University of Southern California Dissertations and Theses

A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system

(USC Thesis Other)

A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

A Series of Longitudinal Analyses of Patient Reported
Outcomes to Further the Understanding of
Care-Management of Comorbid Diabetes and Depression in
a Safety-Net Healthcare System

BY

OLIVIA EVANSON

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the Requirements for the Degree
DOCTOR OF PHILOSOPHY
(INDUSTIRAL AND SYSTEMS ENGINEERING)
May 2020

ii

Acknowledgements

I would like to thank everyone who has mentored and supported me throughout my doctoral
research over the past four and a half years. I would especially like to thank my PhD advisor, Dr.
Shinyi Wu, who has been a great mentor and supporter of my research. Her guidance, patience,
and knowledge helped me throughout all my research and writing of my dissertation. I could not
imagine having a better advisor throughout this process. I would also like to thank my
dissertation committee members, Dr. Haomiao Jin, Dr. Sze-Chuan Suen, Dr. Chih Ping Chou,
and Dr. Julie Higle, for their insightful feedback. Finally, I would like to thank my parents and
my brother for their constant support and belief in me, and my friends for their encouragement
throughout this journey.

iii

Abstract
Introduction
Diabetes and depression are prevalent chronic, often comorbid, diseases in low-income,
primarily minority populations seeking care in resource-constrained safety-net care systems.
Newer care-management models offer an opportunity to increase access to and quality of care for
this population. However, there is a lack of empirical investigation of care experience measures
and mechanisms such as self-management for the newer care-management models to sustain
healthcare triple aim improvements, that is, improving patient experience, reducing costs, and
improving health in this underserved patient population.
Purpose
The purpose of this dissertation is to further our understanding of care-management of comorbid
diabetes and depression among patients in safety-net healthcare system via a series of
longitudinal analyses of patient reported outcomes (PRO). Specifically, this dissertation focuses
on two different patient reported outcomes. First, understanding differences in patient
satisfaction with care across different types of and time varying care-management models to use
this PRO as a measure of quality of care experience. Second, understanding self-management
behavior as a mechanism for sustained care-management effects via investigating dynamic
relationships between behaviors and symptoms, behavior trends, predictive patient factors of
behavior trend, and longitudinal associations between behavior trends and health outcomes.

iv

Methods
This study includes 4 research analyses. The data for all analyses came from the Diabetes and
Depression Care-Management Adoption Trial (DCAT), a multi-site, quasi-experimental,
comorbid depression care-management trial with three trial groups of low-income, primarily
Latino patients with type 2 diabetes from the Los Angeles County Department of Health Services
(LAC-DHS). The LAC-DHS is the second largest safety-net system in the US. The three groups
received usual primary care (UC), team-supported comorbid care management for diabetes and
depression (SC), and automated telephonic depression remote monitoring comorbid care-
management (TC), respectively. The two care management models were offered for a limited
time (6 to 12 months) while patients were followed up to 18 months. Complete cases were used
for each analysis.
Patient satisfaction is analyzed using multi-wave cross-sectional regression models (n=700).
Dynamic relationships between self-management behavior and disease symptoms are analyzed
using cross-lagged path analysis (n=783). Self-management behavior trends are analyzed using
data-driven longitudinal cluster analysis and prediction modeling is used to identify patient
factors associated with self-management behavior trend (n=847). Finally, longitudinal
associations between behavior trend and healthcare triple aim outcomes are analyzed using
longitudinal multilevel regression modeling (n=847).
Results
The first analysis found that patient satisfaction with depression care is significantly higher in the
comorbid depression care-management groups, and patient satisfaction is not influenced by
patient characteristics or disease symptoms. The second analysis found that patient self-
v

management alone is not enough to improve symptoms in this population, and that the care-
management program is helpful in reducing symptoms while in-place, but improvements are not
sustained after patients graduated or left the program. The third analysis found patients in DCAT
can be clustered into three groups of different self-management behavior trend: consistent,
moderate, and struggling in self-management over time. A care-management model (i.e., SC)
along with specific physical, mental, and social health variables are predictive of self-
management behavior trend. The fourth analysis found many baseline but not longitudinal
associations between self-management behavior groups and health and cost measures. The
results also indicate significant although mixed effects of care-management models for the triple
aim outcomes over time. There are significant interaction effects of TC and the struggling in self-
management group on improving mental and social health outcomes.
Conclusion
This dissertation analyzed PROs longitudinally to further the understanding of the newer care-
management models to help safety-net healthcare systems optimize performance of comorbid
diabetes and depression. The findings show patients were more satisfied with the care-
management models than usual care when they were available, which helped them improve
disease symptoms, but the positive experience with SC negatively affected their satisfaction with
usual care in the long-run. Although self-management did not improve disease symptoms at the
population level, the subgroup of TC patients who struggled with self-management improved
mental and social outcomes. This PRO-generated knowledge can help safety-net providers
improve their design and implementation of care-management models to realize the triple aims.

vi

Table of Contents
Acknowledgements………………………………………………………………………………..ii
Abstract…………………………………………………………………………………………...iii
Chapter 1 - Introduction…………………………………………………………………………...1
Chapter 2 - Paper 1: Comparison of Satisfaction with Comorbid Depression Care Models Among
Low-Income Patients with Diabetes………………………………………………………………7
2.1 Introduction……………………………………………………………………………9
2.2 Methods………………………………………………………………………………11
2.2.1 Study Design and Data Source…………………………………………….11
2.2.2 Measures…………………………………………………………………...12
2.2.3 Analysis…………………………………………………………………….13
2.3 Results………………………………………………………………………………..14
2.3.1 Sample Characteristics……………………………………………………..14
2.3.2 Regression Analysis Results……………………………………………….15
2.4 Discussion……………………………………………………………………...…….17
2.4.1 Principal Findings………………………………………………………….17
2.4.2 Comparison to Literature…………………………………………………..18
2.4.3 Limitations and Future Work………………………………………………19
2.5 Conclusion…………………………………………………………………………...20
Appendix for Chapter 2………………………………………………………………….21
Chapter 3 - Paper 2: Longitudinal Analysis of Relationship Between Self-Management
Activities, Care Management, and Symptoms of Diabetes and Depression: Results from A
Comparative Effectiveness Trial…………………………………………………………………23
3.1 Introduction…………………………………………………………………………..25
3.2 Methods………………………………………………………………………………27
3.2.1 Study Design……………………………………………………………….27
3.2.2 Data Source………………………………………………………………...27
3.2.3 Measures…………………………………………………………………...29
3.2.4 Analysis…………………………………………………………………….31

vii

3.3 Results………………………………………………………………………………..33
3.3.1 Model Fit Summary………………………………………………………..33
3.3.2 Research Question 1 Diabetes Self-Management Relationships Results….34
3.3.3 Research Question 2: Care-Management Model Results………………….34
3.3.4 Self-Rated Health Relationships Results…………………………………..35
3.4 Conclusions…………………………………………………………………………..40
Chapter 4 - Paper 3: Investigation of Diabetes Self-Management Behavior Pattern and Trend
Over Time and Predictive Factors Among Low-Income Latino Patients with Type 2
Diabetes………………………………………………………………………………………......45
4.1 Introduction…………………………………………………………………………..47
4.2 Methods………………………………………………………………………………49
4.2.1 Study Design……………………………………………………………….49
4.2.2 Data Source………………………………………………………………...49
4.2.3 Measures…………………………………………………………………...50
4.2.4 Analysis…………………………………………………………………….53
4.3 Results………………………………………………………………………………..54
4.4 Discussion……………………………………………………………………………59
4.4.1 Principal Findings………………………………………………………….59
4.4.2 Comparison to Literature…………………………………………………..61
4.4.3 Limitations and Future Work………………………………………………62
4.5 Conclusion…………………………………………………………………………...63
Chapter 5 - Paper 4: Investigation of Longitudinal Associations Between Self-Management
Behavior Cluster and Triple Aim Outcomes…………………………………………………….64
5.1 Introduction…………………………………………………………………………..66
5.2 Methods………………………………………………………………………………67
5.2.1 Study Design……………………………………………………………….67
5.2.2 Data Source………………………………………………………………...68
5.2.2 Measures…………………………………………………………………...69
5.2.3 Analysis…………………………………………………………………….71
5.3 Results………………………………………………………………………………..72

viii

5.4 Discussion……………………………………………………………………………82
5.4.1 Principal Findings………………………………………………………….82
5.4.2 Comparison to Literature…………………………………………………..83
5.4.3 Limitations and Future Work………………………………………………84
5.5 Conclusion…………………………………………………………………………..85
Appendix for Chapter 5………………………………………………………………….86
Chapter 6 - Conclusion…………………………………………………………………………..95
References………………………………………………………………………………………..99
Chapter 1 References………………………………………………………………….....99
Chapter 2 References…………………………………………………………………...101
Chapter 3 References…………………………………………………………………...104
Chapter 4 References…………………………………………………………………...108
Chapter 5 References…………………………………………………………………...111
Chapter 6 References…………………………………………………………………...113
Appendix - Sample Comparisons………………………………………………………………114

1

Chapter 1
Introduction
Understanding, improving, and validating complex systems are key functions of
industrial and systems engineering (ISE) across many diverse industries. Demand for these ISE
functions has grown exponentially in the healthcare industry. With higher patient demand,
providers are attempting to do more without increasing healthcare costs. One solution is new,
innovative care-management models, such as team-based care and remote symptom monitoring
[1,2]. Due to disparities in access to care, these care-management models are especially valuable
for population health in low-income, minority populations with chronic diseases [3]. ISE assists
in the design and evaluation of these innovative systems [4], and social work aims for social
justice [5] which encompasses working with low-income minority populations to help reduce the
gap in access to healthcare. Thus, this research utilizes engineering+, defined by the Viterbi
Dean’s Office as using engineering innovation to solve real problems that add value in other
industries [6]. Specifically, this research uses industrial and systems engineering and social work
by combining appropriate methods for the evaluation of care-management models to aid in
future provision of more individualized care to better meet patients’ needs in an underserved
population.
Patients are at the center of care provision. A primary goal of care-management models is
to properly allocate care resources to efficiently provide high quality care for addressing patient
health needs [7,8]. One way to ensure patient health care needs are met is through a better
understanding of how patients value care provided by these new models, and what mechanisms
of care are most impactful for maintaining and improving patient health. Patient reported
outcomes (PRO’s) and patient behaviors can help us understand the value of care in care-
2

management models to the patients targeted by these models. Additionally, PRO’s and behaviors
might be reported differently in low-income, minority populations than other populations. To
effectively evaluate care delivery through PRO’s for this population, it is important to understand
how PRO’s operate within this population. This understanding will be valuable for future system
design to better serve low-income, minority populations with chronic disease.
The literature states that longitudinal analyses are needed to determine if PRO
instruments are responsive to changes or differences in care and disease status [9]. A robust way
to assess the responsiveness is to use multiple methods of measurement, including: clinical
endpoints, patient-rated global improvement, change in other PRO measures, or assessing
combinations of clinical and patient-based outcomes [9]. The analyses in this dissertation utilize
all the above methods. Specifically, this dissertation presents four original research articles
which investigate patient reported outcomes, patient behaviors, and the dynamic relationships
between these measures to further our understanding of care-management models for a low-
income, minority population with comorbid diabetes and depression. The same longitudinal data
set was used in all four of the studies that are included in this dissertation. The data comes from
the Diabetes and Depression Care-Management Adoption Trial (DCAT) [10]. DCAT included
1406 patients from the Los Angeles County Department of Health Services. All patients had type
2 diabetes and some patients had comorbid depression. DCAT will be further explained in the
following chapters.
Chapter 2 presents the first original research paper. The paper includes an investigation of
differences in patient satisfaction with three different care-management models over time,
controlling for patient characteristics, group differences, and disease symptoms. The Institute for
Healthcare Improvement’s (IHI) Triple Aim initiative identifies three aims for healthcare
3

interventions: improved health outcomes, reduced costs, and improved patient experience.
Previous DCAT analyses have demonstrated improved health outcomes and lowered costs
[11,12] and have investigated patient experience with the technology of DCAT [13,14]. The
significance of this research was to understand whether patient satisfaction is sensitive to
elements of care-management, or is biased by personal characteristics, depression status, or
diabetes symptoms. This study will help understand the validity of using patient satisfaction to
assess the quality of care experience from newer care-management models, such as comorbid
disease management.
Figure 1. Conceptual Model of Investigation of Patient Satisfaction as a Measure of Care
Experience

Chapter 3 presents the second original research paper. The paper includes an
investigation of the dynamic, longitudinal relationships between care-management models,
patient self-management behaviors and disease symptoms, controlling for patient health status. A
mechanism for short-term care-management models to result in improved long-term health
outcomes for patients with comorbid diabetes and depression is through improved and sustained
patient self-management behaviors to lessen disease symptoms. This study investigates this
4

mechanism using cross-lagged path analysis, a longitudinal approach which estimates
relationships between variables across different time points.
Figure 2. Conceptual Model for Cross-Lagged Path Analysis of Dynamic Relationships between
Self-Management, Care-Management and Disease Symptoms

Chapter 4 presents the third original research paper. The results of the chapter 3 analysis
provided additional research questions about self-management behavior trends and variables that
are associated with self-management behavior trend. This was a two-part analysis. First, data
driven cluster analysis was used to identify different self-management behavior trends in our
study population. Second, prediction modeling was used to identify baseline variables which
predicted self-management behavior trend. The purpose of this analysis was to identify different
self-management behavior trend and identify baseline variables which predict self-management
behavior trend in our study population.
Figure 3. Conceptual Model for Analysis of Predictors of Self-Management Behavior Patterns

5

Chapter 5 presents the fourth and final original research paper of this dissertation. This
analysis was motivated by chapters 3 and 4 of this dissertation. The aim of this analysis was to
investigate longitudinal associations between self-management behavior trend and the IHI’s
triple aim outcomes: reduced healthcare costs, improved outcomes, and improved patient care
experience. This study investigated these associations using longitudinal multilevel regression
modeling.
Figure 4. Conceptual Model for Analysis of Longitudinal Relationships between Self-
Management Behavior Patterns and Triple Aim Outcomes

The four investigations included in this dissertation use patient experience and behavioral
data to further the understanding of comorbid diabetes and depression care-management among
patients in a safety net healthcare system. The conceptual model of the full dissertation,
combining all four analyses, is provided in figure 5. With the availability of newer care-
management models that utilize less face-to-face patient-provider interaction, and a high volume
of patients who require chronic disease care, it is important to: (1) to investigate non-traditional
measures of quality of care experience in newer care-models, and (2) evaluate mechanisms, such
as self-management behaviors, that patients can utilize to monitor and maintain their own health.
Understanding patient experience measures and self-management behavior patterns, associations,
and predictors, allow future care-management to deliver more precise and appropriate care for
6

individual patient needs. This dissertation’s focus on low-income Latino patients with type 2
diabetes is especially relevant to current population health management needs. This population
experiences healthcare access disparity yet has higher prevalence of chronic disease [15].
Latino’s have a higher prevalence of type 2 diabetes than non-Latino whites [15], and low socio-
economic status (SES) is associated with higher HbA1c levels, a measure of blood glucose levels
which is often indicative of diabetes severity [16]. Newer care-management models provide an
opportunity to increase access to care for this safety-net population. Thus, understanding the care
experience and self-management behaviors of chronic disease for patients using these models of
care could be especially beneficial for this population’s health in the implementation of future
care-management interventions. This research advances industrial and systems engineering by
presenting a series of longitudinal analyses of PRO’s to further the understanding of optimizing
comorbid diabetes and depression care-management among patients in a safety-net healthcare
system to achieve triple aim outcomes.
Figure 5. Dissertation Conceptual Model

7

Chapter 2
Comparison of Satisfaction with Comorbid Depression Care Models
Among Low-Income Patients with Diabetes
Abstract
Introduction: Patient satisfaction is a patient-reported outcome with the potential to assess and
improve the quality of newer care management models such as remote patient monitoring using
telecommunication technology.
Objective: To evaluate differences in patient satisfaction among three care-management groups
in a comparative effectiveness trial.
Methods: This study analyzed a comparative effectiveness trial that tested automated remote
assessment technology–facilitated comorbid depression care-management (TC, n=254) in
comparison to team-supported depression care (SC, n=228) and usual primary care (UC, n=218)
among low-income patients with type 2 diabetes. Relationships between patient satisfaction and
care group were evaluated at each 6-month phase up to 18-months using linear regression models
that controlled for depression status, diabetes symptoms, patient characteristics, and study group
differences.
Results: While receiving care-management, SC and TC patients were significantly more satisfied
with depression care than UC patients. No consistently significant associations between patient
satisfaction and patient characteristics or disease symptoms were found.
Conclusions: Patient satisfaction was found to be influenced by elements of care-management,
not by patient characteristics or disease symptoms. Results suggest greater patient satisfaction
with depression care in a care-management model than usual care, whether through clinician
team support or automated remote monitoring technology.
8

Keywords
Patient reported outcomes, patient satisfaction, depression, telecommunication technology, care-
management, type 2 diabetes, safety-net care, comparative effectiveness
Abbreviations
PCP: primary care provider, DMP: disease management program, DMR: disease management
registry, PST: problem-solving therapy, ATA: automated telephonic assessment, DCAT:
Diabetes and Depression Care-Management Adoption Trial.

9

Introduction
With increasing technological capabilities for remote symptom assessment, monitoring,
and patient-provider communication, ambulatory care settings have begun embracing newer care
models outside traditional face-to-face care to improve care management [1,2]. Many studies
have demonstrated the effectiveness of remote patient care-management using
telecommunication technology to increase access to care while decreasing healthcare costs [3–9].
Furthermore, measuring patient satisfaction provides the patients’ perspectives and experience of
healthcare quality in various care-management models [10–12], and analyzing patient
satisfaction with care-management provides an opportunity to understand how patient
satisfaction is influenced by alternative models of care [13,14]. Recently, studies have
successfully evaluated patient satisfaction, post intervention in pre-post studies, as an assessment
of the care received through remote care-management interventions [15,16]. However, we
currently need greater knowledge on how patient satisfaction is affected by variations in care-
management models and what this tells us about the quality of care experience provided.
The Diabetes-Depression Care Management Trial (DCAT) provides an opportunity to
compare two care-management models to usual care. DCAT was a multi-site, quasi-experimental
trial with three trial groups of low-income, primarily Latino patients with type 2 diabetes from
the Los Angeles County Department of Health Services (LAC-DHS) [17] The trial assigned 6
provider teams and their 1400 patients seen in 8 LAC-DHS ambulatory care clinics to one of
three comorbid depression care models:
(1) Usual Care (UC): Patients’ primary care team to provide usual diabetes and
depression care.
10

(2) Supported Care (SC): Patients with poor diabetes control were placed in a protocol-
driven, nurse-led diabetes disease management program (DMP) for a limited time (typically 6–9
months) to improve their diabetes knowledge, self-management, treatment, and follow-up. The
DMP team included physicians, nurse practitioners, and social workers supported by a web-
based disease management registry (DMR) for care-management. During DCAT, DMP clinical
social workers performed depression screening and symptom monitoring as well as problem
solving therapy (PST), while DMP nurse practitioners provided antidepressant treatment
following a depression treatment protocol.
(3) Technology-facilitated Care (TC): The same as SC, plus a care approach using an
automated telephonic assessment (ATA) system for periodic patient remote monitoring. The
ATA system performed depression screening, symptom monitoring, and assessment of important
self-care activities (i.e., physical activity, medication adherence, and/or practicing problem-
solving) once every month for depressed patients and every 3 months for nondepressed patients
at a patient-preferred call time. In addition, the DMR was enhanced by computer algorithms to
notify and task different DMP team members to provide collaborative depression care [18].
Compared to UC, the ATA technology and nurse/social worker–delivered DMP had the
potential to affect how subjects perceive care available to them through increased frequency in
health assessment and likelihood of contact with and support by providers. With the increased
frequency of assessment and access to providers, we hypothesized that, compared to UC, both
TC and SC would increase patient satisfaction with the care available for emotional problems
and diabetes symptoms.
The Institute for Healthcare Improvement’s triple aim initiative identifies three aims for
healthcare interventions: improved health outcomes, reduced costs, and improved patient
11

experience. Previous DCAT analyses have demonstrated improved health outcomes and lowered
costs [19,20] and have investigated patient experience through patient acceptance of the
technology [21,22]. In this research, our goal is to analyze comparative patient experience over
three phases lasting 18 months where every 6 months patients were exposed to different elements
of care-management due to the DMP design and ATA intervention. Using the data collected
from DCAT, this analysis answers the following research question: What is the difference in
patient satisfaction when patients were exposed to different ambulatory care-management
models over time, after controlling for patient differences?
The significance of this research is to understand whether patient satisfaction is sensitive
to elements of care-management, or is biased by personal characteristics, depression status, or
diabetes symptoms [10]. This study will help understand the value of using patient satisfaction to
assess and improve quality of care experience with newer care-management models, such as
comorbid disease management in the context of population health.
Methods
Study Design and Data Source
The study design was reviewed and approved by the USC Health Sciences Institutional
Review Board (IRB), LA Biomedical Research Institute IRB, and the Olive View–UCLA
Education and Research Institute IRB. Participants provided written informed consent.
Participant inclusion criteria included the following: patients were 18 years or older, had a
working phone number, spoke English or Spanish, and could read and understand the consent
form. Exclusion criteria included the following: patients with baseline possible suicide ideation
(Patient Health Questionnaire (PHQ)-9, item 9 response in more than half the days to nearly
every day), cognitive impairment (Short Portable Mental Health Status Questionnaire scores of
12

<5) [23], alcohol abuse (two or more CAGE items from the quantity-frequency index, and
questions about the patient's perception of substance use) [24], or recent lithium/antipsychotic
medication use. Patients were assigned to one of the three study groups based on the clinic from
which they were recruited, thus utilizing a quasi-experimental comparative effectiveness design.
Figure 1 depicts the varying care-management elements over time in the three study groups.
Patients were enrolled in DCAT for 18 months. Data was collected at 6-month intervals; baseline
was in-person at the study sites and the three follow-ups were telephone interviews with a
Spanish-English bilingual, group-blinded interviewer. Only a subset of the DCAT participants,
i.e., 700 participants who completed four waves of the interview data from baseline to 18
months, were included in the analyses.
Figure 1. Varying Care-Management Models Over Time in DCAT’s Three Study Groups
Element of Care
Base to 6
months
6 to 12
months
12 to 18
months
Usual Primary Care for Diabetes
UC UC
Some SC & TC
UC SC TC
Diabetes Disease
Management Program
SC & TC Some SC & TC
Depression Screening and Symptom
Monitoring by Clinical Social
Worker
SC Some SC
Depression Screening and Symptom
Monitoring by ATA System
TC TC

Measures
There were two outcomes of interest: (1) How satisfied are you with the clinical help you
received with emotional problems, and (2) How satisfied are you with the overall healthcare
available to you for your diabetes? The outcomes were recorded on a 5-point Likert scale where
13

1 indicated “very dissatisfied” and 5 indicated “very satisfied.” Measures of patient
demographics and baseline clinical characteristics are listed in Results (Table 1).
The primary independent variable was care-management group. Two of the three groups
(SC and TC) were included as intervention groups, and UC was the control group. There were
many additional control variables in this analysis. The first was depression status, measured
using the PHQ-9, which scores each of the 9 DSM-IV criteria as “0” (not at all) to “3” (nearly
every day); a cumulative score greater than 9 indicates having depression [25]. The second
additional control variable was diabetes symptoms, which was measured with the Whitty-9
Diabetes Symptom Scale, a 9-item questionnaire including: abnormal thirst, blurred vision,
urinated a lot of water during the day, felt unusually hungry, felt shaky, had cold hand and feet,
felt very sleepy during the day, had feeling of pins and needles, and felt faint or fainted, where
mean scores indicate the frequency of symptom experience [26]. These two variables were
included to adjust for patient disease status. A collection of patient characteristics was included
to adjust for patient differences as well as to provide insight into any patient satisfaction
differences by characteristics in this study population [27]. These variables included age, gender,
preferred language spoken (English or Spanish), education level (less than high school or more
than high school), marital status (married or unmarried), and financial stresses (12-item score
measuring financial difficulties). In addition, propensity scores, as developed and used in a
previous DCAT analysis [20], were included to adjust for the quasi-experimental design so
differences in the care-management model could be observed above patient differences by group.
Analysis
Linear regression models were used to estimate comparative treatment effects. Dependent
variables in all regression analysis were the 6-, 12-, and 18-month satisfaction scores. Previous
14

research has shown that modeling Likert scale items as continuous variables, even when
normality assumptions are violated, provide a correct result [28]. Thus, the satisfaction items
were modeled as continuous variables for this analysis even though slight violations of normality
were present.
The satisfaction items were individually regressed on the independent variable, care-
management group, and all control variables: current period depression status, current period
diabetes symptoms, previous period satisfaction score, age, gender, language spoken, education
level, marital status, financial stresses, and propensity score.
After running the linear regression models, global Wald tests were used to test for significant
differences in satisfaction between the care-management groups at all study periods (6, 12, and
18-months). All statistical analyses were conducted at 0.05 significance level (2-tailed) using
SAS 9.4 and Stata 15.
It should be mentioned that repeated measures analysis of variance (ANOVA) may
appear to be a good fit for analyzing this data; however, this was a comparative effectiveness
trial where the intervention was changing at each phase. Thus, repeated measures ANOVA
would not provide the same insight into the effects of the changing intervention at each phase
that the selected methods provide. Additionally, we checked clinic and care team significance
against care-management group, and results showed care-management group was consistently
the most significant predictor of satisfaction.
Results
Sample Characteristics
Table 1 provides sample size and characteristics of subjects in the three DCAT groups.
The study population included primarily low-income Latinos in the Los Angeles area. Most
15

patients were women and approximately one third of patients in each study group were
depressed. Figure 2 provides unadjusted mean satisfaction scores over time.
Table 1. Baseline patient characteristics

Figure 2. Unadjusted Mean Satisfaction Score with (a) Depression Care and (b) Diabetes Care

Regression Analysis Results
Wald tests of study group significance identified significant differences in satisfaction
among the care-management groups in the linear regression models as shown in Table 2. The test
16

identified significant relationships between 6- and 12-month satisfaction with depression care
and study group (p=.0113, p<.0001). Controlling for all other covariates, subjects in TC and SC
both were significantly more satisfied than UC subjects with depression care at 6- and 12-months
(p=.0037, p<.0280 and p<.0001, p<.0001 respectively). The levels of satisfaction in the TC and
SC were not statistically different. The Wald test did not identify any significant relationships
between satisfaction with diabetes care and study group at any phase.
Table 2. Wald Test p-Values of Satisfaction Differences Between Care-Management Groups

Satisfaction with Depression Care Satisfaction with Diabetes Care
Test 6 Months 12 Months 18 Months 6 Months 12 Months 18 Months
Group UC = SC = TC .0113 <.0001 .2314 .2700 .2019 .2777
Group UC = SC .0037 <.0001 .0909 .1211 .0738 .9253
Group UC = TC .0280 <.0001 .2507 .2214 .3239 .1947
Group SC = TC .3472 .9611 .4813 .6534 .3443 .1667

Other noteworthy results include nonsignificant negative associations between
satisfaction with depression care and PHQ-9 depression status, with 18-month follow-up being
an exception in reaching p<.05. There were also negative associations between satisfaction with
diabetes care and Whitty-9 Diabetes Symptom Scale; but the association was only statistically
significant at 6-month follow-up. No patient characteristics consistently displayed associations
with satisfaction with depression care. The only patient characteristic to consistently display a
moderate association with satisfaction with diabetes care was age at 12- and 18-months.
Appendices 1 and 2 provide the complete regression output.
In a post ad-hoc analysis, three additional satisfaction items measured in DCAT were
examined: satisfaction with clinical scheduling services, satisfaction with respect from providers,
17

and satisfaction with medical decision making in care received (results available upon request).
Some significant relationships were found between the service-oriented satisfaction items and
the care-management models, but not with disease symptoms or patient characteristics.
Discussion
Principal Findings
Patient satisfaction with depression care was found to be higher in care-management
intervention groups, and satisfaction with diabetes care was found to have no differences
between groups. No consistently significant associations between patient satisfaction and disease
symptoms or patient characteristics were found. These findings show that patient satisfaction is
responsive to the care processes influenced by the interventions as well as the other aspects of
care (aside from the intervention) that changed during the study period, without heavy affect by
patient characteristics or disease symptoms.
Patient satisfaction measures three intertwined components of the satisfaction construct:
expectations, value, and occurrences [29]. Patient satisfaction measures are affected more by
actual occurrence than perceived-to-be occurrence of interaction with providers; the intervention
improving attributes of care that patients value highly; and more healthcare interactions taking
place, giving patients more opportunities to interact with the care that provides higher value.
Because the DMP providers and social workers were trained to practice the depression care
protocol [17], patients were prone to be more satisfied with the emotional care they received in
the DMP. Patients did not sustain the higher level of satisfaction as they transitioned back to
usual primary care. The ATA technology prompted automated patient depression monitoring in
the TC group, monthly for depressed patients and quarterly for nondepressed patients. While
reducing provider workload in monitoring the entire patient population compared to the SC, the
18

ATA system in TC was able to identify patients in need of care and tasked providers to reach
those patients. Albeit low frequency, the increased contact with providers facilitated by the ATA
was when care was needed. It seemed to increase satisfaction with emotional care for patients in
the TC group equally as the SC group.
Comparison to Literature
A 12-month study of patients with type 2 diabetes and comorbid chronic diseases,
including depression, was conducted with two study arms, usual care and clinician-supported
care. Their study population included primarily white older adults from Washington state; they
were not specified as low income. The researchers found patients in the supported care group
were significantly more satisfied with depression care than usual care at 6-months (87% vs 62%)
and at 12-months (90% vs 55%) [30]. Consistent with those results, we also found that patients
with clinician-supported depression care were significantly more satisfied with depression care at
6 months (58% vs 40%) and 12 months (78% vs 56%). Our analysis additionally investigated the
technology-facilitated assessment and found that patients receiving ATA depression support
were also significantly more satisfied with depression care available to them at 6 months (52%
vs 40%) and 12 months (80% vs 56%), providing support that this newer care-management
approach can equally enhance patient satisfaction with care. The combination of significant
study population differences and similar results between our two analyses provide
generalizability of the findings of higher satisfaction with depression care in a supported-care
model, whether through clinician support or ATA technology.
A different study had a study population most similar to ours study and used technology-
facilitated assessment and follow-up support system for chronic disease; that study also found
high satisfaction rates with the telephone calls for supported care [31]. Although the study
19

population and intervention were similar, the study assessed satisfaction with the technology but
not the satisfaction with care.
Not only has higher patient satisfaction with technology been found in previous DCAT
analysis [22], our study evaluated satisfaction with the care available through the intervention
models and found higher satisfaction rates in both the SC and TC models. In addition, where the
literature shows mixed results of differences in satisfaction by patient characteristics [27], our
study provides further empirical evidence of limited effects of patient characteristics on patient
satisfaction with care received.
Limitations and Future Work
The first limitation of this analysis is the quasi-experimental trial design. This was
addressed by including generalized propensity scores for each subject to account for differences
in patient characteristics between study groups.
Second, attrition of 706 (50%) patients over the 18-month study period was a substantial
reduction in the study sample size. However, there were no significant differences between the
patients who completed the study and those who dropped out aside from by language spoken,
gender, and education level. English-speaking, educated males were more likely to drop out.
This result could have biased the satisfaction levels for the remaining sample. Patients who
remained in the study could have exhibited lower levels of satisfaction due to limited knowledge
on how to advocate for themselves, as this association has previously been found [32].
A third limitation was the duration of time since patients last received care and the study
interview time. Although we see improvements in satisfaction at 6 months and 18 months, there
was a dip in satisfaction between baseline and 6 months. A negative correlation between length
of time since last visit and patient satisfaction was found (satisfaction with depression care ρ =
20

−.024, satisfaction with diabetes care ρ = −.078). Baseline assessment was done in-person
immediately after patients had an appointment in a study clinic, while the follow-up assessment
was done as a telephone interview that did not coincide with the patient’s last medical
appointment. In the meantime, the patient may have received care from medical practitioners
who were not included in the study. Thus, the measurement of patient satisfaction with care
might be biased by memory recall issue, lack of access to care, or other care services than those
intended for the study. Future analyses of patient satisfaction can investigate the occurrence of
the same phenomenon.
Conclusion
This study evaluated patient satisfaction with diabetes and depression comorbid care-
management and ATA technology, controlling for patient characteristics and disease symptoms.
The analysis showed that patient satisfaction is influenced by aspects of care provided by the
intervention, not by patient characteristics or disease symptoms. Care-management models
increased patient satisfaction with depression care compared to usual care, whether through
clinician support or automated remote monitoring technology.

21

Appendices
Appendix 1. Satisfaction with Depression Care Regression Output
Phase 6-months 12-months 18-months
Coef. p Coef. p Coef. p
SC Group 0.30 .004 0.39 <.001 0.14 .091
TC Group 0.21 .028 0.39 <.001 0.09 .251
SC Propensity Score 0.11 .574 0.04 .810 -0.04 .790
TC Propensity Score 0.02 .911 0.31 .117 -0.08 .637
Previous Phase Satisfaction 0.10 .031 0.11 .001 0.18 <.001
Age 0.01 .088 0.004 .240 0.01 .032
Gender (male=1) -0.003 .967 0.04 .535 0.06 .325
Preferred Language
(Spanish=1)
0.06 .589 -0.11 .330 -0.04 .713
Education (< high school) 0.01 .949 0.10 .209 0.07 .355
Marriage Status (1=married) -0.05 .508 -0.005 .945 0.02 .666
Economic Status -0.08 <.0001 -0.004 .755 0.001 .952
Indicator PHQ-9 > 9
(1=depressed)
-0.09 .375 -0.12 .172 -0.26 .002
Whitty-9 Diabetes Symptom
Scale
-0.03 .648 -0.05 .371 0.01 .852
Constant 2.90 <.001 2.99 <.001 2.65 <.001

22

Appendix 2. Satisfaction with Diabetes Care Regression Output
Phase 6-months 12-months 18-months
Coef. p Coef. p Coef. p
SC Group 0.15 .121 0.14 .074 -0.01 .925
TC Group 0.11 .221 0.07 .324 0.10 .195
SC Propensity Score 1.14 .433 0.08 .601 0.19 .204
TC Propensity Score 0.18 .376 0.28 .098 -0.10 .566
Previous Phase Satisfaction 0.08 .140 0.18 <.001 0.26 <.001
Age 0.004 .234 0.008 .010 0.01 .025
Gender (male=1) 0.10 0.159 -0.03 .621 0.04 .461
Preferred Language
(Spanish=1)
-0.07 .515 0.05 .601 0.06 .534
Education (< high school) 0.04 .614 0.05 .513 0.03 .716
Marriage Status (1=married) -0.21 .002 0.04 .526 -0.08 .144
Economic Status -0.05 .001 -0.02 .117 -0.001 .942
Indicator PHQ-9 > 9
(1=depressed)
-0.02 .822 -0.01 .845 -0.07 .392
Whitty-9 Diabetes Symptom
Scale
-0.12 0.043 -0.03 .563 -0.04 .458
Constant 4.05 <.001 2.96 <.001 2.75 <.001

23

Chapter 3
Longitudinal Analysis of Relationship Between Self-Management
Activities, Care-Management, and Symptoms of Diabetes and
Depression: Results from A Comparative Effectiveness Trial
Abstract
Objective. A mechanism for short-term care-management models to result in improved long-term
health outcomes for patients with comorbid diabetes and depression is through improved and
sustained patient self-management behaviors to lesson disease symptoms. This study investigates
this mechanism using a longitudinal cross-lagged path analysis.
Research Design and Methods. Data were obtained from the Diabetes and Depression Care-
Management Adoption Trial (DCAT), which compared 3 care-management models: usual care,
6-month comorbid disease management, and 6-month comorbid disease management with 12-
month remote monitoring technology. The participants were low-income, primarily Latinos
within a safety-net population. The study sample contains 783 participants who had diabetes and
completed four waves of interviews in 6-month intervals (0-18 months). A cross-lagged path
analysis was conducted to investigate relationships between Toobert diabetes self-management
activities, 3 care-management models, the Whitty diabetes symptoms, and PHQ-9 depression
symptoms, controlling for health status.
Results. Only one significant path from diabetes self-management activities to depression
symptoms was found from 12- to 18 months, and no paths to diabetes symptoms were
significant. Significant paths from care-management models to depression symptoms were
identified at the 6-month wave, and to diabetes symptoms at the 12-month wave. Self-rated
health mediated many relationships between diabetes and depression symptoms over time.
24

Conclusions. For patients with more advanced diabetes and potentially comorbid depression,
self-management behaviors are not enough on their own to realize lasting improvements in
symptoms. Care-management models provided additional elements of care which helped
improve symptoms while the program was in place, but symptoms and behaviors regressed when
the program was complete. Future care-management models should work to incorporate
applications of health behavior maintenance themes to promote significant and sustainable self-
management behaviors for improved health outcomes in the long term.

Keywords
Self-management, depression, diabetes, care-management, cross-lagged path analysis, safety-net,
communication and information technology
Abbreviations
PCP: primary care provider, DMP: disease management program, DMR: disease management
registry, PST: problem solving therapy, ATA: automated telephonic assessment, DCAT:
Diabetes and Depression Care-Management Adoption Trial, LAC-DHS: Los Angeles County
Department of Health Services.

25

Introduction
Diabetes doubles the risk of comorbid depression [1] and can increase the risk of other
complications and mortality [2]. Associations between the symptoms of these two diseases have
been shown to be strong [3,4]. Self-management is often used to manage diabetes symptoms, and
common diabetes self-management behaviors such as diet and exercise have been shown to also
improve mental health [5-7]. Although the effects of depression on diabetes self-management
activities have been investigated, there is little research on the effects of diabetes self-
management activities on depression symptoms cross-sectionally or longitudinally [8].
Most self-management improvement interventions focus on teaching self-management
activities through a program [9-12]. In the evaluation of the effectiveness of these interventions,
improvements in self-management behaviors are measured as an outcome simultaneous with
health outcome measurements. Although it is believed that improved self-management behaviors
will be sustained to maintain health outcome improvements, this has not been verified with a
longitudinal study design, such as a cross-lagged path analysis where independent relationships
over time are possible to identify. To understand the value of short-term interventions models on
long-term health outcomes, we need to understand the independent, longitudinal effects of self-
management behaviors which can be sustained after care-management interventions are
complete.
Furthermore, the relationships between diabetes self-management behaviors and disease
symptoms may differ in a predominantly Latino population due to a variety of factors. Previous
research has shown rates of diabetes emotional distress are highest among Latino patients, lower
self-rated health status is reported, and lower satisfaction with help received for diabetes self-
management is reported [13]. A meta-analysis reviewed culturally tailored interventions aiming
26

to enhance diabetes self-management [14], and we find that the independent effects of diabetes
self-management activities were rarely investigated. One study that investigated the mediating
effects of diabetes self-management activities found the effects decreased over time without the
intervention in place, and another did not find self-management behaviors to be predictive of
diabetes outcomes [15,16]. Overall, the current literature states that long-term effects of self-
management education on behaviors and outcomes remain unknown and better understanding of
the effects of diabetes self-management over time in the Latino population is needed [14]. The
Diabetes and Depression Care-Management Adoption Trial (DCAT) was a culturally tailored
trial which allowed the independent longitudinal effects of diabetes self-management activities to
be investigated.
DCAT compared 3 care-management models: usual care, 6-month comorbid disease
management, and 6-month comorbid disease management with 12-month remote monitoring
technology. Follow-up, which includes diabetes and depression symptoms and self-management
activities, was conducted up to 18 months at 6-month intervals. The nature of the data collection
made this data set uniquely appropriate for this analysis.
The goal of this paper is to understand the relationships between diabetes self-
management activities, care-management models, and disease symptoms. The aims of this
analysis are two-fold: (1) to investigate longitudinal relationships between diabetes self-
management activities and symptoms of diabetes and depression, and (2) to investigate whether
the effects of care-management model on symptoms are direct or are mediated by self-
management behaviors.

27

Methods
Study Design
The study design is a cross-lagged path analysis to investigate independent relationships
between diabetes self-management activities, care-management models, and symptoms of
diabetes and depression. DCAT provided the data source for this analysis. Only a subset of the
DCAT participants, those who completed the four waves of interviews (baseline through 18
months) were included in the analysis (n=783).
Data Source
The participants in DCAT were low-income, primarily Latino patients with type 2
diabetes from the Los Angeles County Department of Health Services (LAC-DHS) [17].
Participant inclusion criteria included patients who were 18 years or older, had a working phone
number, spoke English or Spanish, and could read and understand the consent form. Exclusion
criteria included patients with baseline possible suicide ideation (PHQ-9, item 9 response in
more than half the days to nearly every day), cognitive impairment (Short Portable Mental
Health Status Questionnaire scores of <5) [18], alcohol abuse (two or more CAGE items from
the quantity-frequency index, and questions about the patient's perception of substance use) [19],
or recent lithium/antipsychotic medication use. The study protocol was reviewed by the
University of Southern California, Olive View UCLA Medical Center, and Los Angeles
Biomedical Research Institute Human Subjects Review Boards. Utilizing a quasi-experimental
comparative effectiveness design. The trial assigned 6 provider teams and their 1406 patients
seen in 8 LAC-DHS ambulatory care clinics to one of three comorbid depression care models:
1) Usual Care (UC), patients’ primary care team to provide usual diabetes and depression care.
28

2) Supported Care (SC), patients with poor diabetes control were placed in a protocol-driven,
nurse-led diabetes DMP for a limited time (typically 6-9 months) to improve their diabetes
knowledge, self-management, treatment, and follow-up. The DMP team included physicians,
nurse practitioners, and social workers. They were supported by a web-based disease
management registry (DMR) for care-management. During the DCAT, DMP clinical social work
performed depression screening and symptom monitoring as well as problem solving therapy
(PST), while DMP nurse practitioners provided antidepressant treatment following a depression
treatment protocol.
3) Technology-facilitated Care (TC), same as SC, plus a care approach using automated
telephonic assessment (ATA) system for periodic patient remote monitoring. The ATA system
performed depression screening, symptom monitoring, and assessment of important self-care
activities (i.e., physical activity, medication adherence, and/or practicing problem-solving) once
every month for depressed patients and every 3 months for non-depressed patients at patient
preferred call time. In addition, the DMR was enhanced by computer algorithms to notify and
task different team members of the DMP to provide collaborative depression care [17].
Patients were enrolled in DCAT for 18-months. Data was collected at 6-month intervals;
baseline was in-person at the study sites and the three follow-ups were by telephone interviews
with a Spanish-English bilingual, group-blinded interviewer. The three study groups had varying
care-management elements over time including:
• Usual Care: Patients received usual primary care for diabetes for the full 18-months.
• Supported Care: From baseline to 6 months, patients were enrolled in a diabetes
disease management program and received depression screening and monitoring by a
social worker. At 6-months, the majority of patients were transitioned from the
29

diabetes DMP to usual primary care, a small portion of patients continued to receive
these care-management elements for an additional 3-6 months and were then
transitioned back to UC.
• Technology Supported Care: From baseline to 6 months, patients were enrolled in a
diabetes disease management program and received depression screening and
monitoring by ATA. At 6 months, the majority of patients were transitioned from the
diabetes DMP to usual primary care for diabetes, a small proportion continued in the
DMP for an additional 3-6 months. All patients in TC received ATA call for another
6 months. From 12 to 18 months, TC patients received the same care elements as UC.
Measures
Primary variables included care-management model, diabetes self-management activities,
depression symptoms, diabetes symptoms, and self-rated health. Care-management group
variables were indicator variables. Two of the three models, both SC and TC, were included and
the UC group was the control. Care-management model was included through 12 months, as all
patients returned to usual primary care after 12 months.
Mean diabetes self-management activity was computed according to the Toobert diabetes
self-management scale, whose measures include diet, exercise, blood testing, and foot inspection
weekly frequency, and excluding smoking due to low proportion of smokers in study sample
[20]. Each of the ten items had a response from 0 to 7 based on the number of days in the past
week a patient performed each self-management item and mean self-management scores were
obtained through averaging the ten items. Self-management activities were measured at each
phase, baseline through 12 months were included in the analysis.
30

Depression symptoms were measured using the PHQ-9, which scores each of the 9 DSM-
IV criteria as “0” (not at all) to “3” (nearly every day) over the last two weeks, and scores equal
the sum of the 9-item responses [21]. This analysis utilized a categorical PHQ-9 variable
according to depression diagnostic status where scores from 0-4 = “none,” 5-9 = “mild,” 10-14 =
“moderate,” and 15+ = “moderate to severe.” Diabetes symptoms were measured with the
Whitty 9-item questionnaire, where questions asked patients how often they have experienced
each symptom over the past month and responses included: 1 = never, 2 = one or a few days, 3 =
on several days, 4 = on most days, and 5 = every day. The 9 items, including: abnormal thirst,
blurred vision, urinated a lot of water during the day, felt unusually hungry, felt shaky, had cold
hand and feet, felt very sleepy during the day, had feeling of pins and needles, and felt faint or
fainted, were averaged to obtain the mean score [22]. HbA1c would be the obvious choice for the
diabetes outcomes variable; however, in our sample there was a significant amount of data
missing (72%) not at random, making it a nonviable variable choice. The disease symptom
variables were included through 18 months since the effects of care management group and self-
management on these variables were of interest.
A patient’s general health is associated with symptom burden and emotional distress [23],
so we controlled for patients’ general health using the single item measure, self-rated health [24].
Patient health status is measured by the single item self-rated health. Self-rated health was
measured using a 5-point Likert scale, where 1 = poor and 5 = excellent. Self-rated health was
measured at each phase, baseline through 12-months were included in the analysis.

31

Analysis
Cross-lagged path analysis was used to investigate relationships between the variables of
interest. Paths from care-management model variables were included to all 6- and 12-month
variables. These paths were included to understand two different types of relationships:
intervention effects of the SC and TC models from baseline to 6 months, and transition of care
from intervention group to usual primary care effects from 6 to 12 months.
Paths from diabetes self-management activities to consecutive phases of diabetes and
depression symptoms were included to identify any relationships between diabetes self-
management activities and disease symptoms. Paths between diabetes and depression symptoms
were included to account for any bidirectional relationships between disease symptoms in the
DCAT study population.
A control variable in cross-sectional analysis, self-rated health, was treated as an
independent variable due to the statistical modeling methods used and the repeated collection of
self-rated health responses from patients. Paths from self-rated health to consecutive phases of
self-management activities and disease symptoms were included to capture any dynamic
relationships between diabetes self-management activities and disease symptoms. The outcome
variables, depression level and diabetes symptoms, were included through 18-months to capture
all longitudinal effects of care-management model, self-management activities, and self-rated
health. The independent variables, self-management activities and self-rated health, were
included through 12-months as this allowed for capture of 12-month effects on 18-month
outcome variables without excess variables in the model.
Propensity scores, as developed and used in previous DCAT analysis [25], were included
to control for group differences due to the quasi-experimental design. Propensity score paths led
32

to care-management models and baseline variables to capture the adjustment. Patient
characteristics including age, gender, and language spoken were included to control for patient
level differences. Paths from patient characteristics to all baseline through 18-month variables
were included. Analysis was performed using SAS 9.4 and Mplus, and significance was
recognized at p<.05. Figure 1 depicts the path model described above.
Figure 1. Analytical Model

33

Results
Table 1 provides sample size and comparative characteristics of subjects in the three
DCAT groups. Most subjects were female, Latino, and mean age greater than 50 years old. After
controlling for group differences with propensity scores, as done in previous DCAT analysis
[25], differences between models were accounted for.
Table 1. Baseline Patient Characteristics

a
Number of respondents.
b
Values are numbers (percentages) for categorical variables, mean (sd) for continuous variables.
c
Chi-square test for categorical variables, ANOVA F-test for continuous variables.
d
Assessed by the 9-item diabetes symptoms scale, scores ranged from 1-5, higher scores indicate
more severe diabetes symptoms [(22)].
e
Assessed by the 9-item Patient Health Questionnaire, scored ranged from 0-27, higher scores
indicate worse depressive symptoms [21].

Model Fit Summary. The path model was created using complete cases (n=783). There
were 21 variables in the model and 169 parameters. The model was found to be a good fit to the
data. This was indicated by a goodness of fit index (GFI) of 0.9882 (adjusted GFI = 0.9560) and
a RMSEA estimate equal to 0.0278 (90% CI: 0.0107, 0.0377). The Chi-square test indicates a
34

rejection of the null hypothesis with a Chi-square statistic of 99.5 and 62 degrees of freedom
(p=.0018). This is due to the large size of the model and inability to add additional paths for
improved model fit from violation of theory.
In this longitudinal analysis, the total, direct, and indirect effect paths provided the
dynamic relationships between all the variables in the model. All the direct significant paths are
visually displayed in Figure 2 below. Complete path model direct effect results are available in
table 2. The total, direct, and indirect paths to diabetes and depression symptom variables are
provided in Table 3. All variables had significant positive paths between all incremental phases,
and to verify the stability of the paths of variables over time, paths from baseline to all
subsequent phases were included. These results are also available upon request.
Research Question 1: Diabetes Self-Management Activities Relationships. There was a
single significant path from self-management activities to disease symptoms. The path was a
significant negative direct path from 12-month self-management activities and 18-month
depression symptoms. The significance of this path was not strong enough to mediate any
significant paths between care-management group and depression symptoms.
Research Question 2: Care-Management Model Effects. Significant negative total effects
of both care-management models on 6-month depression symptoms were found. There were no
significant effects of care-management model on 6-month diabetes symptoms. At 12 months, the
total effects of the SC care-management group on diabetes symptoms were significant and
negative. The significant effect was the direct effect of the SC care-management model on
diabetes symptoms, indirect effects were not significant. There were no other significant effects
of care-management models on disease symptoms at 12 months. Similarly, there were no
significant total effects of care-management models on disease symptoms at 18 months. In
35

addition to care-management effects on disease symptoms, there were significant positive total
effects of both care-management models on 6-month self-management activities. These results
were reversed during transition of care where there was a significant positive total effect of both
care management models to 12-month self-management activities. Although care-management
models had significant effects on self-management activities, the effects were not strong enough
to allow self-management activities to act as a mediator of care-management effects on disease
symptoms.
Self-Rated Health Relationships. There were no significant direct or indirect paths
between self-rated health and diabetes self-management activities. However, self-rated health
acted as a mediator between diabetes and depression symptoms over time, and baseline self-rated
health had many indirect effects on 12- and 18-month symptoms. Self-rated health had
significant negative total effects on both depression and diabetes symptoms at all phases, 6, 12,
and 18 months.
In a post-hoc path analysis, diabetes symptoms were replaced with HbA1c (results
available upon request). We found diabetes self-management activities were not significantly
related to HbA1c at any phase except for 12 to 18 months. The 12- to 18-month change was
clinically significant; however, HbA1c levels worsened. We do not report the path model
including HbA1c in this analysis due to the significant amount of missing data (72% missing).
Non-missing data indicates patients were attending visits at a clinic, thus it is possible this
significant relationship is a result of access to care instead of self-management activities.

36

Figure 2. Path Model Visual Results

37

Table 2. Complete Path Model Results
Dependent
Variable
Independent Variable Base to 6-months 6- to 12-months 12- to 18-months
Self-
Management
Care-Management
Model (SC, TC)
0.08 (.0287),
0.12 (.0010)
-0.16 (<.0001),
-0.18 (<.0001)
Self-Management 0.33 (<.0001) 0.40 (<.0001)
Depression Symptoms -0.12 (.0028) -0.02 (.5207)
Diabetes Symptoms -0.05 (.1813) 0.01 (.7599)
Self-Rated Health 0.02 (.5365) 0.04 (.4018)
Depression
Symptoms
Care-Management
Model (SC, TC)
-0.09 (.0103),
-0.10 (.0035)
0.04 (.2427),
0.07 (.0140)
Self-Management 0.04 (.2584) 0.03 (.2495) -0.06 (.0274)
Depression Symptoms 0.43 (<.0001) 0.46 (<.0001) 0.36 (<.0001)
Diabetes Symptoms 0.11 (.0024) 0.15 (<.0001) 0.05 (.1493)
Self-Rated Health -0.12 (.0003) -0.04 (.1779) -0.06 (.0381)
Diabetes
Symptoms
Care-Management
Model (SC, TC)
0.01 (.6827),
0.04 (.2565)
-0.07 (.0280),
-0.05 (.0930)
Self-Management -0.03 (.3334) 0.01 (.7209) -0.01 (.7469)
Depression Symptoms 0.20 (<.0001) 0.15 (<.0001) 0.08 (.0153)
Diabetes Symptoms 0.42 (<.0001) 0.45 (<.0001) 0.27 (<.0001)
Self-Rated Health -0.09 (.0061) -0.04 (.1337) -0.08 (.0051)
Self-Rated
Health
Care-Management
Model (SC, TC)
0.02 (.5487),
0.03 (.5018)
0.07 (.0540),
0.02 (.6517)
Self-Management 0.02 (.6339) -0.01 (.8118)
Depression Symptoms -0.17 (<.0001) -0.16 (<.0001)
Diabetes Symptoms -0.08 (.0553) -0.13 (.0006)
Self-Rated Health 0.23 (<.0001) 0.28 (<.0001)

38

Table 3. Path Model Standardized Total, Direct and Indirect Results
Baseline 6 Months 12 Months 18 Months
Total
Effect p
Direct
Effect p
Total
Indirect p
Self-
Management
Depression
Symptoms

0.035 0.269 - - - -
Depression
Symptoms
Depression
Symptoms

0.428 <.001 - - - -
Diabetes
Symptoms
Depression
Symptoms

0.113 0.002 - - - -
Self-Rated
Health
Depression
Symptoms

-0.118 <.001 - - - -
SC
Depression
Symptoms

-0.091 0.019 - - - -
TC
Depression
Symptoms

-0.1 0.003 - - - -
Self-
Management
Diabetes
Symptoms

-0.029 0.353 - - - -
Diabetes
Symptoms
Diabetes
Symptoms

0.417 <.001 - - - -
Depression
Symptoms
Diabetes
Symptoms

0.198 <.001 - - - -
Self-Rated
Health
Diabetes
Symptoms

-0.085 0.004 - - - -
SC
Diabetes
Symptoms

0.014 0.682 - - - -
TC
Diabetes
Symptoms

0.038 0.207 - - - -
Self-
Management

Depression
Symptoms

0.022 0.257 - - 0.022 0.257
Depression
Symptoms

Depression
Symptoms

0.418 <.001 0.19 <.001 0.229 <.001
Diabetes
Symptoms

Depression
Symptoms

0.116 <.001 - - 0.116 <.001
Self-Rated
Health

Depression
Symptoms

-0.075 <.001 - - -0.075 <.001
SC

Depression
Symptoms

-0.002 0.961 0.036 0.256 -0.038 0.067
TC

Depression
Symptoms

0.038 0.294 0.075 0.019 -0.037 0.046
Self-
Management

Diabetes
Symptoms

-0.005 0.79 - - -0.005 0.79
Depression
Symptoms

Diabetes
Symptoms

0.16 <.001 - - 0.16 <.001
Diabetes
Symptoms

Diabetes
Symptoms

0.401 <.001 0.193 <.001 0.208 <.001
39

Self-Rated
Health

Diabetes
Symptoms

-0.066 <.001 - - -0.066 <.001
SC

Diabetes
Symptoms

-0.074 0.038 -0.067 0.027 -0.007 0.698
TC

Diabetes
Symptoms

-0.049 0.169 -0.051 0.104 0.002 0.904
Self-
Management

Depression
Symptoms -0.006 0.724 - - -0.006 0.724
Depression
Symptoms

Depression
Symptoms 0.389 <.001 0.137 <.001 0.252 <.001
Diabetes
Symptoms

Depression
Symptoms 0.09 0.001 - - 0.09 0.001
Self-Rated
Health

Depression
Symptoms -0.064 <.001 - - -0.064 <.001
SC

Depression
Symptoms -0.02 0.374 - - -0.02 0.374
TC

Depression
Symptoms -0.003 0.896 - - -0.003 0.896
Self-
Management

Diabetes
Symptoms -0.01 0.57 - - -0.01 0.57
Depression
Symptoms

Diabetes
Symptoms 0.142 <.001 - - 0.142 <.001
Diabetes
Symptoms

Diabetes
Symptoms 0.376 <.001 0.142 0.001 0.234 <.001
Self-Rated
Health

Diabetes
Symptoms -0.06 <.001 - - -0.06 <.001
SC

Diabetes
Symptoms -0.023 0.251 - - -0.023 0.251
TC

Diabetes
Symptoms -0.002 0.933 - - -0.002 0.933

40

Conclusions
In this analysis, minimal effects of changes in self-management activities on disease
symptoms were found; however, there were significant effects from care-management model on
disease symptoms and some mediating relationships between self-rated health and disease
symptoms. It was shown in this analysis it is possible for care-management interventions to have
positive influence on self-management behaviors when in place, but change is not sustainable
without proper mechanisms and support integrated to promote maintenance of self-management
behaviors. DCAT provided a suitable sample for this analysis due to the longitudinal nature of
the data, the sample size, and the three different care-management models. Cross-lagged path
analysis provided a unique, longitudinal method to evaluate the independent relationships
between all variables.
As mentioned, our study found only one significant relationship between self-
management activities and depression symptoms which occurred between 12 and 18 months. In
current literature, self-management activities are believed to be beneficial in improving health
outcomes [9-12]. However, these results evaluate the effects of self-management activities as a
result of care-management interventions, often without investigation of behavioral maintenance
at follow-up after programs. Additionally, these results are not consistent across study
populations. For example, in two reviews like our study population, inconclusive and
inconsistent relationships between self-management activities and health outcomes were
identified for patients with chronic illness and low socio-economic status and minority patients
[26,27]. Another study has shown sustained improvements in self-management activities as well
as improved health outcomes, but self-management activities were not evaluated as independent
predictors of improved health outcomes [28]. In our analysis of diabetes self-management
41

activities as an independent predictor of disease symptoms, improved self-management activities
did not improve disease symptoms aside from the relationship between 12-month self-
management activities and 18-month depression symptoms. The significant path may suggest
that someone who already had decent self-management behavior and continued to improve
would eventually see some improvement in depression symptoms. Future analysis should aim to
alleviate the disconnect between self-management activities and disease symptoms in low-
income minority patients, by understanding the factors that inhibit self-management maintenance
in this population.
Our analysis shows it was the care-management programs that were accountable for
improvements in symptoms which is similar to other studies [9-12, 29]. During the intervention
period (baseline to 6-months) we found that mechanisms of the intervention models directly
improved diabetes self-management activities and depression symptoms, but the changes in self-
management activities were not strong enough to have a significant mediating effect improving
disease symptoms, as previously mentioned. During transition of care (6 to 12 months), our
analysis shows diminishing effects of care-management model on diabetes self-management
activities and depression symptoms as patients were transitioned back to usual care, which agrees
with previous findings [30,31]. With many interventions missing behavioral maintenance
support, patients return to primary care where there are limited resources to provide continuous
support like in the intervention, and this often leads to diminished improvements post-
intervention. Opposite of the previous results we found no improvements in diabetes symptoms
when the care-management models were in place (baseline to 6 months), and we found
improvements in diabetes symptoms during transition of care (6 to 12 months). In an
intervention like ours, with a different study population, they also found a delayed effect on
42

diabetes symptoms where after 9 months patients experienced reduced diabetes symptom distress
[32]. Our results show the direct effects of care-management model are responsible for the
improvement in diabetes symptoms at 12 months, but it also highlights the importance of
depression care management for patients with diabetes. With significant effects of depression
symptoms on diabetes symptoms at each phase, depression care-management that reduces
depression symptoms will in turn reduce the impact on diabetes symptoms.
This analysis identified the difficulty of sustaining self-management activities without an
intervention in place. This challenge has also been identified in a previous analysis of a self-
management intervention on a similar population to our study population [33]. One thing we’ve
noted in a post-hoc analysis (results available upon request) of individual self-management
activities is that diet and exercise behaviors seem to fluctuate the most with transitions of care,
improving during care-management interventions and regressing afterwards. Previous analyses
on populations similar to our study population have attempted to improve diet and exercise
behaviors through inclusion of social support via spouses and other family member support [34-
36]. Social support via family and education for diet and exercise behavioral change are elements
of DCAT that could be missing for supporting behavior change; however, maintenance of health-
related behaviors is a more complex challenge. A systematic review has identified five
interconnected themes for maintenance of health-related behaviors including: maintenance
motives, self-regulation, resources, habits, and contextual influences [37]. These five themes
should be integrated into care-management intervention through applications. Some examples
include: maintaining positive motives through emphasizing the positive outcomes of new
behaviors, facilitating self-regulation through developing strategies to overcome behavioral
barriers that lead to relapse, and providing social support for behavioral maintenance [37].
43

As mentioned, we controlled for patients’ general health using the single item self-rated
health measure because it is associated with patient symptom burden and emotional distress [24].
We found that self-rated health, as an independent variable, mediated longitudinal symptoms
relationships within disease and between diseases. We did not find relationships between care-
management model and self-rated health, but the significant relationships between self-rated
health and symptoms suggest that future care-management models that can affect patient health
could benefit outcomes in the long-term. Due to the significance of self-rated health, it is of
interest to explore a more comprehensive method for controlling for general health, such as
inclusion of physical, mental, and social factors [24].
As a robustness check, subgroup path models for each group of DCAT were evaluated.
Results are available upon request. Stronger relationships between diabetes self-management
activities and symptoms, and self-rated health and symptoms were observed in the SC and TC
models. Overall, results of this analysis suggest multiple patient factors and mechanisms of
DCAT can be attributed to longitudinal changes observed in disease symptoms.
There are a few limitations of this study. The first limitation is that this study was a quasi-
experimental design and was not completely randomized by group. To account for group
differences, we used propensity scores. The second limitation is path analysis requires wide
format data, so only complete cases could be used. This lead 623 patients to be excluded from
analysis, and since missing data is likely not random, the results could be biased by the sample.
The third limitation is the construction of the diabetes symptoms variable. The measurement
scale used does not have the ability to distinguish, for example, between one symptom
“everyday” or four symptoms “one or a few days.” Fourth is the measurement of self-
management activities. We measured self-management activities as a snapshot of behavior,
44

asking about patient behavior over the past week, yet this measure was only recorded once every
6-months. There could be a better way to collect more accurate and consistent data, such as the
use of a mobile phone app. The final limitation is the use of a single item, self-rated health, to
control for patients’ general health. This variable displayed mediating relationships with
symptoms, however, no relationships with self-management activities were identified indicating
opportunity to investigate other health factors that inhibit self-management activities in future
work.
For patients with more advanced diabetes and potentially comorbid depression, self-
management behaviors are not enough on their own to realize lasting improvements in
symptoms. Care-management models provided additional elements of care, including access to
providers and medication, which helped improve symptoms while the program was in place.
When the program was complete, symptoms and behaviors regressed. Self-rated health displayed
longitudinal effects on depression and diabetes symptoms, indicating some predictive power of
disease symptoms. Future care-management models should work to incorporate applications of
health behavior maintenance themes to promote significant and sustainable self-management
behaviors for improved health outcomes in the long term.

45

Chapter 4
A Cluster Analysis of Diabetes Self-Management Behavior Over Time
and Predictive Factors Among Low-Income Latino Patients with Type 2
Diabetes
Abstract
Objective: The literature has identified patients with complex health conditions are likely to have
different self-management needs; however, there is a lack of empirical evidence supporting this
claim and identifying the factors affecting self-management. This analysis aims to cluster
patients by self-management behavior pattern and trend over time and determine factors that
predict self-management behavior pattern and trend.
Research Design and Methods: The study design is a two-part analysis that includes: (1) data
driven clustering analysis for self-management behavior over time and (2) self-management
behavior cluster prediction modeling. Data for this analysis came from DCAT, and only
complete cases were used (n=847).
Results: The cluster analysis identified three self-management behavior clusters: consistent
(n=172), moderate (n=573), and struggling (n=102) with self-management. Many factors were
significantly different between cluster. A subset of these were found to be most predictive of
self-management behavior cluster including: care-management group, diabetes complications,
economic stressors, education level, and dysthymia diagnosis.
Conclusion: This research found empirical evidence of three clusters of patient self-management
behaviors over time, and that multiple health factors, including physical, mental, and social
health factors, differ across the patient self-management behavior clusters. Moreover, a subset of
46

physical, mental, and social factors was found to predict self-management behavior cluster. This
research furthers our understanding of different self-management behavior groups and predictors
of long-term self-management behaviors within this patient population. This knowledge will help
providers anticipate the long-term behaviors of their patients to provide appropriate levels of
support.

Keywords
Self-management, depression, diabetes, Latino, care-management, safety-net, cluster analysis,
prediction modeling
Abbreviations
DCAT: Diabetes and Depression Care-Management Adoption Trial, UC: Usual primary Care,
SC: Supported Care, TC: Technology supported Care, LAC-DHS: Los Angeles County
Department of Health Services, QOL: Quality of Life.

47

Introduction
In a previous Diabetes and Depression Care-Management Adoption Trial (DCAT)
analysis, minimal relationships were found between self-management and diabetes symptoms,
depression symptoms, and health status [Chapter 3]. The results of this analysis inspired new
research questions about clusters of patients’ self-management behaviors over time and factors
that are associated with distinct self-management behavior clusters.
The current literature has identified differences in characteristics of self-management
among patients with complex health needs vs traditional self-management recommendations. A
review has identified socio-economic conditions as strongly influential in the self-management
of complex conditions, often forcing patients to emphasize care of one condition over another
allowing their overall health to suffer in the process [1]. Additionally, many of the studies in the
review found that self-management of multi-morbidity is influenced by many different health
factors. However, most of the studies of patient self-management of complex conditions involve
the analysis of a small sample, use qualitative methods, or use cross-sectional surveys. The
review concludes that additional empirical evidence is required to verify these claims as there is
a current lack of empirical investigation of relationships between self-management and outcomes
in patients with complex conditions.
Another patient population with unique self-management needs are Latino patients living
with type 2 diabetes who face different self-management challenges than other populations. It is
important for us to understand differences in self-management behaviors and needs in this
population because Latino’s have a higher prevalence of type 2 diabetes than non-Latino whites
[2]. A review of community and health care self-management barriers among Latinos with
diabetes identified that there is a lack of work being conducted among Latino patients with
48

diabetes with regards to socioeconomic conditions [3]. They point out the finding that
neighborhoods characterized by limited resources or inadequate health plans pose self-
management challenges. This points out a specific hole in the literature, the needs and barriers to
self-management among low-income Latinos with type 2 diabetes. Furthermore, it has also been
identified that strategies for sustaining self-management behaviors among this population are
needed [4]. One study attempted to fill this hole. A qualitative study of factors that influence
diabetes self-management in Latino patients living with type 2 diabetes in California identified
four themes as either enhancers or inhibitors of self-management: access to resources, struggle
with diet, self-efficacy, and social support [5]. This is a start, but in order to improve and sustain
behaviors, it is important to empirically investigate which factors are associated with diabetes
self-management behavior in low income Latino populations.
The Diabetes and Depression Care-Management Adoption Trial (DCAT) was a quasi-
experimental trial of comorbid depression care-management with three study groups on a low-
income predominately Latino patients with type 2 diabetes [6]. This trial offers a proper study
population to empirically investigate self-management in a population with complex health
needs and challenging socio-economic conditions due to the longitudinal nature of data
collection, the large sample size, and the complex health conditions of patients enrolled.
Previous analyses have focused on qualitative analysis or cross-sectional survey analysis
of patients’ complex condition self-management [7-16] and there is a lack of empirical
investigation of self-management behavior of patients with complex chronic conditions. This
investigation offers a longitudinal analysis of a large sample of low-income primarily Latino
patients with type 2 diabetes to help understand (1) self-management clustering by behavior
patterns and trend over time and (2) predictors of self-management behavior cluster. Thus, the
49

aim of this analysis is to answer two research questions: (1) Are DCAT patients clustered by
different self-management behaviors over time and (2) what variables predict diabetes self-
management behavior cluster? The importance of this research is two-fold. First, it will provide
understanding of different self-management clusters that exist in this population. Second, we will
gain understanding of factors that differ by self-management behavior cluster and factors that
predict self-management behavior cluster among patients in this population.
Methods
Study Design
The study design is a two-part, ordered analysis that includes: (1) data driven clustering
analysis for self-management behavior cluster and (2) self-management behavior cluster
prediction modeling. The two parts of this analysis were performed in order. The first part of the
analysis includes three steps: first, identify the appropriate number of clusters using fit statistics
and appropriate visual plots, second, identify the number of patients assigned to each cluster, and
third, identify baseline patient differences between the clusters. After these three steps of part 1
were complete, the prediction modeling of part 2 was performed. The data used in this analysis
came from the DCAT. DCAT study design is described below in Data Source. Only a subset of
the DCAT participants who completed four waves of interviews from baseline to 18 months
(n=847) will be included in the analysis.
Data Source
The participants in DCAT were low-income, primarily Latino patients with type 2
diabetes from the Los Angeles County Department of Health Services (LAC-DHS) [6].
Participant inclusion criteria included patients who were 18 years or older, had a working phone
number, spoke English or Spanish, and could read and understand the consent form. Exclusion
50

criteria included patients with baseline possible suicide ideation (PHQ-9, item 9 response in
more than half the days to nearly every day), cognitive impairment (Short Portable Mental
Health Status Questionnaire scores of <5) [17], alcohol abuse (two or more CAGE items from
the quantity-frequency index, and questions about the patient's perception of substance use) [18],
or recent lithium/antipsychotic medication use. The study protocol was reviewed by the
University of Southern California, Olive View UCLA Medical Center, and Los Angeles
Biomedical Research Institute Human Subjects Review Boards. Utilizing a quasi-experimental
comparative effectiveness design, the trial assigned 6 provider teams and their 1406 patients seen
in 8 LAC-DHS ambulatory care clinics to one of three comorbid depression care models: Usual
Care (UC), Supported Care (SC), Technology Supported Care (TC). More detailed descriptions
of the three care-management models can be found in chapter 3 of this dissertation and the
DCAT study design paper [6]. Data was collected at 6-month intervals; baseline was in-person at
the study sites and the three follow-ups were by telephone interviews with a Spanish-English
bilingual, group-blinded interviewer.
Measures
Cluster Analysis Measures. Growth models were developed using longitudinal self-
management behavior data. Mean diabetes self-management behavior was recorded at each
phase, baseline through 18 months, and computed according to the Toobert’s Summary of
Diabetes Self-Care Activities measure (SDSCA). Scale measures include diet, exercise, blood
testing, and foot inspection weekly frequency, and excluding smoking due to low proportion of
smokers in study sample [19]. Each of the ten items had a response from 0 to 7 based on the
number of days in the past week a patient performed each self-management item and mean self-
management scores were obtained through averaging the ten items.
51

From the latent growth analysis, individual intercepts and slopes were obtained for each
patient. The intercept represented the individual self-management behavior level at baseline, and
slope represented the patient’s self-management behavior trend throughout DCAT.
Prediction Modeling Measures. Once clusters had been assigned, many variables were
tested as potential predictors of patient self-management behavior cluster. Candidate variables
were guided by previous DCAT results and were selected from patient characteristics, healthcare
utilization and cost measures, physical health measures, psychological health and quality of life
(QOL) measures, and social health measures [20]. Candidate variables include the measures in
Table 1, and as described below, with two exceptions: (1) clinic due to collinearity with study
group, and (2) baseline self-management due to the dependent variable, self-management trend,
being developed from this measure.
Patient characteristics included: age, gender, education level via an indicator variable of
less than high school education, and marriage status via an indicator variable of married or not
married. The nominal variable for care-management group, which has three groups: UC, SC, and
TC, was also included.
Healthcare utilization and cost measures included: healthcare utilization cost which was
computed as the sum of lab, outpatient visit, emergency room use, and inpatient visit costs.
Medication costs were included independent of the other costs and includes: insulin, oral
diabetes medication, antidepressant medication, and other prescription medication costs. Lastly,
number of outpatient visits were included.
Physical health measures included: HbA1c, Body Mass Index (BMI), total cholesterol
level, diabetes symptoms, diabetes complications, and diabetes distress. Diabetes symptoms were
measured with the Whitty 9-item questionnaire, where questions asked patients how often they
52

have experienced each symptom over the past month and responses included: 1 = never, 2 = one
or a few days, 3 = on several days, 4 = on most days, and 5 = every day. The 9 items, including:
abnormal thirst, blurred vision, urinated a lot of water during the day, felt unusually hungry, felt
shaky, had cold hand and feet, felt very sleepy during the day, had feeling of pins and needles,
and felt faint or fainted, were averaged to obtain the mean score [21]. Diabetes complications
measured if patients experienced: visions, problems, loss of feeling in legs or feet, kidney
problems, foot ulcer or infection, amputation, sexual impairment, and heart attack or cardiac
procedure. The diabetes distress screening scale items included diabetes emotional burden and
diabetes related distress [22]. Items were scored 1-6; higher scores indicated a higher level of
diabetes distress.
Psychological health and QOL measures included: depression symptom, anxiety level,
dysthymia status, diagnosis of major depressive disorder (MDD), functional disability, and
mental and physical QOL. Depression symptoms were measured using the PHQ-9 questionnaire.
Each individual DSM-IV criteria had possible outcomes from “0” (not at all) to “3” (nearly every
day) over the last two weeks, and the total PHQ-9 score equals the sum of the 9-item responses
[23]. Dysthymia was measured with an indicator variable of 1= diagnosed with dysthymia and 0
= not. History of MDD was measured with an indicator of 1 = history of MDD and 0 = not. Brief
Symptom Inventory score (BSI) assessed by the brief symptoms inventory [24], scored 0-24;
higher scores indicate worse anxiety. Sheehan Disability Scale (SDS) was scored 0-30 [25,26];
higher scores indicate more functional impairment. Mental and Physical quality of life (QOL)
assessed by the SF-12 Physical and Mental component scores.
Social health measures included: economic stresses, general stresses, stress levels, and
satisfaction with healthcare. Economic Stressors were assessed by 12 general and health-related
53

economic stresses, scored 0-12; higher scores indicate a higher level of economic stress. Lastly,
general stresses measured work, unemployment, financial, marital, family, children, caregiver,
cultural, legal, immigration, illness of family, and community worry stresses. Each of the items
were recorded on a scale of 0 = “no stress” to 10 = “the most stress you can imagine.” Both
number of stresses and total stress level were included in the analysis.
Care experience measures included two patient satisfaction items were included,
specifically: (1) How satisfied are you with the clinical help you received with emotional
problems, and (2) How satisfied are you with the overall healthcare available to you for your
diabetes? The outcomes were recorded on a 5-point Likert scale where 1 indicated “very
dissatisfied” and 5 indicated “very satisfied.”
Analysis
There were two phases of analysis: (1) data driven clustering analysis to determine
patient self-management behavior cluster and (2) self-management behavior cluster prediction
modeling.
Analysis was conducted using the mixed model procedure in SAS. A random intercept
and slope linear time model was run on self-management behavior, as shown in equation set 1.
Baseline through 18-months were included. Individual slope and intercept were collected from
this model.
Equation Set 1. Growth Analysis Model
Level 1: Self-Management ~ β0i + β1i(Time)ti (1)
Level 2: β0i ~ γ00 + U0i (2)
β1i ~ γ10 + U1i (3)
54

Data driven cluster analysis was conducted using the cluster procedure in SAS.
Standardized individual level intercept, slope, and sum of squared residuals were obtained via
the ACECLUS procedure in SAS. Patients with extreme outlier sum of squared residuals were
removed (n=5). Data was clustered by the individual level, standardized intercept and slope
obtained through growth modeling analysis. Statistics including R-square statistics and Cubic
Clustering Criterion (CCC), pseudo f-statistic, and T
2
plots were used to determine the optimal
number of clusters. Cluster plots were also used to visually assess the appropriateness of the
cluster assignment.
Finally, multinomial logistic regression prediction models were used to assess which of
the candidate variables predict self-management behavior cluster over time. The SAS logistic
procedure with stepwise selection was used due to ability to efficiently handle many predictors.
Robustness of this selection method was verified with forward and backward selection, and
methods agreed. Criteria for entry and exit were set to p<0.2. The Hosmer-Lemeshow goodness
of fit (GOF) test was used to assess quality of model fit. Significant predictors were used to fit a
descriptive model and obtain odds ratios for group assignment. Predictors included patient
characteristics, healthcare utilization and cost measures, and baseline physical, psychological and
QOL, and social health measures as described above.
Equation Set 2. Multinomial Logistic Regression Prediction Model
𝑆𝑒𝑙𝑓 − 𝑀𝑎𝑛𝑎𝑔𝑒𝑚𝑒𝑛𝑡 𝐵𝑒 ℎ𝑎𝑣𝑖𝑜𝑟 𝑇𝑟𝑒𝑛𝑑 ~ 𝛽 0
+ ∑ 𝛽 𝑖 𝑃𝑟𝑒𝑑𝑖𝑐𝑡𝑜𝑟 𝑖 + 𝑒 𝑖 𝑛 𝑖 =1
(4)
Results
Cluster analysis resulted in three clusters of patients based on initial self-management
behavior differences and trend throughout DCAT. Number of clusters was assessed by the CCC,
pseudo F statistic, and Pseudo T
2
plots, as well as R
2
values and self-management cluster plots
55

for visual separation of the clusters. Cluster statistics are available in Table 1. The CCC had a
peak at 4 clusters. The pseudo F statistic plot, which represents the separation of clusters at the
current cluster level (large is good), had a similar value between 2 and 3 clusters and then
continued to rise until n=15 clusters; however, 15 is not a reasonable number of clusters for this
sample. The pseudo T
2
plot, which represents the separation of the cluster most recently joined
(large is good), showed large values for 2,3 and 4 clusters. The R
2
statistics for 2, 3, and 4
clusters are R
2
= 0.486, 0.651, 0.786, respectively. Finally, cluster plots were used to visually
assess the appropriateness of the cluster assignment, as seen in Figure 1. The larger R
2
, T
2
, and
pseudo F values, as well as the spike in the CCC, suggest that 4 clusters is a better fit than 3.
However, when we visually assess the clusters in Figure 1, we can see clusters 3 and 4 represent
the highest and lowest self-management performing patients, so changes in statistics between
n=3 and n=4 clusters is due to the extreme case differences. Three clusters make sense based on
the interpretation of the measure and sample sizes of each cluster. Thus, we utilized the 3-cluster
assignment. The three clusters are defined as: consistent (n=172, 20.3%), moderate (n=573,
67.7%), and struggling (n=102, 12.0%) with self-management. Cluster names were given based
on the general pattern and trend of the patients classified to the cluster. Consistent patients
displayed higher baseline self-management values and most had a positive trend over time.
Moderate patients displayed average baseline self-management and behavior remained steady
over time. Struggling patients displayed poor baseline self-management and either remained
steady or regressed over time. Cluster size and statistics are provided in table 1.
Table 2 provides sample size, patient characteristics, and baseline statistics for all
variables included in the analysis by cluster. Most subjects were female, Latino, and mean age
greater than 50 years old.
56

Table 1. Cluster Fit Statistics
Number per Cluster Fit Statistics
No. of Clusters 1 2 3 4 CCC F t
2
R
2

2 Clusters 102 745

-12 798 727 .49
3 Clusters 102 573 172

-15 786 299 .65
4 Clusters 86 573 172 16 -6.3 1033 930 .77

Figure 1. Plots of Self-Management Intercept vs Slope by Cluster, for N = 2, 3, and 4 Clusters
(Key: 1 = struggling, 2 = moderate, 3 = consistent self-management behavior)

57

Table 2. Self-Management Behavior Cluster Statistics
Cluster Stats
Consistent
Self-Management
(n=172)
Moderate
Self-Management
(n=573)
Struggling with
Self-Management
(n=102) p
Arm

**
UC 50 (29.07) 213 (37.17) 37 (36.27)

SC 71 (41.28) 166 (28.97) 29 (28.43)

TC 51 (29.65) 194 (33.86) 36 (35.29)

Clinic

**
UC Site 1 30 (17.44) 87 (15.18) 17 (16.67)

UC Site 2 20 (11.63) 126 (21.99) 20 (19.61)

SC Site 1 25 (14.53) 60 (10.47) 15 (14.71)

SC Site 2 (Hospital) 24 (13.95) 52 (9.08) 4 (3.92)

SC Site 3 22 (12.79) 54 (9.42) 10 (9.80)

TC Site 1 31 (18.02) 122 (21.29) 24 (23.53)

TC Site 2 (Hospital) 9 (5.23) 37 (6.46) 2 (1.96)

TC Site 3 11 (6.40) 35 (6.11) 10 (9.80)

Socio-Demographics

Age 54.24 (8.61) 53.43 (9.29) 51.40 (9.31) **
Female 108 (62.79) 387 (67.54) 57 (58.88)

Latino 155 (90.64) 532 (93.01) 92 (90.20)

Married 95 (55.23) 332 (57.94) 53 (51.96)

Spanish pref. lang. 148 (86.05) 513 (89.53) 76 (74.51)

Less than HS Edu. 122 (70.93) 440 (76.92) 64 (62.75) ***
Unemployed 118 (68.60) 374 (65.27) 72 (70.59)

Cost and Utilization Measures

Predicted Future Care Cost 6471.68 (3683.48) 6674.37 (4040.76) 6244.55 (3548.01)

Healthcare Utilization Costs
α
1520.15 (4981.09) 1229.43 (3711.98) 636.67 (1601.49)

Pharmacy Costs
α
396.72 (462.87) 438.62 (605.93) 398.28 (500.80)

Number of Clinic Visits 2.95 (3.13) 2.56 (1.92) 2.36 (1.66) *
A1C Tested 161 (93.60) 532 (91.27) 92 (90.20) ***
Diabetes Measures

BMI
α
31.81 (6.41) 32.81 (7.41) 34.43 (7.57) **
A1C
α
9.23 (1.90) 9.10 (2.14) 9.12 (2.16)

Cholesterol
α
179.55 (51.07) 176.93 (48.62) 183.03 (43.09)
Age of diabetes onset 43.78 (10.07) 43.15 (10.09) 42.55 (11.12)

Insulin use 102 (59.30) 300 (52.36) 47 (46.08)

Num diabetes Complications .92 (.98) 1.31 (1.14) 1.37 (1.23) *
Whitty-9 Diabetes Symp Scale 1.44 (.46) 1.66 (.61) 1.83 (.68) ***
Diabetes Emotional Burden 2.62 (1.91) 2.99 (2.08) 3.47 (2.03) ***
Diabetes Regimen Distress 2.46 (1.94) 2.87 (2.02) 3.41 (2.12) ***
SDSCA (Self-management) 5.51 (.71) 4.33 (1.07) 2.70 (1.05) ***
Psychological and QOL Measures

PHQ-9 4.67 (4.92) 6.65 (5.85) 8.69 (6.10) ***
Dysthymia 12 (6.98) 103 (17.98) 25 (24.51) ***
58

Prev. Diagnosis of MDD 4 (2.33) 41 (7.16) 14 (13.73) ***
BSI .71 (2.31) 1.17 (2.82) 1.73 (3.53) **
SDS 1.39 (2.36) 2.05 (2.86) 2.87 (3.29) ***
SF-12 Mental QOL 53.63 (10.13) 49.75 (13.16) 45.79 (14.23) ***
SF-12 Physical QOL 46.69 (10.24) 44.47 (11.17) 41.61 (11.14) ***
Chronic pain 32 (18.60) 142 (24.78) 31 (30.39) *
Self-Rated Health 2.55 (.82) 2.31 (.80) 2.05 (.86) ***
Stress Measures

Econ. Stressors 3.84 (2.26) 3.92 (2.13) 4.61 (2.21) ***
Number of Stressors 1.97 (2.07) 2.44 (2.23) 3.26 (2.38) ***
Sum of Stress Level
α
13.40 (16.30) 17.03 (17.69) 23.23 (19.05) **
Satisfaction Measures

Stsf. with Diabetes Care 4.77 (.49) 4.70 (.61) 4.61 (.66) *
Stsf. with Depression Care 4.60 (.73) 4.50 (.80) 3.78 (.85) *
α
Indicates variable was transformed to satisfy normality for cluster mean comparisons, raw mean and
standard deviation are reported
* : p<.10, **: p<.05, ***: p<.01

The full stepwise selection predictive modeling method found 7 predictors to fit the
criteria for entry into the model: care-management group, diabetes complications, economic
stresses, less than high school education indicator, dysthymia diagnosis, age, and BMI (Hosmer-
Lemeshow GOF p=.097). The refined model was fit to a multinomial logistic regression model
to obtain the odds ratio estimates. Age and BMI were no longer significant (p>.05) and were
removed. The final model includes: care-management group, diabetes complications, economic
stresses, less than high school education indicator, and dysthymia diagnosis. Significant results
include: For patients in the SC group, the odds of performing consistent self-management are
2.16 that of performing moderate self-management. For each one unit increase in diabetes
complications, the odds of consistent self-management are .738 that of moderate self-
management. For each one unit increase in economic stress, the odds of struggling with self-
management are 1.157 that of performing moderate self-management. For patients who have less
than high school education, the odds of struggling with self-management are 0.475 that of
performing moderate self-management. Lastly, for patients who have dysthymia, the odds of
59

performing consistent self-management are .345 that of performing moderate self-management.
Complete model results can be found in table 3.
Table 3. Prediction Model Odds Ratios
Cluster Predictor Odds Ratio 95% Confidence Limits
p
Struggling
with Self-
Management
Less than HS Edu 0.475 0.301 0.748
*
Economic Stressors 1.157 1.052 1.272
*
Dysthymia Diagnosis 1.439 0.835 2.481

SC vs UC .882 .508 1.531

TC vs UC .974 .586 1.619

Diabetes Complications 1.010 .834 1.222

Consistent
with Self-
Management

SC vs UC 2.160 1.403 3.326
*
TC vs UC 1.106 .710 1.723

Diabetes Complications .738 .616 .885
*
Dysthymia Diagnosis .345 .180 .663
*
Less than HS Edu .801 .540 1.189

Economic Stressors 1.012 .932 1.098

*: indicates p<.05
Discussion
Principle Findings
This analysis found that DCAT patients are clustered by different initial self-management
behavior with moderate rates of change over time. Specifically, three clusters were identified:
consistent, moderate and struggling with self-management. Most patients displayed moderate
self-management (67.7%) with smaller proportions of patients performing consistent self-
management (20.3%) or struggling with self-management (12.0%). There are many significant
baseline differences between the clusters with a smaller subset of variables significantly
predicting diabetes self-management behavior cluster including: care-management group,
diabetes complications, economic stresses, education level, and dysthymia.
First, we saw that patients in the SC group were more consistently performing self-
management over time, and cluster proportions were similar among the UC and TC groups.
Since self-management behavior cluster represents baseline behavior pattern and trend over time,
60

this result could mean that patients in SC were performing better self-management from the
beginning or the care received via the care-management model helped patients improve self-
management over time. Patients in the SC care-management group were more likely to have
more severe diabetes due to many recruited from the hospital vs clinic [6]. Thus, it is likely a
combination of both: (1) it is critical for patients with more severe diabetes to perform consistent
self-management to maintain health and (2) the SC care-management model provided the
support for patients to perform self-management consistently. The most notable differences in
patient socio-demographics are that there were significantly younger and more highly educated
patients struggling with self-management. Among utilization measures, patients performing
consistent self-management had significantly more primary care visits and A1C testing. This
result highlights significant differences in access to care measures among self-management
behaviors, which indicates that self-management behaviors could be associated with patient
access to care. Patients struggling with self-management had the highest BMI, number of
complications, number of symptoms, and diabetes emotional burden and regimen distress.
Similarly, patients struggling with self-management displayed significantly worse mental health
than the other self-management behavior groups, reporting more severe depression, higher rates
of dysthymia, history of major depressive disorder and chronic pain, more severe functional
disability, and worse health beliefs were all reported. Lastly, patients struggling with self-
management reported higher levels of economic, social, and disease related stress. These results
empirically reveal that among patients with complex chronic conditions, there are significant
differences in patients’ physical, mental, and social health between different self-management
behavior cluster.
61

Predictive modeling told the same story as the descriptive analysis with a condensed set
of variables. First, we found that the odds of performing consistent self-management behaviors
vs moderate self-management were higher among SC patients than UC patients. As mentioned,
this result is likely a combination of patients with more severe diabetes being required to perform
more consistent self-management and the SC care-management model providing the support to
perform consistent self-management. Differences were not significant between TC and UC
groups. The odds of performing consistent self-management behaviors vs moderate self-
management decreased with increased number of diabetes complications. The odds of struggling
with self-management behavior vs moderate self-management increased with increasing
economic stress. Patients with lower education had lower odds of struggling with self-
management than perform moderate self-management. Finally, patients who were diagnosed
with dysthymia had lower odds of performing consistent self-management behaviors than
moderate self-management behaviors. Baseline patient health measures across multiple
dimensions of health, physical, mental and social, were found to significantly predict self-
management behavior cluster. These findings suggest multiple factors contribute to patient self-
management behavior patterns and trends over time, and future care-models should consider
these factors in providing self-management education and support to patients with complex
chronic conditions.
Comparison to Literature
Previous investigation has identified thematically that patients with complex health
conditions have different self-management requirements [1]; however, few studies have
empirically investigated predictive factors of self-management behaviors over time for patients
with complex health conditions. When empirical investigation has been conducted on patient
62

populations like ours or with similar research objectives, the investigations have been limited by
small sample sizes, qualitative data, or cross-sectional data. These analyses include: a qualitative
investigation of factors that are influential of self-management (n=27) which found four factors:
access to resources, diet, self-efficacy, and social support [5]. A cross-sectional regression
analysis of factors associated with self-management (n=117) found diabetes distress and
economic challenges to be negatively associated with self-management [27]. Finally, a survey
investigation of patients with type 2 diabetes in South Korea (n=200) found factors affecting
self-management differed across different levels of self-management [28]. In comparison to
these analyses, out analysis provides and empirical, data-driven, longitudinal analysis of a large
patient sample with availability of many health measures. This research provides patient self-
management behavior cluster classification, identifies baseline differences between cluster, and
identifies predictive variables of self-management cluster.
Limitations and Future Work
Due to the nature of the analysis, complete cases were required. Thus, one limitation of
this analysis was the attrition of patients in later phases of DCAT reducing the sample size and
potentially distorting proportion of patients in each self-management group. Among baseline
characteristics of missing patients, there were higher A1C levels, cholesterol levels, and
proportion of patients diagnosed with dysthymia. Additionally, the missing group has a higher
proportion of more highly education patients. Results found that more highly educated patients
had higher odds of struggling with self-management. Thus, although it was found that only 12%
of patients were struggling with self-management, there are likely many more patients struggling
with self-management who dropped out of the study.
63

Since this analysis identified many relationships between baseline measures and self-
management behavior cluster, future work could consider investigating the associations between
self-management behavior cluster and health measures longitudinally. This analysis would
provide insight into the level of association self-management behavior has with health measures
over time.
Conclusion
This research found empirical evidence of three clusters of patient self-management
behaviors over time, and that multiple health factors, including physical, mental, and social
health factors, differ across the patient self-management behavior clusters. Moreover, a subset of
physical, mental, and social factors was found to predict self-management behavior cluster. This
research furthers our understanding of different self-management behavior groups and predictors
of long-term self-management behaviors within this patient population. This knowledge will help
providers anticipate the long-term behaviors of their patients to provide appropriate levels of
support.
64

Chapter 5:
Investigation of Longitudinal Associations Between Self-Management
Behavior Cluster and Triple Aim Outcomes
Abstract
Objective: The aim of this research is to investigate longitudinal associations between self-
management behavior cluster and triple aim outcomes.
Research Design and Methods: The study design is a longitudinal multilevel regression analysis
of triple aim outcomes on self-management behavior cluster. The data used in this analysis came
from DCAT. Only a subset of the DCAT participants, those who met the inclusion criteria for the
previous chapter’s analysis, were included (n=847).
Results: Many associations between self-management behavior cluster and health measures were
at baseline. A few significant associations over time were identified between self-management
behavior cluster and health outcomes, including healthcare utilization costs, diabetes symptoms,
and mental quality of life. Additional significant longitudinal associations were found between
the subgroup of TC patients struggling with self-management over time and the health outcomes:
healthcare utilization costs, medication costs, depression symptoms, mental quality of life,
functional disability, and stress levels.
Conclusion: Health outcomes were primarily associated with self-management behavior cluster
at baseline, as patient behaviors did not change greatly over time. The care-management models
provided by DCAT moderated the associations between self-management behavior cluster and
healthcare costs, mental health, and social health outcomes over time. This knowledge can help
providers improve design and implementation of future care-management models.

65

Keywords
Self-management, depression, diabetes, Latino, care-management, safety-net, multi-level
regression modeling, longitudinal analysis
Abbreviations
DCAT: Diabetes and Depression Care-Management Adoption Trial, UC: Usual primary Care,
SC: Supported Care, TC: Technology supported Care, LAC-DHS: Los Angeles County
Department of Health Services. QOL: quality of life

66

Introduction
This research investigation is motivated by the previous chapter of this dissertation. The
investigation found that patients are clustered into three groups based on self-management
behavior pattern and trend over time: consistent, moderate, and struggling with self-management
behavior. It was also found that there are many significant differences in baseline patient
characteristics and health outcomes between the three clusters, with a small subset of baseline
variables significantly predicting self-management behavior cluster. These results left the
following research question unanswered: what is the level of association between self-
management behavior cluster and outcomes over time in this study population?
The current literature has investigated associations between self-management behavior
and health outcomes, primarily investigating improvements in outcomes as a result of self-
management education programs simultaneous with health outcome measurements [1-4].
Additionally, there are a limited set of outcomes included in the investigation. The Institute for
Healthcare Improvement’s triple aim initiative identifies three aims for healthcare interventions:
improved health outcomes, reduced costs, and improved patient experience. Current
interventions are relying on self-management as a mechanism for improved outcomes without
fully investigating the longitudinal effects of self-management on all triple aim outcomes.
Additionally, there is lack of investigation of self-management of complex chronic
conditions. A review has identified socioeconomic challenges and comorbid conditions as factors
influencing self-management of complex chronic conditions [5]. Socioeconomic challenges and
comorbid conditions often lead to prioritization of one condition over the other and conflicting
self-management information from care providers [5]. This could lead to increased healthcare
costs for patients, worse healthcare outcomes, and a negative care experience. It is possible that a
care-management intervention could work with self-management to improve these triple aim
67

outcomes, but there is currently a lack of empirical longitudinal investigation of these
relationships.
One population that this knowledge could be especially beneficial for are low-income
Latino patients with type 2 diabetes. It is important for us to understand differences in self-
management behaviors and needs in this population because Latino’s have a higher prevalence
of type 2 diabetes than non-Latino whites and face disparities in access to care [6]. A review of
community and healthcare self-management barriers among Latinos with diabetes identified that
there is a lack of work being conducted among Latino patients with diabetes with regards to
living conditions [7], and that strategies for sustaining self-management behaviors among this
population are needed [8].
The Diabetes and Depression Care-Management Adoption Trial (DCAT) offers a proper
data set for this investigation due to the sample size, the longitudinal nature of the data
collection, the enrollments of primarily low-income Latino patients, and the complex conditions
of the patients. The aim of this investigation is to answer the research question: what longitudinal
associations exist between self-management behavior cluster and triple aim outcomes? The
importance of this research is to understand longitudinal relationships with self-management
behavior trend and healthcare cost, health outcomes, and care experience to effectively apply
self-management education and support in future care-management interventions for Latino
patients with complex health conditions.
Methods
Study Design
The study design is a multi-level regression analysis of triple aim outcomes on self-management
behavior trend, as previously developed in chapter 4 of this dissertation. The data used in this
68

analysis came from DCAT. DCAT study design is described below in Data Source. Only a
subset of the DCAT participants, those who met the inclusion criteria for the previous chapter’s
analysis, were included (n=847).
Data Source
The participants in DCAT were low-income, primarily Latino patients with type 2
diabetes from the Los Angeles County Department of Health Services (LAC-DHS) [9].
Participant inclusion criteria included patients who were 18 years or older, had a working phone
number, spoke English or Spanish, and could read and understand the consent form. Exclusion
criteria included patients with baseline possible suicide ideation (PHQ-9, item 9 response in
more than half the days to nearly every day), cognitive impairment (Short Portable Mental
Health Status Questionnaire scores of <5) [10], alcohol abuse (two or more CAGE items from
the quantity-frequency index, and questions about the patient's perception of substance use) [11],
or recent lithium/antipsychotic medication use. The study protocol was reviewed by the
University of Southern California, Olive View UCLA Medical Center, and Los Angeles
Biomedical Research Institute Human Subjects Review Boards. Utilizing a quasi-experimental
comparative effectiveness design. The trial assigned 6 provider teams and their 1406 patients
seen in 8 LAC-DHS ambulatory care clinics to one of three comorbid depression care models:
Usual Care (UC), Supported Care (SC), Technology Supported Care (TC). More detailed
descriptions of the three care-management models can be found in chapter 3 of this dissertation
and the DCAT study design paper [9]. Data was collected at 6-month intervals; baseline was in-
person at the study sites and the three follow-ups were by telephone interviews with a Spanish-
English bilingual, group-blinded interviewer.

69

Measures
The dependent variables in this analysis included a variety of outcomes representative of
the IHI’s Triple Aims: reduced healthcare costs, improved outcomes, and improved patient care
experience.
Healthcare costs were assessed by: (1) healthcare utilization cost which was computed as
the sum of lab, outpatient visit, emergency room use, and inpatient visit costs, and (2) medication
cost which was included independent of the other costs and includes: insulin, oral diabetes
medication, antidepressant medication, and other prescription medication costs.
Health outcomes were assessed by many common diabetes and depression physical,
mental, and social outcomes. A physical bio measure included HbA1c in mmol/mol units.
Diabetes symptoms were measured with the Whitty 9-item questionnaire, where questions asked
patients how often they have experienced each symptom over the past month and responses
included: 1 = never, 2 = one or a few days, 3 = on several days, 4 = on most days, and 5 = every
day. The 9 items, including: abnormal thirst, blurred vision, urinated a lot of water during the
day, felt unusually hungry, felt shaky, had cold hand and feet, felt very sleepy during the day,
had feeling of pins and needles, and felt faint or fainted, were averaged to obtain the mean score
[12]. Diabetes complications measured number of complications patients experienced including:
visions, problems, loss of feeling in legs or feet, kidney problems, foot ulcer or infection,
amputation, sexual impairment, and heart attack or cardiac procedure. The diabetes distress
screening scale items included diabetes emotional burden and diabetes related distress [13].
Items were scored 1-6; higher scores indicated a higher level of diabetes distress. Depression
symptoms were measured using the PHQ-9 questionnaire. Each individual DSM-IV criteria had
possible outcomes from “0” (not at all) to “3” (nearly every day) over the last two weeks, and the
70

total PHQ-9 score equals the sum of the 9-item responses [14]. Brief Symptom Inventory score
(BSI) assessed by the brief symptoms inventory [15], scored 0-24; higher scores indicate worse
anxiety. Mental and Physical quality of life (QOL) assessed by the SF-12 Physical and Mental
component scores which are computed using the scores of twelve questions and range from 0 to
100, where a zero score indicates the lowest level of health measured by the scales and 100
indicates the highest level of health. Sheehan Disability Scale (SDS) was scored 0-30 [16,17];
higher scores indicate more functional disability. Economic Stressors were assessed by 12
general and health-related economic stresses, scored 0-12; higher scores indicate a higher level
of economic stress. General stresses measured work, unemployment, financial, marital, family,
children, caregiver, cultural, legal, immigration, illness of family, and community worry stresses.
Each of the items were recorded on a scale of 0 = “no stress” to 10 = “the most stress you can
imagine.” Both number of stresses and total stress level were included in the analysis.
Finally, patient care experience was assessed by two patient satisfaction with care items.
Specifically, the items were: (1) How satisfied are you with the clinical help you received with
emotional problems, and (2) How satisfied are you with the overall healthcare available to you
for your diabetes? The outcomes were recorded on a 5-point Likert scale where 1 indicated “very
dissatisfied” and 5 indicated “very satisfied.”
The primary independent variable was the self-management behavior trend. Self-
management behavior cluster is a nominal variable with three values: consistent, moderate, and
struggling with self-management. The variable’s development and patient group assignment is
described in the previous chapter, Chapter 4. The second independent variable of interest was
care-management group. Again, this was a nominal variable with three groups: usual primary
71

care (UC), supported care (SC), and technology supported care (TC). The care management
groups are described in further detail in Chapter 3 of this dissertation.
Analysis
The analysis method utilized was multi-level regression analysis. All models were tested
for necessity of a random intercept, random slope, and quadratic time variable. The analysis was
conducted in SAS using the mixed model procedure for continuous outcomes and the GLIMMIX
procedure for over-dispersed count outcomes. Building the models was an iterative process. To
determine the best fit for each outcome, models were run in the following order: an empty
model, a linear time model, a linear time model with random intercept, a linear time model with
random slope, and a quadratic time model. The most appropriate fitting model from this list was
then fit with all other independent variables. The dependent variables included level 1 variables,
level 2 variables, and cross-level interactions.
We tested the level 1 direct effects via linear and quadratic time variables. We tested
level 2 direct effects via self-management behavior trend and care-management group, which
were included in equation (2), the grand mean of the control groups. Lastly, we tested cross-level
interaction effects by including self-management behavior trend, care-management group, and
the self-management behavior trend and care-management interaction term in equation (1) which
will include an interaction between these level 2 variables and the level 1, linear time variable.
Thus, each dependent variable was regressed on linear time, quadratic time (when appropriate),
self-management behavior trend, care-management group, the interaction of self-management
behavior trend and time, the interaction of care-management group and time, and the interaction
of care-management group, self-management behavior trend, and time. The equations are
provided in equation set 1.
72

Equation Set 1. Multi-Level Model Equations
Level 1: Health Outcome ~ β0i + β1i(Time)ti + β2i(Time
2
)ti (1)
Level 2: β0i ~ γ00 + γ01(Care-Mgmt)i + γ02(Self-Mgmt Cluster)i + U0i (2)
β1i ~ γ10 + γ11(Care-Mgmt)i + γ12(Self-Mgmt Cluster)i
+ γ13(Care-Mgmt)i*(Self-Mgmt Cluster)i + U1i (3)
β2i ~ γ20 + U2i (4)
Results
Table 1 provides sample size, patient characteristics, and baseline statistics for all
variables included in the analysis by self-management trajectory cluster. Most subjects were
female, Latino, and mean age greater than 50 years old. Significant differences between the
clusters are discussed in Results of Chapter 4 of this dissertation.

73

Table 1. Sample Characteristics (Mean (sd) for continuous variables, Frequency (%) for discrete
variables)

Consistent Self-
Management
(n=172)
Moderate Self-
Management
(n=573)
Struggling w/
Self-Management
(n=102) p
Arm

**
UC 50 (29.07) 213 (37.17) 37 (36.27)

SC 71 (41.28) 166 (28.97) 29 (28.43)

TC 51 (29.65) 194 (33.86) 36 (35.29)

Demographics/Bio Metrics

Age 54.24 (8.61) 53.43 (9.29) 51.40 (9.31) **
Female 108 (62.79) 387 (67.54) 57 (58.88) *
Latino 155 (90.64) 532 (93.01) 92 (90.20)

Married 95 (55.23) 332 (57.94) 53 (51.96)

Spanish pref. lang. 148 (86.05) 513 (89.53) 76 (74.51) ***
BMI
α
31.81 (6.41) 32.81 (7.41) 34.43 (7.57) **
Total Cholesterol
α
179.55 (51.07) 176.93 (48.62) 183.03 (43.09)

A1C
α

Baseline 9.23 (1.90) 9.10 (2.14) 9.12 (2.16)

6 months 8.11 (1.55) 8.29 (1.72) 8.37 (1.78)

12 Months 8.43 (1.79) 8.59 (1.86) 8.57 (1.83)

18 Months 8.45 (1.78) 8.66 (1.80) 8.81 (2.04)

Socioeconomic Variables

Less than HS Edu. 122 (70.93) 440 (76.92) 64 (62.75) ***
Unemployed 118 (68.60) 374 (65.27) 72 (70.59)

Econ. Stress mean

Baseline 3.84 (2.26) 3.92 (2.13) 4.61 (2.21) ***
6 months 3.51 (2.08) 4.05 (2.23) 4.81 (2.58) ***
12 Months 3.19 (1.99) 3.87 (2.45) 4.52 (2.69) ***
18 Months 2.87 (1.92) 3.42 (2.07) 4.18 (2.43) ***
Number of Stressors

Baseline 1.97 (2.07) 2.44 (2.23) 3.26 (2.38) ***
6 months 1.66 (1.68) 2.58 (2.07) 2.97 (2.29) ***
12 Months 1.77 (1.83) 2.08 (1.72) 2.60 (1.98) ***
18 Months 1.22 (1.44) 1.75 (1.66) 2.17 (1.76) ***
Sum of Stress Level
α

Baseline 13.40 (16.30) 17.03 (17.69) 23.23 (19.05) **
6 months 10.17 (11.68) 16.09 (15.12) 21.19 (18.49) ***
12 Months 10.67 (12.97) 13.50 (12.50) 18.19 (16.32) **
18 Months 7.01 (9.78) 10.53 (11.42) 14.69 (13.65) **
Utilization and Cost Measures

A1C Tested 161 (93.60) 532 (91.27) 92 (90.20) ***
Number of Clinic Visits 2.96 (3.13) 2.56 (1.92) 2.36 (1.66)

Predicted Future Care Cost 6471.68 (3683.48) 6674.37 (4040.76) 6244.55 (3548.01)

Utilization Cost
α

74

Baseline 1520.15 (4981.08) 1229.43 (3711.98) 636.67 (1601.49)

6 months 1034.94 (2702.45) 1287.07 (3691.67) 802.77 (3287.18)

12 Months 738.09 (1987.35) 1120.53 (4116.81) 886.33 (2986.33)

18 Months 711.94 (2419.42) 534.14 (1976.66) 732.11 (2252.47)

Pharmacy Costs
α

Baseline 396.72 (462.87) 438.62 (605.93) 398.28 (500.80)

6 months 589.81 (603.10) 557.29 (744.68) 519.76 (590.91)

12 Months 498.09 (588.12) 809.53 (5851.77) 468.66 (607.54) **
18 Months 312.85 (394.34) 615.64 (5923.99) 278.21 (424.16)

Diabetes Measures

Age of diabetes onset 43.78 (10.07) 43.15 (10.09) 42.55 (11.12)

Insulin use 102 (59.30) 300 (52.36) 47 (46.08) *
Num. Diabetes Complications
β

Baseline .92 (.98) 1.31 (1.14) 1.37 (1.23) ***
6 months 1.20 (1.14) 1.50 (1.19) 1.66 (1.25) ***
12 Months .99 (1.01) 1.25 (1.10) 1.23 (1.21) **
18 Months 1.08 (1.05) 1.19 (1.16) 1.36 (1.25)

Whitty-9 Diabetes Symp Scale

Baseline 1.44 (.46) 1.66 (.61) 1.83 (.68) ***
6 months 1.40 (.42) 1.70 (.62) 1.80 (.63) ***
12 Months 1.40 (.47) 1.63 (.61) 1.77 (.59) ***
18 Months 1.47 (.51) 1.60 (.59) 1.77 (.65) ***
Diabetes Emotional Burden

Baseline 2.62 (1.91) 2.99 (2.08) 3.47 (2.03) ***
6 months 1.85 (1.47) 2.59 (1.96) 3.47 (2.16) ***
12 Months 1.73 (1.31) 2.37 (1.91) 3.37 (2.22) ***
18 Months 1.52 (1.17) 2.09 (1.70) 2.85 (2.05) ***
Diabetes Regimen Distress

Baseline 2.46 (1.94) 2.87 (2.02) 3.41 (2.12) ***
6 months 1.31 (.87) 2.07 (1.54) 2.76 (2.02) ***
12 Months 1.31 (.90) 2.01 (1.69) 3.22 (2.20) ***
18 Months 1.24 (.81) 1.72 (1.45) 2.62 (1.97) ***
Diabetes selfcare 5.51 (.71) 4.33 (1.07) 2.70 (1.05) ***
Mental Health Measures

Dysthymia Diagnosis 12 (6.98) 103 (17.98) 25 (24.51) ***
Previous Diagnosis of MDD 4 (2.33) 41 (7.16) 14 (13.73) ***
PHQ-9

Baseline 4.67 (4.92) 6.65 (5.85) 8.69 (6.10) ***
6 months 2.77 (3.87) 4.99 (5.57) 7.23 (5.87) ***
12 Months 2.97 (3.93) 5.42 (5.86) 7.76 (6.29) ***
18 Months 2.87 (3.90) 5.04 (5.45) 7.75 (6.86) ***
BSI

Baseline .71 (2.31) 1.17 (2.82) 1.73 (3.53) **
6 months 1.10 (2.42) 2.00 (3.61) 3.61 (5.00) ***
12 Months .97 (2.26) 1.84 (3.24) 2.80 (4.29) ***
75

18 Months 1.08 (2.50) 1.69 (3.09) 2.62 (4.26) ***
SDS

Baseline 1.39 (2.36) 2.05 (2.86) 2.87 (3.29) ***
6 months 1.67 (2.55) 2.40 (2.83) 3.39 (2.88) ***
12 Months 1.41 (2.35) 2.25 (3.03) 2.78 (3.33) ***
18 Months 1.31 (2.45) 1.81 (2.68) 2.81 (3.10) ***
SF-12 Mental QOL

Baseline 53.63 (10.13) 49.75 (13.16) 45.79 (14.23) ***
6 months 54.61 (8.79) 50.42 (11.21) 45.57 (12.92) ***
12 Months 54.48 (8.79) 51.70 (10.72) 47.46 (13.28) ***
18 Months 54.73 (8.44) 52.27 (10.00) 46.68 (12.64) ***
SF-12 Physical QOL

Baseline 46.69 (10.24) 44.47 (11.17) 41.61 (11.14) ***
6 months 46.01 (10.15) 43.56 (11.28) 39.97 (11.97) ***
12 Months 46.35 (9.94) 43.15 (11.37) 40.92 (10.90) ***
18 Months 46.40 (10.28) 43.82 (11.33) 41.26 (11.16) ***
Chronic pain 32 (18.60) 142 (24.78) 31 (30.39) *
Self-Rated Health 2.55 (.82) 2.31 (.80) 2.05 (.86) ***
Satisfaction Measures

Stsf. with Diabetes Care

Baseline 4.77 (.49) 4.70 (.61) 4.61 (.66) *
6 months 3.76 (.85) 4.18 (.87) 4.16 (.89) **
12 Months 4.30 (.64) 4.19 (.71) 4.13 (.84)

18 Months 4.27 (.66) 4.19 (.75) 4.18 (4.43)

Stsf. with Depression Care

Baseline 4.60 (.73) 4.50 (.80) 3.78 (.85) *
6 months 3.51 (2.08) 3.51 (.90) 4.81 (2.58) ***
12 Months 3.90 (.82) 3.84 (.80) 3.77 (.90)

18 Months 4.16 (.69) 4.03 (.75) 3.99 (.75) *
‘*’ indicates p<.10; ‘**’; indicates p<.05; ‘***’ indicates p<.01
Note: Baseline through 18-mo. reported for outcome variables, baseline only reported for all other var
α
Indicates variable was transformed to satisfy normality for cluster mean comparisons, raw mean and
standard deviation are reported
β
Indicates over-dispersed count data

Economic stresses, general stresses, medication costs, diabetes symptoms, PHQ-9, SF-12
Physical QOL, and SDS displayed a better fit with a fixed slope. All other outcomes displayed a
better fit with a random intercept, random slope model. Healthcare utilization costs, medication
costs, A1C, diabetes symptoms, PHQ-9, BSI, SDS, and satisfaction with diabetes and depression
care displayed a better fit with a quadratic time variable. All other outcomes displayed a better fit
with a linear time variable only. For most outcomes, approximately equal proportions of variance
76

were explained by clustering and the longitudinal effect of time; however, for healthcare
utilization costs, diabetes emotional burden and regimen distress, and satisfaction with care a
larger proportion of the variance was explained by the longitudinal effects of time than by
clustering. Intra-Class Correlation (ICC) values can be found in table 2.
Table 2. ICC Coefficient: Proportion of Variance Explained by Clustering
Outcome ICC
Utilization Costs .25
Medication Costs .61
A1C .54
Diab Symptoms .54
Diab. Complications .48
Diab. Emo. Burden .33
Diab. Reg. Distress .23
PHQ-9 .59
BSI .44
SF-12 Mental .48
SF-12 Physical .55
SDS .51
Economic Stressors .30
General Stresses .36
Total Stress Level .40
Satisfaction with Diabetes Care .14
Satisfaction with Depression Care .05

Among the utilization and cost outcomes, struggling with self-management was
significantly associated with lower baseline healthcare utilization costs. Over time, performing
consistent self-management was significantly associated with higher healthcare utilization costs.
Both SC and TC care-management groups were significantly associated with reduced healthcare
utilization costs over time. Furthermore, patients performing consistent self-management in the
TC group observed further reduction in utilization costs over time. Finally, performing consistent
77

self-management behaviors among SC patients was significantly associated with reduced
medication costs over time.
Among the physical health diabetes outcomes, struggling with self-management was
significantly associated with more baseline diabetes symptoms, emotional burden, and regimen
distress. Alternatively, performing consistent self-management was significantly associated with
less baseline diabetes symptoms, complications, emotional burden and regimen distress. Over
time, performing consistent self-management behavior was significantly associated with more
diabetes symptoms.
Among mental health and QOL outcomes, struggling with self-management was
significantly associated with higher baseline levels of depression and anxiety symptoms, greater
functional disability, and worse mental and physical QOL. Alternatively, performing consistent
self-management was significantly associated with less severe baseline depression and anxiety
symptoms, less functional disability, and higher mental and physical QOL. Performing consistent
self-management is significantly associated with lower levels of mental QOL over time. For
patients in the SC care-management group, performing consistent self-management was
significantly associated with less depression symptoms over time; whereas struggling with self-
management was significantly associated with greater depression symptoms over time. Patients
in the TC care-management group and struggling with self-management behaviors observed
significant negative associations with depression symptoms and functional disability (SDS), and
a positive association with mental QOL.
Among the socioeconomic health outcomes, consistent self-management was
significantly associated with lower baseline levels of economic stresses, general stresses, and
total stress level. Alternatively, struggling with self-management was associated with higher
78

baseline levels of economic stresses, general stresses, and total stress level. Significant negative
associations were found between patients who were struggling with self-management over time
in the TC group and general stresses and total stress level. Lastly, among the satisfaction
measures there were no significant associations between self-management behavior and
satisfaction with diabetes and depression care. Complete regression output can be found in table
3 and trend plots of the MLM results for select outcomes are included in figure 1 (the appendix
includes the trend plots for each outcome using the MLM results).
79

Table 3. Longitudinal Multi-Level Model Results

* indicates p<.05
α
indicates outcome was log transformed to satisfy normal distribution
Note: sample size is included for outcomes that had missing values and were regressed on less than n=847 observations
80

Figure 1. MLM Trend Plots of Select Outcomes
Healthcare Utilization Costs:
Level 1, intercept and time, and level 2, self-management cluster (struggling only) and care-management group,
direct effects were significant. Significant cross-level interactions included care-management group over time,
the consistent self-management group over time, and the three-way interaction of consistent self-management
and TC care-management over time.

Depresion Symptoms (PHQ-9):
Level 1, intercept and time, and level 2, self-management cluster, direct effects were significant. Significant
cross-level interactions included the TC care-management group over time and the three-way interactions
between self-management cluster and care-management group over time (excluding consistent TC patients).

81

Sheehan Disability Scale:
Level 1, intercept and time, and level 2, self-management cluster, direct effects were significant. Significant
cross-level interactions included the three-way interaction of patients struggling with self-management in the
TC care-management group over time.

General Stresses (Level):
Level 1, intercept and time, and level 2, self-management cluster and care-management group (SC only), direct
effects were significant. Significant cross-level interactions included the three-way interaction of patients
struggling with self-management in the TC care-management group over time.
82

Discussion
Principle Findings
This analysis identified many associations between self-management behavior patterns
and health outcomes in our study population. Many of the significant associations were identified
between self-management behavior trend and baseline outcomes; however, some associations
were identified between a subgroup of patients, those in the TC group who were struggling with
self-management, and mental and social health outcomes over time.
Most of the significant results with self-management trend were with the intercept,
without significant interaction with time. This was expected, as it was found in the previous
analysis that the most significant factor in the self-management behavior clustering was patient
baseline self-management level. Among these results, effects were as expected. Performing
consistent self-management was associated with less diabetes symptoms, diabetes complications,
diabetes emotional burden and regimen distress, depression and anxiety symptoms, functional
disability, and a higher mental and physical QOL. Alternatively, struggling with self-
management was associated with greater diabetes symptoms, diabetes complications, diabetes
emotional burden and regimen distress, depression and anxiety symptoms, functional disability,
and a lower mental and physical QOL. The small subset of outcomes which were significantly
associated with self-management behavior trend over time included: positive associations with
healthcare utilization costs and diabetes symptoms, and a negative association with mental
quality of life. The positive association with healthcare utilization over time is likely related to
patients in the consistent self-management cluster having a higher mean number of primary care
visits. With more doctor visits, it is logical that expenses are greater. The positive association
with diabetes symptoms could be due to patients who were experiencing symptoms turning to
self-management in attempt to reduce symptoms. The negative association with mental quality of
83

life could be due to patients who feel better believing they do not need to perform as much self-
management, because they believe their disease status to be less severe.
Interesting results included the three-way interactions of self-management cluster, care-
management group and time. Patients performing consistent self-management in the intervention
groups had significant associations with reduced costs over time, medication costs for patients in
SC and healthcare utilization costs for patients in TC. The remaining significant association
between SC patients and self-management trend was with depression symptoms; patients
performing consistent self-management were associated with improved depression symptoms
over time, whereas struggling with self-management was associated with worsened depression
symptoms over time. This result could be due to patients in SC having more severe conditions
than patients in the other groups, or SC patients responding to transitions of care. This result
brings awareness to the need for a better understanding of the dynamic relationships between
patient behaviors and care-management programs over time, especially for patients with complex
conditions. Finally, among patients in the TC group who were struggling with self-management,
significant associations were found with improved depression symptoms, improved mental QOL,
reduced functional disability, and reduced stress levels over time. These results highlight the
ability for self-management behaviors and care-management programs to work together over
time to achieve triple aim outcomes.
Comparison to Literature
Previous longitudinal investigations of associations between self-management behaviors
and health outcomes among patients with type 2 diabetes have focused on the outcomes of A1C
without investigation of all triple aim outcomes [18,19]. Additionally, these analyses have been
limited by small sample sizes or pre-post data sources. Results of these analyses have found
84

long-term associations between self-management behaviors and changes in A1C. Our analysis
contributes a longitudinal, empirical investigation of associations between self-management
behavior and care-management group on triple aim outcomes. We identify significant
associations between self-management behavior, care-management group and many different
outcomes, as well as moderating effects of care-management groups. This understanding will
help inform providers in future care-management assignment and implementation.
Limitations and Future Work
Due to the nature of the analysis, complete cases were required. Thus, one limitation of
this analysis was the attrition of patients in later phases of DCAT reducing the sample size and
potentially distorting proportion of patients in each self-management group. Among baseline
characteristics of missing patients, there were higher A1C levels, cholesterol levels, and
proportion of patients diagnosed with dysthymia. Additionally, the missing group has a higher
proportion of more highly education patients. Results found that more highly educated patients
had higher odds of struggling with self-management. Thus, although it was found that only 12%
of patients were struggling with self-management, there are likely many more patients struggling
with self-management who dropped out of the study. A second limitation of this analysis is some
outcome variables had missing data which further reduced the sample size. A1C had 3% missing,
healthcare utilization cost had 16% missing, and medication cost had 29% missing. All outcomes
were important to include, A1C is a primary measure of diabetes severity and cost measures
provide insight into patient use of care and prescription regimen adherence.
Future work could consider further exploration of associations between self-management
behavior cluster and health outcomes over time on other patient populations. It would be of
interest to see if the associations found in our study population are also found in other patient
85

populations. This investigation would identify if associations between self-management behavior
cluster and health outcomes are similar or different across different patient populations.
Conclusion
Health outcomes were primarily associated with self-management behavior cluster at
baseline, as patient behaviors did not change greatly over time. The care-management models
provided by DCAT moderated the associations between self-management behavior cluster and
healthcare costs, mental health, and social health outcomes over time. This knowledge can help
providers improve design and implementation of future care-management models.

86

Appendix. Health Outcome Trend Plots Using Multi-Level Model Results
Healthcare Utilization Costs:
Level 1, intercept and time, and level 2, self-management cluster (struggling only) and care-management group,
direct effects were significant. Significant cross-level interactions included care-management group over time,
the consistent self-management group over time, and the three-way interaction of consistent self-management
and TC care-management over time.

Medication Costs:
Level 1, intercept and time, and level 2, care-management group (TC only), direct effects were significant.
Significant cross-level interactions included the three-way interaction of consistent self-management and the SC
care-management group over time.

87

HbA1c:
Level 1, intercept and time, and level 2, care-management group, direct effects were significant. Significant
cross-level interactions included the TC care-management group over time.

88

Diabetes Symptoms:
Level 1, intercept, and level 2, self-management cluster and care-management group (SC only), direct effects
were significant. Significant cross-level interactions included consistent self-management behavior over time.

Diabetes Complications:
Level 1, intercept, and level 2, self-management cluster (consistent only), direct effects were significant. There
were no significant cross-level interactions.

89

Diabetes Emotional Burden (1 = Not a problem to 6 = A serious problem):
Level 1, intercept and time, and level 2, self-management cluster and care-management group (SC only), direct
effects were significant. Significant cross-level interactions included care-management groups over time.

Diabetes Regimen Distress (1 = Not a problem to 6 = A serious problem):
Level 1, intercept and time, and level 2, self-management cluster and care-management group (SC only), direct
effects were significant. Significant cross-level interactions included care-management groups over time.

90

Depresion Symptoms (PHQ-9):
Level 1, intercept and time, and level 2, self-management cluster, direct effects were significant. Significant
cross-level interactions included the TC care-management group over time and the three-way interactions
between self-management cluster and care-management group over time (excluding consistent TC patients).

Anxiety Symptoms (BSI):
Level 1, intercept and time, and level 2, self-management cluster and care-management group (TC only), direct
effects were significant. There were no significant cross-level interactions.

91

SF-12 Mental:
Level 1, intercept and time, and level 2, self-management cluster and care-management group (SC only), direct
effects were significant. Significant cross-level interactions included consistent and struggling with self-
management behaviors over time, and the three-way interaction between patients struggling with care-
management in the TC group over time.

SF-12 Physical:
Level 1, intercept, and level 2, self-management cluster and care-management group (SC only), direct effects
were significant. Significant cross-level interactions included SC care-management over time.

92

Sheehan Disability Scale:
Level 1, intercept and time, and level 2, self-management cluster, direct effects were significant. Significant
cross-level interactions included the three-way interaction of patients struggling with self-management in the
TC care-management group over time.

Economic Stresses:
Level 1, intercept and time, and level 2, self-management cluster (struggling only), direct effects were
significant. Significant cross-level interactions included SC care-management over time.

93

General Stresses (Number):
Level 1, intercept and time, and level 2, self-management cluster and care-management group (SC only), direct
effects were significant. Significant cross-level interactions included the three-way interaction of patients
struggling with self-management in the TC care-management group over time.

General Stresses (Level):
Level 1, intercept and time, and level 2, self-management cluster and care-management group (SC only), direct
effects were significant. Significant cross-level interactions included the three-way interaction of patients
struggling with self-management in the TC care-management group over time.

94

Satisfaction with Depression Care (5-point Likert Scale, 1=Very Dissatisfied to 5=Very Satisfied):
Level 1, intercept and time, and level 2, care-management group, direct effects were significant. Significant
cross-level interactions included the SC care-management group over time.

Satisfaction with Diabetes Care (5-point Likert Scale, 1=Very Dissatisfied to 5=Very Satisfied):
Level 1, intercept and time, and level 2, care-management group, direct effects were significant. Significant
cross-level interactions included the SC care-management group over time.

95

Chapter 6

Conclusion

This dissertation used patient reported outcomes to answer four research questions: How
does patient satisfaction differ across care-management groups, and is patient satisfaction a valid
indicator of quality of care experience? Is self-management behavior a mechanism for improved
disease symptoms or are disease symptom improvements direct effects of care-management
models? What are the different self-management behavior groups in this population, and what
factors predict self-management behavior group in this population? Lastly, what longitudinal
associations exist between self-management behavior group and triple aim outcomes, and are
associations moderated by care-management model?
Chapter 2, the investigation of patient satisfaction, found patient satisfaction with
depression care is significantly higher in the comorbid depression care-management groups, and
was not influenced by patient characteristics or disease symptoms. Patient satisfaction was
influenced by aspects of care provided by the intervention, reflecting care experience. Chapter 3,
the longitudinal investigation of relationships between self-management, disease symptoms, and
care-management model, found self-management on its own is not enough to improve symptoms
in this population. Care-management was helpful in reducing symptoms while in place, but
improvements were not sustained after patient graduated or left the program. Chapter 4,
investigation of self-management behavior trends, found patients in DCAT can be clustered by
initial self-management differences and rates of change over time into three groups of different
self-management behavior clusters: consistent, moderate, and struggling with self-management
over time. A care-management model (i.e., SC) along with specific physical, mental, and social
health variables are predictive of self-management behavior cluster. Chapter 5, the longitudinal
96

investigation of associations between self-management behavior cluster and triple aim outcomes,
found many baseline but not longitudinal associations between self-management behavior cluster
and healthcare costs and health outcomes. The results also indicate significant although mixed
effects of care-management models for the triple aim outcomes over time. There are significant
interaction effects of TC and the struggling in self-management cluster on improving mental and
social health outcomes.
The literature states that longitudinal analyses are needed to determine if PRO
instruments are responsive to changes or differences in care and disease status [1]. A robust way
to assess the responsiveness is to use multiple methods of measurement, including: clinical
endpoints, patient-rated global improvement, change in other PRO measures, or assessing
combinations of clinical and patient-based outcomes [1]. In the analyses included in this
dissertation, all methods are included in the investigation of PRO’s by including objective
clinical measures in the analyses, utilizing longitudinal PRO data, and investigating both
objective clinical measures and PRO measures in the same analysis. Overall, the results of these
analyses of PRO’s agree with traditional clinical measures such as A1C. In some cases,
improvements in PRO’s were more significant than traditional measures. Since behaviors and
health outcomes were both reported by the patient the correlation between the two is higher. This
is a limitation that is discussed below. Another explanation is that many PRO’s have the ability
to change more quickly than some clinical measures such as A1C which is measured once every
3-months. The results suggest PRO’s can capture similar information as traditional clinical
measures and can be used to inform providers of patient condition and response to care-
management in the absence of face-to-face care.
97

A recent report discusses our abilities for PRO data collection and our current
understanding of PRO’s, as well as the ability for PRO’s to turn “patient-centered care” into a
reality via PRO ability to support the analysis of effectiveness and quality of care [2]. This
dissertation provides a series of analyses which demonstrate PRO’s ability to improve patient-
centered care. Specifically, this research advances our knowledge of different patient behaviors,
predictors of behaviors, and relationships between self-management behaviors and triple aim
outcomes over time for safety net patients. This knowledge can be used by safety-net providers
to inform them of the relative proportion of each patient behavior group they can anticipate
treating, to predict patient behavior over time given baseline assessment, and to assign
appropriate level care-management based on patients predicted behavior over time.
One limitation of this dissertation is the use of complete cases (non-missing data for any
phase baseline through 18-months) in each analysis which reduced the sample size used. It was
not a valid option to interpolate missing data as data was not missing at random. In a comparison
of complete to incomplete cases for each analysis, differences in physical, mental and social
health were observed between the two groups. In the patient satisfaction analysis (chapter 2)
sample, complete cases had more diabetes, depression, and economic and general stressors. This
comparison shows that the patients with more severe diseases and more external stressors
remained. These patients are likely whose satisfaction with care would be influenced by disease
status or characteristics, yet we do not find that in our analysis. Thus, our results show patient
satisfaction is a valid measure of quality of care experience. In the samples in the following
analyses (chapter 3-5), complete cases had fewer diabetes symptoms, depression symptoms, and
economic and general stressors. This comparison shows that our results are more conservative
than if complete data for all patients had been available. We find that care-management model
98

effects interact with self-management behavior trend to improve outcomes. If patients with more
severe diseases remained in the analysis it would have increased the observed effects in our
results. Among the objective measures of diabetes and depression in these chapters, missing
cases had significantly higher A1C levels, cholesterol levels, proportion of patients diagnosed
with dysthymia, and proportion of more highly educated patients. According to the results that
patients who are more highly educated have higher odds of struggling with self-management,
there is likely an underestimate of patients who are struggling with self-management in these
analyses. Strategies for retention of patients who are struggling with self-management could be
an important aim for safety-net providers in the future.
Overall, this dissertation analyzed PRO’s longitudinally to further the understanding of
newer care-management models to help safety-net healthcare systems optimize performance of
comorbid diabetes and depression. The findings of these analyses show patient satisfaction with
care increased when care-management models were available, but the positive experience with
SC negatively affected their satisfaction with usual care after the care-management program
ended. Although self-management did not improve disease symptoms at the population level, the
subgroup of TC patients who struggled with self-management improved mental and social
outcomes. This PRO-generated knowledge can help safety-net providers optimize their design
and implementation of care-management models to realize the triple aims.

99

References
Chapter 1 References
1. Milani R, Lavie C. Health Care 2020: Reengineering Health Care Delivery to Combat
Chronic Disease. The American Journal of Medicine 2015; 128(4): 337-343.
2. Hanlon P, Daines L, Campbell C, et al. Telehealth Interventions to Support Self-
Management of Long-Term Conditions: A Systematic Metareview of Diabetes, Heart
Failure, Asthma, Chronic Obstructive Pulmonary Disease, and Cancer. Journal of
Medical Internet Research 2017; 19(5): E172.
3. Kim K, Choi J, Choi, E, et al. Effects of Community-Based Health Worker Interventions
to Improve Chronic Disease Management and Care Among Vulnerable Populations: A
Systematic Review. American Journal of Public Health 2016; 106(4): E3-28.
4. Kopach-Konrad R, Criswell M, Hasan I, et al. Applying Systems Engineering Principles
in Improving Health Care Delivery. Journal of General Internal Medicine 2007;
22(Suppl3): 431.
5. National Association of Social workers. Read the Code of Ethics.
https://www.socialworkers.org/About/Ethics/Code-of-Ethics/Code-of-Ethics-English
(2019, accessed 17 April 2019).
6. USC Viterbi School of Engineering. Engineering +.
ttps://viterbischool.usc.edu/engineering-plus/ (2018, accessed 16 April 2019).
7. Stellefson M, Dipnarine K, Stopka C. The Chronic Care Model and Diabetes
Management in US Primary Care Settings: A Systematic Review. Preventing Chronic
Diseases 2013; 10: E26.
8. Neumeyer-Gromen A, Lampert T, Stark K, et al. Disease Management Programs for
Depression: A Systematic Review and Meta-Analysis of Randomized Controlled Trials.
Medical Care 2004; 42(12): 1211-1221.
9. Revicki D, Hays R, Cella D, et al. Recommended Methods for Determining
Responsiveness and Minimally Important Differences for Patient-Reported Outcomes.
Journal of Clinical Epidemiology 2008; 61(2): 102-109.
10. Wu S, Ell K, Gross-Schulman SG, et al. Technology-facilitated depression care
management among predominantly Latino diabetes patients within a public safety net
care system: comparative effectiveness trial design. Contemp Clin Trials 2014;37(2):342-
354.
11. Hay J, Lee P, Jin H, et al. Cost-Effectiveness of a Technology-Facilitated Depression
Care Management Adoption Model in Safety-Net Primary Care Patients with Type 2
Diabetes. Value Health 2018; 21(5): 561-568.
100

12. Wu S, Ell K, Jin H, et al. Comparative Effectiveness of a Technology-Facilitated
Depression Care Management Model in Safety-Net Primary Care Patients with Type 2
Diabetes: 6-Month Outcomes of a Large Clinical Trial. J Med Internet Res 2018; 20(4):
e147.
13. Ramirez M, Wu S, Jin H, et al. Automated Remote Monitoring of Depression:
Acceptance Among Low-Income Patients in Diabetes Disease Management. J Med
Internet Res Mental Health 2016; 3(1): e6.
14. Vidyanti I, Wu B, Wu S, et al. Low-Income Minority Patient Engagement with
Automated Telephonic Depression Assessment and Impact on Health Outcomes. Qual
Life Res 2014; 24(5): 1119–1129.
15. Lanting LC, Joung IMA, Mackenbach JP, Lamberts SWJ, Bootsma AH. Ethnic
differences in mortality, end-stage complications, and quality of care among diabetic
patients: a review. Diabetes Care 2005;28(9):2280–8 10.2337/diacare.28.9.2280
16. Bijlsma-Ruttte A, Rutters F, Elders P, et al. Socio-Economic Status and HbA1c in Type 2
Diabetes: A Systematic Review and Meta-Analysis. Diabetes Metab Res Rev.
2018;34:e3008. https://doi.org/10.1002/dmrr.3008

101

Chapter 2 References
1. Niznik J, He H, Kane-Gill S. Impact of Clinical Pharmacist Services Delivered Via
Telemedicine in the Outpatient or Ambulatory Care Setting: A Systematic Review. Res
Social Adm Pharm 2018; 14(8): 707-717.
2. Shah M, Wasserman E, Gillespie S, et al. High-Intensity Telemedicine Decreases Emergency
Department Use for Ambulatory Care Sensitive Conditions by Older Adult Senior Living
Community Residents. J Am Med Dir Assoc 2015; 16(12): 1077-1081.
3. Chen Q, Carbone E. Functionality, Implementation, Impact, and the Role of Health Literacy
in Mobile Phone Apps for Gestational Diabetes: Scoping Review. J Med Internet Res 2017;
2(2).
4. Darkins, Adam, et al. Reduced Cost and Mortality Using Home Telehealth to Promote Self-
Management of Complex Chronic Conditions: A Retrospective Matched Cohort Study of
4,999 Veteran Patients. Telemed J E Health 2015; 21(1): 70–76.
5. Luxton, D., Pruitt, D., Wagner, A., et al. Home-Based Telebehavioral Health for U.S.
Military Personnel and Veterans with Depression: A Randomized Controlled Trial. J.
Consult. Clin. Psychol 2016; 84(11): 923-934.
6. Orozco-Beltran D, Sanchez-Molla M, et al. Telemedicine in Primary Care for Patients with
Chronic Conditions: The ValCronic Quasi-Experimental Study. J Med Internet Res 2017;
19(12): e400.
7. Polisena J, Tran K, Cimon K, et al. Home telemonitoring for congestive heart failure: a
systematic review and meta-analysis. J Telemed Telecare 2010; 16(2): 68-76.
8. Rasmussen, Ole W, et al. Telemedicine Compared with Standard Care in Type 2 Diabetes
Mellitus: A Randomized Trial in an Outpatient Clinic. J Telemed Telecare 2016; 22(6): 363–
368.
9. Watts, Sharon A, et al. Improved Glycemic Control in Veterans with Poorly Controlled
Diabetes Mellitus Using a Specialty Care Access Network-Extension for Community
Healthcare Outcomes Model at Primary Care Clinics. J Telemed Telecare 2015; 22(4): 221–
224.
10. Al-Abri R, Al-Balushi A. Patient Satisfaction as a Tool Towards Quality Improvement.
Oman Med J 2014; 29(1): 3-7.
11. Chow A, Mayer E, et al. Patient-reported Outcome Measures: The Importance of Patient
Satisfaction in Surgery. Surgery 2009; 146(3): 435-43.
12. Gros, Daniel F, et al. Treatment Satisfaction of Home-Based Telehealth versus in-Person
Delivery of Prolonged Exposure for Combat-Related PTSD in Veterans. J Telemed Telecare
2016; 24(1): 51–55.
102

13. Kruse CS, Krowski N, Rodriguez B, et al. Telehealth and Patient Satisfaction: a Systematic
Review and Narrative Analysis. BMJ Open 2017; 7(8): 1-12.
14. Mair F, Whitten P. Systematic Review of Studies of Patient Satisfaction with Telemedicine.
BMJ 2000; 320:1517-20.
15. Hulsbosch A, Nugter M, Tarnis P, et al. Videoconferencing in a Mental Health Service in
The Netherlands: A Randomized Control Trial in Patient Satisfaction and Clinical Outcomes
for Outpatients with Severe Mental Illness. J Telemed Telecare 2017; 23(5): 513-520.
16. Pani D, Piga M, Barabino G, et al. Home Tele-Rehabilitation for Rheumatic Patients: Impact
and Satisfaction of Care Analysis. J Telemed Telecare 2017; 23(2): 292-300.
17. Wu S, Ell K, Gross-Schulman SG, et al. Technology-Facilitated Depression Care-
Management Among Predominantly Latino Diabetes Patients within a Public Safety Net
Care System: Comparative Effectiveness Trial Design. Contemp Clin Trials 2014; 37(2):
342-354.
18. Wu S, Vidyanti I, Liu P, et al. Patient-Centered Technological Assessment and Monitoring of
Depression for Low-Income Patients. J Ambulatory Care Manage 2014; 37(2): 138-147.
19. Hay J, Lee P, Jin H, et al. Cost-Effectiveness of a Technology-Facilitated Depression Care
Management Adoption Model in Safety-Net Primary Care Patients with Type 2 Diabetes.
Value Health 2018; 21(5): 561-568.
20. Wu S, Ell K, Jin H, et al. Comparative Effectiveness of a Technology-Facilitated Depression
Care Management Model in Safety-Net Primary Care Patients with Type 2 Diabetes: 6-
Month Outcomes of a Large Clinical Trial. J Med Internet Res 2018; 20(4): e147
21. Ramirez M, Wu S, Jin H, et al. Automated Remote Monitoring of Depression: Acceptance
Among Low-Income Patients in Diabetes Disease Management. J Med Internet Res Mental
Health 2016; 3(1): e6.
22. Vidyanti I, Wu B, Wu S, et al. Low-Income Minority Patient Engagement with Automated
Telephonic Depression Assessment and Impact on Health Outcomes. Qual Life Res 2014;
24(5): 1119–1129.
23. Pfeiffer E. A Short Portable Mental Status Questionnaire for the Assessment of Organic
Brain Deficit in Elderly Patients. J Am Geriatr Soc 1975; 23(10):433-41.
24. Ewing J. Detecting Alcoholism the CAGE Questionnaire. JAMA. 1984; 252(14): 1905-1907.
25. Koenke K, Spitzer RL, Williams JB. The PHQ-9: Validity of a Brief Depression Severity
Measure. J Gen Intern Med 2001; 16(9): 606-13.
26. Whitty P, Steen N, Eccles M, et al. A New Self-Completion Outcome Measure for Diabetes:
is it Responsive to Change? Qual Life Res 1997; 6(5): 407-13.
103

27. Batbaatar E, Dorjdagva J, Ariunbat L, et al. Determinants of Patient Satisfaction: A
Systematic Review. Perspectives in Public Health 2016; 137(2): 89-101.
28. Sullivan G, Artino A. Analyzing and Interpreting Data from Likert-Type Scales. JGME
2013; 5(4): 541-542.
29. Linder-Pelz, S. Toward a Theory of Patient Satisfaction. Social Science & Medicine 1985;
16(5): 577–582.
30. Katon W, Lin E, Von Korff M, et al. Collaborative Care for Patients with Depression and
Chronic Illnesses. N Engl J Med 2010; 363: 2611-2620.
31. Piette J, Marinec N, Gallegos-cabriales E, et al. Spanish-Speaking Patients’ Engagement in
Interactive Voice Response (IVR) Support Calls for Chronic Disease Self-Management: Data
from Three Countries. J Telemed Telecare 2013; 19(2): 89-94.
32. Salazar, L. The Effect of Patient Self-Advocacy on Patient Satisfaction: Exploring Self-
Compassion as a Mediator. Journal Communication Studies 2018; 69(5): 567-582.

104

Chapter 3 References
1. Li C, Ford ES, Strine TW, et al. Prevalence of Depression Among US Adults with
Diabetes: Findings from the 2006 Behavioral Risk Factor Surveillance System. Diabetes
Care 2008;31(1):105–7.
2. Katon W, Rutter C, Simon G, et al. The Association of Comorbid Depression with
Mortality in Patients with Type 2 Diabetes. Diabetes Care 2005;28:2668-2673.
3. Golden S, Lazo M, Carnethon M, et al. Examining a Bidirectional Association Between
Depressive Symptoms and Diabetes. JAMA 2008;299(23):2751-9.
4. Semenkovich K, Brown M, Svrakic D, et al. Depression in Type 2 Diabetes Mellitus:
Prevalence, Impact, and Treatment. Drugs 2015;75(6):577-587.
5. Callaghan P. Exercise: A Neglected Intervention in Mental Health Care? Journal of
Psychiatric and Mental Health Nursing 2004; 11(4): 476-483.
6. Jacka F, O’Niel A, Opie R, et al. A Randomized Controlled Trial of Dietary
Improvement for Adults with Major Depression (the SMILES Trial). BMC Medicine
2017; 15(23): online.
7. Polivy J, Herman CP. Mental Health and Eating Behaviors: A Bi-directional Relation.
Canadian Journal of Public Health 2005; 96(suppl 3): S43-S46.
8. Sumlin LL, Garcia TJ, Brown SA, et al. Depression and Adherence to Lifestyle Changes
in Type 2 Diabetes: A Systematic Review. Diabetes Educ 2014;40:731–744
pmid:24939883.
9. Ory M, Ahn S, Jian L, et al. Successes of a National Study of the Chronic Disease Self-
Management Program Meeting the Triple Aim of Health Care Reform. Medical Care.
2013;51(11):992-998.
10. Stenberg U, Haaland-Overby M, Fredriksen K, et al. A Scoping Review of the Literature
on Benefits and Challenges of Participating in Patient Education Programs Aimed at
Promoting Self-Management for People Living with Chronic Illness. Patient Education
and Counseling 2016;99(11):1759-71.
11. Markle-Reid M, Ploeg J, Fraser K, et al. Community Program Improves Quality of Life
and Self-Management in Older Adults with Diabetes Mellitus and Comorbidity. Clinical
Investigation 2018; 66(2): 263-73.
12. Powers M, Bardsley J, Cypress Marjorie, et al. Diabetes Self-Management Education and
Support in Type 2 Diabetes: A Joint Position Statement of the American Diabetes
Association, the American Association of Diabetes Educators, and the Academy of
Nutrition and Dietetics. The Diabetes Educator 2017;43(1):40-53.
105

13. Naranjo D, Hessler D, Deol r, et al. Health and Psychosocial Outcomes in U.S. Adult
Patients with Diabetes from Diverse Ethnicities. Current Diabetes Reports 2012; 12(6):
729-738.
14. Ricci-Cabello I, Ruiz-Pérez I, RojasGarcia A, Pastor G, Rodríquez-Barranco M,
Goncalves DC. Characteristics and Effectiveness of Diabetes Self-Management
Educational Programs Targeted to Racial/ Ethnic Minority Groups: A Systematic
Review, Meta-Analysis and Meta-Regression. BMC Endocr Disord 2014;14:60.
15. Rosal MC, Ockene IS, Restrepo A, White MJ, Borg A, Olendzki B, Scavron J, Candib L,
Welch G, Reed G: Randomized Trial of a Literacy-Sensitive, Culturally Tailored
Diabetes Self-Management Intervention for Low-Income Latinos: Latinos en Control.
Diabetes Care. 2011, 34 (4): 838-844.
16. Two FJ, Kieffer EC, Palmisano G, et al. Racial and Ethnic Approaches to Community
Health (REACH) Detroit partnership: improving diabetes-related outcomes among
African American and Latino adults. Am J Public Health. 2005, 95 (9): 1552-1560.
17. Wu S, Ell K, Gross-Schulman SG, et al. Technology-Facilitated Depression Care-
Management Among Predominantly Latino Diabetes Patients within a Public Safety Net
Care System: Comparative Effectiveness Trial Design. Contemp Clin Trials
2014;37(2):342-354.
18. Pfeiffer E. A Short Portable Mental Status Questionnaire for the Assessment of Organic
Brain Deficit in Elderly Patients. J Am Geriatr Soc 1975;23(10):433-441.
19. Ewing J.A. Detecting Alcoholism: the CAGE questionnaire. JAMA 1984;252:1905-1907.
20. Toobert D, Hampson S, Glasglow R. The Summary of Diabetes Self-Care Activities
Measure. Diabetes Care 2000;23:943-950.
21. Koenke K, Spitzer RL, Williams JB. The PHQ-9: Validity of a Brief Depression Severity
Measure. J Gen Intern Med 2001;16(9):606-13.
22. Whitty P, Steen N, Eccles M, et al. A New Self-Completion Outcome Measure for
Diabetes: is it Responsive to Change? Qual Life Res 1997;6(5):407-13.
23. Lange L, Piette J. Perceived Health Status and Perceived Diabetes Control: Psychological
Indicators and Accuracy. Journal of Psychosomatic Research 2005;58(2):129-137.
24. DeSalvo K, Fan V, McDonnel M, Fihn S. Predicting Mortality and Healthcare Utilization
with a Single Question. Health Services Research. 2005;40(4):1234-46.
25. Wu S, Ell K, Jin H, et al. Comparative Effectiveness of a Technology-Facilitated
Depression Care Management Model in Safety-Net Primary Care Patients with Type 2
Diabetes: 6-Month Outcomes of a Large Clinical Trial. J Med internet Res. 2018; 20(4):
e147.
106

26. Van Hecke A, Heinen M, Fernandez-Ortega P, et al. Systematic Literature Review on
Effectiveness of Self-Management Support Interventions in Patients with Chronic
Conditions and Low Socio-Economic Status. Journal of Advanced Nursing 2016;73(4):
775-793.
27. Pena-Purcell N, Han G, Smith M, et al. Impact of Diabetes Self-Management Education
on Psychological Distress and Health Outcomes Among African Americans and
Hispanics/Latinos With Diabetes. Diabetes Spectrum 2019; 32(2): Advanced Online
Publication https://spectrum.diabetesjournals.org/content/early/2019/07/05/ds18-0081
28. Tang T, Funnell M, Oh M. Lasting Effects of a 2-Year Diabetes Self-Management
Support Intervention: Outcomes at 1-Year Follow-Up. Preventing Chronic Disease
2012;9:E109.
29. Wong K, Wong F, Yeung W, Chang K. The Effect of Complex Interventions on
Supporting Self-Care Among Community-Dwelling Older Adults: A Systematic Review
and Meta-Analysis. Age and Ageing 2018;47(2):185-193.
30. Lin E, Katon W, Von Korff M, et al. Relationship of Depression and Diabetes Self-Care,
Medication Adherence, and Preventive Care. Diabetes Care 2004;27(9):2154-2160.
31. Tang PC, Overhage JM, Chan AS, et al. Online Disease Management of Diabetes:
Engaging and Motivating Patients Online with Enhanced Resources-Diabetes
(EMPOWER-D), a Randomized Controlled Trial. Journal of the American Medical
Informatics Association 2012;20(3):526-534.
32. Lamers F, Jonkers C, Bosma H, et al. Treating Depression in Diabetes Patients: Does a
Nurse-Administered Minimal Psychological Intervention Affect Diabetes-Specific
Quality of Life and Glycaemic Control? A Randomized Controlled Trial. Journal of
Advanced Nursing 2011;67(4):788-779.
33. Hu J, Amirehsani K, Wallace D, et al. A Family-Based, Culturally Tailored Diabetes
Intervention for Hispanics and Their Family Members. The Diabetes Educator 2016;
42(3): 299-314.
34. Wen L, Shepherd M, Parchman M. Family Support, Diet, and Exercise Among Older
Mexican Americans with Type 2 Diabetes. The Diabetes Educator 2004; 30(6): 980-993.
35. Beverly E, Miller C, Wray L. Spousal Support and Food-Related Behavior Change in
Middle-Aged and Older Adults Living with Type 2 Diabetes. Health Education and
Behavior 2006; 35(5):707-720.
36. Mayberry L, Harper K, Osborn C. Family Behaviors and Type 2 Diabetes: What to
Target and How to Address in Interventions for Adults with Low Socioeconomic Status.
Chronic Illness 2016; 12(3): 199-215.
107

37. Kwasnicka D, Dombrowski S,White M, et al. Theoretical Explanations for Maintenance
of Behaviour Change: A Systematic Review of Behaviour Theories. Health Psychology
Review 2016; 10(3): 277-296.

108

Chapter 4 References
1. Gobeil-Lavoie AP, Chouinard MC, Danish A, et al. Characteristics of Self-Management
Among Patients with Complex Health Needs: a Thematic Analysis Review. BMJ Open 2019;
9(5): e028344. doi:10.1136/ bmjopen-2018-028344
2. Lanting LC, Joung IMA, Mackenbach JP, et al. Ethnic Differences in Mortality, End-Stage
Complications, and Quality of Care Among Diabetic Patients: a Review. Diabetes Care
2005;28(9):2280–8 10.2337/diacare.28.9.2280
3. Lopez-Class M, Jurkowski J. The Limits of Self-Management: Community and Health Care
System Barriers Among Latinos With Diabetes, Journal of Human Behavior in the Social
Environment 2010; 20(6): 808-826. DOI: 10.1080/10911351003765967
4. Rosal M, Ockene I, Restrepo A, et al. Randomized Trial of a Literacy-Sensitive, Culturally
Tailored Diabetes Self-Management Intervention for Low-Income Latinos. Diabetes Care
2011, 34(4): 838-844.
5. Ramal, E., Petersen, A.B., Ingram, K.M. et al. Factors that Influence Diabetes Self-
Management in Hispanics Living in Low Socioeconomic Neighborhoods in San Bernardino,
CaliforniaJ Immigrant Minority Health (2012) 14: 1090. https://doi.org/10.1007/s10903-012-
9601-y
6. Wu S, Ell K, Gross-Schulman SG, et al. Technology-Facilitated Depression Care
Management Among Predominantly Latino Diabetes Patients Within a Public Safety Net
Care System: Comparative Effectiveness Trial Design. Contemp Clin Trials 2014;37(2):342-
354.
7. Bayliss E, Ellis J, Steiner J. Barriers to Self-Management and Quality-of-Life Outcomes in
Seniors with Multimorbidities. Annals of Family Medicine 2007; 5(5):395-402.
8. Sevick MA, Trauth JM, Ling BS, et al. Patients with Complex Chronic Diseases:
Perspectives on Supporting Self-Management. J Gen Intern Med 2007;22(Suppl 3):438–44.
9. Roberts R, Adams B, Warner B. Effects of Chronic Illness on Daily Life and Barriers to Self-
Care for Older Women: A Mixed-Methods Exploration. J Women Aging 2017; 29(2): 126-
136.
10. Bower P, Hann M, Rick J, et al. Multimorbidity and Delivery of Care for Long-Term
Conditions in the English National Health Service: Baseline Data from a Cohort Study. J
Health Serv Res Policy 2013;18(2 Suppl):29–37.
11. Coventry PA, Fisher L, Kenning C, et al. Capacity, Responsibility, and Motivation: a Critical
Qualitative Evaluation of Patient and Practitioner Views About Barriers to Self-Management
in People with Multimorbidity. BMC Health Serv Res 2014;14:536.
12. Liddy C, Blazkho V, Mill K. Challenges of Self-Management when Living with Multiple
109

Chronic Conditions: Systematic Review of the Qualitative Literature. Can Fam Physician
2014;60:1123–33
13. Mc Sharry J, Bishop FL, Moss-Morris R, et al. 'The Chicken and Egg Thing': Cognitive
Representations and Self-Management of Multimorbidity in People with Diabetes and
Depression. Psychol Health 2013;28:103–19
14. Fortin M, Bravo G, Hudon C, et al. Psychological Distress and Multimorbidity in Primary
Care. Ann Fam Med 2006;4:417–22
15. Ribu L, Ronnvig M, Corbin J. People with type 2 Diabetes Struggling for Self-Management:
A Part Study from the Randomized Control Trial in RENEWING HEALTH. Nursing Open
2019; 6(3): 1088-1096.
16. Eton DT, Ridgeway JL, Egginton JS, et al. Finalizing a Measurement Framework for the
Burden of Treatment in Complex Patients with Chronic Conditions. Patient Relat Outcome
Meas 2015;6:117–26
17. Pfeiffer E. A Short Portable Mental Status Questionnaire for the Assessment of Organic
Brain Deficit in Elderly Patients. J Am Geriatr Soc 1975;23(10):433-441.
18. Ewing J.A. Detecting Alcoholism: the CAGE questionnaire. JAMA 1984;252:1905-1907
19. Toobert D, Hampson S, Glasglow R. The Summary of Diabetes Self-Care Activities
Measure. Diabetes Care 2000;23:943-950.
20. Wu S, Ell K, Jin H, et al. Comparative Effectiveness of a Technology-Facilitated Depression
Care Management Model in Safety-Net Primary Care Patients with Type 2 Diabetes: 6
Month Outcomes of a Large Clinical Trial. JMIR 2018; 20(4): e147.
21. Whitty P, Steen N, Eccles M, et al. A New Self-Completion Outcome Measure for Diabetes:
is it Responsive to Change? Qual Life Res 1997;6(5):407-13.
22. Fisher L, Glasgow RE, Mullan JT, Skaff MM, Polonsky WH. Development of a Brief
Diabetes Distress Screening Instrument. Ann Fam Med 2008;6(3):246-252 [FREE Full text]
[CrossRef] [Medline]
23. Koenke K, Spitzer RL, Williams JB. The PHQ-9: Validity of a Brief Depression Severity
Measure. J Gen Intern Med 2001;16(9):606-13.
24. Derogatis LR, Savitz KL. The SCL-90-R Brief Symptom Inventory (BSI) in Primary Care.
In: Marvish ME, editor. Handbook of Psychological Assessment in Primary Care Settings.
Mahwah, NJ: Erlbaum; 2000.
25. Sheehan DV, Harnett-Sheehan K, Raj BA. The Measurement of Disability. Int Clin
Psychopharmacol 1996 Jun;11 Suppl 3:89-95. [Medline]
110

26. Sheehan KH, Sheehan DV. Assessing Treatment Effects in Clinical Trials with the Discan
Metric of the Sheehan Disability Scale. Int Clin Psychopharmacol 2008 Mar;23(2):70-83.
[CrossRef] [Medline]
27. Hernandez, R., Ruggiero, L., Riley, B. B., et al. Correlates of Self-Care in Low-Income
African American and Latino Patients with Diabetes. Health Psychology 2014; 33(7), 597-
607. doi:http://dx.doi.org.libproxy2.usc.edu/10.1037/hea0000043
28. Kim M, Lee E. Factors Affecting Self-Care Behavior Levels Among Elderly Patients with
Type 2 Diabetes: A Quantile Regression Approach. Medicina 2019; 55(7): 340.

111

Chapter 5 References
1. Markle-Reid M, Ploeg J, Fraser K, et al. Community Program Improves Quality of Life and
Self-Management in Older Adults with Diabetes Mellitus and Comorbidity. Clinical
Investigation 2018; 66(2): 263-73.
2. Ory M, Ahn S, Jian L, et al. Successes of a National Study of the Chronic Disease Self-
Management Program Meeting the Triple Aim of Health Care Reform. Medical Care.
2013;51(11):992-998.
3. Powers M, Bardsley J, Cypress Marjorie, et al. Diabetes Self-Management Education and
Support in Type 2 Diabetes: A Joint Position Statement of the American Diabetes
Association, the American Association of Diabetes Educators, and the Academy of Nutrition
and Dietetics. The Diabetes Educator 2017;43(1):40-53.
4. Stenberg U, Haaland-Overby M, Fredriksen K, et al. A Scoping Review of the Literature on
Benefits and Challenges of Participating in Patient Education Programs Aimed at Promoting
Self-Management for People Living with Chronic Illness. Patient Education and Counseling
2016;99(11):1759-71.
5. Gobeil-Lavoie AP, Chouinard MC, Danish A, et al. Characteristics of Self-Management
Among Patients with Complex Health Needs: a Thematic Analysis Review. BMJ Open 2019;
9(5): e028344. doi:10.1136/ bmjopen-2018-028344
6. Lanting LC, Joung IMA, Mackenbach JP, Lamberts SWJ, Bootsma AH. Ethnic Differences
in Mortality, End-Stage Complications, and Quality of Care Among Diabetic Patients: a
Review. Diabetes Care 2005;28(9):2280–8 10.2337/diacare.28.9.2280
7. Lopez-Class M, Jurkowski J. The Limits of Self-Management: Community and Health Care
System Barriers Among Latinos With Diabetes, Journal of Human Behavior in the Social
Environment 2010; 20(6): 808 826. DOI: 10.1080/10911351003765967
8. Rosal M, Ockene I, Restrepo A, et al. Randomized Trial of a Literacy-Sensitive, Culturally
Tailored Diabetes Self-Management Intervention for Low-Income Latinos. Diabetes Care
2011, 34(4): 838-844.
9. Wu S, Ell K, Gross-Schulman SG, et al. Technology-Facilitated Depression Care
Management Among Predominantly Latino Diabetes Patients within a Public Safety Net
Care System: Comparative Effectiveness Trial Design. Contemp Clin Trials 2014;37(2):342-
354.
10. Pfeiffer E. A Short Portable Mental Status Questionnaire for the Assessment of OrganicBrain
Deficit in Elderly Patients. J Am Geriatr Soc 1975;23(10):433-441.
11. Ewing J.A. Detecting Alcoholism: the CAGE Questionnaire. JAMA 1984;252:1905-1907
112

12. Whitty P, Steen N, Eccles M, et al. A New Self-Completion Outcome Measure for Diabetes:
is it Responsive to Change? Qual Life Res 1997;6(5):407-13.
13. Fisher L, Glasgow RE, Mullan JT, Skaff MM, Polonsky WH. Development of a Brief
Diabetes Distress Screening Instrument. Ann Fam Med 2008;6(3):246-252 [FREE Full text]
[CrossRef] [Medline]
14. Koenke K, Spitzer RL, Williams JB. The PHQ-9: Validity of a Brief Depression
SeverityMeasure. J Gen Intern Med 2001;16(9):606-13.
15. Derogatis LR, Savitz KL. The SCL-90-R Brief Symptom Inventory (BSI) in Primary Care.
In: Marvish ME, editor. Handbook of Psychological Assessment in Primary Care Settings.
Mahwah, NJ: Erlbaum; 2000.
16. Sheehan DV, Harnett-Sheehan K, Raj BA. The Measurement of Disability. Int Clin
Psychopharmacol 1996 Jun;11 Suppl 3:89-95. [Medline]
17. Sheehan KH, Sheehan DV. Assessing Treatment Effects in Clinical Trials with the Discan
Metric of the Sheehan disability scale. Int Clin Psychopharmacol 2008 Mar;23(2):70-83.
[CrossRef] [Medline]
18. Aikens J, Piette J. Longitudinal Associations Between Medication Adherence and Glycaemic
Control in Type 2 Diabetes. Diabetic Medicine 2012; 30(3): 338-344.
19. Chen L, Chuang L, Chang C, et al. Evaluation Self-Management Behaviors of Diabetic
Patients in a Telehealthcare Program: Longitudinal Study over 18 Months. J Med Internet
Res 2013; 15(12): e266.

113

Chapter 6 References
1. Revicki D, Hays R, Cella D, et al. Recommended Methods for Determining Responsiveness
and Minimally Important Differences for Patient-Reported Outcomes. Journal of Clinical
Epidemiology 2008; 61(2): 102-109.
2. Basch E. Patient-Reported Outcomes – Harnessing Patient Voices to Improve Clinical Care.
N Engl J Med 2017; 376:105-108.

114

Appendix

Sample Comparisons

Different samples were used in each separate analysis (chapter) of this dissertation. Due to the
nature of the analyses used, complete longitudinal cases were required. Complete cases were
determined through having complete longitudinal data for the dependent variable of interest.
Incomplete cases were if one or more phases of data measurement are missing for said outcome.
Specific sample inclusion and exclusion criteria for each analysis are described below.
Paper 1 (Chapter 2)
Paper 1 inclusion criteria is as follows: complete data for (1) satisfaction with diabetes care and
(2) satisfaction with depression care, and (3) all independent variables in the regression models
for all four waves of DCAT (baseline, 6- , 12-, and 18-months). The sample included 700
patients. The table below provides mean (sd) for continuous variables and frequency (%) for
discrete variables.
Table 1. Paper 1 Sample Comparisons
Cluster Stats
Complete
(n=700)
Incomplete
(n=706) p
Arm <.001*
UC 254 (31.14) 266 (37.68)
SC 228 (32.57) 252 (35.69)
TC 254 (36.29) 188 (26.63)
Socio-Demographics
Age 53.16 (9.25) 53.37 (9.23) .671
Female 462 (66.00) 430 (60.91) .047*
Latino 649 (92.98) 605 (85.82) <.001*
Spanish pref. lang. 614 (87.71) 549 (77.76) <.001*
Married 392 (56.00) 380 (53.82) .412
Less than HS Edu. 522 (74.68) 455 (64.45) <.001*
Unemployed 473 (67.57) 471 (66.71) .732
US Born 65 (9.29) 113 (16.03) <.001*
In the US for 10+ Years 593 (93.98) 548 (93.52) .739
Cost and Utilization Measures
115

Predicted Future Care Cost 6611.44 (4017.63) 6687.42 (3837.57) .717
Healthcare Utilization Costs
α
1531.42 (4351.99) 1177.94 (3670.21) .126
Pharmacy Costs
α
410.08 (552.96) 432.46 (566.10) .464
Diabetes Measures
A1C
α
9.19 (2.09) 9.29 (2.15) .377
BMI
α
32.69 (7.20) 32.77 (7.37) .845
Cholesterol
α
179.01 (49.34) 181.87 (51.47) .313
Age of diabetes onset 42.95 (10.28) 43.21 (10.34) .628
Insulin use 387 (55.29) 355 (50.28) .060
Num diabetes Complications 1.31 (1.13) 1.23 (1.17) .167
Whitty-9 Diabetes Symp Scale 1.69 (.62) 1.62 (.59) .036*
Diabetes Emotional Burden 3.10 (2.08) 2.91 (1.99) .086
Diabetes Regimen Distress 2.97 (2.07) 2.80 (1.99) .102
Diabetes Self-Management 4.34 (1.28) 4.31 (1.34) .653
Psychological and QOL Measures
PHQ-9 7.05 (5.97) 6.29 (6.00) .017*
Dysthymia 132 (18.86) 124 (17.59) .538
BSI 1.26 (3.00) 1.16 (2.98) .525
SDS 2.23 (2.96) 2.05 (2.85) .258
SF-12 Mental 49.01 (13.50) 50.66 (12.53) .018*
SF-12 Physical 43.89 (11.33) 44.66 (10.80) .192
Self-Rated Health 2.28 (.82) 2.32 (.83) .269
Stress Measures
Econ. Stressors 4.14 (2.18) 3.89 (2.20) .033*
Number of Stressors 2.53 (2.26) 2.26 (2.15) .021*
Sum of Stress Level
α
17.90 (18.18) 15.78 (17.25) .025*
Satisfaction Measures
Stsf. with Diabetes Care 4.71 (.61) 4.70 (.59) .886
Stsf. with Depression Care 4.51 (.79) 4.48 (.80) .541

116

Paper 2 (Chapter 3)
Paper 2 inclusion criteria is as follows: complete data for (1) self-management behavior, (2)
depression symptoms (via PHQ-9), (3) diabetes symptoms, and (4) self-rated health for all four
waves of DCAT (baseline, 6- , 12-, and 18-months). The sample included 783 patients. The table
below provides mean (sd) for continuous variables and frequency (%) for discrete variables.
Table 2. Paper 2 Sample Comparisons
Cluster Stats
Complete
(n=783)
Incomplete
(n=623) p
Arm .004*
UC 275 (35.12) 209 (33.55)
SC 240 (30.65) 240 (38.52)
TC 268 (34.23) 174 (27.93)
Socio-Demographics
Age 53.17 (9.27) 53.38 (9.21) .676
Female 507 (64.75) 385 (61.80) .253
Latino 719 (92.06) 535 (86.01) <.001*
Spanish pref. lang. 679 (86.72) 484 (77.69) <.001*
Married 448 (57.22) 324 (52.01) .051
Less than HS Edu. 580 (74.07) 397 (63.83) <.001*
Unemployed 512 (65.39) 432 (69.34) .117
US Born 78 (9.96) 100 (16.08) <.001*
In the US for 10+ Years 663 (94.85) 478 (92.28) .067
Cost and Utilization Measures
Predicted Future Care Cost 6510.76 (3910.70) 6824.08 (3943.68) .137
Healthcare Utilization Costs
α
1410.83 (4169.44) 1279.88 (3837.06) .573
Pharmacy Costs
α
422.34 (571.56) 419.97 (544.25) .939
Diabetes Measures
A1C
α
9.13 (2.08) 9.38 (2.16) .034*
BMI
α
32.76 (7.14) 32.69 (7.47) .853
Cholesterol
α
178.02 (48.47) 183.59 (52.72) .051
Age of diabetes onset 43.13 (10.30) 43.02 (10.32) .840
Insulin use 419 (53.51) 323 (51.85) .534
Num diabetes Complications 1.21 (1.13) 1.35 (1.17) .023*
Whitty-9 Diabetes Symp Scale 1.62 (.60) 1.69 (.62) .032*
Diabetes Emotional Burden 2.93 (2.06) 3.10 (2.01) .115
Diabetes Regimen Distress 2.83 (2.04) 2.95 (2.02) .254
Diabetes Self-Management 4.38 (1.27) 4.26 (1.36) .100
Psychological and QOL Measures
PHQ-9 6.20 (5.72) 7.26 (6.28) .001*
Dysthymia 125 (15.96) 131 (21.06) .014*
117

BSI 1.09 (2.83) 1.37 (3.17) .088
SDS 1.96 (2.81) 2.36 (3.01) .010*
SF-12 Mental 50.59 (12.69) 48.89 (13.43) .015*
SF-12 Physical 44.77 (10.96) 43.65 (11.19) .059
Self-Rated Health 2.33 (.82) 2.26 (.83) .079
Stress Measures
Econ. Stressors 3.95 (2.19) 4.10 (2.19) .198
Number of Stressors 2.42 (2.22) 2.37 (2.20) .703
Sum of Stress Level
α
16.74 (17.60) 16.97 (17.95) .809
Satisfaction Measures
Stsf. with Diabetes Care 4.70 (.57) 4.70 (.64) .934
Stsf. with Depression Care 4.50 (.77) 4.49 (.82) .753

118

Papers 3 and 4 (Chapters 4 and 5)
Paper 3 and 4 utilized the same study sample due to the nature of the data set. Papers 3 and 4
inclusion criteria is as follows: complete data for self-management behavior for all four waves of
DCAT (baseline, 6- , 12-, and 18-months). Additionally, a small sample of extreme outlier
patients were removed to avoid skewing the results (n=5). The sample included 847 patients.
Most of the measures are not significant; however, there are some differences between three
objective measures: A1C, cholesterol, and dysthymia diagnosis. This is one limitation of this
analysis. Missing patients have higher A1C, cholesterol, and proportion of patients diagnosed
with dysthymia. Additionally, the missing group has a higher proportion of more highly
education patients. Results found that more highly educated patients had higher odds of
struggling with self-management. Thus, although it was found that only 12% of patients were
struggling with self-management, there are likely many more patients struggling with self-
management who dropped out of the study. Aside from these measures, there were no significant
baseline differences in self-management levels, diabetes symptoms or complications, depression
symptoms, and stress levels. The table below provides mean (sd) for continuous variables and
frequency (%) for discrete variables.
Table 3. Paper 3 and 4 Sample Comparisons
Cluster Stats
Complete
(n=847)
Incomplete
(n=559) p
Arm .026*
UC 300 (35.42) 184 (32.92)
SC 266 (31.40) 214 (38.28)
TC 281 (33.18) 161 (28.80)
Socio-Demographics
Age 53.34 (9.17) 53.14 (9.33) .684
Female 552 (65.17) 340 (60.82) .098
Latino 552 (65.17) 475 (85.13) <.001*
Spanish pref. lang. 737 (87.01) 426 (76.21) <.001*
Married 480 (56.67) 292 (52.24) .102
119

Less than HS Edu. 626 (74.00) 351 (62.79) <.001*
Unemployed 564 (66.59) 380 (67.98) .587
US Born 84 (9.92) 94 (16.85) .0001*
In the US for 10+ Years 715 (94.45) 426 (92.61) .198
Cost and Utilization Measures
Predicted Future Care Cost 6581.45 (3912.40) 6752.84 (3950.37) .423
Healthcare Utilization Costs
α
1414.09 (4098.78) 1259 (3909.98) .513
Pharmacy Costs
α
425.53 (566.77) 415.53 (548.60) .761
Diabetes Measures
A1C
α
9.13 (2.09) 9.40 (2.15) .020*
BMI
α
32.81 (7.27) 32.62 (7.31) .640
Cholesterol
α
178.20 (48.52) 183.93 (53.11) .049*
Age of diabetes onset 43.20 (10.21) 42.89 (10.47) .581
Insulin use 449 (53.01) 293 (52.42) .827
Num diabetes Complications 1.24 (1.13) 1.32 (1.18) .168
Whitty-9 Diabetes Symp Scale 1.64 (.60) 1.68 (.61) .166
Diabetes Emotional Burden 2.97 (2.05) 3.05 (2.01) .504
Diabetes Regimen Distress 2.85 (2.03) 2.93 (2.03) .472
Diabetes Self-Management 4.37 (1.27) 4.26 (1.37) .116
Psychological and QOL Measures
PHQ-9 6.49 (5.81) 6.94 (6.27) .174
Dysthymia 140 (16.53) 116 (20.79) .043*
BSI 1.14 (2.83) 1.32 (3.21) .276
SDS 2.02 (2.85) 2.32 (2.98) .056
SF-12 Mental 50.06 (12.92) 49.51 (13.23) .440
SF-12 Physical 44.57 (11.06) 43.82 (11.08) .213
Self-Rated Health 2.32 (.82) 2.26 (.83) .170
Stress Measures
Econ. Stressors 3.98 (2.18) 4.05 (2.21) .563
Number of Stressors 2.44 (2.24) 2.32 (2.18) .315
Sum of Stress Level
α
17.04 (17.78) 16.52 (17.71) .315
Satisfaction Measures
Stsf. with Diabetes Care 4.70 (.59) 4.70 (.61) .943
Stsf. with Depression Care 4.50 (.79) 4.48 (.80) .632

Abstract (if available)

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Clinical prediction models to forecast depression in patients with diabetes and applications in depression screening policymaking

PDF

Depression severity, self-care behaviors, and self-reported diabetes symptoms and daily functioning among low-income patients receiving depression care

PDF

Using a human factors engineering perspective to design and evaluate communication and information technology tools to support depression care and physical activity behavior change among low-inco...

PDF

Simulation modeling to evaluate cost-benefit of multi-level screening strategies involving behavioral components to improve compliance: the example of diabetic retinopathy

PDF

Investigation of health system performance: effects of integrated triple element method of high reliability, patient safety, and care coordination

PDF

Feature and model based biomedical system characterization of cancer

PDF

Barriers to practice: primary care patient experience and quality health outcomes in the Veterans Health Administration

PDF

Understanding primary nonadherence to medications and its associated healthcare outcomes: a retrospective analysis of electronic medical records in an integrated healthcare setting

PDF

Essays on improving human interactions with humans, algorithms, and technologies for better healthcare outcomes

PDF

Understanding the quality of geriatric healthcare and family caregiver advocacy for patients with dementia

PDF

Integrated management of atrial fibrillation in women in an underserved, safety-net health care system: a multicenter, single health system randomized control efficacy trial protocol

PDF

Investigating a physiological pathway for the effect of guided imagery on insulin resistance

Asset Metadata

Creator Evanson, Olivia (author)

Core Title A series of longitudinal analyses of patient reported outcomes to further the understanding of care-management of comorbid diabetes and depression in a safety-net healthcare system

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Industrial and Systems Engineering

Publication Date 02/06/2020

Defense Date 11/20/2019

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

Tag care-management,cluster analysis,comparative effectiveness,cross-lagged path analysis,Depression,Latino,longitudinal analysis,multi-level regression modeling,OAI-PMH Harvest,patient reported outcomes,patient satisfaction,prediction modeling,safety-net care,self-management,telecommunication technology,type 2 diabetes

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Wu, Shinyi (committee chair), Jin, Haomiao (committee member), Suen, Sze-Chuan (committee member)

Creator Email oevanson@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-266151

Unique identifier UC11673385

Identifier etd-EvansonOli-8150.pdf (filename),usctheses-c89-266151 (legacy record id)

Legacy Identifier etd-EvansonOli-8150.pdf

Dmrecord 266151

Document Type Dissertation

Rights Evanson, Olivia

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA