcome and provides the magnitude of
effects. These analyses utilized the elastic net method, which combines least absolute shrinkage
and selection operator (LASSO) and ridge regression for regularization (i.e., for variable
selection and to prevent over-fitting; Zou & Hastie, 2005). LASSO penalizes predictors that are
not statistically significant, limiting results only to predictors with non-zero estimated effects,
and ridge regression focuses only on shrinking the estimated effect; the elastic net method used
in these analyses identify the amount of penalization towards ridge vs. LASSO in a combined
model.
The specific random forest and regularized regression models used varied based on the
dependent variable. Namely, analyses predicting frequent pain involved random forest classifiers
and regularized logistic regression, which are appropriate for binary classification problems (i.e.,
analyses of whether or not someone experienced frequent pain). Analyses of pain intensity

70
increases involved random survival forests and regularized Cox regression (Ishwaran et al.,
2008; Kattan, 2003), which are survival models (i.e., analyses of time until an event, such as an
increase in pain intensity).
For each outcome, we began with a random forest model that included all possible
predictors, to establish the rank order of importance of possible predictors and begin a process of
dimension reduction (reducing the total number of predictors under consideration to make
models easier to interpret and reduce overfitting, wherein the model performs well on the
original sample but poorly on new data). In line with common practices, we set the number of
predictors to test in each node as the square root of the total number of predictors (Rigatti, 2017).
To select the node size (the minimum number of observations in a terminal node, which
determines when the model stops testing additional splits), we considered the out-of-bag error
rate (the error rate when testing the random forest’s decision trees on the observations that were
not used to construct those trees). Given the total sample size available, the minimum node size
tested was five to avoid overfitting the model to the data. The number of decision trees in the
final random forest model for each outcome was set based on when the out-of-bag error rate
stabilized. Random forest models were fit with the R package randomForestSRC.
The top 15 (most important) predictors were taken from the random forest model results
for each outcome and included as candidate predictors when fitting the regularized regression
model. K-fold cross validation was used to select model parameters, where the data is partitioned
into K subsets, the model is trained on K-1 subsets, and validated on the remaining subset; this
process is repeated multiple times to reduce variability and ensure robustness of the model across
different samples. Regularized regression model parameters included the alpha parameter

71
(selects the balance between ridge vs. LASSO regularization) and the lambda parameter (selects
the degree of regularization). Regularized regression models were fit with the R package glmnet.
Several methodological decisions were made in part to address the relatively small study
sample, such as cross validation, out-of-bag testing, regularization, and dimension reduction.
Machine learning is often employed with large samples; as it is a data-driven approach to
identifying complex associations between many candidate predictors and an outcome, larger
samples are said to provide a more diverse set of examples for the model to learn from, leading
to improved performance, accuracy, and reduced overfitting (Domingos, 2012). Machine
learning methodology also often involves testing models on an unseen subsample of participants
(i.e., a “test set”) to validate the model and reduce overfitting, further necessitating additional
data (Angra & Ahuja, 2017). While our sample was not large enough to allow for a separate test
set, K-fold cross validation and out-of-bag testing can help mitigate overfitting by selecting
model parameters that generalize well across multiple subsamples of data (Hastie et al., 2001).
Dimension reduction and regularization also reduce model complexity and overfitting by
eliminating irrelevant predictors and promoting simpler models, which is especially crucial with
smaller samples (Hastie et al., 2001).
Missing data was handled as follows. When calculating average daily pain intensity
scores, if a participant had only one pain score on a particular day, that single score was
considered their pain score for the day. For analyses of pain intensity increases, when a
participant did not have any pain scores on a particular day, they were considered to not have a
pain intensity increase on that day. When a participant was lost to follow up, that participant’s
data was censored following their last datapoint. When calculating the frequent pain outcome,
available data was used such that for each participant the number of days they reported moderate

72
pain were divided by the number of days they provided pain data. For aggregates of predictors
from days 1-30, participants’ available data were used to compute their average score or
frequency of endorsing an item on the days they provided data. Missingness on baseline and day
1-30 aggregate predictors was handled by sorting cases by demographic and military variables
(i.e., sex, minority race/ethnicity, and number of deployments) and using the 5 cases before and
after each missing value to compute the mode. Baseline or day 1-30 aggregate predictors that
were considered to be included in models but had greater than 20% missingness were ultimately
excluded from models, thus proximal changes in sleep quantity and quality during days 24-30
were considered and omitted as 23% of participants were missing on both variables.
Results
Frequent pain
In the full sample, 54 individuals (73%) did not experience moderate pain during the
majority of days 31-88 (greater that 50% of days; frequent pain), and 20 individuals (27%) did
experience moderate pain during the majority of days 31-88.
Random forest classifier results. A random forest classifier predicting frequent pain was
trained in which the number of variables tested at each node split was six (the square root of the
total number of predictors). Node sizes between 5-20 were tested when tuning the model (in
increments of 1). Considering both the out-of-bag sample rate when tuning the model and
concerns about overfitting, the minimum leaf node was set to 5. The final random forest model
size was 10,000 trees as the out-of-bag error rate stabilized. The misclassification rate in the final
model was .16, and the area under the curve was .87, suggesting adequate model performance.

73
Figure 3.1 presents the rank order of importance for the top 15 variables predicting
frequent pain from the random forest classifier model, where the variable importance metric (VI)
is interpreted as the increase in prediction error when permuting (shuffling) that variable. The top
two predictors were average pain intensity during days 1-30 (VI = .283, ranked 1st) and pain
interference (the extent to which pain interferes with day-to-day life) at baseline (VI = .069,
ranked 2nd). Proximal changes (days 24-30) in mental health made up four of the next top five
predictors, including proximal changes in anxiety (VI = 0.031, ranked 3rd), stress (VI = 0.023,
ranked 4th), depression (VI = 0.017, ranked 6th), and negative affect (VI = 0.017, ranked 7th).
Baseline anxiety sensitivity was ranked 5th (VI = 0.021). Average substance use during days 1-30
also appeared twice in the top 15 variables, including average number of drinks with alcohol (VI
= 0.012, ranked 8th) and average number of times using cannabis per day (VI = 0.012, ranked
12th).
Regularized logistic regression results. The top 15 variables from the random forest
classifier model were included when training the regularized logistic regression predicting
frequent pain. Five-fold cross validation was used to tune model parameters, as increasing the
number of folds was infeasible given the sample size resulting in a lack of positive (frequent
pain) cases in some of the folds. Alpha values between 0-1 were tested when tuning the model
(in increments of 0.1); for each alpha value, the glmnet algorithm automatically selects a range
of lambda values to test by generating a sequence from a high value that makes all model
coefficients zero to a lower value that results in a less restricted model. The final regularized
logistic regression model had an Alpha of 0 indicating a pure ridge model and a lambda of .47
indicating a moderate degree of regularization. The area under the curve was .91 suggesting good
model performance.

74
Figure 3.2 presents odds ratios from the regularized logistic regression model predicting
frequent pain. Variables with a relatively high magnitude positive association with frequent pain
included proximal changes in anxiety (OR = 1.20 for a 1SD increase during days 24-30),
proximal changes in stress (OR = 1.18 for a 1SD increase during days 24-30), daily average pain
(OR = 1.17 for a 1-point increase on average during days 1-30), and proximal changes in
depression (OR = 1.16 for a 1SD increase during days 24-30). Variables with a relatively high
magnitude negative association with frequent pain included posttraumatic cognitions (negative
beliefs about oneself or the world in the aftermath of trauma) (OR = 0.94 for a 1-point increase at
baseline), proximal changes in positive affect (OR = 0.93 for a 1SD increase during days 24-30),
and daily average number of drinks with alcohol (OR = 0.82 for a 1-drink increase on average
during days 1-30).
Pain intensity increases
Figure 3.3 presents the survival plot for clinically significant pain intensity increases in
the full sample. This plot depicts the probability over the course of the study period of not
experiencing a pain intensity increase for all individuals in our sample. Thus, there was a
relatively consistent increase in the probability of experiencing an increase in pain intensity over
the study period. By the end of the study period (day 88), roughly half of the sample was not
expected to have a pain intensity increase, and half the sample was expected to have a pain
intensity increase. The average time to a clinically significant pain intensity increase was 15.83
days.
Random survival forest results. A random survival forest model predicting time to
clinically meaningful increase in pain was trained in which the number of variables tested at each
node split was six (the square root of the total number of predictors). Node sizes between 5-20

75
were tested when tuning the model (in increments of 1). Considering both the out-of-bag sample
rate when tuning the model and concerns about overfitting, the minimum leaf node was set to 5.
The final random forest model size was 40,000 trees as the out-of-bag error rate stabilized. The
out-of-bag error rate in the final model was .55. This relatively poor accuracy speaks to the
limitation of the sample size. The relatively large number of trees required to reach stability
indicates the random survival forest method had challenges identifying consistent patterns in the
data given the sample size and possible noise - thus the high out-of-bag error rate.
Figure 3.4 presents the rank order of importance for the top 15 variables predicting pain
intensity increases from the random survival forest model, where the VI metric is interpreted as
the increase in prediction error when permuting (shuffling) that variable. The top two predictors
were baseline posttraumatic cognitions (VI = 0.024, ranked 1st), and proximal changes (days 24-
30) in positive affect (VI = 0.011, ranked 2nd). Social interaction variables made up five of the
top 15 predictors, including text messaging frequency during days 1-30 (VI = 0.010, ranked 3rd),
proximal changes in phone call frequency (VI = 0.007, ranked 5th), proximal changes in
videoconferencing frequency (VI = 0.003, ranked 7th), videoconferencing frequency during days
1-30 (VI = 0.002, ranked 9th), and phone call frequency during days 1-30 (VI = 0.002, ranked
11th). Proximal changes in PTSD symptoms was also relatively highly ranked (VI = 0.007,
ranked 4th), as was proximal changes in number of drinks with alcohol (VI = 0.005, ranked 6th).
Regularized Cox regression results. The top 15 variables from the random survival
forest model were included when training the regularized Cox regression predicting pain
intensity increases. Ten-fold cross validation was used to tune model parameters and resulted in
improved model accuracy over five-fold cross validation. Alpha values between 0-1 were tested
when tuning the model (in increments of 0.1); for each alpha value, the glmnet algorithm

76
automatically selects a range of lambda values to test by generating a sequence from a high value
that makes all model coefficients zero to a lower value that results in a less restricted model. The
final regularized Cox regression model had an Alpha of 0.7 indicating a model incorporating a
blend of LASSO and ridge regularization, and a lambda of .03 indicating a low degree of
regularization. The c-index (concordance index) was .73 suggesting adequate model performance
(the c-index is a measure of how often the model correctly predicts the survival order for two
randomly selected individuals).
Figure 3.5 presents hazard ratios from the regularized Cox regression model predicting
pain intensity increases. Variables with a relatively high magnitude positive association with pain
intensity increases included female sex (HR = 1.58 relative to male sex), proximal changes in
PTSD symptoms (HR = 1.52 for a 1SD increase during days 24-30), and posttraumatic
cognitions (HR = 1.23 for a 1-point increase at baseline). Variables with a relatively high
magnitude negative association with frequent pain included proximal changes in positive affect
(HR = 0.77 for a 1SD increase during days 24-30) and proximal changes in number of drinks
with alcohol (HR = 0.79 for a 1SD increase during days 24-30).
Prototypical plots. Figure 3.6 depicts pain increase survival plots for subsamples based
on scores on the top two predictors from the random survival forest model: posttraumatic
cognitions and proximal changes in positive affect. This plot depicts the probability over the
course of the study period of not experiencing a pain intensity increase for each subsample. Thus,
those with relatively high posttraumatic cognitions (their composite score on the Posttraumatic
Cognitions Inventory indicated that they agreed with all scale items on average) and who did not
experience a proximal increase in positive affect had the highest probability of a pain increase
during the study period. However, those with relatively high posttraumatic cognitions but who

77
did experience a proximal increase in positive affect had a pain increase probability that was in
line with the rest of the study sample (i.e., those who did not have relatively high posttraumatic
cognitions).
Discussion
The present study exemplifies how novel data and methodology can extend knowledge of
physical and behavioral health problems such as pain, and potentially point to novel treatment
targets and strategies. The biopsychosocial model and prior empirical work suggest that many
factors influence pain (Engel, 1981; Wade & Halligan, 2017), making it challenging to
disentangle the key determinants of pain especially among a population like veterans with PTSD
who may have complex clinical and psychosocial characteristics (Giordano et al., 2018). While
most prior work on the determinants of pain incorporates measures of pain intensity averaged
over one or more weeks, much less is known about the predictors of pain’s temporal qualities
(i.e., time until an increase in pain, or relatively frequent pain), which are consequential for how
they might influence behavioral health (Salamon et al., 2014; Suzuki et al., 2020). Thus, the
present study utilizes data collected daily (daily diary data) to examine predictors of day-to-day
increases in pain intensity and of experiencing pain relatively frequently. We analyze this data
utilizing machine learning models, which allow us to examine many potential predictors of pain
outcomes concurrently in line with the biopsychosocial model of pain (Lötsch & Ultsch, 2018).
In analyses of frequent pain (i.e., whether someone experienced pain of at least moderate
intesnity during a majority of days 31-88), pain interference at baseline (the extent to which pain
interferes with day-to-day physical and psychological functioning) emerged as the second
highest predictor after average pain during days 1-30 (included as a control) in the random forest

78
model and had a relatively large magnitude of effects in the regularized regression model. While
prior research has demonstrated a positive association between pain interference and pain
intensity (Jensen et al., 2017), our analysis also suggests that, even after controlling for pain
intensity, pain interference also predicts subsequent frequent pain. This may be because those
high in pain interference may find it difficult to engage in help seeking and/or recreational and
social activities that might otherwise help them cope with pain, leading them to experience pain
on a greater number of days. Proximal changes in mental health (during days 24-30, just before
pain frequency was measured), also emerged as several of the top predictors of frequent pain in
the random forest model; proximal changes in anxiety, stress, depression, and negative affect had
relatively large positive associations with frequent pain, and proximal changes in positive affect
had a relatively large negative association with frequent pain in the regularized regression model.
There is a robust association between mental health and pain outcomes in prior literature
(Hooten, 2016), and our results are notable for demonstrating that relatively short-term changes
in mental health could influence daily pain frequency over a subsequent two-month period. This
may be because poor mental health may precipitate pain catastrophizing (i.e., very negative
emotions and beliefs about pain), which has been linked with worse pain outcomes (Quartana et
al., 2009). Our result suggests targeting mental health and enhancing positive affect could pay
large dividends among those who experience frequent pain, in line with existing integrative
interventions for multi-morbid problems (Dworkin et al., 2008; Lumley & Schubiner, 2019).
In analyses of pain intensity increases (i.e., the number of days until a participant
experienced a clinically significant pain increase in pain intensity), posttraumatic cognitions
(negative beliefs about oneself and the world in the aftermath of trauma) and proximal changes
in positive affect emerged as the top two predictors in the random forest model. In the

79
regularized regression models these variables had relatively large associations with pain intensity
increases in the opposite and expected directions: posttraumatic cognitions were positively
associated with pain intensity increases and proximal changes in positive affect were negatively
associated with pain intensity increases. Findings are in line with recent work on either
posttraumatic cognitions (Curry et al., 2019) or positive affect (Haun et al., 2023) and pain
outcomes among veterans and suggest an interplay between psychological beliefs, affect, and
pain. In fact, plotting survival curves for subsamples based on these predictors demonstrates how
recent increases in positive affect might protect against pain intensity increases particularly
among those high in posttraumatic cognitions (see Figure 3.6). Integrative treatments for pain
that target both pain and affect and show promise with veterans (Haun et al., 2023) may be
especially useful for those with PTSD and concomitant posttraumatic cognitions. Indeed,
proximal changes in PTSD symptoms also emerged as a top predictor and demonstrated a large
positive association with pain intensity increases. Models at times provided unique information:
while sex was somewhat low in importance in the random forest model (as the 14th most
important predictor), female sex demonstrated the largest (positive) association with pain
intensity increases in the regularized regression model. Our result is in line with recent work
demonstrating greater pain among female veterans (Naylor et al., 2017, 2019; Saba et al., 2023).
Several biopsychosocial factors have been put forth as possibly explaining sex-related
differences in pain, with some suggesting that differences in physiological mechanisms
contribute to more pain among women, and others hypothesizing that men may under-report pain
because of social pressure (Melchior et al., 2016). Sex-related differences in pain among veterans
demand further attention from researchers and health systems. Differences in results for this
variable also exemplify how random forest and regularized regression can provide

80
complementary information when used in tandem, due likely to their different ways of handling
interactions and non-linearity, addressing collinearity, and interpreting feature importance (Angra
& Ahuja, 2017; Davis et al., 2021; Davis, Rao, et al., 2022).
Given the ability of machine learning models to include many variables in a single
model, our analyses demonstrate associations between several types of variables and both
frequent pain and increases in pain intensity. Two additional types of variables that emerged as
important predictors in random forest models for both outcomes included substance use and
social interaction. Average cannabis use quantity was positively associated with both frequent
pain and increases in pain intensity in regularized regression models. Conversely, average
drinking quantity had a large magnitude negative association with frequent pain, and proximal
changes in drinking had a large magnitude negative association with pain intensity increases. It is
thought provoking that drinking and cannabis use had opposite associations with our outcomes,
suggesting drinking might reduce pain among our sample but cannabis use might make it worse.
As problematic cannabis use was an inclusion criterion of the parent study, it is possible that
elevated pain is a consequence of an over-reliance on cannabis in this sample, in line with selfmedication models (Khantzian, 1997). While there is growing evidence from clinical studies that
cannabis can be an efficacious medical treatment for pain in some cases (Nugent et al., 2017),
recent studies also suggest that overuse of cannabis might lead to hyperalgesia where individuals
become more sensitive to pain long term (Romero-Sandoval et al., 2018; Zhang-James et al.,
2023). As the present sample was not recruited as a heavy drinking sample, perhaps alcohol does
have an analgesic effect, especially as it could also help with many of the mental health
symptoms shown to predict pain in this analysis (Khantzian, 1997).

81
Social interaction variables displayed counter-intuitive associations with study outcomes
in regularized regression models. That is, text messaging frequency was positively associated
with both frequent pain and pain intensity increases, and proximal changes in videoconferencing
frequency were also positively associated with pain intensity increases. While social interaction
is thought to be protective against pain (Che et al., 2018), it is possible that in this sample of
veterans with PTSD, communicating frequently via technology is more burdensome than helpful
or is not a sufficient replacement for other forms of interaction. Some research with non-veterans
has shown that text messaging (Holtzman et al., 2017) and videoconferencing (Noone et al.,
2020) may be less beneficial to well-being than in-person interaction, and these forms of
communication should be more closely studied among those with pain and multi-morbid
behavioral health problems. It is also possible that many the virtual interactions captured in this
analysis were stressful, and future research could analyze the content of interactions to determine
whether positive interactions are of value for protecting against pain.
Our study includes several limitations that should be considered. Results from this
exploratory study may not generalize to all U.S. veterans, particularly those who do not use
technology or who are less likely to engage in a research study. Predictors of pain intensity
increases and frequent pain may also differ in veterans who have less severe behavioral health
symptomology (i.e., those without PTSD symptoms and/or problematic levels of cannabis use)
or more severe pain (i.e., those who are recruited from a pain clinic). While a strength of using
daily diary data is that surveying participants daily may reduce recall bias (i.e., participants may
be better able to remember pain and other clinical symptoms and behavior), data are still selfreported, and study designs may benefit from incorporating objective measures such as those that
are available from wearable devices. Also, while most models demonstrated relatively good

82
performance considering cross validation and out-of-bag metrics (aside from the random survival
forest predicting pain intensity increases), the number of participants in this analysis was too
small to allow for the testing of our machine learning models on an unseen subsample of
participants (i.e., a “test set”) as is often done in applications of machine learning (Angra &
Ahuja, 2017), and models may have lacked statistical power to identify associations among some
variables. Future work should include model validation and/or replication in larger samples with
an a priori power analysis. Random survival forest model for pain intensity increases in
particular may need information from larger samples, given that this model took a relatively
large number of trees to reach stability (compared to the random forest classifier for frequent
pain). Perhaps model performance for pain intensity increases could be improved in mixed
models assessing day-level increases as the outcome (yes/no for each day with many repeated
measurements per individual) and using prior- and/or same-day symptoms and behaviors as
predictors. Additionally, while proponents of the biopsychosocial model increasingly consider
the influence of spirituality on health (Saad et al., 2017), we were not able to include such factors
in our models as those data were not collected in the parent study. Finally, clear causal
conclusions cannot be drawn from these models, as there could be additional unmeasured factors
that explain associations between the included predictors and pain.
Still, our results point to patterns of associations between multiple constructs (including
demographics, clinical characteristics, and changes in clinical characteristics) and pain. Given
that these associations were identified even with a relatively small sample, results speak to the
clinical importance of biopsychosocial variables. This exploratory study can thus also be
considered a proof-of-concept for the benefits of utilizing machine learning with novel data to
explore complex health phenomena such as pain through the lens of the biopsychosocial model.

83
We provide evidence for several potential intervention and prevention targets among veterans
with multi-morbid pain and PTSD, including individual differences in how people might
experience pain (i.e., pain interference) and PTSD (i.e., posttraumatic cognitions). As proximal
changes in mental health symptoms were implicated in both frequent pain and pain intensity
increases, providers should prioritize integrative interventions that address mental health and
particularly aim to support positive affect. Health systems should also remain vigilant in
identifying and addressing pain and multi-morbid problems among women veterans, and to this
end researchers should continue to elucidate sex-related differences in pain and its treatment.

84
Table 3.1 Predictor variables descriptions, descriptive statistics, and missingness
Variable Description Type
Category
n, or
M(SD)
Missing
%
Demographic characteristics (Baseline survey)
Minority status Participant is non-White race and/or Hispanic/Latinx ethnicity
(Yes/No) Binary 39 / 33 2.7%
Sex Self-reported biological sex at birth (Male/Female) Binary 58 / 14 2.7%
LGBT Participant identifies as lesbian, gay, bisexual, and/or
transgender (Yes/No) Binary 13 / 59 2.7%
Marital status Participant reports being married (Yes/No) Binary 29 / 43 2.7%
Education Highest level of education completed (Some college or
below/College graduate) Ordinal 46 / 26 2.7%
Income Recoded to midpoint of participant's income category Continuous
M = 63838
(SD =
51854.59)
4.1%
Unemployed Participant is unemployed and not retired (Yes/No) Binary 17 / 55 2.7%
Retired Participant is retired (Yes/No) Binary 12 / 60 2.7%
Event exposure (Baseline survey)
Number of
deployments Number of times participant has deployed Count
M = 2.55
(SD =
2.93)
4.1%
Traumatic
events
Count of types of events participant reports has "happened to
[them]" on Life Events Checklist; Gray et al., 2004) (0-17) Count
M = 7.79
(SD =
3.00)
2.7%
Traumatic brain
injury (TBI)
Positive screen for TBI on Defense and Veterans Brain Injury
Center Screening Tool (Schwab et al., 2006) (Yes/No) Binary 38 / 34 2.7%
Clinical traits (Baseline survey)
Pain
interference
PROMIS Pain Interference Short Form, measures the extent to
which pain interferes with day-to-day life (Chen, 2018) (4-20) Continuous
M = 10.32
(SD =
4.91)
2.7%
Pain anxiety Pain Anxiety Symptom Scale Short Form, measures anxiety
related to pain (McCracken 2002) (0-100) Continuous
M = 37.54
(SD =
21.85)
6.8%
Anxiety
sensitivity
Anxiety Sensitivity Index-3, measures fear of symptoms of
anxiety (Taylor, 2007) (0-64) Continuous
M = 29.52
(SD =
16.39)
4.1%
Posttraumatic
cognitions
Posttraumatic Cognitions Inventory, measures negative beliefs
about oneself and the world in the aftermath of trauma (Foa,
1999) (1-7)
Continuous
M = 3.84
(SD =
16.51)
4.1%
Self-reported clinical symptoms and behavior, aggregated (Daily diary surveys, days 1-30)
Pain Control variable - single item Numerical Rating Scale (0-10);
average of responses during days 1-30 Continuous
M = 3.08
(SD =
2.27)
0.0%
PTSD symptoms PTSD Checklist for DSM-5 short form (Zuromski et al., 2019)
(0-16); average of responses during days 1-30 Continuous
M = 4.04
(SD =
3.19)
0.0%
Stress Single item ("How stressed did you feel?") (0-5); average of
responses during days 1-30 Continuous
M = 3.00
(SD =
1.05)
0.0%

85
Depression
symptoms
Single item ("How depressed did you feel?") (0-5); average of
responses during days 1-30 Continuous
M = 2.81
(SD =
1.08)
0.0%
Anxiety
symptoms
Single item ("How anxious did you feel?") (0-5); average of
responses during days 1-30 Continuous
M = 2.52
(SD =
1.09)
0.0%
Negative affect
Positive and Negative Affect Schedule - Short Form subscale
(Thompson, 2007) (5-25); average of responses during days 1-
30
Continuous
M = 9.22
(SD =
3.36)
0.0%
Positive affect
Positive and Negative Affect Schedule - Short Form subscale
(Thompson, 2007) (5-25); average of responses during days 1-
30
Continuous
M = 11.78
(SD =
3.66)
0.0%
Drinking
quantity
Single item (Number of drinks containing alcohol); average of
responses during days 1-30 Continuous
M = 0.62
(SD =
0.85)
0.0%
Cannabis use
quantity
Single item (Number of times used cannabis products); average
of responses during days 1-30 Continuous
M = 2.26
(SD =
4.07)
0.0%
Sleep quality Single item ("How would you rate the quality of your sleep last
night?") (0-5); average of responses during days 1-30 Continuous
M = 2.89
(SD =
0.50)
8.1%
Sleep duration Single item ("How many hours did you sleep last night?");
average of responses during days 1-30 Continuous
M = 5.92
(SD =
1.25)
8.1%
Social
interaction -
Text messaging
Single item ("Did you text message with people you have
social relationships with?" Yes/No); Proportion of days with a
Yes response during days 1-30
Continuous
M = 57%
(SD =
29%)
0.0%
Social
interaction -
Talking by
phone
Single item ("Did you talk on the phone with people you have
social relationships with?" Yes/No); Proportion of days with a
Yes response during days 1-30
Continuous
M = 35%
(SD =
29%)
0.0%
Social
interaction -
Spending time
in person
Single item ("Did you spend time in person with people you
have social relationships with" Yes/No); Proportion of days
with a Yes response during days 1-30
Continuous
M = 61%
(SD =
25%)
0.0%
Social
interaction -
Video call
Single item ("Did you video call with people you have social
relationships with?" Yes/No); Proportion of days with a Yes
response during days 1-30
Continuous
M = 10%
(SD =
17%)
0.0%
Proximal change in self-reported clinical symptoms and behavior (Daily diary surveys, days 24-30)
Δ PTSD
symptoms
Calculated as the difference between a participant's day 24-30
average PTSD symptoms and their day 1-30 average, in terms
of their day 1-30 standard deviation
Continuous
M = 0.08
(SD =
0.75)
16.2%
Δ Stress
Calculated as the difference between a participant's day 24-30
average stress and their day 1-30 average, in terms of their day
1-30 standard deviation
Continuous
M = -0.09
(SD =
0.67)
16.2%
Δ Depression
symptoms
Calculated as the difference between a participant's day 24-30
average depression symptoms and their day 1-30 average, in
terms of their day 1-30 standard deviation
Continuous
M = 0.02
(SD =
0.65)
16.2%
Δ Anxiety
symptoms
Calculated as the difference between a participant's day 24-30
average anxiety symptoms and their day 1-30 average, in terms
of their day 1-30 standard deviation
Continuous
M = 0.08
(SD =
0.60)
16.2%

86
Δ Negative affect
Calculated as the difference between a participant's day 24-30
average negative affect and their day 1-30 average, in terms of
their day 1-30 standard deviation
Continuous
M = 0.08
(SD =
0.75)
16.2%
Δ Positive affect
Calculated as the difference between a participant's day 24-30
average positive affect and their day 1-30 average, in terms of
their day 1-30 standard deviation
Continuous
M = -0.08
(SD =
0.68)
16.2%
Δ Drinking
quantity
Calculated as the difference between a participant's day 24-30
average drinking quantity and their day 1-30 average, in terms
of their day 1-30 standard deviation
Continuous
M = 0.04
(SD =
0.57)
16.2%
Δ Cannabis use
quantity
Calculated as the difference between a participant's day 24-30
average cannabis use quantity and their day 1-30 average, in
terms of their day 1-30 standard deviation
Continuous
M = 0.02
(SD =
0.56)
16.2%
Δ Social
interaction -
Text messaging
Calculated as the percent difference in proportion of days with
a Yes response during days 24-30, compared with their day 1-
30 proportion
Continuous
M = -17%
(SD =
52%)
14.9%
Δ Social
interaction -
Talking by
phone
Calculated as the percent difference in proportion of days with
a Yes response during days 24-30, compared with their day 1-
30 proportion
Continuous
M = -12%
(SD =
82%)
14.9%
Δ Social
interaction -
Spending time
in person
Calculated as the percent difference in proportion of days with
a Yes response during days 24-30, compared with their day 1-
30 proportion
Continuous
M = -23%
(SD =
41%)
14.9%
Δ Social
interaction -
Video call
Calculated as the percent difference in proportion of days with
a Yes response during days 24-30, compared with their day 1-
30 proportion
Continuous
M = -22%
(SD =
88%)
14.9%

87
Figure 3.1 Variable importance metrics from random forest classifier predicting frequent pain
Note: The variable importance metric is interpreted as the increase in prediction error when
permuting (shuffling) that variable
0 0.05 0.1 0.15 0.2 0.25 0.3
Text messaging frequency (Proportion of days)
Positive affect (Proximal change)
Number of times used cannabis (Daily average)
Stress (Daily average)
PTSD symptoms (Proximal change)
Posttraumatic cognitions (Baseline)
Positive affect (Daily average)
Number of drinks with alcohol (Daily average)
Negative affect (Proximal change)
Depression (Proximal change)
Anxiety sensitivity (Baseline)
Stress (Proximal change)
Anxiety (Proximal change)
Pain interference (Baseline)
Pain (Daily average)
VARIABLE IMPORTANCE

88
Figure 3.2 Odds ratios from regularized logistic regression predicting frequent pain
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
Number of drinks with alcohol (+1 drink, daily average)
Positive affect (+1SD proximal change)
Posttraumatic cognitions (+1 point increase, baseline)
PTSD symptoms (+1SD proximal change)
Anxiety sensitivity (+1 point increase, baseline)
Number of times used cannabis (+1 time, daily average)
Positive affect (+1 point increase, daily average)
Stress (+1 point increase, daily average)
Text messaging frequency (+.1 proportion of days)
Pain interference (+1 point increase, baseline)
Negative affect (+1SD proximal changes)
Depression (+1SD proximal change)
Pain (+1 point increase, daily average)
Stress (+1SD proximal change)
Anxiety (+1SD proximal change)
0.9 1.0 1.1 1.2
Odds Ratio

89
Figure 3.3 Survival plot depicting probability of pain intensity increases over time for the full
sample
Note: Depicts the probability over the course of the study period of not experiencing a pain
intensity increase for all individuals in our sample. Thus, there was a relatively consistent
increase in the probability of experiencing an increase in pain intensity over the study period. By
the end of the study period (day 88), roughly half of the sample was not expected to have a pain
intensity increase, and half the sample was expected to have a pain intensity increase.
0.00
0.25
0.50
0.75
1.00
0 20 40 60
Days
Survival probability

90
Figure 3.4 Variable importance metrics from random survival forest predicting pain intensity
increases
Note: The variable importance metric is interpreted as the increase in prediction error when
permuting (shuffling) that variable.
0 0.005 0.01 0.015 0.02 0.025 0.03
Number of times used cannabis (Daily average)
Sex
Number of deployments
Traumatic event exposure
Phone call frequency (Proportion of days)
Depression symptoms (Daily average)
Videoconferencing frequency (Proportion of days)
Income
Videoconferencing frequency (Proximal change)
Number of drinks with alcohol (Proximal change)
Phone call frequency (Proximal change)
PTSD symptoms (Proximal change)
Text messaging frequency (Proportion of days)
Positive affect (Proximal change)
Posttraumatic cognitions (Baseline)
VARIABLE IMPORTANCE

91
Figure 3.5 Hazard ratios from regularized Cox regression predicting pain intensity increases
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
Positive affect (+1SD proximal change)
Number of drinks with alcohol (+1SD proximal change)
Traumatic event exposure (+1 event)
Number of deployments (+1 deployment)
Phone call frequency (+.1 proportion of days)
Phone call frequency (+10% proximal change)
Videoconferencing frequency (+.1 proportion of days)
Videoconferencing frequency (+10% proximal change)
Number of times used cannabis (+1 time, daily average)
Income (+$10,000)
Depression symptoms (+1 point increase, daily average)
Text messaging frequency (+.1 proportion of days)
Posttraumatic cognitions (+1 point increase, baseline)
PTSD symptoms (+1SD proximal change)
Sex (Female)
0.8 1.0 1.2 1.4 1.6
Hazard Ratio

92
Figure 3.6 Survival plots depicting probability of pain intensity increases over time for
subsamples based on top two predictors from random survival forest model
Note: High posttraumatic cognitions is defined as scoring greater than four on the Posttraumatic
Cognitions Inventory, indicating agreement with all scale items on average. Plot depicts the
probability over the course of the study period of not experiencing a pain intensity increase.
Thus, the group with high posttraumatic cognitions and without recent increases in positive
affect (depicted in blue) had the highest probability of experiencing a pain intensity increase.

93
Chapter 4: “My body will remember what my mind wants to forget”: Towards a biocultural vulnerability model of veteran multi-morbidity (Paper 3)
Abstract
Introduction: Military veterans have high rates of multi-morbid problems, including physical
pain, posttraumatic stress disorder (PTSD), and problematic substance use. Prior work suggests
the etiology of multi-morbid problems among veterans is complex and involves both challenging
military experiences and military cultural factors. Qualitative research methods are promising for
extending understanding of vulnerability for multi-morbid problems by eliciting the subjective
experiences of veterans with such problems.
Methods: Participants were U.S. veterans of post-9/11 conflicts reporting pain, PTSD
symptomology, and problematic substance use (n = 20). Semi-structured interviews focused on
participants’ perceptions of how their multi-morbid problems developed and how they cope with
such problems. Interviewees were also prompted to describe their high point, low point, and
turning point experiences during their time as a military service member. Interview transcripts
were analyzed using a grounded theory approach to develop a novel explanatory model of
veteran multi-morbidity.
Results: In terms of the types of military experiences that left veterans vulnerable to developing
multi-morbid problems, participants cited (1) military deployments, (2) specific discrete events
such as physical injuries and psychological stressors, and (3) the accumulation of physical and
psychological stressors over time. Veterans also described how military cultural factors could
moderate the impact of military experiences and leave veterans more or less vulnerable to multimorbid problems. Military cultural factors that were said to increase vulnerability for multi-

94
morbid problems included (4) the significance of losing one’s physical and mental fitness for
serving personnel, (5) a tendency to ignore or minimize pain, (6) military drinking norms
including self-medication and a glorification of alcohol, and (7) stigma about seeking treatment.
Military cultural factors that were said to decrease vulnerability for multi-morbid problems
included (8) camaraderie between veterans and (9) the importance of service to others or to a
larger mission.
Discussion: Veteran participants revealed how military experiences and culture may interact to
determine vulnerability for multi-morbid problems. Results are in line with bio-cultural models
of health, which suggest that while cultural beliefs and practices can help people adapt to
challenging experiences, they can also diminish one’s ability to cope. Clinical psychosocial
assessments should capture the wide range of challenging military experiences and cultural risk
and protective factors, as well as veterans’ perceptions of how these influence their
symptomology. Integrative treatment approaches for multi-morbid problems may be tailored to
address military-cultural barriers to treatment, and to leverage strengths by fostering camaraderie
and service among veterans.

95
Introduction
It is well-established from survey and chart review (quantitative) studies that military
veterans have high rates of multi-morbid physical and behavioral health problems (Nahin, 2017;
Saba et al., 2021). Three such problems are physical pain, posttraumatic stress disorder (PTSD),
and problematic substance use. Quantitative research has demonstrated high overlap between
PTSD and pain among veterans, with 66% of those in treatment for PTSD experiencing pain
(Shipherd et al., 2007), and 49% of those seeking pain treatment meeting criteria for PTSD (Otis
et al., 2010). Additionally, pain and PTSD are both associated with problematic use of
substances, including alcohol (Saba et al., 2021; Tiet & Moos, 2021) and cannabis (Bryan et al.,
2021; Metrik et al., 2022), among veteran populations. Theoretical and empirical work suggests a
litany of factors, such as physical, psychological, and cultural factors, that make veterans
vulnerable to developing multi-morbid problems (Asmundson et al., 2002). Since most research
on multi-morbid problems among veterans has been quantitative in nature, it has generally
involved using narrowly selected data to demonstrate the influence of relatively few factors
(Chui et al., 2022; Raines et al., 2022). Qualitative research, on the other hand, is promising for
expanding what is known about vulnerability for multi-morbid problems by exploring veterans’
subjective experiences during and after military service and how these relate with physical and
behavioral health (Lacoursiere et al., 1980; Williamson et al., 2020). To date, however, little
qualitative research has explored vulnerability for multi-morbid pain, PTSD, and problematic
substance use among veterans.
Theoretical models of shared vulnerability have been developed to explain risk for multimorbid problems among at-risk populations such as veterans. Such models suggest that problems
like pain, PTSD, and problematic substance use are often multi-morbid because there are

96
predisposing factors that raise risk for each condition, concurrently. These models suggests that
multi-morbid conditions share certain risk factors in common, and thus they occur in tandem in
individuals and groups with these risk factors. Prior research on shared vulnerability among
veterans has primarily focused on psychological factors such as anxiety sensitivity and stress
reactivity. Theoretical and empirical work on shared vulnerability between PTSD and pain for
example has identified anxiety sensitivity (i.e., worry about anxiety symptoms) as a risk factor
that may amplify both problems (Asmundson et al., 2002; Raines et al., 2021). Heightened stress
reactivity (i.e., having higher levels of perceived stress in response to the same stressor) has also
been identified as factor that confers shared vulnerability for multi-morbid PTSD and
problematic substance use among veterans in a recent review (María-Ríos & Morrow, 2020).
Empirical research, however, points to a range of additional factors beyond psychological
factors that may increase vulnerability for multi-morbid problems among veterans. Namely,
military service often includes experiences that are physically and/or psychologically challenging
such as physical injuries and exposure to trauma, and such experiences may precipitate multimorbid problems such as pain, PTSD, and problematic substance use. Large-scale studies of
veterans have shown combat exposure to be a robust correlate of each of these problems (Ang et
al., 2006; Chui et al., 2022). Additionally, chart review studies conducted in the Veterans Health
Administration (VHA) system have demonstrated that combat injuries involving a blow or jolt to
the head (i.e., traumatic brain injuries) predict both pain and PTSD among veterans (Cifu et al.,
2013). Other types of traumatic experiences relatively common among service members,
including military sexual trauma, may raise risk for multi-morbid problems: in a chat review
study of nearly 500,000 VHA patients, those who experienced military sexual trauma had higher

97
rates of multi-morbid PTSD and substance use disorders than those who did not (Gilmore et al.,
2016).
There is also evidence that it is not only direct experiences during military service but
military culture (i.e., beliefs, behaviors, values, and norms that are shared among serving
personnel; Cohen, 2009) that can influence vulnerability for multi-morbid pain and behavioral
health problems. Service members are often socialized to conform to traditional ideas of
masculinity such as emotional control and concealment of weakness, in part, because such ideas
are construed to be adaptive for completing challenging missions (Abraham et al., 2017).
However, traditional masculinity has been associated with heightened PTSD symptomology
among veterans: In a survey study of 45 male veterans, scores on a measure of traditional
masculinity were positively associated with worse PTSD symptoms (Neilson et al., 2020).
Additionally, a military cultural norm of heavy alcohol use has been documented as a means of
building cohesion among service members and relieving stress (Ames & Cunradi, 2004), and
service members who believe to a greater extent that such a drinking culture exists are more
likely to drink excessively and evidence greater consequences due to drinking (Meadows et al.,
2022).
In fact, it is likely the case that military experiences interact with military culture to
influence risk for multi-morbid pain and behavioral health problems, as military culture could
influence how serving personnel cope with challenging military experiences. In the medical
anthropology literature, bio-cultural models indeed suggest that culture interacts with
challenging experiences to influence long-term health and functioning (McElroy, 1990;
Zuckerman, 2018). These models suggest that people develop cultural beliefs and practices to
adapt to challenging experiences (i.e., such as conforming to traditional masculinity to help with

98
completing a mission or drinking to build cohesion and relieve stress), but sometimes these
beliefs and practices can make the impact of challenging experiences worse. One can imagine,
for example, how emotional control and concealment of weakness (i.e., traditional masculinity)
could be adaptive for a service member during an active-duty combat mission but impede coping
with the impact of trauma or injury after that service member has transitioned to being a veteran.
In a survey study of 349 veterans (both males and females), those who scored higher on
measures of traditional masculinity indeed displayed higher levels of stigma against seeking
mental health treatment (McDermott et al., 2017). Traditional masculinity may also serve as a
barrier to benefitting from treatment for those who do seek treatment: In studies of veterans in
treatment for PTSD, those who demonstrated a higher degree of traditional masculinity showed a
poorer treatment response (less improvement in PTSD symptoms) (O’Loughlin et al., 2022,
2023). On the other hand, military cultural norms such as camaraderie (i.e., the bonds between
service members in the same unit or across the military in general; Oliver et al., 1999) may be
protective in the aftermath of challenging military experiences, as survey research has shown that
camaraderie may reduce the likelihood of PTSD symptom development after combat exposure
(Nevarez et al., 2017).
The present study utilizes qualitative methods to deepen the understanding of how
military experiences and culture increase vulnerability for multi-morbid pain, PTSD, and
problematic substance use among veterans. Qualitative studies conducted with veterans provided
some of the earliest evidence for the multi-morbid impact of both traumatic (Lacoursiere et al.,
1980) and morally injurious (V. Williamson et al., 2020) experiences (i.e., those involving
perceived moral transgressions, such as killing or witnessing the death of non-combatants or
being betrayed by leadership; (Griffin et al., 2019) among veterans. Qualitative interviews have

99
also highlighted specific ways that military culture can create barriers to seeking healthcare, as
service members have reported hiding their injuries out of concern for being viewed as “weak”
and unable to perform their job duties (Cogan et al., 2021). Narrative life history studies, a novel
qualitative method where researchers ask participants to recount descriptions of specific life
experiences (i.e., high points, such as being recognized for one’s accomplishments; low points,
such as witnessing the death of a fellow service member; and turning points, such as joining or
leaving the military), have shown promise for deepening knowledge of how such experiences
relate with behavioral health outcomes including substance use and PTSD in studies with nonveterans (Adler et al., 2016; Dunlop & Tracy, 2013; McCoy & Dunlop, 2017). Given known
links between military service and both pain and behavioral health problems (Chui et al., 2022;
Gilmore et al., 2016; V. Williamson et al., 2020) these methods could help us understand what
kinds of military experiences are especially salient when it comes to vulnerability for multimorbid problems. Narrative life history methodology could also prompt veterans to discuss
sensitive aspects of military culture, such as drinking and masculinity, without direct questions
about these topics. While most prior work on this topic has relied on quantitative survey
instruments, narrative life history interviews could elicit the nuances of experiences that stand
out in veterans’ own minds. Thus, the present study involves semi-structured and narrative life
history interviews with 20 veterans with multi-morbid symptomology to explore how veterans
perceive their experiences in the military and its culture relate with their pain, PTSD, and
substance use. This study contextualizes prior quantitative research findings and applies a
grounded theory approach (Strauss & Corbin, 1994) to develop a preliminary explanatory model
of multi-morbidity among veterans.

100
Methods
Participants and procedures
Veterans were selected from the participant pools of either the MAVERICK study
(described in Papers 1 and 2 of this dissertation) or an intervention study of mobile MindfulnessBased Relapse Prevention (mMBRP) for veterans with PTSD and alcohol use disorder.
Participants in both parent studies had to be U.S. veterans from the Air Force, Army, Marine
Corps, or Navy, and age 18 to 40. All parent study participants had to experience a traumatic
event and have at least some PTSD symptomology (assessed with the Primary Care PTSD
Screen; Cameron & Gusman, 2003). Participants all evidenced problematic substance use, as
they had to meet criteria for hazardous cannabis use (score of 8+ on the CUDIT-R; Loflin et al.,
2018) to be eligible for MAVERICK, which focused on veterans with PTSD who used cannabis,
or hazardous drinking (score of 8+ on the AUDIT; Babor et al., 2001) to enroll in the mMBRP
study, which focused on veterans with PTSD and alcohol use disorder. Because the purpose of
this study was to explore vulnerability for multi-morbid PTSD symptoms, cannabis use, and
pain, extreme case purposive sampling (Etikan et al., 2016) was used to select parent study
participants who also had pain of at least moderate intensity. That is, parent study participants
with baseline scores of 9+ (average of “moderate” on all items) on the PROMIS-Pain Intensity
measure (Stone et al., 2016) were included in a potential participant pool for the present study.
Those participants who had also previously consented to be contacted about future
research opportunities were contacted by a research assistant to request their participation in a
qualitative interview. The research assistant administered an informed consent form to those who
were interested, and scheduled them for an interview with the lead author. Interviews were
conducted via Zoom, recorded with participants’ consent, and transcribed without any participant

101
identifying information. Recordings were then immediately deleted. Participants were provided a
$35 gift card of their choice for completing a qualitative interview. Study procedures for the
present qualitative study and for both parent studies (MAVERICK and mMBRP) were approved
by the Institutional Review Board at the University of Southern California.
Twenty veterans completed interviews ranging from 35 minutes to 1 hour and 6 minutes
in length, as there was variability in interview time due to varying response length on semistructured and narrative life history questions, as well as the extent to which interviewees raised
new related topics. Eight participants were recruited from the MAVERICK study, and 12 were
recruited from the mMBRP study. The average age was 38.05 (SD = 6.1), 70% were male, and
75% were White. The average PTSD score was 43.35 (SD = 16.72) on the PCL-5 (indicating the
average participant met criteria for PTSD) and the average pain score was 10 (SD = 1.8) on the
PROMIS Pain Intensity measure (indicating moderate pain intensity on average). Participants
reported drinking an average of 10.35 days (SD = 9.24) and using cannabis an average of 8.65
days (SD = 12.43) over the prior 30 days. T-tests and proportion tests revealed no significant
differences in demographics and clinical characteristics between the participants recruited from
the MAVERICK study and the participants recruited from the mMBRP study. See Table 4.1 for
additional information on participant characteristics.
Qualitative Interview Protocol
A semi-structured interview protocol was developed under the guidance of the
mentorship team. Interviews focused on topics such as the participant’s perception of how their
pain and PTSD developed; how they perceive their multi-morbid symptoms are related to one
another; their use of substances to cope with multi-morbid problems; and the other ways in
which they cope with multi-morbid problems. The protocol also enabled exploration of

102
additional or unexpected themes that arose during interviews. Interviews also included a
narrative life history portion, in which participants were prompted to speak for 3-5 minutes
about three key scenes that were relevant to their physical or mental health during their time as a
military service member or veteran: their high point, low point, and turning point.
Quantitative Measures
Several variables were obtained from participants’ parent study baseline survey
responses, to characterize their demographic and clinical characteristics.
Demographics. Participants self-reported their age, race, ethnicity, and biological sex at
birth.
PTSD symptoms. PTSD symptomology was assessed with the 20-item PTSD Checklist
for DSM-5 (Bovin et al., 2016). Participants indicated how often they were bothered by 20 PTSD
symptoms during the past month, from 0 (not at all) to 4 (extremely). Items (e.g., “repeated,
disturbing, unwanted memories” and “feeling jumpy or easily startled”) were summed for a
composite score ranging from 0 to 80. The scale evidenced high reliability in the study sample (α
= .95).
Pain intensity. Pain was assessed using the three-item Patient Reported Outcomes
Measurement Information System (PROMIS) Pain Intensity (short form) measure (Stone et al.,
2016). Participants indicated how intense their pain was at its worst and on average over the past
seven days, and their pain level at time of assessment, from 1 (no pain) to 5 (very severe). Items
were summed for a composite score ranging from 3 to 15. The scale evidenced adequate
reliability in the study sample (α = .71).
Substance use. Participants were asked how many days over the prior thirty days they
had at least one drink containing alcohol, and how many days they used cannabis products that

103
contain tetrahydrocannabinol (THC, the ingredient in some cannabis products that gets users
high).
Qualitative Data Analysis
The present study utilized a modified grounded theory approach and followed thematic
development methods outlined by Crabtree and Miller (1992) for analyzing qualitative and
narrative life history interview data. During coding and thematic development, military
experiences and cultural factors influencing multi-morbidity were used as sensitizing concepts to
guide analyses in line with study aims (Bowen, 2006). A codebook was first developed based on
prior theory and empirical work on multi-morbidity. Transcripts were then coded using Atlas.ti
version 22. The lead author and a research assistant co-coded two interview transcripts and
discussed their code definitions and decisions after each transcript in order to refine the
codebook and application of codes. The lead author then coded the remaining transcripts.
Passages with shared and substantive meaning were identified within coded excerpts to develop
an initial set of 13 themes. In an iterative process, themes were refined and grouped into a final
set of nine themes, and these were linked with one another to develop an explanatory model of
multi-morbidity among veterans. Each step of the analysis was conducted under the guidance of
the mentorship team who each had expertise in veteran behavioral health and/or qualitative
methods.
Results
Participants’ military experiences commonly included physical injuries (e.g., being
“blown up,” shot, stabbed, carrying heavy equipment for too long, or experiencing a training
accident), and/or psychological stressors (e.g., combat trauma, military sexual trauma, and

104
morally injurious events [i.e.., perceived moral transgressions, such as killing or witnessing the
death of non-combatants or being betrayed by leadership]).
When analyzing how veterans understand the links between their military service and
vulnerability for multi-morbid pain and behavioral health problems, several themes emerged. In
terms of the types of military experiences that left them vulnerable to developing multi-morbid
problems, veterans commonly discussed (1) military deployments, (2) specific discrete events
such as physical injuries and psychological stressors, and (3) the accumulation of physical and
psychological stressors over time (such as over a military career). Veterans also commonly
described how military cultural factors influenced their military experiences and how these
factors were related to multi-morbid problems. Military cultural factors that were said to increase
vulnerability for multi-morbid problems included (4) the significance of losing one’s physical
and mental fitness for serving personnel, (5) a tendency to ignore or minimize pain, (6) military
drinking norms including self-medication and a glorification of alcohol, and (7) stigma about
seeking treatment. Military cultural factors that were said to decrease vulnerability for multimorbid problems included (8) camaraderie between veterans and (9) the importance of service
to others or to a larger mission. See Figure 4.1 for a visual depiction of emergent themes and
veterans’ conceptualization of how they were linked to vulnerability for multi-morbid problems.
In results below, veterans’ names and ages are modified to protect their confidentiality.
Military experience types and vulnerability for multi-morbid problems
Theme: Military deployments. Among our sample, veterans commonly perceived that
their multi-morbid pain and behavioral health problems were precipitated by a military
deployment or deployments. That is, when asked if their multi-morbid problems all developed in
tandem due to the same precipitating event (i.e., such as a single physical injury or psychological

105
stressor), many indicated it was not a single discrete event but being on deployment in general
that left them vulnerable to developing multi-morbid problems.
Sean, a 34-year-old who served in the Army, was one such veteran who attributed his
multi-morbid pain and behavioral health problems to his experience of being deployed. He stated
during his interview that he developed pain and PTSD “both in direct relation to combat
deployment in Iraq.” While he shared that he suffered from pain after his deployment due to
carrying “half [his] body weight in gear,” he attributed his PTSD symptoms to experiencing
several friends’ deaths during combat while on that same deployment. “So, I don’t want to say
[pain and PTSD] happened at the same time,” Sean explained, “but I was deployed, and it all
happened.” After transitioning to civilian life, Sean began to drink and use other drugs
excessively to cope with both the pain and PTSD that were consequences of his deployment: “It
was just chasing that high to numb the pain and everything that I endured.”
Theme: Specific discrete events. Other veterans could identify specific discrete events
during their time as a service member that heightened vulnerability for both pain and behavioral
health problems concurrently. Specific military experiences that were psychologically stressful
and caused a physical injury, namely combat injuries or accidents involving heavy machinery,
were often said to directly precipitate both behavioral health problems and pain. This was
especially true of traumatic brain injuries: Multiple veterans who experienced a traumatic brain
injury during combat shared that this precipitated both lasting PTSD symptomology and pain
(particularly headaches).
Bethany, a 48-year-old Army veteran, attributed both her pain and behavioral health
symptoms to a single physical injury she sustained during a basic training incident. “I got left in
the pushup position for 45 minutes by a drill sergeant,” she stated. “The next day, I could hardly

106
walk. And I’ve had problems with my neck and back ever since.” Bethany believes she also has
PTSD, which she attributes to this incident as well as the life-altering physical disability that
resulted from it. “It’s caused me to be disabled and discharged from the military and changed the
course of my life. It’s caused me a lot of problems.” Recently, Bethany began self-medicating
with cannabis to cope with both pain and PTSD symptoms that have persisted since the incident
and her discharge.
Some veterans shared that psychologically stressful military experiences can have
profound downstream effects on behavioral health as well as pain, even if these experiences do
not immediately cause a physical injury or pain. For many of these veterans the physical and/or
psychological sequalae of PTSD, such as increased stress, hypervigilance, and traumatic
memories, were often said to act as mechanisms in the development and maintenance of pain
after a traumatic and/or morally injurious event. Some participants indicated they have hurt
themselves or perpetuated an existing injury by tensing or abruptly moving their body when they
were triggered by a loud sound or traumatic memory. In addition, many participants with multimorbid pain and PTSD cited memories of traumatic events when explaining their substance use,
as they often drank or used other drugs to “forget” or “drown away” such memories.
The story of Charlotte, a 34-year-old who served in the Navy, exemplifies the physical
impact that traumatic events can have via PTSD symptomology. Charlotte shared how key
experiences during her military service have contributed to both PTSD and pain, and she cited
being sexually assaulted by a higher up as a low point during her military career. For Charlotte,
recalling the memories of this traumatic event elicited a physical response, including pain, even
during our interview:

107
When I think about a lot of this, it feels like when your lymph nodes swell up in your
throat, like you're getting strep throat. I start getting a very noticeable pain on both sides
of my throat. Thinking about my shoulders and stuff right now, I can feel it like, the
tightness and strain is there. It’s not just one thing, it’s like this cascading effect of your
body just reacting to fear.
Charlotte went on to explain that she is triggered to experience especially severe pain
when her PTSD symptoms are at their highest, which she calls her “overwhelmed” state. She has
a history of problematic alcohol use, which began after she realized that drinking helped her feel
less overwhelmed. “I felt normal. That’s the best way I can describe it.”
The case of Brian, a 38-year-old Army veteran, demonstrates how chronically heightened
stress may be another mechanism linking psychologically stressful military experiences to pain
later in life. Brian shared that a turning point during his military career was taking the life of an
enemy combatant, in line with conceptions of moral injury. “It’s like the human mind and the
human body is not meant to take another man’s life,” Brian stated. “It changes a person.” Brian
later began to experience pain, much of which he attributes to heightened stress beginning to
affect his body physically. “[Taking a life] is like added stress. Stress deteriorates your body. I
think that’s what a lot of [veterans] go through. We’re all deteriorating.” Brian began to drink
heavily shortly after experiencing his turning point, and he continued to drink after returning
from deployment to cope with both his mental health symptoms and pain.
Theme: Accumulation of physical and psychological stressors. For many participants,
it was not a single event but multiple physical and psychological stressors that accumulated over
the course of their military careers that left them vulnerable to developing multi-morbid

108
problems. Veterans explained that training, combat operations, and work during deployments are
often physically demanding and can cause bodily wear and tear over time. Multiple veterans
spoke about having to carry heavy armor or machinery for long periods and cited this as
contributing to pain post-deployment. Many veterans also attributed their mental health
challenges, including PTSD, to experiencing several traumatic and/or morally injurious events
over time during their military careers. Some in the sample experienced physical injuries and/or
traumatic events before or after their military career as well, and these veterans tended to
attribute their pain and behavioral health problems to the combination of their experiences both
in and out of the military.
Jack, a 37-year-old Marine Corps veteran, described the accumulating physical demands
of military service and their potential impact on both pain and behavioral health. Jack
experiences pain in his knees and shoulders, and he was one of several veterans who cited
carrying heavy body armor and equipment over a long period of time as the primary reason for
their pain: “I was wearing a full flak jacket with bullet proof plates, which all weighed like 40
pounds, and then 100 pounds of ammunition on top of it just tearing down my body.” Jack went
on to share that the pain he experiences as a veteran can amplify his PTSD symptoms such as
anger and stress, because pain is a “heavy reminder” of “very physically tough things.”
Reflecting on his active-duty experiences, Jack stated, “Most people don’t beat the shit out of
their bodies the way we had to … The wear and tear on your body during active duty is similar to
being a pro athlete, without the resources of a pro athlete.” Jack has a history of using alcohol to
cope with his PTSD symptoms, which he has deemed problematic, though he finds cannabis to
be helpful for his pain.

109
For 41-year-old Tyler, an Army Special Forces veteran, an accumulation of experiences
that were not just physically demanding but also traumatic appeared to heighten risk for multimorbid problems. When asked when he began to experience pain and PTSD symptoms, Tyler
spoke of being injured many times over his 20-year military career: “I've got shrapnel in my leg,
my right leg behind my knee. I've had breaks in my feet from jumping [as a paratrooper] and
different things that have happened to me. I've had burns on my face and neck and hands.” For
Tyler, the fact that several of these injuries also constituted traumatic events made them even
more impactful. He shared that as he was “exposed to more trauma” he experienced more severe
pain, worse PTSD symptomology, and consumed larger amounts of alcohol to “self-medicate”
for his “mental health issues.”
Cultural moderators of the impact of military experiences on multi-morbid problems
Theme: The significance of losing one’s physical and mental fitness. Several aspects
of military culture were said to said to magnify (i.e., moderate) the impact of challenging
military experiences on multi-morbid problems including pain and behavioral health problems.
One such factor is the military cultural belief in the importance of physical or mental fitness and
the significance of losing one’s fitness. That is, since being of fit mind and body is so vital for
serving military personnel, experiences that negatively affected their physical function or mental
health could be even more distressing and affect their sense of identity. Several veterans
expressed feeling attached to their physically demanding roles in the military and said that it was
“depressing” or “disappointing” for them to retire earlier than they had planned due to a physical
disability. One veteran who had “loved” her work as a mechanic while on active duty but had to
retire early due to the physical strain of her job reported her lowest point was when she realized
she had to take pain medication as a young veteran, because this had made her feel like she was

110
“weak.” This illustrates how loss of fitness may be stigmatized among service members and
veterans.
As Tyler’s injuries accumulated over the course of his 20-year career, he became acutely
aware of his diminishing physical fitness. These changes were especially salient because of what
they meant about his ability to do his job and be an effective team member. He felt shame about
the idea of letting down his fellow service members and a loss of identity. Tyler shared, “It’s
another strain on the mind because [your fitness] is what you think you are, that’s what defines
you.” As Tyler’s mental health deteriorated and his alcohol use increased, his concern that his
team would be upset about his inability to perform represented a barrier to seeking treatment.
Tyler remembers worrying, “This is catastrophic, because they are leaning on me to do my job.
We're a team and they need me to do it.”
Theme: Ignoring or minimizing pain. Veterans also described a culture of ignoring or
minimizing physical and emotional pain in the military and suggested this can influence coping
in the aftermath of challenging military experiences. Veterans reported that during military
service, they and their command at times ignored pain and mental health problems, prioritizing
instead completing the mission. Ignoring pain is “trained into you,” one veteran exclaimed.
Another veteran shared an often repeated saying that summarizes the culture of ignoring pain for
the purpose of completing a mission: “We say, ‘Embrace the suck.’ You know, everything sucks,
you embrace the suck, and you push through.” Thus, some veterans ignored medical advice to
rest or delay deployment after a physical injury, and others described how their command
minimized injuries or accused them of trying to get out of work. Ignoring pain often led to
perpetuating an injury or reinjuring oneself and contributed to worse or more persistent pain over
time.

111
Similarly, the tendency to ignore psychological suffering among service members and
their command led some veterans to not seek or be denied mental health care after experiencing a
traumatic event. This was especially commonly among female veterans in the sample who
reported experiencing military sexual trauma. Most women who reported to a higher up that they
were assaulted were not believed or helped. This may explain why virtually all women
interviewed either never reported or hesitated to report military sexual trauma. One woman
shared, “I felt like I wouldn't be believed, it wouldn't be taken seriously. And maybe it would just
cause more issues for me.”
The experience of Adam, a 43-year-old Army veteran, illustrates that while veterans learn
that ignoring pain can be adaptive for their military careers, it can ultimately contribute to poor
behavioral health outcomes. Adam attributes his pain to being “blown up” several times on
combat deployments. When asked how he tends to cope with pain, he describes “blocking it out
of his mind” and “focusing on the things that need to be done,” even if he continues to
experience some pain. This is what he was taught in the military, he states: “The mission still has
to get done. You can recover and rest at night." At the same time, Adam has at times felt like his
multi-morbid problems are beyond his ability to cope, which left him susceptible to selfmedicating with alcohol. Left with few alternatives, he began drinking heavily while in the
military to relax and “numb the pain.”
Theme: Military drinking norms. Several other veterans reported that heavy drinking
and self-medicating with alcohol after challenging experiences are common in the military.
While drinking is not allowed during deployment, it is often common in garrison (in U.S. or
overseas military installations where they are stationed when not on deployment). One
participant explained “that was all you had over there” to cope with difficulties, and thus

112
multiple veterans engaged in what they called “alcoholic” drinking behavior during their time in
the military. One veteran attributed this to how alcohol is “glorified” in the military, and he
shared, “You have to ply young men with alcohol and tales of bravery to get them to do the
things that you want them to do.” For many in our sample, the military’s drinking culture left
them vulnerable to problematic drinking as veterans to cope with their military experiences and
multi-morbid problems. Some indicated this behavior was perpetuated by ideas of masculinity
and the tendency to ignore pain more broadly, as these can dissuade help-seeking and promote
poor coping.
Damon, a 36-year-old veteran of the Air Force, exemplifies how military cultural norms
around drinking can heighten risk for self-medication after challenging military experiences.
Damon reports that he has PTSD from his experiences as a military police officer: “I’ve seen a
lot of different things during that job.” His low point in the military was responding to a hit and
run vehicular accident on base, after a child was struck and killed. Damon began drinking
heavily to “escape reality” and “get past the memory” of what he had seen. When asked why he
believes he gravitated towards alcohol, he attributed this, in part, to ideas of masculinity: “For
men, we’re taught, ‘Don’t show emotions.’ You gotta be tough. We’re not gonna go seek help.”
Theme: Treatment-related stigma. Another aspect of military culture that can impede
coping and perpetuate multi-morbid symptomology is treatment-related stigma. Some
participants shared that they had stigmatizing experiences from higher ups when trying to seek
care for mental health problems and/or pain while in the service, such as being told mental health
treatment is for poor performers or being called a “baby” when asking to rest or care for an
injury. Especially common were concerns that having a record of seeking treatment that would
negatively affect career prospects, such as by making it challenging to get security clearances.

113
Several participants had internalized these concerns and carry them into civilian life, creating a
barrier to seeking support from loved ones or professionals. Some participants reported worrying
they would be judged by fellow veterans for seeking help as they felt this would constitute an
admission of weakness. Others continued to worry about having a record of poor health as
veterans: “We’re not going to tell [providers] the truth. We’re so used to being monitored.”
Treatment-related stigma critically impede Carrie’s ability to cope after she experienced a
traumatic event. A 39-year-old Army veteran, Carrie’s low point was seeing body parts of
deceased fellow service members after an attack while she was deployed in Iraq. She struggled to
cope and felt herself “deteriorating” emotionally. When she asked for help obtaining treatment,
her command told her to “man up,” and “not be a pussy.” After she began to have nightmares
about her traumatic event, she again asked her command for help and was called “dishonorable.”
These experiences stuck with her, leaving her hesitant to seek treatment for her multi-morbid
problems as a veteran. “That’s why there’s veterans that still stigmatize [treatment],” she stated.
“Because when we sought help, we were treated as broken and not repairable.” As with other
veterans we interviewed, Carrie’s untreated behavioral health problems began to affect her
physically, leading to pain: “My body will remember what my mind wants to forget.”
For Tyler, having a non-stigmatizing experience was vital for his ability to finally come
to terms with his diminishing physical ability and treat his multi-morbid problems. Though he
had been avoiding asking for help due to his fears of being perceived as weak and letting others
down, his turning point was hitting “rock bottom” due to how badly PTSD and alcohol use were
affecting his life, resulting in him finally seeking treatment. After he divulged his behavioral
health problems to his command he lost his security clearance, effectively ending his career. He
was surprised, however, to find support in his command and fellow service members, who

114
helped him find help. “That helped my mental health quite a bit,” Tyler shared. That is when all
these things started to come into order.”
Theme: Camaraderie between veterans. Some aspects of military culture that were
instilled during military service were protective for coping with challenging experiences and
coping with multi-morbid problems. One protective factor was the importance placed on
camaraderie among fellow service members and veterans. Participants often spoke of the
strength of the bonds they developed with those they served with. They highlighted how
continued relationships with these “battle buddies” help them cope with mental health symptoms
and allow them to share information about medical and psychological treatments among others
with similar problems. Several discussed that it is easier to talk with other veterans than nonveterans about their military experiences and shared that this can make veterans a particularly
important source of psychological support.
Camaraderie with fellow service members eventually became a key source of resilience
for Brian in the aftermath of his morally injurious experience. He shared he “locked [his]
emotions away” after his first tour and took years to talk about his experiences with anyone.
Finally, he sought support from fellow veterans. “I think it’s easier because we all went through
the same things.” For Brian, talking with other veterans is now one of the primary ways he copes
with mental health symptoms, and he and his friends also sometimes discuss pain management
strategies and treatments with one another. Reflecting on his relationships with other veterans, he
stated, “When you’re a soldier, you’re a soldier for life.”
Sean’s experience was similar. After he returned from deployment, he reached his low
point as his drinking continued to “spiral.” Finally, he found help by reaching out to a veteranserving organization, and his substance use began to improve. Sean shared that one of the most

115
helpful aspects of this organization is that many of the staff are veterans. “You’re with people
that are just like you,” Sean exclaimed. “The camaraderie you build through these programs is
very similar to the camaraderie you have while serving.” Sean shared that this organization has
helped him “get out of [his] shell” and “get away from the drugs and alcohol.”
Theme: The importance of service. The importance that service members place on
serving others or carrying out a larger mission was revealed as another source of protection after
challenging military experiences. Veterans described how engaging in mission-driven volunteer
work was a way to come to terms with challenging experiences and cope with subsequent multimorbid problems. Participants most often described the benefits of engaging in volunteer work
that benefitted other veterans. For Sean, one of the most valuable aspects of becoming involved
with the veteran-serving organization was eventually taking a leadership role and helping other
veterans: “That fulfills me and gives me a new meaning and a new mission.” For Charlotte, who
continued to feel disillusioned for some time after being sexually assaulted, her turning point was
standing up to an officer to advocate for other female service member victims of military sexual
trauma.
Paul, a 41-year-old Army veteran, illustrates how service to other veterans can help
veterans not only cope with multi-morbid problems, but heal from the challenging experiences
that precipitated such problems. Paul’s experienced a traumatic brain injury in 2004, leading to
PTSD, severe pain to, and the loss of function in one leg. He has used a wheelchair for several
years, which he states has been “tough” and can drive problematic drinking: “It’s kind of that
feeling of being useless now.” However, Paul finds respite in serving other veterans through
various organizations and activities. For his turning point, he described his work leading service
dog training retreats for other veterans and their families. He shared the profound benefits he

116
experiences, and particularly how leading these retreats is helping him come to terms with his
own physical and psychological difficulties:
It's healing for me to be able to watch these other families heal. Because had I not been
in the wheelchair, had I not needed the service dog, had I not gone through all of these
things with my own family and separation, I wouldn't have understood the impact that it
could make on these other veterans.
Discussion
Many risk factors of various types have been empirically linked with multi-morbid
problems among veterans, including psychological factors (Asmundson et al., 2002; Raines et
al., 2021), combat and other deployment experiences (Chui et al., 2022), and aspects of military
culture (O’Loughlin et al., 2023). It is therefore likely that the etiology of multi-morbid problems
among veterans is complex and multi-factorial (i.e., due to several, interacting determinants),
however, theoretical models explaining vulnerability for multi-morbidity have focused on a
relatively narrow set of factors (Asmundson et al., 2002). This paper examines the subjective
experiences of military veterans during and after deployment using qualitative methods to
develop a novel, shared vulnerability model explaining multi-morbidity among veterans. Veteran
participants revealed how military experiences and culture likely interact to determine
vulnerability for multi-morbid problems. Importantly, veterans discussed not only how some
military cultural factors could confer harm, but how others might confer protection, and these
may be leveraged in novel intervention and prevention strategies addressing multi-morbidity.

117
Several themes emerged to capture the types of direct experiences that veteran
participants attributed their multi-morbid problems to. In line with prior work on traumatic injury
(Cifu et al., 2013), some veterans described how a single military service-related discrete event
that both resulted in a physical injury and was psychologically traumatic led to them having both
pain and behavioral health problems such as PTSD and substance use. Some veterans linked TBI
to their multi-morbid problems, a known risk factor for such problems (Cifu et al., 2013). Other
veteran cited injuries that have been less explored in the existing literature, such as a traumatic
training injury and an injury from jumping out of a plane as a paratrooper. The field should
perhaps continue to explore traumatic injury among veterans to ensure its conception of such
injuries are broad enough to better target their effects. For other veterans, psychologically
traumatic events could lead to downstream substance use and pain – even if they did not
immediately cause injury or pain – because symptoms of PTSD could later become mechanisms
of pain and self-medication. Veterans in our sample reported traumatic memory, increased stress,
and hypervigilance as such mechanisms linking PTSD, pain, and substance use. This finding is
in line with mutual maintenance models (Sharp & Harvey, 2001) suggesting that shared
mechanisms between these conditions in part explain multi-morbidity among them. Like another
qualitative study demonstrating traumatic memory as a shared mechanism for PTSD and pain
among non-veterans (B. Taylor et al., 2013) ours demonstrates how rich interview data can
clarify nuanced mechanistic pathways – in this case, among veterans with PTSD, pain, and
problematic substance use.
For other veterans, it was not a discrete event but multiple exposures over time that led to
vulnerability for multi-morbidity. Some veterans identified military deployment as the factor that
heightened risk for both pain and behavioral health problems and elaborated by saying that

118
deployment included exposure to multiple physical and psychological stressors impacting multimorbid problems such as combat injuries, trauma, and morally injurious events. Some veterans
explicitly described how the accumulation of multiple exposures over time (during a
deployment, multiple deployments, or over the course of their military careers) led to multimorbid problems because of physical and psychological wear and tear. That is, veterans
described how carrying heavy machinery for too long, “beating the shit out of [their] bodies,”
and exposure to multiple traumatic events were risk factors for pain and behavioral health
problems. These observations are in line with cumulative stress and stress sensitization models
(Davis et al., 2023; Dube et al., 2009) which state that experiencing multiple stressors over time
has a profound negative effect on health by making individuals more sensitive to the effects of
stress. Some recent quantitative research has also provided evidence in support of these models
among veterans (Davis et al., 2023), suggesting that those who have experienced combat trauma,
military sexual trauma, and childhood trauma in combination experience higher stress and PTSD
symptomology. The field may benefit from developing an increasingly nuanced typology of the
various psychological and physical stressors that serving personnel commonly experience
(similar to typologies of adverse childhood experiences; Felitti, 2009), and determining how
specific types, combinations, and frequencies of exposure are associated with both behavioral
health and physical health outcomes post-deployment.
It was not only veterans’ direct experiences, but also aspects of military culture that
influenced how they thought about and coped with such experiences, that left them vulnerable to
multi-morbid problems. Namely, veterans described how their belief in the importance of
physical and mental fitness magnified their distress in the face of a physical injury or
psychological challenges that resulted in the loss of fitness. Relatedly, participants reported that

119
they and their higher ups often ignored physical and psychological distress (i.e., pain or
behavioral health problems) while deployed and this influenced how they coped with such
problems as veterans. Veterans conveyed how their identities can be tied to their physical and
mental fitness and they can feel they are letting their fellow service members down when pain
and/or mental health problems impinge their ability to serve on a team, reflecting over-arching
stigma surrounding pain, mental health, and disability. Treatment-related stigma was also cited as
a barrier to addressing multi-morbid problems, and a culture of heavy drinking in the military left
veterans vulnerable to self-medicating to cope with such problems. While some quantitative
studies have demonstrated associations between cultural factors and pain and/or behavioral
health problems (Herrera et al., 2013; Meadows et al., 2022), fewer have explored how such
factors can moderate the multi-morbid impact of challenging experiences (Nevarez et al., 2017).
Results of our study are in line with bio-cultural models of health (McElroy, 1990; Zuckerman,
2018), which suggest that while many cultural beliefs and practices develop to help people adapt
to stressful experiences, they can end up diminishing one’s ability to cope. Veterans discussed
how their tendency to ignore physical and psychological distress, for example, was “trained into”
them to enable them to effectively carry out their work in the military activities, but this can
often leave them without effective strategies to deal with multi-morbid problems down the line.
Other aspects of military culture were consistently protective, as they helped veterans
cope with their challenging military experiences and resultant multi-morbidity. One such
protective factor was camaraderie amongst service members. Veterans commonly shared that it is
easier to connect with other veterans and that doing so benefits them psychologically. Veterans
specifically reported discussing with one another about their military experiences and their
experiences with pain treatments, as well as supporting one another in recovery from a substance

120
use disorder. These results expand on findings from a single survey study demonstrating
camaraderie among veterans may protect against the development of PTSD symptoms in the
aftermath of combat exposure (Nevarez et al., 2017). Relatedly, veterans conveyed how the value
they place on serving others or a greater mission can be protective. Veterans discussed how
helping other veterans who have gone through challenging experiences, ranging from military
sexual trauma to physical disabilities, especially helps them cope with and heal from their own
similar experiences. Interventions that involve receiving support from other veterans show
promise for behavioral health problems (Hall et al., 2020) and pain (Matthias et al., 2016), and
the veterans in our study suggest similar benefits may extend to those providing support to other
veterans.
We note some potential limitations of this study. First, while we set out to explore
vulnerability for multi-morbid problems among veterans, results might differ if we recruited
veterans with physical and behavioral health problems other than the ones we focused on in this
study (pain, PTSD, and cannabis or alcohol use). As we did not recruit an opioid using sample
nor ask explicitly about opioid use (aside from asking about pain treatment and substance use
generally), opioid use may be critical to understand among many veterans with pain and
behavioral health problems. Veterans from other countries or eras also likely differ in the kinds
of military experiences they are exposed to and the relevance of specific cultural factors.
Additionally, interviews were conducted via Zoom with a subset of veterans who were
previously participants in one of two mobile app studies. Responses might differ among veterans
who are less likely to engage in research studies or who have less comfort with technology.
Though we provide evidence for a compelling explanatory model, future research with larger
samples is necessary to validate and build upon this work. Additional qualitative research may

121
establish greater nuance in describing how specific military cultural factors moderate specific
experiences; while our preliminary model implies moderating factors play similar roles across
different types of experiences, it could be the case, for example, that losing fitness is especially
significant for those who experience a discrete injury, or that camaraderie with other veterans
one deployed with could protect against an otherwise challenging deployment.
Our results carry important implications for clinicians and clinical researchers who work
with veterans. First, psychosocial assessment and screening tools should allow clinicians to
capture the wide range of military-related challenging experiences and cultural risk and
protective factors, as well as veterans’ perceptions of how these factors influence their
symptomology. Fortunately, integrative and multi-disciplinary treatments that target both the
physical and behavioral health effects of challenging military experiences exist (Chiesa et al.,
2014; Lumley & Schubiner, 2019; Otis et al., 2009) but these should be more widely
disseminated given known gaps in available treatment (Becker et al., 2017; Cheney et al., 2018).
Clinicians and researchers should further tailor their approaches to address military-cultural
barriers to treatment receipt and efficacy, such as treatment-related stigma and a heavy drinking
culture. Interventions may seek to modify challenging beliefs about the importance of physical
activity or the tendency to ignore pain, and they may leverage cultural strengths by fostering
camaraderie and service among veterans.

122
Table 4.1 Qualitative participant characteristics (n = 20)
M (SD) or n (%)
Age 38.05 (6.1)
Sex
Men 14 (70%)
Women 6 (30%)
Race
White 15 (75%)
Black 2 (10%)
Mixed/other 3 (15%)
Ethnicity
Hispanic/Latinx 5 (25%)
Non-Hispanic/Latinx 15 (75%)
PTSD symptoms 42.35 (17.72)
Pain intensity 10 (1.38)
Drinking days (past 30) 10.35 (9.24)
Cannabis days (past 30) 8.65 (12.43)
Note: Participant’s characteristics collected during baseline survey of the
parent study from which they were recruited. Sex is biological sex at birth
reported by the participant. Mixed/other race category includes one individual
who reported their race as “Asian, White”, one individual who reported their
race as “Native American/Alaska Native, White,” and one individual who
reported their race as “Other.” PTSD symptoms were measured by the PTSD
Checklist for DSM-6 (range: 0-80). Pain was measured by the PROMIS Pain
Intensity scale (range: 3-15). Drinking and cannabis days are number of days
participant drank or used cannabis during the past 30 days.

123
Figure 4.1 Preliminary, bio-cultural vulnerability model of veteran multi-morbidity

124
Chapter 5: Conclusion
Multi-morbid physical and behavioral health problems, including pain, PTSD, and
problematic substance use, are exceedingly prevalent among U.S. military veterans (Kerns &
Brandt, 2020; Norman et al., 2018; Stecker et al., 2010). This is a population that is often
exposed to physically and psychologically challenging experiences during military service
(Davis et al., 2023; Mitchell et al., 2011; Wisco et al., 2017), and it is likely numerous
biopsychosocial factors interact in complex ways to influence vulnerability for multi-morbid
pain, PTSD, and problematic substance use (Engel, 1981; Wade & Halligan, 2017). Existing
theory also suggests symptoms of each of these problems may make the other problems worse
(Asmundson et al., 2002). Better understanding the etiology of multi-morbid problems in terms
of how they develop and impact one another among veterans is crucial. This three-paper
dissertation sought to advance theory and knowledge of multi-morbid problems among veterans,
by exploring (1) the day-to-day associations between pain, PTSD symptoms, and cannabis use,
(2) biopsychosocial predictors of pain among veterans with PTSD, and (3) veterans' subjective
perceptions of how military experiences and culture influence with their multi-morbid problems.
Review of major findings and integration with existing literature
In paper 1, DSEM was applied to daily diary data on pain, PTSD symptoms, and
cannabis use to explore the lagged associations between these constructs at the daily level.
Results did not provide evidence in favor of mutual maintenance between pain and PTSD
symptoms at the daily level; pain and PTSD did not predict one another from one day to the next,
in contrast to some prior studies that tested associations between these problems using quarterly
measurement intervals (McAndrew et al., 2019; Ravn et al., 2018). However, there was a

125
significant positive correlation between same-day pain and PTSD symptoms, suggesting there
may be day-varying “third variables” (e.g., problems sleeping or exposure to daily stressors; Ivey
et al., 2018; Messman et al., 2022) influencing both problems concurrently. Cannabis use and
pain were negatively associated at the daily level such that heightened cannabis use predicted
lower pain the subsequent day, supporting and extending prior work demonstrating analgesic
effects of cannabis (Romero-Sandoval et al., 2018). Pain also predicted reduced next-day
cannabis use, which seems to contradict the self-medication model (Khantzian, 1997) but could
be due to veterans reducing their cannabis use when they become tolerant to its effects on pain
(Romero-Sandoval et al., 2018). Anxiety sensitivity, a psychological factor that has been
theorized to confer shared vulnerability for pain and PTSD (Asmundson et al., 2002), did not
moderate associations between these problems or with cannabis use at the daily level. As there
were between-person associations between pain and PTSD, this suggests a need to explore
additional individual-level factors that might confer shared vulnerability.
In paper 2, machine learning methods were applied to examine biopsychosocial
predictors of frequent pain and increases in pain intensity among veterans with PTSD. Pain
interference emerged as a top predictor of frequent pain, extending prior research on how pain
interference can influence other pain outcomes (Naylor et al., 2019). Proximal changes in mental
health symptoms also predicted frequent pain, building on prior work linking mental health and
pain (Jensen & Turk, 2014) and suggesting short-term changes in mental health symptoms may
have a relatively long-term influence on pain. Posttraumatic cognitions and proximal changes in
positive affect emerged as the top predictors of pain intensity increases, and post-hoc analyses
suggested increasing positive affect may protect against pain among those with posttraumatic
cognitions in particular. Sex had a relatively large magnitude association with pain intensity

126
increases, supporting recent work demonstrating disparities in pain among female veterans
(Naylor et al., 2017, 2019). While these analyses should be replicated in larger samples, they
exemplify how machine learning can provide a more holistic picture of the determinants of pain
in line with biopsychosocial models (Engel, 1981).
In paper 3, qualitative interviews were conducted with 20 veterans to understand their
perceptions of how military experiences and culture relate with multi-morbid pain, PTSD, and
problematic substance use. Veterans described several types of military experiences that left them
vulnerable to multi-morbid problems in line with prior work, ranging from singular events that
caused injury and distress (Afari et al., 2009; C. J. Bryan et al., 2015; Nelson, 2002) to an
accumulation of stressors over time (Davis et al., 2023; Macia et al., 2020). Veterans conveyed
how military cultural factors interacted with these experiences to influence coping and multimorbidity. Factors said to increase vulnerability included the significance of losing one's physical
and mental fitness, a tendency to ignore pain, military drinking norms, and treatment stigma,
building upon prior quantitative studies with veterans linking cultural factors with pain and
behavioral health problems (Cogan et al., 2021; Meadows et al., 2022). Factors said to decrease
vulnerability included camaraderie between veterans and engaging in service to other veterans,
expanding upon a prior survey study showing camaraderie may protect against PTSD after
combat exposure (Nevarez et al., 2017). Results provide support for a novel theoretical model of
veteran vulnerability for multi-morbidity, in line with bio-cultural models of health developed by
medical anthropologists (McElroy, 1990; Zuckerman, 2018).

127
Summary of theoretical implications
This dissertation tested and sought to extended current theories of multi-morbidity among
veterans. In terms of the mutual maintenance model (Sharp & Harvey, 2001), paper 1 did not
find support for bidirectional associations between pain and PTSD symptoms at the day-tosubsequent-day level. There was a same-day correlation between pain and PTSD symptoms,
suggesting day-varying factors may maintain both problems concurrently. Proximal changes in
PTSD symptoms also predicted pain intensity increases over a longer time frame (the next eight
weeks) in paper 2, and veterans described several shared mechanisms (traumatic memory, stress,
hypervigilance) between pain and PTSD in paper 3. In terms of the self-medication model
(Khantzian, 1997), in paper 1 prior day cannabis use predicted subsequent reductions in pain,
suggesting veterans may be using cannabis for pain relief. The fact that prior day pain predicted
subsequent reductions in cannabis use is on the surface difficult to reconcile with the selfmedication model (which would generally also suggest heightened pain would precede increased
cannabis use), but this result could be due to veterans reducing their use when cannabis stops
having an analgesic effect. In paper 2, cannabis use does predict worse pain over a longer time
period (the subsequent eight weeks), though alcohol use predicts better pain outcomes, which
might be explained by different baseline levels of cannabis vs. alcohol. Veterans also consistently
self-reported a history of self-medicating with both alcohol and cannabis for pain and PTSD in
paper 3. As for the shared vulnerability model (Asmundson et al., 2002), paper 1 did not identify
anxiety sensitivity as a predisposing factor, as there was no evidence anxiety sensitivity
moderates associations between multi-morbid problems. Paper 3 however extends conceptions of
shared vulnerability by highlighting how military experiences and cultural factors can interact
with one another to affect vulnerability for multi-morbid problems.

128
Taken together, these findings suggest that theories of multi-morbidity should
increasingly (1) consider the role of different time intervals when assessing symptom
associations, moderators, and mechanisms, (2) incorporate a wider array of biopsychosocial
moderators including symptom severity (e.g., associations might differ between those with
differential levels of use of different substances, pain, and behavioral health problems, as well as
different pain conditions), and (3) continue to integrate the cultural context of multi-morbidity.
Future research should test refined theoretical models that incorporate these considerations, using
methods that can capture the dynamic and multifaceted nature of multi-morbidity among
veterans.
Summary of clinical implications
Results of this dissertation carry several implications for clinical practice with veterans.
As paper 1 revealed days with heightened pain were likely to be days with heightened PTSD,
providers should prioritize transdiagnostic treatments like mindfulness that target shared
mechanisms such as daily stressors (Dahm et al., 2015; Santorelli et al., 2017). As results also
suggest veterans are self-medicating for pain, providers should stay informed on the changing
scientific and legal landscape when it comes to cannabis use and be prepared to discuss the
potential benefits and consequences with their veteran patients. Given evidence for daily
symptom associations in general, integrating real-time symptom monitoring tools such as mobile
apps directly into systems of care could allow for increasingly personalized and time-sensitive
interventions. Paper 2 provides additional evidence for the importance of integrative treatments
and especially points to positive affect as a potentially salient treatment target for veterans with
multi-morbid PTSD and pain. Health systems should also remain vigilant in addressing sex-

129
related disparities in pain. Paper 3 suggests psychosocial assessments should capture the wide
range of military-related risk and protective factors as well as veterans' perceptions of how such
factors influence their multi-morbid symptomology. Clinicians should address military-cultural
barriers to treatment such as stigma and the tendency to ignore pain, and interventions may
leverage cultural strengths by fostering camaraderie and engaging veterans in serving other
veterans. Findings also reinforce ongoing calls for trauma-informed care in both behavioral and
physical healthcare settings, where providers understand and acknowledge the potentially
profound impact of veterans' military experiences (Butler et al., 2011).
Future research directions
Several additional avenues of future research should follow from this dissertation.
Researchers should continue to explore multivariate associations between multi-morbid
problems among veterans, perhaps prioritizing ecological momentary assessment to better
understand same-day (rather than day to day) symptom associations, mechanisms, and
moderators. In combination with mobile and wearable device data, there are compelling
possibilities for clarifying the proximal determinants of multi-morbid problems, including
moment-to-moment triggers and physiological states, and social interaction of various types and
quality (Matsangidou et al., 2021). Advances in data science and machine learning can equip
researchers well to leverage such data; these methods show promise not only for extending
etiological knowledge but also for informing improvement in health systems and practices, such
as the development of decision support tools that help clinicians select personalized interventions
for at-risk individuals with pain and behavioral health problems (Matsangidou et al., 2021). In
terms of person-level vulnerability for multi-morbid problems, our preliminary bio-cultural

130
model provides a novel framework for understanding such problems among veterans, and this
model should be tested and refined among larger samples of veterans with a wider range of
problems. Intervention researchers should develop integrative interventions for pain and
behavioral health problems among veterans that target salient constructs from this dissertation
(i.e., positive affect) and other ongoing research, and these interventions should be co-designed
by veterans to leverage their perspectives and military cultural strengths. This dissertation also
points to opportunities for measure development, such as low burden multidimensional pain
scales, and screening and assessment tools to capture the range of challenging military
experiences. Finally, research on multi-morbidity should focus on at-risk subgroups of veterans,
including women and racial and ethnic minorities, as well as aging post-9/11 veterans who may
increasingly become at risk of new onset or worsening pain (Carlson et al., 2018; McClendon et
al., 2021; U.S. Department of Veteran’s Affairs, 2015). Additional important subgroups are
veterans recruited from pain clinics, those with different types and causes of pain, and those with
problematic opioid use. The field should prioritize increasingly interdisciplinary research to
match the complexity of multi-morbid problems among veterans, and to develop and disseminate
well-matched intervention and prevention strategies.

131
References
Abraham, T., Cheney, A. M., & Curran, G. M. (2017). A Bourdieusian analysis of US military
culture ground in the mental help-seeking literature. American Journal of Men’s Health,
11(5), 1358–1365.
Adler, J. M., Lodi-Smith, J., Philippe, F. L., & Houle, I. (2016). The incremental validity of
narrative identity in predicting well-being: A review of the field and recommendations for
the future. Personality and Social Psychology Review, 20(2), 142–175.
Afari, N., Harder, L. H., Madra, N. J., Heppner, P. S., Moeller‐Bertram, T., King, C., & Baker, D.
G. (2009). PTSD, combat injury, and headache in veterans returning from Iraq/Afghanistan.
Headache: The Journal of Head and Face Pain, 49(9), 1267–1276.
Allan, N. P., Raines, A. M., Capron, D. W., Norr, A. M., Zvolensky, M. J., & Schmidt, N. B.
(2014). Identification of anxiety sensitivity classes and clinical cut-scores in a sample of
adult smokers: Results from a factor mixture model. Journal of Anxiety Disorders, 28(7),
696–703.
Ames, G., & Cunradi, C. (2004). Alcohol use and preventing alcohol-related problems among
young adults in the military. Alcohol Research & Health, 28(4), 252.
Ang, D. C., Peloso, P. M., Woolson, R. F., Kroenke, K., & Doebbeling, B. N. (2006). Predictors
of incident chronic widespread pain among veterans following the first Gulf War. The
Clinical Journal of Pain, 22(6), 554–563.
Angra, S., & Ahuja, S. (2017). Machine learning and its applications: A review. 2017
International Conference on Big Data Analytics and Computational Intelligence (ICBDAC),
57–60. https://doi.org/10.1109/ICBDACI.2017.8070809
Asmundson, G. J. G., Coons, M. J., Taylor, S., & Katz, J. (2002). PTSD and the experience of
pain: research and clinical implications of shared vulnerability and mutual maintenance
models. The Canadian Journal of Psychiatry, 47(10), 930–937.
Asparouhov, T., Hamaker, E. L., & Muthén, B. (2018). Dynamic structural equation models.
Structural Equation Modeling: A Multidisciplinary Journal, 25(3), 359–388.
Babor, T., Higgins-Biddle, J. C., Saunders, J. B., & Monteiro, M. G. (2001). The Alcohol Use
Disorders Identification Test: Guidelines for use in primary care. Geneva: World Health
Organization, 1–40.
http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:The+Alcohol+Use+Disor
ders+Identification+Test:+Guidelines+for+Use+in+Primary+Care#9
Babson, K. A., & Feldner, M. T. (2010). Temporal relations between sleep problems and both
traumatic event exposure and PTSD: a critical review of the empirical literature. Journal of
Anxiety Disorders, 24(1), 1–15.

132
Becker, W. C., Dorflinger, L., Edmond, S. N., Islam, L., Heapy, A. A., & Fraenkel, L. (2017).
Barriers and facilitators to use of non-pharmacological treatments in chronic pain. BMC
Family Practice, 18(1), 41. https://doi.org/10.1186/s12875-017-0608-2
Benedict, T. M., Keenan, P. G., Nitz, A. J., & Moeller-Bertram, T. (2020a). Post-Traumatic Stress
Disorder Symptoms Contribute to Worse Pain and Health Outcomes in Veterans With PTSD
Compared to Those Without: A Systematic Review With Meta-Analysis. Military Medicine.
https://doi.org/10.1093/milmed/usaa052
Benedict, T. M., Keenan, P. G., Nitz, A. J., & Moeller-Bertram, T. (2020b). Post-Traumatic Stress
Disorder Symptoms Contribute to Worse Pain and Health Outcomes in Veterans With PTSD
Compared to Those Without: A Systematic Review With Meta-Analysis. Military Medicine.
https://doi.org/10.1093/milmed/usaa052
Berghoff, C. R., McDermott, M. J., & Dixon-Gordon, K. L. (2018). Psychological flexibility
moderates the relation between PTSD symptoms and daily pain interference. Personality
and Individual Differences, 124, 130–134.
https://doi.org/https://doi.org/10.1016/j.paid.2017.12.012
Black, D. S. (2009). A brief definition of mindfulness. Behavioral Neuroscience, 7(2), 109.
Boden, M. T., & Hoggatt, K. J. (2018). Substance use disorders among veterans in a nationally
representative sample: prevalence and associated functioning and treatment utilization.
Journal of Studies on Alcohol and Drugs, 79(6), 853–861.
Boehnke, K. F., Scott, J. R., Litinas, E., Sisley, S., Williams, D. A., & Clauw, D. J. (2020). Highfrequency medical cannabis use is associated with worse pain among individuals with
chronic pain. The Journal of Pain, 21(5–6), 570–581.
Bolger, N., & Laurenceau, J.-P. (2013). Intensive longitudinal methods: An introduction to diary
and experience sampling research. Guilford Press.
Bourn, L. E., Sexton, M. B., Porter, K. E., & Rauch, S. A. M. (2016). Physical activity moderates
the association between pain and PTSD in treatment-seeking veterans. Pain Medicine
(United States), 17(11), 2134–2141. https://doi.org/10.1093/pm/pnw089
Bovin, M. J., Marx, B. P., Weathers, F. W., Gallagher, M. W., Rodriguez, P., Schnurr, P. P., &
Keane, T. M. (2016). Psychometric properties of the PTSD checklist for diagnostic and
statistical manual of mental disorders-fifth edition (PCL-5) in veterans. Psychological
Assessment, 28(11), 1379–1391. https://doi.org/10.1037/pas0000254
Bowe, A., & Rosenheck, R. (2015). PTSD and substance use disorder among veterans:
characteristics, service utilization and pharmacotherapy. Journal of Dual Diagnosis, 11(1),
22–32.
Bowen, G. A. (2006). Grounded Theory and Sensitizing Concepts. International Journal of
Qualitative Methods, 5(3), 12–23. https://doi.org/10.1177/160940690600500304

133
Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32.
Bryan, C. J., Bryan, A. O., Anestis, M. D., Anestis, J. C., Green, B. A., Etienne, N., Morrow, C.
E., & Ray-Sannerud, B. (2015). Measuring Moral Injury: Psychometric Properties of the
Moral Injury Events Scale in Two Military Samples. Assessment, 1, 14.
Bryan, J. L., Hogan, J., Lindsay, J. A., & Ecker, A. H. (2021). Cannabis use disorder and posttraumatic stress disorder: The prevalence of comorbidity in veterans of recent conflicts.
Journal of Substance Abuse Treatment, 122, 108254.
Butler, L. D., Critelli, F. M., & Rinfrette, E. S. (2011). Trauma-informed care and mental health.
Directions in Psychiatry, 31(3), 197–212.
Cameron, R. P., & Gusman, D. (2003). The primary care PTSD screen (PC-PTSD): development
and operating characteristics. Primary Care Psychiatry, 9(1), 9–14.
Caniglia, E. C., Stevens, E. R., Khan, M., Young, K. E., Ban, K., Marshall, B. D. L., Chichetto,
N. E., Gaither, J. R., Crystal, S., Edelman, E. J., Fiellin, D. A., Gordon, A. J., Bryant, K. J.,
Tate, J., Justice, A. C., & Braithwaite, R. S. (2020). Does Reducing Drinking in Patients
with Unhealthy Alcohol Use Improve Pain Interference, Use of Other Substances, and
Psychiatric Symptoms? Alcoholism: Clinical and Experimental Research, 44(11), 2257–
2265. https://doi.org/https://doi.org/10.1111/acer.14455
Carlson, M., Endlsey, M., Motley, D., Shawahin, L. N., & Williams, M. T. (2018). Addressing
the impact of racism on veterans of color: A race-based stress and trauma intervention.
Psychology of Violence, 8(6), 748.
Che, X., Cash, R., Ng, S. K., Fitzgerald, P., & Fitzgibbon, B. M. (2018). A Systematic Review of
the Processes Underlying the Main and the Buffering Effect of Social Support on the
Experience of Pain. The Clinical Journal of Pain, 34(11).
https://journals.lww.com/clinicalpain/fulltext/2018/11000/a_systematic_review_of_the_pro
cesses_underlying.10.aspx
Cheney, A. M., Koenig, C. J., Miller, C. J., Zamora, K., Wright, P., Stanley, R., Fortney, J.,
Burgess, J. F., & Pyne, J. M. (2018). Veteran-centered barriers to VA mental healthcare
services use. BMC Health Services Research, 18(1), 1–14. https://doi.org/10.1186/s12913-
018-3346-9
Chiesa, A., Anselmi, R., & Serretti, A. (2014). Psychological mechanisms of mindfulness-based
interventions: what do we know? Holistic Nursing Practice, 28(2), 124–148.
Chui, Z., Fear, N. T., Greenberg, N., Jones, N., Jones, E., & Goodwin, L. (2022). Combat
exposure and co-occurring mental health problems in UK Armed Forces personnel. Journal
of Mental Health, 31(5), 624–633.
Cifu, D. X., Taylor, B. C., Carne, W. F., Bidelspach, D., Sayer, N. A., Scholten, J., & Campbell,
E. H. (2013). Traumatic brain injury, posttraumatic stress disorder, and pain diagnoses in

134
OIF/OEF/OND Veterans. Journal of Rehabilitation Research and Development, 50(9),
1169–1176.
Clark, M. E., Gironda, R. J., & Young, R. W. (2003). Development and validation of the Pain
Outcomes Questionnaire-VA. Journal of Rehabilitation Research and Development, 40(5),
381–395. https://doi.org/10.1682/JRRD.2003.09.0381
Cogan, A. M., Haines, C. E., & Devore, M. D. (2021). Intersections of US military culture,
hegemonic masculinity, and health care among injured male service members. Men and
Masculinities, 24(3), 468–482.
Cohen, A. B. (2009). Many forms of culture. American Psychologist, 64(3), 194.
Curran, P. J., Howard, A. L., Bainter, S. A., Lane, S. T., & McGinley, J. S. (2014). The separation
of between-person and within-person components of individual change over time: a latent
curve model with structured residuals. Journal of Consulting and Clinical Psychology,
82(5), 879.
Curry, I., Malaktaris, A. L., Lyons, R., Herbert, M. S., & Norman, S. B. (2019). The Association
Between Negative Trauma‐Related Cognitions and Pain‐Related Functional Status Among
Veterans With Posttraumatic Stress Disorder and Alcohol Use Disorder. Journal of
Traumatic Stress, 32(2), 317–322. https://doi.org/10.1002/jts.22394
Dahm, K. A., Meyer, E. C., Neff, K. D., Kimbrel, N. A., Gulliver, S. B., & Morissette, S. B.
(2015). Mindfulness, Self-Compassion, Posttraumatic Stress Disorder Symptoms, and
Functional Disability in U.S. Iraq and Afghanistan War Veterans. Journal of Traumatic
Stress, 28(5), 460–464. https://doi.org/10.1002/jts.22045
Daniëls, N. E. M., Hochstenbach, L. M. J., van Zelst, C., van Bokhoven, M. A., Delespaul, P. A.
E. G., & Beurskens, A. J. H. M. (2021). Factors That Influence the Use of Electronic
Diaries in Health Care: Scoping Review. JMIR Mhealth Uhealth, 9(6), e19536.
https://doi.org/10.2196/19536
Davis, J. P., Eddie, D., Prindle, J., Dworkin, E. R., Christie, N. C., Saba, S., DiGuiseppi, G. T.,
Clapp, J. D., & Kelly, J. F. (2021). Sex differences in factors predicting post-treatment
opioid use. Addiction, 116(8), 2116–2126.
Davis, J. P., Prindle, J., Saba, S., Lee, D. S., Leightley, D., Tran, D. D., Sedano, A., Fitzke, R.,
Castro, C. A., & Pedersen, E. R. (2023). Childhood adversity, combat experiences, and
military sexual trauma: A test and extension of the stress sensitization hypothesis.
Psychological Medicine, 53(9), 4055–4063.
Davis, J. P., Prindle, J., Saba, S., Tran, D., Lee, D., Sedano, A., Castro, C. C., & Pedersen, E. R.
(2022). Changes in perceived stress during the COVID‐19 pandemic among American
veterans. Stress and Health.
Davis, J. P., Rao, P., Dilkina, B., Prindle, J., Eddie, D., Christie, N. C., DiGuiseppi, G., Saba, S.,
Ring, C., & Dennis, M. (2022). Identifying individual and environmental predictors of

135
opioid and psychostimulant use among adolescents and young adults following outpatient
treatment. Drug and Alcohol Dependence, 233, 109359.
Domingos, P. (2012). A few useful things to know about machine learning. Communications of
the ACM, 55(10), 78-87.
Dube, S. R., Fairweather, D., Pearson, W. S., Felitti, V. J., Anda, R. F., & Croft, J. B. (2009).
Cumulative childhood stress and autoimmune diseases in adults. Psychosomatic Medicine,
71(2), 243.
Duckworth, M. P., & Iezzi, T. (2010). Physical Injuries, Pain, and Psychological Trauma:
Pathways to Disability. Psychological Injury and Law, 3(3), 241–253.
https://doi.org/10.1007/s12207-010-9086-8
Dunlop, W. L., & Tracy, J. L. (2013). Sobering stories: Narratives of self-redemption predict
behavioral change and improved health among recovering alcoholics. Journal of
Personality and Social Psychology, 104(3), 576.
Dworkin, R. H., Turk, D. C., Wyrwich, K. W., Beaton, D., Cleeland, C. S., Farrar, J. T.,
Haythornthwaite, J. A., Jensen, M. P., Kerns, R. D., & Ader, D. N. (2008). Interpreting the
clinical importance of treatment outcomes in chronic pain clinical trials: IMMPACT
recommendations. The Journal of Pain, 9(2), 105–121.
Eather, A., Kenardy, J., Healy, K. L., Haynes, M., & Sterling, M. (2019). How are pain and
traumatic stress symptoms related in acute whiplash--associated disorders? An investigation
of the role of pain-related fear in a daily diary study. Pain, 160(9), 1954–1966.
Egli, M., Koob, G. F., & Edwards, S. (2012). Alcohol dependence as a chronic pain disorder.
Neuroscience & Biobehavioral Reviews, 36(10), 2179–2192.
https://doi.org/https://doi.org/10.1016/j.neubiorev.2012.07.010
Engel, G. L. (1981). The Clinical Application of the Biopsychosocial Model. The Journal of
Medicine and Philosophy: A Forum for Bioethics and Philosophy of Medicine, 6(2), 101–
124. https://doi.org/10.1093/jmp/6.2.101
Etikan, I., Musa, S. A., & Alkassim, R. S. (2016). Comparison of convenience sampling and
purposive sampling. American Journal of Theoretical and Applied Statistics, 5(1), 1–4.
Falkenström, F., Finkel, S., Sandell, R., Rubel, J. A., & Holmqvist, R. (2017). Dynamic models
of individual change in psychotherapy process research. Journal of Consulting and Clinical
Psychology, 85(6), 537.
Felitti, V. J. (2009). Adverse childhood experiences and adult health. Academic Pediatrics, 9(3),
131–132. https://doi.org/10.1016/j.acap.2009.03.001
Fulton, J. J., Calhoun, P. S., Wagner, H. R., Schry, A. R., Hair, L. P., Feeling, N., Elbogen, E., &
Beckham, J. C. (2015). The prevalence of posttraumatic stress disorder in Operation
Enduring Freedom/Operation Iraqi Freedom (OEF/OIF) Veterans: A meta-analysis. In

136
Journal of Anxiety Disorders (Vol. 31, pp. 98–107). Elsevier Ltd.
https://doi.org/10.1016/j.janxdis.2015.02.003
Garland, E. L., Thomas, E., & Howard, M. O. (2014). Mindfulness-oriented recovery
enhancement ameliorates the impact of pain on self-reported psychological and physical
function among opioid-using chronic pain patients. Journal of Pain and Symptom
Management, 48(6), 1091–1099. https://doi.org/10.1016/j.jpainsymman.2014.03.006
Gatchel, R. J., Peng, Y. B., Peters, M. L., Fuchs, P. N., & Turk, D. C. (2007). The
biopsychosocial approach to chronic pain: scientific advances and future directions.
Psychological Bulletin, 133(4), 581.
Geier, T. J., Hunt, J. C., Hanson, J. L., Heyrman, K., Larsen, S. E., Brasel, K. J., & deRoon‐
Cassini, T. A. (2020). Validation of Abbreviated Four‐and Eight‐Item Versions of the PTSD
Checklist for DSM‐5 in a Traumatically Injured Sample. Journal of Traumatic Stress, 33(3),
218–226.
Gilmore, A. K., Brignone, E., Painter, J. M., Lehavot, K., Fargo, J., Suo, Y., Simpson, T., Carter,
M. E., Blais, R. K., & Gundlapalli, A. V. (2016). Military sexual trauma and co-occurring
posttraumatic stress disorder, depressive disorders, and substance use disorders among
returning Afghanistan and Iraq veterans. Women’s Health Issues, 26(5), 546–554.
Giordano, N. A., Bader, C., Richmond, T. S., & Polomano, R. C. (2018). Complexity of the
relationships of pain, posttraumatic stress, and depression in combat‐injured populations:
An integrative review to inform evidence‐based practice. Worldviews on Evidence‐Based
Nursing, 15(2), 113–126.
Goebel, J. R., Compton, P., Zubkoff, L., Lanto, A., Asch, S. M., Sherbourne, C. D., Shugarman,
L., & Lorenz, K. A. (2011). Prescription sharing, alcohol use, and street drug use to manage
pain among veterans. Journal of Pain and Symptom Management, 41(5), 848–858.
https://doi.org/10.1016/j.jpainsymman.2010.07.009
Goldstein, P., Ashar, Y., Tesarz, J., Kazgan, M., Cetin, B., & Wager, T. D. (2020). Emerging
clinical technology: application of machine learning to chronic pain assessments based on
emotional body maps. Neurotherapeutics, 17(3), 774–783.
Griffin, B. J., Purcell, N., Burkman, K., Litz, B. T., Bryan, C. J., Schmitz, M., Villierme, C.,
Walsh, J., & Maguen, S. (2019). Moral Injury: An Integrative Review. Journal of Traumatic
Stress, 32(3), 350–362. https://doi.org/10.1002/JTS.22362
Gunthert, K. C., & Wenze, S. J. (2012). Daily diary methods. In Handbook of research methods
for studying daily life. (pp. 144–159). The Guilford Press.
Hall, S., Flower, M., Rein, L., & Franco, Z. (2020). Alcohol Use and Peer Mentorship in
Veterans. Journal of Humanistic Psychology, 63(6), 893–913.
https://doi.org/10.1177/0022167820905952

137
Haun, J. N., Fowler, C. A., Venkatachalam, H. H., Saenger, M. S., Alman, A. C., Smith, B. M.,
Schneider, T., Benzinger, R., Stroupe, K., & French, D. D. (2023). Empower Veterans
Program (EVP): a chronic pain management program demonstrates positive outcomes
among veterans. BMC Health Services Research, 23(1), 431.
https://doi.org/10.1186/s12913-023-09327-5
Hawn, S. E., Cusack, S. E., & Amstadter, A. B. (2020). A Systematic Review of the SelfMedication Hypothesis in the Context of Posttraumatic Stress Disorder and Comorbid
Problematic Alcohol Use. Journal of Traumatic Stress, 33(5), 699–708.
https://doi.org/https://doi.org/10.1002/jts.22521
Hastie, T., Tibshirani, R.,, Friedman, J. (2001). The Elements of Statistical Learning. New York,
NY, USA: Springer New York Inc.
Herrera, C. J., Owens, G. P., & Mallinckrodt, B. (2013). Traditional machismo and caballerismo
as correlates of posttraumatic stress disorder, psychological distress, and relationship
satisfaction in Hispanic veterans. Journal of Multicultural Counseling and Development,
41(1), 21–35.
Hoerster, K. D., Lehavot, K., Simpson, T., McFall, M., Reiber, G., & Nelson, K. M. (2012).
Health and Health Behavior Differences: U.S. Military, Veteran, and Civilian Men.
American Journal of Preventive Medicine, 43(5), 483–489.
https://doi.org/https://doi.org/10.1016/j.amepre.2012.07.029
Holtzman, S., DeClerck, D., Turcotte, K., Lisi, D., & Woodworth, M. (2017). Emotional support
during times of stress: Can text messaging compete with in-person interactions? Computers
in Human Behavior, 71, 130–139. https://doi.org/https://doi.org/10.1016/j.chb.2017.01.043
Hooten, W. M. (2016). Chronic Pain and Mental Health Disorders: Shared Neural Mechanisms,
Epidemiology, and Treatment. Mayo Clinic Proceedings, 91(7), 955–970.
https://doi.org/https://doi.org/10.1016/j.mayocp.2016.04.029
Huang, M., Bliwise, D. L., Hall, M. H., Johnson, D. A., Sloan, R. P., Shah, A., Goldberg, J., Ko,
Y.-A., Murrah, N., Levantsevych, O. M., Shallenberger, L., Abdulbagki, R., Bremner, J. D.,
& Vaccarino, V. (2022). Association of Depressive Symptoms with Sleep Disturbance: A
Co-twin Control Study. Annals of Behavioral Medicine, 56(3), 245–256.
https://doi.org/10.1093/abm/kaab040
Ishwaran, H., Kogalur, U. B., Blackstone, E. H., & Lauer, M. S. (2008). Random survival forests.
The Annals of Applied Statistics, 2(3), 841–860.
Ivey, K., Allen, R. S., Liu, Y., Parmelee, P. A., & Zarit, S. H. (2018). Immediate and Lagged
Effects of Daily Stress and Affect on Caregivers’ Daily Pain Experience. The Gerontologist,
58(5), 913–922. https://doi.org/10.1093/geront/gnx099
Jakob, J. M. D., Lamp, K., Rauch, S. A. M., Smith, E. R., & Buchholz, K. R. (2017). The impact
of trauma type or number of traumatic events on PTSD diagnosis and symptom severity in
treatment seeking veterans. The Journal of Nervous and Mental Disease, 205(2), 83–86.

138
Jensen, M. P., Tomé-Pires, C., de la Vega, R., Galán, S., Solé, E., & Miró, J. (2017). What
Determines Whether a Pain is Rated as Mild, Moderate, or Severe? The Importance of Pain
Beliefs and Pain Interference. The Clinical Journal of Pain, 33(5).
https://journals.lww.com/clinicalpain/fulltext/2017/05000/what_determines_whether_a_pai
n_is_rated_as_mild,.4.aspx
Jensen, M. P., & Turk, D. C. (2014). Contributions of psychology to the understanding and
treatment of people with chronic pain: why it matters to ALL psychologists. American
Psychologist, 69(2), 105.
Juwara, L., Arora, N., Gornitsky, M., Saha-Chaudhuri, P., & Velly, A. M. (2020). Identifying
predictive factors for neuropathic pain after breast cancer surgery using machine learning.
International Journal of Medical Informatics, 141, 104170.
Kamper, S. J., Michaleff, Z. A., Campbell, P., Dunn, K. M., Yamato, T. P., Hodder, R. K.,
Wiggers, J., & Williams, C. M. (2019). Back pain, mental health and substance use are
associated in adolescents. Journal of Public Health, 41(3), 487–493.
https://doi.org/10.1093/pubmed/fdy129
Kattan, M. W. (2003). Comparison of Cox regression with other methods for determining
prediction models and nomograms. The Journal of Urology, 170(6S), S6–S10.
Kerns, R. D., & Brandt, C. A. (2020). NIH-DOD-VA Pain Management Collaboratory:
Pragmatic Clinical Trials of Nonpharmacological Approaches for Management of Pain and
Co-occurring Conditions in Veteran and Military Health Systems: Introduction. Pain
Medicine, 21(Supplement_2), S1–S4. https://doi.org/10.1093/PM/PNAA358
Khantzian, E. J. (1997). The self-medication hypothesis of substance use disorders: A
reconsideration and recent applications. Harvard Review of Psychiatry, 4(5), 231–244.
https://doi.org/10.3109/10673229709030550
Khazaie, H., Ghadami, M. R., & Masoudi, M. (2016). Sleep disturbances in veterans with
chronic war-induced PTSD. Journal of Injury and Violence Research, 8(2), 99.
Lacoursiere, R. B., Godfrey, K. E., & Ruby, L. M. (1980). Traumatic neurosis in the etiology of
alcoholism: Viet Nam combat and other trauma. The American Journal of Psychiatry.
Lee, S. Y., Finkelstein-Fox, L., Park, C. L., Mazure, C. M., Huedo-Medina, T. B., & Hoff, R.
(2019). Bidirectionality of pain interference and PTSD symptoms in military veterans: Does
injury status moderate effects? Pain Medicine (United States), 20(5), 934–943.
https://doi.org/10.1093/pm/pny133
Lew, H. L., Otis, J. D., Tun, C., Kerns, R. D., Clark, M. E., & Cifu, D. X. (2009). Prevalence of
chronic pain, posttraumatic stress disorder, and persistent postconcussive symptoms in
OIF/OEF veterans: polytrauma clinical triad. Journal of Rehabilitation Research &
Development, 46(6).

139
Loetsch, J., Alfredsson, L., & Lampa, J. (2020). Machine-learning–based knowledge discovery
in rheumatoid arthritis–related registry data to identify predictors of persistent pain. Pain,
161(1), 114–126.
Loflin, M., Babson, K., Browne, K., & Bonn-Miller, M. (2018). Assessment of the validity of the
CUDIT-R in a subpopulation of cannabis users. The American Journal of Drug and Alcohol
Abuse, 44(1), 19–23.
Lötsch, J., & Ultsch, A. (2018). Machine learning in pain research. Pain, 159(4), 623.
Lumley, M. A., & Schubiner, H. (2019). Psychological Therapy for Centralized Pain: An
Integrative Assessment and Treatment Model. Psychosomatic Medicine, 81(2), 114–124.
https://doi.org/10.1097/PSY.0000000000000654
Macia, K. S., Moschetto, J. M., Wickham, R. E., Brown, L. M., & Waelde, L. C. (2020).
Cumulative Trauma Exposure and Chronic Homelessness Among Veterans: The Roles of
Responses to Intrusions and Emotion Regulation. Journal of Traumatic Stress, 33(6), 1017–
1028. https://doi.org/10.1002/JTS.22569
Mahdavi, S. B., Riahi, R., Vahdatpour, B., & Kelishadi, R. (2021). Association between
sedentary behavior and low back pain; A systematic review and meta-analysis. Health
Promotion Perspectives, 11(4), 393.
Mannes, Z. L., Malte, C. A., Olfson, M., Wall, M. M., Keyes, K. M., Martins, S. S., Cerdá, M.,
Gradus, J. L., Saxon, A. J., & Keyhani, S. (2022). Increasing risk of cannabis use disorder
among US veterans with chronic pain: 2005-2019. Pain, 10–1097.
María-Ríos, C. E., & Morrow, J. D. (2020). Mechanisms of Shared Vulnerability to Posttraumatic Stress Disorder and Substance Use Disorders . In Frontiers in Behavioral
Neuroscience (Vol. 14). https://www.frontiersin.org/articles/10.3389/fnbeh.2020.00006
Matsangidou, M., Liampas, A., Pittara, M., Pattichi, C. S., & Zis, P. (2021). Machine Learning in
Pain Medicine: An Up-To-Date Systematic Review. Pain and Therapy, 10(2), 1067–1084.
https://doi.org/10.1007/s40122-021-00324-2
Matthias, M. S., Kukla, M., McGuire, A. B., & Bair, M. J. (2016). How Do Patients with Chronic
Pain Benefit from a Peer-Supported Pain Self-Management Intervention? A Qualitative
Investigation. Pain Medicine, 17(12), 2247–2255. https://doi.org/10.1093/pm/pnw138
McAndrew, L. M., Lu, S.-E., Phillips, L. A., Maestro, K., & Quigley, K. S. (2019). Mutual
maintenance of PTSD and physical symptoms for Veterans returning from deployment.
European Journal of Psychotraumatology, 10(1), 1608717.
McClendon, J., Essien, U. R., Youk, A., Ibrahim, S. A., Vina, E., Kwoh, C. K., & Hausmann, L.
R. M. (2021). Cumulative Disadvantage and Disparities in Depression and Pain Among
Veterans With Osteoarthritis: The Role of Perceived Discrimination. Arthritis Care &
Research, 73(1), 11–17. https://doi.org/https://doi.org/10.1002/acr.24481

140
McCoy, T. P., & Dunlop, W. L. (2017). Down on the upside: Redemption, contamination, and
agency in the lives of adult children of alcoholics. Memory, 25(5), 586–594.
McDermott, R. C., Currier, J. M., Naylor, P. D., & Kuhlman, S. T. W. (2017). Student veterans’
self-stigma of seeking help: Contributions of painful self-conscious emotions, traditional
masculine norms, and war-zone service. Psychology of Men & Masculinity, 18(3), 226.
McElroy, A. (1990). Biocultural Models in Studies of Human Health and Adaptation. Medical
Anthropology Quarterly, 4(3), 243–265.
https://doi.org/https://doi.org/10.1525/maq.1990.4.3.02a00010
McNeish, D., & Hamaker, E. L. (2020). A primer on two-level dynamic structural equation
models for intensive longitudinal data in Mplus. In Psychological Methods (Vol. 25, Issue 5,
pp. 610–635). American Psychological Association. https://doi.org/10.1037/met0000250
Meadows, S. O., Beckman, R., Engel, C. C., & Jeffery, D. D. (2022). The Culture of Alcohol in
the U.S. Military: Correlations With Problematic Drinking Behaviors and Negative
Consequences of Alcohol Use. Armed Forces & Society, 49(2), 531–555.
https://doi.org/10.1177/0095327X211069162
Melchior, M., Poisbeau, P., Gaumond, I., & Marchand, S. (2016). Insights into the mechanisms
and the emergence of sex-differences in pain. Neuroscience, 338, 63–80.
https://doi.org/https://doi.org/10.1016/j.neuroscience.2016.05.007
Messman, B. A., Rafiuddin, H. S., Slavish, D. C., Weiss, N. H., & Contractor, A. A. (2022).
Examination of daily-level associations between posttraumatic stress disorder symptoms
and COVID-19 worries. Psychological Trauma: Theory, Research, Practice, and Policy,
14(3), 497–506. https://doi.org/10.1037/tra0001170
Metrik, J., Bassett, S. S., Aston, E. R., Jackson, K. M., & Borsari, B. (2018). Medicinal versus
recreational cannabis use among returning veterans. Translational Issues in Psychological
Science, 4(1), 6.
Metrik, J., Stevens, A. K., Gunn, R. L., Borsari, B., & Jackson, K. M. (2020). Cannabis use and
posttraumatic stress disorder: prospective evidence from a longitudinal study of veterans.
Psychological Medicine, 1–11.
Metrik, J., Stevens, A. K., Gunn, R. L., Borsari, B., & Jackson, K. M. (2022). Cannabis use and
posttraumatic stress disorder: prospective evidence from a longitudinal study of veterans.
Psychological Medicine, 52(3), 446–456. https://doi.org/DOI:
10.1017/S003329172000197X
Mitchell, M. M., Gallaway, M. S., Millikan, A., & Bell, M. R. (2011). Combat stressors
predicting perceived stress among previously deployed soldiers. Military Psychology, 23(6),
573–586.

141
Mun, C. J., Suk, H. W., Davis, M. C., Karoly, P., Finan, P., Tennen, H., & Jensen, M. P. (2019).
Investigating intraindividual pain variability: methods, applications, issues, and directions.
PAIN, 160(11).
Muthén, L. K., & Muthén, B. O. (2017). Mplus User’s Guide. Eighth Edition. Los Angeles, CA:
Muthén & Muthén. https://doi.org/10.1111/j.1600-0447.2011.01711.x
Nahin, R. L. (2017). Severe Pain in Veterans: The Effect of Age and Sex, and Comparisons With
the General Population. Journal of Pain, 18(3), 247–254.
https://doi.org/10.1016/j.jpain.2016.10.021
Nampiaparampil, D. E. (2008). Prevalence of chronic pain after traumatic brain injury: a
systematic review. Jama, 300(6), 711–719.
Naylor, J. C., Ryan Wagner, H., Brancu, M., Shepherd-Banigan, M., Elbogen, E., Kelley, M.,
Fecteau, T., Goldstein, K., Kimbrel, N. A., Marx, C. E., Group, V. A. M.-A. M. W., Group,
V. A. M.-A. M. W. V. W., & Strauss, J. L. (2017). Self-Reported Pain in Male and Female
Iraq/Afghanistan-Era Veterans: Associations with Psychiatric Symptoms and Functioning.
Pain Medicine, 18(9), 1658–1667. https://doi.org/10.1093/pm/pnw308
Naylor, J. C., Wagner, H. R., Johnston, C., Elbogen, E. E., Brancu, M., Marx, C. E., Beckham, J.
C., Calhoun, P. S., Dedert, E., & Fairbank, J. A. (2019). Pain intensity and pain interference
in male and female Iraq/Afghanistan-era veterans. Women’s Health Issues, 29, S24–S31.
Neilson, E. C., Singh, R. S., Harper, K. L., & Teng, E. J. (2020). Traditional masculinity
ideology, posttraumatic stress disorder (PTSD) symptom severity, and treatment in service
members and veterans: A systematic review. Psychology of Men & Masculinities, 21(4),
578.
Nelson, S. (2002). Physical symptoms in sexually abused women: somatization or undetected
injury? Child Abuse Review: Journal of the British Association for the Study and Prevention
of Child Abuse and Neglect, 11(1), 51–64. https://doi.org/10.1002/car.721
Nevarez, M. D., Yee, H. M., & Waldinger, R. J. (2017). Friendship in War: Camaraderie and
Prevention of Posttraumatic Stress Disorder Prevention. Journal of Traumatic Stress, 30(5),
512–520. https://doi.org/https://doi.org/10.1002/jts.22224
Noone, C., McSharry, J., Smalle, M., Burns, A., Dwan, K., Devane, D., & Morrissey, E. C.
(2020). Video calls for reducing social isolation and loneliness in older people: a rapid
review. Cochrane Database of Systematic Reviews, 5.
https://doi.org/10.1002/14651858.CD013632
Norman, S. B., Haller, M., Hamblen, J. L., Southwick, S. M., & Pietrzak, R. H. (2018). The
burden of co-occurring alcohol use disorder and PTSD in U.S. military veterans:
Comorbidities, functioning, and suicidality. Psychology of Addictive Behaviors, 32(2), 224–
229. https://doi.org/10.1037/adb0000348

142
Nugent, S. M., Morasco, B. J., O’Neil, M. E., Freeman, M., Low, A., Kondo, K., Elven, C.,
Zakher, B., Motu’apuaka, M., Paynter, R., & others. (2017). The effects of cannabis among
adults with chronic pain and an overview of general harms: a systematic review. Annals of
Internal Medicine, 167(5), 319–331.
Ocañez, K. L. S., Kathryn McHugh, R., & Otto, M. W. (2010). A meta‐analytic review of the
association between anxiety sensitivity and pain. Depression and Anxiety, 27(8), 760–767.
Oliver, L. W., Harman, J., Hoover, E., Hayes, S. M., & Pandhi, N. A. (1999). A quantitative
integration of the military cohesion literature. Military Psychology, 11(1), 57–83.
O’Loughlin, J. I., Cox, D. W., Castro, C. A., & Ogrodniczuk, J. S. (2022). Disentangling the
Individual and Group Effects of Masculinity Ideology on PTSD Treatment. Counselling
Psychology Quarterly, 35(3), 587–604. https://doi.org/10.1080/09515070.2021.1922359
O’Loughlin, J. I., Cox, D. W., Ogrodniczuk, J. S., & Castro, C. A. (2023). Traditional
masculinity ideology and psychotherapy treatment outcome for military service veteran
men. Psychology of Men & Masculinities, 24(1), 16.
Olsen, M. F., Bjerre, E., Hansen, M. D., Hilden, J., Landler, N. E., Tendal, B., & Hróbjartsson, A.
(2017). Pain relief that matters to patients: systematic review of empirical studies assessing
the minimum clinically important difference in acute pain. BMC Medicine, 15(1), 1–18.
Otis, J. D., Gregor, K., Hardway, C., Morrison, J., Scioli, E., & Sanderson, K. (2010). An
examination of the co-morbidity between chronic pain and posttraumatic stress disorder on
U.S. Veterans. Psychological Services, 7(3), 126–135. https://doi.org/10.1037/a0020512
Otis, J. D., Keane, T. M., Kerns, R. D., Monson, C., & Scioli, E. (2009). The development of an
integrated treatment for veterans with comorbid chronic pain and posttraumatic stress
disorder. Pain Medicine, 10(7), 1300–1311. https://doi.org/10.1111/j.1526-
4637.2009.00715.x
Pedersen, E. R., Naranjo, D., & Marshall, G. N. (2017). Recruitment and retention of young
adult veteran drinkers using Facebook. PLoS One, 12(3), e0172972.
Quartana, P. J., Campbell, C. M., & Edwards, R. R. (2009). Pain catastrophizing: a critical
review. Expert Review of Neurotherapeutics, 9(5), 745–758.
Qureshi, A. R., Patel, M., Neumark, S., Wang, L., Couban, R. J., Sadeghirad, B., Bengizi, A., &
Busse, J. W. (2023). Prevalence of chronic non-cancer pain among military veterans: a
systematic review and meta-analysis of observational studies. BMJ Military Health,
e002554. https://doi.org/10.1136/military-2023-002554
Raines, A. M., Primeaux, S. J., Ennis, C. R., Walton, J. L., Slaton, K. D., Vigil, J. O., Allan, N. P.,
Paulus, D. J., Zvolensky, M. J., Schmidt, N. B., & Franklin, C. L. (2022). Posttraumatic
Stress Disorder and Pain in Veterans: Indirect Association Through Anxiety Sensitivity.
Cognitive Therapy and Research, 46(1), 209–216. https://doi.org/10.1007/s10608-021-
10230-8

143
Raines, A. M., Primeaux, S. J., Ennis, C. R., Walton, J. L., Slaton, K. D., Vigil, J. O., Allan, N. P.,
Paulus, D. J., Zvolensky, M. J., Schmidt, N. B., & others. (2021). Posttraumatic Stress
Disorder and Pain in Veterans: Indirect Association Through Anxiety Sensitivity. Cognitive
Therapy and Research, 1–8.
Raines, A. M., Walton, J. L., McManus, E. S., Cuccurullo, L.-A. J., Chambliss, J., Uddo, M., &
Franklin, C. L. (2017). Associations between lower order anxiety sensitivity dimensions and
DSM-5 posttraumatic stress disorder symptoms. Cognitive Behaviour Therapy, 46(2), 162–
173.
Raja, S. N., Carr, D. B., Cohen, M., Finnerup, N. B., Flor, H., Gibson, S., Keefe, F., Mogil, J. S.,
Ringkamp, M., & Sluka, K. A. (2020). The revised IASP definition of pain: concepts,
challenges, and compromises. Pain, 161(9), 1976.
Ravn, S. L., Hartvigsen, J., Hansen, M., Sterling, M., & Andersen, T. E. (2018). Do posttraumatic pain and post-traumatic stress symptomatology mutually maintain each other? A
systematic review of cross-lagged studies. PAIN, 159(11).
Rigatti, S. J. (2017). Random Forest. Journal of Insurance Medicine, 47(1), 31–39.
https://doi.org/10.17849/insm-47-01-31-39.1
Romero-Sandoval, E. A., Fincham, J. E., Kolano, A. L., Sharpe, B. N., & Alvarado-Vázquez, P.
A. (2018). Cannabis for Chronic Pain: Challenges and Considerations. Pharmacotherapy:
The Journal of Human Pharmacology and Drug Therapy, 38(6), 651–662.
https://doi.org/https://doi.org/10.1002/phar.2115
Saad, M., De Medeiros, R., & Mosini, A. C. (2017). Are We Ready for a True Biopsychosocial–
Spiritual Model? The Many Meanings of “Spiritual.” In Medicines (Vol. 4, Issue 4).
https://doi.org/10.3390/medicines4040079
Saba, S. K., Davis, J. P., Bricker, J. B., Christie, N. C., & Pedersen, E. R. (2023). Pain
Trajectories among U.S. Veterans During COVID-19. The Journal of Pain, 24(11), 2093–
2102. https://doi.org/https://doi.org/10.1016/j.jpain.2023.06.018
Saba, S. K., Davis, J. P., Prindle, J. J., Castro, C. A., & Pedersen, E. R. (2021). Associations
between symptoms of posttraumatic stress disorder, pain, and alcohol use disorder among
OEF/OIF/OND veterans. Addictive Behaviors, 122, 107031.
https://doi.org/https://doi.org/10.1016/j.addbeh.2021.107031
Saconi, B., Polomano, R. C., Compton, P. C., McPhillips, M. v, Kuna, S. T., & Sawyer, A. M.
(2021). The influence of sleep disturbances and sleep disorders on pain outcomes among
veterans: A systematic scoping review. Sleep Medicine Reviews, 56, 101411.
Salamon, K. S., Davies, W. H., Fuentes, M. R., Weisman, S. J., & Hainsworth, K. R. (2014). The
pain frequency-severity-duration scale as a measure of pain: preliminary validation in a
pediatric chronic pain sample. Pain Research and Treatment, 2014.

144
Santorelli, S. F., Kabat-Zinn, J., Blacker, M., Meleo-Meyer, F., & Koerbel, L. (2017).
Mindfulness-based stress reduction (MBSR) authorized curriculum guide. Center for
Mindfulness in Medicine, Health Care, and Society (CFM). University of Massachusetts
Medical School.
Schuler, K., Ruggero, C. J., Mahaffey, B., Gonzalez, A., L. Callahan, J., Boals, A., Waszczuk, M.
A., Luft, B. J., & Kotov, R. (2021). When hindsight is not 20/20: ecological momentary
assessment of PTSD symptoms versus retrospective report. Assessment, 28(1), 238–247.
Sharp, T. J., & Harvey, A. G. (2001). Chronic pain and posttraumatic stress disorder: Mutual
maintenance? In Clinical Psychology Review (Vol. 21, Issue 6, pp. 857–877).
https://doi.org/10.1016/S0272-7358(00)00071-4
Shiffman, S., Stone, A. A., & Hufford, M. R. (2008). Ecological Momentary Assessment. Annual
Review of Clinical Psychology, 4(1), 1–32.
https://doi.org/10.1146/annurev.clinpsy.3.022806.091415
Shipherd, J. C., Keyes, M., Jovanovic, T., Ready, D. J., Baltzell, D., Worley, V., Gordon-Brown,
V., Hayslett, C., & Duncan, E. (2007). Veterans seeking treatment for posttraumatic stress
disorder: What about comorbid chronic pain? Journal of Rehabilitation Research &
Development, 44(2), 153–166. https://doi.org/10.1682/JRRD.2006.06.0065
Short, N. A., Allan, N. P., & Schmidt, N. B. (2017). Sleep disturbance as a predictor of affective
functioning and symptom severity among individuals with PTSD: An ecological momentary
assessment study. Behaviour Research and Therapy, 97, 146–153.
https://doi.org/https://doi.org/10.1016/j.brat.2017.07.014
Smith, S. M., Goldstein, R. B., & Grant, B. F. (2016). The association between post-traumatic
stress disorder and lifetime DSM-5 psychiatric disorders among veterans: Data from the
National Epidemiologic Survey on Alcohol and Related Conditions-III (NESARC-III).
Journal of Psychiatric Research, 82, 16–22.
Stecker, T., Fortney, J. R., Owen, M. P., Mcgovern, S., & Williams, B. S. (2010). Co-Occurring
Medical, Psychiatric, and Alcohol-Related Disorders Among Veterans Returning From Iraq
and Afghanistan. Psychosomatics, 51, 503–507. https://doi.org/10.1016/S0033-
3182(10)70743-X
Stone, A. A., Broderick, J. E., Junghaenel, D. U., Schneider, S., & Schwartz, J. E. (2016).
PROMIS fatigue, pain intensity, pain interference, pain behavior, physical function,
depression, anxiety, and anger scales demonstrate ecological validity. Journal of Clinical
Epidemiology, 74, 194–206.
Stratton, K. J., Clark, S. L., Hawn, S. E., Amstadter, A. B., Cifu, D. X., & Walker, W. C. (2014).
Longitudinal interactions of pain and posttraumatic stress disorder symptoms in US Military
service members following blast exposure. The Journal of Pain, 15(10), 1023–1032.
Strauss, A., & Corbin, J. (1994). Grounded theory methodology: An overview. In Handbook of
qualitative research. (pp. 273–285). Sage Publications, Inc.

145
Suzuki, H., Aono, S., Inoue, S., Imajo, Y., Nishida, N., Funaba, M., Harada, H., Mori, A.,
Matsumoto, M., Higuchi, F., Nakagawa, S., Tahara, S., Ikeda, S., Izumi, H., Taguchi, T.,
Ushida, T., & Sakai, T. (2020). Clinically significant changes in pain along the Pain
Intensity Numerical Rating Scale in patients with chronic low back pain. PLOS ONE, 15(3),
e0229228. https://doi.org/10.1371/journal.pone.0229228
Tan, G., Dao, T. K., Farmer, L., Sutherland, R. J., & Gevirtz, R. (2011). Heart rate variability
(HRV) and posttraumatic stress disorder (PTSD): a pilot study. Applied Psychophysiology
and Biofeedback, 36(1), 27–35.
Tan, G., Fink, B., Dao, T. K., Hebert, R., Farmer, L. S., Sanders, A., Pastorek, N., & Gevirtz, R.
(2009). Associations among pain, PTSD, mTBI, and heart rate variability in veterans of
Operation Enduring and Iraqi Freedom: a pilot study. Pain Medicine, 10(7), 1237–1245.
Taylor, B., Carswell, K., & Williams, A. C. de C. (2013). The Interaction of Persistent Pain and
Post-Traumatic Re-Experiencing: A Qualitative Study in Torture Survivors. Journal of Pain
and Symptom Management, 46(4), 546–555.
https://doi.org/https://doi.org/10.1016/j.jpainsymman.2012.10.281
Taylor, S. (2003). Anxiety sensitivity and its implications for understanding and treating PTSD.
Journal of Cognitive Psychotherapy, 17(2), 179–186.
Taylor, S., Zvolensky, M. J., Cox, B. J., Deacon, B., Heimberg, R. G., Ledley, D. R.,
Abramowitz, J. S., Holaway, R. M., Sandin, B., & Stewart, S. H. (2007). Robust dimensions
of anxiety sensitivity: development and initial validation of the Anxiety Sensitivity Index-3.
Psychological Assessment, 19(2), 176.
Thake, J., & Davis, C. G. (2011). Assessing problematic cannabis use. Addiction Research &
Theory, 19(5), 448–458.
Tiet, Q. Q., & Moos, R. H. (2021). Strong associations among PTSD, pain, and alcohol and drug
use disorders in VA primary care patients. Drug and Alcohol Dependence, 223, 108699.
https://doi.org/https://doi.org/10.1016/j.drugalcdep.2021.108699
Turna, J., & MacKillop, J. (2021). Cannabis use among military veterans: A great deal to gain or
lose? Clinical Psychology Review, 101958.
U.S. Department of Veteran’s Affairs. (2015). Study of Barriers for Women Veterans to VA Health
Care Final Report. https://www.womenshealth.va.gov/docs/Womens Health
Services_Barriers to Care Final Report_April2015.pdf
Wade, D. T., & Halligan, P. W. (2017). The biopsychosocial model of illness: a model whose
time has come. Clinical Rehabilitation, 31(8), 995–1004.
https://doi.org/10.1177/0269215517709890
Wang, J., & Wang, X. (2019). Structural Equation Modeling: Applications Using Mplus. John
Wiley & Sons, Incorporated. https://doi.org/10.1002/9781119422730

146
Wei, Y., Blanken, T. F., & Van Someren, E. J. W. (2018). Insomnia Really Hurts: Effect of a Bad
Night’s Sleep on Pain Increases With Insomnia Severity. Frontiers in Psychiatry, 9.
https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2018.00377
Whitworth, J. W., Scioli, E. R., Keane, T. M., & Marx, B. P. (2022). Physical inactivity, cigarette
smoking, and psychiatric comorbidity among veterans with posttraumatic stress disorder.
Health Psychology, 41(3), 169.
Williamson, A., & Hoggart, B. (2005). Pain: a review of three commonly used pain rating scales.
Journal of Clinical Nursing, 14(7), 798–804.
Williamson, V., Murphy, D., Stevelink, S. A. M., Allen, S., Jones, E., & Greenberg, N. (2020).
The impact of trauma exposure and moral injury on UK military veterans: a qualitative
study. European Journal of Psychotraumatology, 11(1), 1704554.
https://doi.org/10.1080/20008198.2019.1704554
Wisco, B. E., Marx, B. P., May, C. L., Martini, B., Krystal, J. H., Southwick, S. M., & Pietrzak,
R. H. (2017). Moral injury in U.S. combat veterans: Results from the national health and
resilience in veterans study. Depression and Anxiety, 34(4), 340–347.
https://doi.org/10.1002/DA.22614
Wolfgang, A. S., & Hoge, C. W. (2023). Cannabis and cannabinoids for pain and posttraumatic
stress disorder in military personnel and veterans. JAMA Psychiatry, 80(9), 869–870.
Zale, E. L., Maisto, S. A., & Ditre, J. W. (2015). Interrelations between pain and alcohol: An
integrative review. In Clinical Psychology Review (Vol. 37, pp. 57–71). Elsevier Inc.
https://doi.org/10.1016/j.cpr.2015.02.005
Zhang-James, Y., Wyon, E., Grapsas, D., & Johnson, B. (2023). Daily cannabis use may cause
cannabis-induced hyperalgesia. The American Journal on Addictions, 32(6), 532–538.
https://doi.org/https://doi.org/10.1111/ajad.13456
Zou, H., & Hastie, T. (2005). Regularization and variable selection via the elastic net. Journal of
the Royal Statistical Society: Series B (Statistical Methodology), 67(2), 301–320.
Zuckerman, M. K. (2018). Biocultural models. The International Encyclopedia of Biological
Anthropology, 1–8.
Zvolensky, M. J., Rogers, A. H., Manning, K., Hogan, J. B. D., Paulus, D. J., Buckner, J. D.,
Mayorga, N. A., Hallford, G., & Schmidt, N. B. (2018). Anxiety sensitivity and cannabis
use problems, perceived barriers for quitting, and fear of quitting. Psychiatry Research, 263,
115–120.

Abstract (if available)

Abstract Military veterans experience a high burden of multi-morbid physical and behavioral health problems, including physical pain, posttraumatic stress disorder, and problematic substance use. This three-paper dissertation seeks to extend knowledge of multi-morbid problems and their etiology among veterans. Paper 1 uses dynamic structural equation modeling with daily diary data to explore trivariate associations between pain, PTSD symptoms, and problematic cannabis use among a sample of veterans (n = 74); this paper seeks to test and extend existing theories of multi-morbidity. Given the breadth of additional biopsychosocial factors that are known to influence pain, paper 2 uses machine learning with the daily diary sample to clarify the predictors of pain – particularly of pain intensity increases and frequent pain – among veterans with multi-morbid PTSD. Finally, paper 3 engages veterans (n = 20) in qualitative and narrative life history interviews to understand their subjective experiences of multi-morbid problems and how military experiences and culture influence such problems.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Homelessness and substance use treatment: using multiple methods to understand risks, consequences, and unmet treatment needs among young adults

PDF

Tobacco use change among formerly homeless supportive housing residents: socioecological barriers and facilitators to cessation

PDF

Couch-surfing among youth experiencing homelessness: an examination of HIV risk

PDF

Exploring the social determinants of health in a population with similar access to healthcare: experiences from United States active-duty army wives

PDF

Understanding emotional regulation and mood of young adults in the context of homelessness using geographic ecological momentary assessment

PDF

Integrative care strategies for older adults experiencing co-occurring substance use and mental health disorders (I-CARE)…

PDF

Impact of change in sexual identity on mental health risks among sexual minority adolescents

PDF

An examination of sexual harassment, gender discrimination, stalking, and sexual assault among female and male veterans and associations with PTSD and depression

PDF

TREAT SAW: an integrated model of MDMA-assisted therapy and Save A Warrior’s evidence-based practices for holistic healing of complex post-traumatic stress (C-PTS) in California’s returning veterans

PDF

Homeless female veterans—silent epidemic

PDF

Data Synthesis as Catalyst for Equity

PDF

Chronically informed: hope for people with chronic illness

PDF

HUB_mv: an evidence-based non-police led MH/SUD diversion program in rural MA

PDF

Warrior tribe: veteran utilization of aquaponics to fight loneliness

PDF

Mindfulness and resilience: an investigation of the role of mindfulness in post-9/11 military veterans' mental health-related outcomes

PDF

A comprehensive analysis of police brutality and a conclusion based on empirical data regarding the impact of empathy on these occurrences

PDF

Complicated relationships: the prevalence and correlates of sexual assault, dating violence victimization, and minority stress among sexual minority adolescents throughout the United States

PDF

Substance use disorder treatment clinician and director challenges working with collaborative justice court and child welfare systems: A multisite qualitative study

PDF

Atender, live! Strategies to acknowledge and cope with cultural taboos in the Latinx community

PDF

Military and veteran

Asset Metadata

Creator Saba, Shaddy (author)

Core Title Pain and multi-morbidity among veterans: theory-guided, data-driven, and narrative approaches

School Suzanne Dworak-Peck School of Social Work

Degree Doctor of Philosophy

Degree Program Social Work

Degree Conferral Date 2024-08

Publication Date 06/19/2024

Defense Date 06/03/2024

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

Tag OAI-PMH Harvest,Pain,posttraumatic stress disorder,substance use,veterans

Format theses (aat)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Castro, Carl (committee chair), Davis, Jordan (committee member), Henwood, Benjamin (committee member), Pedersen, Eric (committee member), Prindle, John (committee member)

Creator Email ssaba@usc.edu,ssaba87@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC113996WQG

Unique identifier UC113996WQG

Identifier etd-SabaShaddy-13121.pdf (filename)

Legacy Identifier etd-SabaShaddy-13121

Document Type Dissertation

Format theses (aat)

Rights Saba, Shaddy

Internet Media Type application/pdf

Type texts

Source 20240619-usctheses-batch-1171 (batch), 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 author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

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

Repository Email cisadmin@lib.usc.edu