Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Three essays in health economics
(USC Thesis Other)
Three essays in health economics
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
1
Three Essays in Health Economics
Zhiwen (Richard) Xie
Doctor of Philosophy (ECONOMICS)
FACULTY OF THE USC GRADUATE SCHOOL
University of Southern California
August 2018
2
Contents
Chapter 1 - Innovation in Heart Failure Treatment: Life Expectancy, Disability, and Health Disparities ... 3
Text ............................................................................................................................................................... 5
Introduction ............................................................................................................................................... 5
Methods .................................................................................................................................................... 6
Results ....................................................................................................................................................... 9
Discussion ............................................................................................................................................... 12
Conclusions ............................................................................................................................................. 16
Clinical Perspectives ............................................................................................................................... 16
Tables and Figures ...................................................................................................................................... 18
Chapter 2 - Racial and Ethnic Disparities in Medication Adherence Among Privately Insured Patients in
the United States ......................................................................................................................................... 27
Text ............................................................................................................................................................. 29
Introduction ............................................................................................................................................. 29
Methods .................................................................................................................................................. 30
Results ..................................................................................................................................................... 33
Discussion ............................................................................................................................................... 35
Public Health Implications ...................................................................................................................... 36
Tables and Figures ...................................................................................................................................... 38
Chapter 3 – The Effects of Extended-Release Formulations on Adherence, A1C Control and Hospital
Outcomes Among Patients Treated with Type 2 Diabetes Therapies ......................................................... 45
Text ............................................................................................................................................................. 46
Introduction ............................................................................................................................................. 46
Methods .................................................................................................................................................. 47
Results ..................................................................................................................................................... 51
Discussion ............................................................................................................................................... 53
Tables and Figures ...................................................................................................................................... 56
REFERENCES ........................................................................................................................................... 66
3
Chapter 1 - Innovation in Heart Failure Treatment: Life
Expectancy, Disability, and Health Disparities
4
Abbreviations List
ADL Activities of Daily Living
BMI Body Mass Index
CHF Congestive Heart Failure
EQ-5D EuroQol five dimensions questionnaire
FEM Future Elderly Model
HRS Health and Retirement Survey
IADL Instrumental Activities of Daily Living
MEPS Medical Expenditure Panel Survey
MCBS Medicare Current Beneficiary Survey
QALY Quality-Adjusted Life Year
DFLY Disability-Free Life Year
5
Text
Introduction
There is much concern about the increasing share of national income devoted to health care,
1
but
spending increases have been accompanied by significant health improvements. Most significantly, age-
standardized death rates from all causes have fallen 43% since 1969 (from 1,279 deaths per 100,000 in
1969 to 730 in 2013).
2
Better cardiovascular outcomes have driven much of this improvement, with age-
adjusted deaths from heart disease falling from 520 per 100,000 in 1969 to 168.5 in 2015.
2,3
Evidence-
based treatment of associated risk factors has been credited with contributing to these declines.
4
However, progress may be slowing, and — in some disease areas like congestive heart failure (CHF) —
may even be reversing. An estimated 5.7 million American adults suffer from CHF, and CHF is a
contributing factor in 1 in 9 U.S. deaths.
4
The Centers for Disease Control reports that between 2011 and
2014, age-adjusted death rates from heart failure rose from 16.9 to 18.6 per 100,000.
5
This trend may also exacerbate existing racial health disparities. African Americans develop heart failure
earlier than whites, and are more likely to be admitted to the hospital for it.
6,7
In addition, the 5-year risk-
adjusted all-cause mortality rate for CHF patients is 34% higher for African Americans than whites.
8,9
Given these existing racial disparities, the fact that the age-adjusted death rates from CHF are increasing
is particularly alarming.
Adding to the personal toll of CHF—premature death, disability and loss of quality of life—its economic
costs are substantial: almost $32 billion annually for U.S. treatment costs and lost productivity.
7
Fortunately, recent treatment innovations suggest that the future impact of CHF on patient outcomes,
6
economic productivity, and overall social value could be reduced, perhaps even in a way that mitigates
health disparities.
10-12
This paper models the potential benefits to population health from continued innovation in CHF
treatment. Using US population-wide simulations, we estimate trends in CHF prevalence, and how much
improved CHF treatments could improve overall social value, and reduce racial and gender differences in
health outcomes.
Methods
To illustrate the potential benefits of improved CHF treatment, we adapted the Future Elderly Model
(FEM), an established economic-demographic microsimulation that has been used to study a wide variety
of health policy questions. The FEM has been developed over time with support from the National
Institute on Aging, the Department of Labor, the MacArthur Foundation, and the Centers for Medicare &
Medicaid Services to study health care innovation in a wide variety of contexts, including heart disease.
13-
17
Overview.
The FEM simulates health and medical spending for Americans aged 51 years and older. The model uses
initial demographic characteristics and health conditions for each individual to project their medical
spending, health conditions and behaviors, disability status, and quality of life. A key advantage of the
FEM is that it tracks individual-level health trajectories and patient outcomes, which allows us to consider
the impact of innovation by characteristics such as gender and race.
7
The FEM’s core module uses individuals’ current characteristics to calculate transition probabilities
among health states, including mortality, functional status, body mass index (BMI) and six disease
conditions: diabetes, high blood pressure, heart disease (including CHF), cancer (excluding skin cancer),
stroke, and lung disease. The model uses inputs from three nationally representative datasets: The Health
and Retirement Survey (HRS), a biennial survey of the American population aged 51+ which has been
conducted since 1992; the Medical Expenditure Panel Survey (MEPS), a set of large-scale surveys of the
non-institutionalized US population, and the Medicare Current Beneficiary Survey (MCBS), a survey of
Medicare beneficiaries about their health status, healthcare use and insurance coverage. More detail on the
model and data sources is provided in the technical appendix.
18
Prevalence and incidence of CHF.
To predict which individuals have or will get CHF during the simulation, we use HRS historical data to
build a two-year CHF incidence model based on predictors including age, sex, education, race, age-race
interactions, BMI, smoking behavior, marital status, and the six disease conditions modeled. This is a
first-order Markov model in which time-varying components enter via their status two years prior. For
example, diabetes status in the prior wave of the survey is a predictor of incident CHF in the current
wave. All transition models in the simulation have this structure.
CHF status is included as a predictor of other outcomes of interest, including mortality, functional
limitations (ADL), and instrumental activities of daily living (IADL) limitations. Mortality is estimated
as a two-year probit model, controlling for age, race, sex, education, widowhood, smoking status, the six
chronic diseases, ADLs, and IADLs. The number of functional limitations is estimated as an ordered
probit with four categories: none, one, two, and three or more. This ADL model controls for the same set
8
of variables as the mortality model, plus BMI. IADL limitations are also modeled with the same
predictors, as an ordered probit with three categories: none, one, and two or more.
Valuing health benefits.
To value health benefits, we predict quality-adjusted life years (QALYs) using the EQ-5D, a widely-used
health-related quality of life index. The EQ-5D instrument includes five questions regarding the extent of
problems in mobility, self-care, daily activities, pain, and anxiety/depression, and has been widely used in
both Europe and the United States.
19,20
Using the 2001 MEPS, we estimate a linear model fitting EQ-5D
scores as a function of six chronic conditions and functional status (details in technical appendix). We
predict a QALY measure for every person in the simulation in every year based on their simulated health
and functional status.
We simulate outcomes for a representative cohort of 51- and 52-year-olds beginning in 2016 (n=13,040).
This cohort, described in the technical appendix, is based on respondents from the HRS. In each year, the
spending module predicts medical expenditures over the next two years (the HRS is biennial) based on
each individual’s current ‘state’. The health module is then used to predict who will survive to year 2018,
and their obesity status, disease, and functional state, and a predicted QALY for that year. The spending
module is then used to predict that period's health care resource use. The simulation iterates in this
manner until everyone in the 2016 cohort has died. We repeat the simulation 500 times for each scenario
and report the average outcomes and resulting confidence intervals. Primary outcomes are life
expectancy, quality-adjusted life expectancy, and lifetime medical spending. All costs and QALYs are
discounted using a 3% annual discount rate as suggested by Gold et al.
21
9
Scenarios.
To predict the prevalence of CHF from 2016 to 2030, we simulate the population aged 50+ in 2016 and
beyond, accounting for projected demographic trends over time. To examine the burden of CHF, we
model the life trajectories of the cohort of individuals aged 51-52 in 2016 to construct a baseline scenario,
and compare this to a “No CHF” scenario in which no individuals in the cohort develop CHF throughout
the simulation, maintaining all other transition dynamics. While completely preventing CHF might seem
unrealistic, it provides an upper bound for the potential social gain from a medical innovation. For
comparison, we perform similar analyses eliminating, in turn, diabetes, high blood pressure, lung disease,
cancer, obesity and stroke. To calculate total QALYs added in each scenario, we calculate the number of
individuals aged 51-52 in 2016 and multiply this by the average QALYs added.
Implementation.
Transition models are estimated using the RAND HRS version P, using nationally representative waves
from 1998-2012 (n=114,489 person-waves). Medical costs are estimated using MEPS 2007-2010 for the
non-Medicare population and MCBS 2007-2012 for the Medicare population. All estimations performed
with Stata 14.0. See the technical appendix for estimates.
18
Results
Figure 1 shows the prevalence of CHF through 2030—years 1996 through 2010 represent data from the
HRS, while years 2012 and beyond reflect simulation estimates. Among 65- to-70-year-olds, the
prevalence of CHF is expected to increase from 4.29% in 2010 to 8.45% (95% CI: 8.03-8.87%) in 2030.
10
Analysis of the 2010-2012 HRS data among patients with cardiovascular disease shows that the age-
adjusted incidence of CHF is higher among blacks than whites, and highest for black females (4.8%)
(black males: 4.1%, white females: 4.0%, white males: 3.5%).
We also considered the impact of CHF on disability status, and how this varies with race and gender.
Using 2000-2012 HRS data, we identified all patients without CHF in one period who went on to develop
CHF in the subsequent period, and their reported ability to perform five ADLs: eating, bathing, dressing,
walking across a room, and getting in or out of bed. Figure 2 reports the age-adjusted proportion of these
patients who report limitations in three or more ADLs before and after CHF diagnosis.
Immediately before CHF diagnosis, 9.6% of patients report three or more limitations, rising to 17.4%
after CHF diagnosis. The onset of significant disability with CHF diagnosis is particularly severe among
black men: Before diagnosis, 7.4% of black males who will develop CHF report three or more
limitations, increasing to 20% immediately after diagnosis. Among black females who develop CHF, the
proportion reporting three or more limitations is 20.3% before diagnosis and 30.2% afterwards. The
proportion of the population that did not develop CHF across two consecutive waves saw no significant
changes in age-adjusted disability.
Medical expenditures follow a similar pattern (Figure 3). In the 2000-2012 HRS data we found that, prior
to diagnosis, patients who will develop CHF are somewhat sicker than the average person of the same
age, with medical expenditures 25-30% higher than those of people without CHF. After diagnosis, CHF
patients have medical expenditures 50-56% higher. The increment is especially large among black
females.
11
Increasing prevalence of a disease such as CHF, with significant mortality, disability and expenditure
implications for the overall population and differential implications by race and gender underscores the
potential benefits from improved treatment. We explored these benefits by simulating scenarios in which
we eliminate seven diseases—CHF, cancer, diabetes, high blood pressure, lung disease, obesity, and
stroke—and compare the resulting gains in life expectancy, QALYs and disability-free life years
(DFLYs). Affected patients retain all the other characteristics and comorbidities of a patient with the
disease in question; they do not return to “average” health. Figure 4 presents these results: Among
patients who otherwise would have developed CHF, eliminating the disease increases average life
expectancy by 1.92 (95% CI: 1.91-1.93) years, increases the average time lived without a disability by
0.78 (0.78-0.79) years and increases quality-adjusted life expectancy by 1.43 (1.42-1.44) years. Only
eliminating cancer, lung disease and diabetes generate greater life expectancy increases for their affected
populations. Table 1 presents simulation results including the lifetime risk of each condition, and the
impact of eliminating each on the life expectancy for the entire population, which combines the impact on
the affected population with prevalence.
If an innovation to eliminate heart failure is applied to the 4.1 million individuals aged 51-52 in 2016, it
could generate nearly 2.9 million additional life years, 2.1 million QALYs, and 1.2 million DFLYs.
Depending on the value of each additional QALY, the population health benefits of such an innovation
range from $210 to $420 billion.
Figure 5 Panel A shows that eliminating heart failure increases average life expectancy among those
affected by 2.10 (95% CI: 2.06-2.14) years for black males, 1.90 (1.88-1.92) for white males, 2.18 (2.15-
2.22) for black females, and 1.84 (1.82-1.86) for white females. Panel B shows that eliminating CHF adds
12
0.86 (0.84-0.88) disability-free life years for black males, 0.88 (0.87-0.89) for white males, 0.76 (0.74-
0.78) for black females, and 0.72 (0.71-0.73) for white females. Panel C shows QALY gains: 1.52 (1.50-
1.55) for black males, vs. 1.44 (1.43-1.46) for white males, and 1.55 (1.53-1.58) for black females
compared to 1.37 (1.36-1.39) for white females.
Discussion
CHF prevalence and lifetime risk.
Our estimates of the future prevalence of CHF are generally higher than in other studies.
22,23
This is in
part because our estimates focus on prevalence among the older population, while other estimates report
prevalence among the entire US population. However, our simulation also incorporates trends in the risk
factors that lead to CHF, which are themselves increasing.
For example, Heidenreich et al. project CHF prevalence increasing from 2.4% in 2012 to 3.0% in 2030.
22
Their estimates are driven by changes in the size of subpopulations—defined by age, gender, and
ethnicity—but they do not allow the prevalence within a subpopulation to change over time. Our
subpopulation-specific CHF prevalence estimates incorporate projected trends in comorbidities and other
health indicators that accompany CHF, including obesity, hypertension, etc. Thus, increasing obesity
over time will increase the prevalence of CHF even within demographic subpopulations, leading to higher
CHF prevalence estimates than those using the Heidenreich methods.
22,23
Our model projects a lifetime risk of CHF incidence of 35% for patients aged 51-52, similar to other
lifetime risk estimates based on large-scale population studies.
24
Our findings of disparities between
13
blacks and whites in the risk of CHF in the HRS data mirrors results of previous studies. Most notably,
the Atherosclerosis Risk in Communities Study found that the lifetime risk of CHF for those aged 45-75
was higher for blacks than for whites, and highest for black females (24% vs. 21% for black males, 19%
for white males, 13% for white females).
7
Disability and disparities.
People with CHF often have other serious medical conditions, such as arthritis (62%) or diabetes (38%);
are unable to walk two to three blocks or walk up 10 steps (57%); need help with activities of daily living
(11%); and take 6.4 prescription medications on average.
25
Such factors may affect CHF patients’ ability
to live independently, with needs ranging from help from an informal caregiver to moving to a nursing
facility.
We show (Figure 2) that disability outcomes vary by race—black patients with CHF diagnoses are much
more likely to report limitations in three or more ADLs than their white counterparts. Among men
diagnosed with CHF, roughly the same fractions of black and white patients report three or more ADLs
before diagnosis (7.4% vs. 7.1%), but after diagnosis, that fraction increases by more than 170% for black
males, and only 62% for white males. Similar trends appear in the expenditure data (Figure 3).
While the correlation between CHF and disability may be the result of CHF occurring in patients who are
already very sick and disabled, our findings suggest that CHF may also play a causal role in patients’
decline: For many, the onset of CHF can be relatively sudden, and preceded by relatively good health, but
disability may progress rapidly after diagnosis. If so, then perhaps the appearance of significant disability
could be forestalled if a diagnosis of CHF could be delayed or eliminated. And if health disparities are
14
driven by differential disability outcomes by race and gender, then more effective treatments for CHF
could reduce health disparities.
Value of innovation in CHF.
Until recently, there has been relatively little innovation in CHF treatment, with standard care involving
medications to treat symptoms, including angiotensin-converting enzyme inhibitors, beta blockers, and
diuretics.
11
New drugs for heart failure have recently been approved that reduced the risk of death or
hospitalization from heart failure by 20 percent in clinical trials.
26,27
While these new treatments do not
eliminate CHF, they point at the potential for significant innovation in this disease area.
Our results demonstrate that eliminating CHF, even without changing patients’ underlying health
characteristics, would add 1.92 years to each affected patient’s life—more than affected patients would
gain by eliminating stroke, obesity, or high blood pressure. Improvements in CHF treatment can also
enhance patients’ quality of life, with elimination adding 1.43 QALYs or 0.78 DFLYs to the average CHF
patient’s life. These also compare favorably with innovations to eliminate high blood pressure, a
condition that receives far more public health attention than CHF.
Health disparities.
We estimate that eliminating CHF could narrow the disparity between black and white average life
expectancy. In our baseline scenario, for the subgroup that developed CHF, white males live 5.1 years
longer than black males, and white females 3.5 years longer than black females, on average. Curing CHF
reduces this gap by 0.1 and 0.3 years, respectively.
15
Limitations.
We note several limitations. First, our results are derived from simulations estimated by the FEM, which
uses simplifications of the dynamic relationships driving outcomes in the real world, and parameterizes
those relationships using estimates from the literature. If these simplifications are incomplete, or the
parameter estimates are imperfect, model results may not correspond with actual outcomes.
Second, our “No-CHF” scenario assumes an innovation that eliminates CHF, although medical
innovation is unlikely to completely eliminate CHF in the near term. Instead, actual innovations are more
likely to reduce CHF incidence or lessen its severity, but not eliminate it completely. This scenario does
however also help answer the question: what is the social cost of CHF in older populations?
Third, we do not explicitly model the social determinants of health disparities, such as greater poverty,
poorer access to care, and lower health literacy among black patients with CHF.
28
Our “No-CHF”
scenario implicitly assumes that a cure for CHF is applied equally to all CHF patients, regardless of social
determinants, without modeling how that penetration would happen. Nevertheless, understanding the
social value that such a cure, uniformly applied, would unlock helps inform what kinds of policies would
be most beneficial.
Finally, our model of the relationship between CHF and outcomes (disability, quality-of-life, and
mortality) is based on associations observed in cohorts of thousands of older Americans followed over
decades. We have not demonstrated a direct causal link.
Policy Implications.
16
Policy programs to improve public health have often focused on interventions for high-prevalence
diseases and conditions such as diabetes, obesity, and hypertension. Cancer has also received much
policy attention, including the recent Cancer Moonshot, with the goal of hastening cures, despite limited
progress. Our work suggests that similar emphasis, focus and investment in finding ways to eliminate
CHF could have as much or more impact in terms of adding life years, QALYs and DFLYs, and
potentially reducing racial disparities among the population.
Conclusions
Heart failure is one example of the growing disease burden older Americans bear as they live longer but
face growing risks of disability.
29
From a societal standpoint, policymakers and other decision makers
must balance competing aims to benefit all people generally and disadvantaged groups specifically to
achieve goals of both efficiency and equity. Innovations that improve disease outcomes—not just
eliminate them—can improve efficiency by increasing benefits to society through longer, healthier, more
productive lives. Some treatment innovations also can improve equity by narrowing longstanding health
disparities among minorities and women.
Innovation in CHF deserves scientific and policy attention not simply because it can extend lives and
reduce disability and decline in older Americans but also because it could ameliorate some racial and
gender disparities in health outcomes associated with cardiovascular disease.
Clinical Perspectives
Competency in Medical Knowledge: CHF prevalence will increase substantially over the next two
decades, affecting black Americans more than whites. A CHF diagnosis coincides with significant
increase in disability and medical expenditures, particularly among blacks compared with whites.
17
Improving CHF treatment could generate significant social value, and reduce existing racial/ethnic health
disparities.
Translational Outlook: Future research to identify more effective treatment and prevention of CHF
could both improve the quality and length of life for patients with CHF, and reduce disparities among
patients affected by CHF.
18
Tables and Figures
Figure Legends
Figure 1: Prevalence of Congestive Heart Failure Among U.S. 65-70 Year Olds
Sources: Health & Retirement Survey, and the Future Elderly Model Simulation
Figure 2: Age-Adjusted Percent of CHF Population Reporting Limitations in 3 or More Activities of
Daily Living, Before and After CHF Diagnosis, by Race and Gender
Source: Health & Retirement Survey 2000 to 2012 Data
Figure 3: Annual Per-Capita Medical Expenditures, by Race and Gender
Source: Health & Retirement Survey 2000 to 2012 Data
Figure 4: Average Gain in Selected Outcomes from Eliminating Seven Conditions (Among Those
Affected)
Panel A –Life Expectancy in Years
Panel B – Disability-Free Life Years
Panel C – Quality-Adjusted Life Years
Source: The Future Elderly Model Analysis
Figure 5: Average Gain in Selected Outcomes from Eliminating Congestive Heart Failure (Among Those
Affected), by Race and Gender
Panel A –Life Expectancy in Years
Panel B –Disability-Free Life Years
Panel C – Quality-Adjusted Life Years
Source: The Future Elderly Model Analysis
19
Table 1:
Lifetime risk of seven diseases, and gain in population-wide life expectancy
from eliminating each one (95% confidence intervals in parentheses)
Disease/Condition
for Intervention*
Lifetime Risk/
Percentage of
Population Affected
LE gain for the
entire
population†
LE gain for the sub-
population affected
by the intervention
Cancer 39.2% (38.8 – 39.6) 1.09 (1.08 – 1.10) 2.75 (2.71 – 2.79)
CHF 36.2% (35.9 – 36.4) 0.70 (0.69 – 0.71) 1.92 (1.91 – 1.93)
Diabetes 57.4% (56.8 – 58.0) 1.13 (1.11 – 1.15) 1.96 (1.93 – 2.00)
High Blood Pressure 89.0% (88.8 – 89.3) 0.64 (0.63 – 0.65) 0.72 (0.70 – 0.73)
Lung Disease 27.8% (27.4 – 28.3) 0.74 (0.73 – 0.75) 2.63 (2.57 – 2.68)
Obesity 42.1% (40.9 – 43.2) 0.59 (0.56 – 0.61) 1.39 (1.34 – 1.44)
Stroke 37.5% (37.3 – 37.7) 0.51 (0.51 – 0.52) 1.36 (1.35 – 1.37)
* The intervention eliminates the disease or condition
† Entire population refers to the cohort of 51-52 year old in the year 2016
Source: The Future Elderly Model Simulation
20
21
22
23
24
25
26
27
Chapter 2 - Racial and Ethnic Disparities in Medication
Adherence Among Privately Insured Patients in the
United States
28
Abbreviations and Acronyms
SES – Socioeconomic Status
PDC – Proportion of Days Covered
MPR – Medication Possession Ratio
T2DM – Type 2 Diabetes Mellitus
OOP – Out-of-Pocket
29
Text
Introduction
Chronic medications must be taken as prescribed to be effective, yet poor adherence is endemic.
30-39
Nearly half of all patients that were prescribed pharmaceutical therapies do not take sufficient doses to
experience therapeutic benefits.
40
Suboptimal medication-taking behavior, consisting of poor adherence
among existing users and discontinuation of therapy, is particularly acute in racial/ethnic minority
populations. Minorities typically have higher prevalence of chronic disease, worse access to medical
care, and greater financial constraints.
41-44
Past studies suggest that Blacks and Hispanics are at least 50
percent more likely to have suboptimal adherence rates than Whites, and that differences in
socioeconomic status (SES) may play a significant role in explaining these deficits.
34,45
A large number of U.S. studies have documented suboptimal medication adherence and discontinuation
across a wide range of patient populations and disease conditions, but few have examined the separate
roles of race/ethnicity and SES in explaining disparities in adherence.
38,39,42,45
Studies that rely on
pharmacy claims usually lack information on a patient’s race/ethnicity and socioeconomic well-being, at
best inferring differences in SES from averages of large geographic units such as zip codes or census
tracts.
35,36,45
By contrast, studies that rely on survey data include more accurate and detailed
sociodemographic information, but rely on self-reported adherence measures, which are shown to
overestimate actual adherence.
41,46-48
In this paper, we linked longitudinal pharmacy and medical claims to detailed demographic and
socioeconomic data on privately insured patients to compare race-specific rates of medication adherence
and discontinuation to commonly prescribed medications used to treat hypertension, hyperlipidemia, and
30
Type 2 diabetes mellitus (T2DM). Identifying the independent effects of race/ethnicity and SES on
medication adherence can inform appropriate interventions to reduce disparities in adherence and health
outcomes.
Methods
Data & Sample Selection
This study used a 25% sample of De-identified Clinformatics® Data Mart (OptumInsight, Eden Prairie,
MN), a large-scale claims dataset of privately insured patients with both medical and pharmacy coverage
from a large commercial insurer in the United States from 2011 to 2013.
We linked enrollment data to medical and pharmacy claims over time for commercially insured members
and their dependents. The enrollment data included information on patients’ plan type, coverage period,
gender, and year of birth. The prescription claims included information on the date of each prescription,
type of drug, brand and generic name, active ingredient, and days of supply, as well as financial
information such as the copayment and deductible associated with each claim. The medical claims
captured standard information, including date of service, diagnosis codes, and duration of hospital stays.
Our study focused on chronic medications treating T2DM, hypertension, and hyperlipidemia, three of the
most prevalent and costly chronic conditions in the United States.
49
We identified patients taking
medications for these conditions by examining their prescription drug claims and enrollment files. To
ensure at least a year of follow-up, we included individuals age 18 to 64 who filled two or more
prescriptions in the therapeutic class in 2011 or 2012 and were continuously enrolled in the same benefit
plans through December 31
st
, 2013. Based on these inclusion criteria, our study sample consisted of
56,720 (3.9 percent) beneficiaries taking oral antidiabetics, 156,468 (10.9 percent) beneficiaries taking
31
antihypertensive medications, and 144,673 (10.1 percent) beneficiaries taking antihyperlipidemic
medications (See Table A1 for details of sample selection). The average follow-up period was 2.4, 2.5,
and 2.5 years for patients on oral antidiabetic, antihypertensive, and antihyperlipidemic medications,
respectively.
Socioeconomic Variables & Race/Ethnicity
In addition to the standard demographic information, our linked data contained a detailed set of SES
measures, including race/ethnicity, education level, and household income. Race and household income
were derived by the KMB Group based on a combination of self-reports, public records, purchase
transactions and consumer surveys.
50
Education at the census block group level was derived from U.S.
Census data.
1
Adherence Measures – Proportion of Days Covered (PDC) and Discontinuation
Poor adherence can be a product of three different behavioral pathways: reduced initiation of treatment,
worse adherence among existing users, and/or discontinuation of therapy. Since we could not infer
primary nonadherence (patients who were prescribed a medication but never initiated therapy), we
focused on secondary adherence of patients after they initiated a therapy.
51-53
We used proportion of days
covered (PDC) to measure adherence behavior of current users. PDC is a well-established method for
measuring adherence using claims data
51,52,54,55
and offers a more conservative estimate compared to the
medication possession ratio (MPR).
51,56
We measured adherence at the ingredient level. In cases where a
patient was on multiple medications within the same therapeutic class, the patient was considered
adherent for a particular day if he or she was covered by at least one of the medications within the
1
Census block group is the smallest unit on which the Census bureau publishes sample data. It has a population of
600 to 3,000 people. It is smaller than a Census Track, and larger than a Census Block.
32
therapeutic class. We defined the observation period from the date of first fill within the therapeutic class
to the last day of 2013, excluding days when the patient was hospitalized.
To more accurately reflect adherence among active users, we identified patients who discontinued therapy
altogether. Patients were considered to have discontinued a therapy if they were not in possession of any
drug in the class for at least 180 days. By this definition, 24 to 26 percent of patients discontinued
treatment in the approximately 30-month follow-up period (Table 1). For patients that discontinued a
therapy, we measured their adherence only for the period when they were taking the medication, defined
as the period from the date of the first observed fill to the last day of possession in the follow-up period.
Statistical Analysis – Multivariate Regressions
We first examined differences in adherence by race/ethnicity in each therapeutic class. To understand the
contribution of different observable factors in explaining racial/ethnic disparities in adherence rates,
57
we
performed stepwise linear regressions. We started with a baseline specification with patient-level PDC as
the dependent variable and binary indicators for each race/ethnicity. In turn, we added: patient
demographics (age in log form, gender and geographic area of residence defined by census division),
comorbid conditions (Charlson comorbidity index, and other chronic drugs taken within the observation
period
2
), out-of-pocket (OOP) costs (calculated as the average copay and deductible per day of supply of
medication), education (categorical variable), and household income (categorical variable). Throughout
this process, we observed how racial/ethnic differences changed with additional controls, and whether
these changes varied across the three therapeutic classes. We hypothesized that controlling for a richer set
of socioeconomic variables would yield smaller, yet more accurate estimates of the independent effects of
race/ethnicity on medication adherence.
2
On an active ingredient level.
33
As a second step, we performed two sets of logistic regression models, by therapeutic class, to quantify
and compare the relative risk of suboptimal adherence (PDC < 0.8) and the likelihood of discontinuing
therapy altogether. These models adjusted for all the covariates included in the previous analysis. The
regression coefficients were then used to predict odds ratios by race/ethnicity of suboptimal adherence
and discontinuation.
All statistical analyses were conducted using Stata version 14.1.
Results
Table 1 shows the characteristics of the study sample, by therapeutic class. Overall, 56 percent of patients
in the sample were male, with an average age of 52.6 years. While the majority were white, Blacks and
Hispanics made up a disproportionate share of those taking oral antidiabetic medications. Users of oral
antidiabetic medications had lower levels of educational attainment and household income, and more
comorbidities than those taking antihypertensive or cholesterol-lowering medications. Nearly three-
fourths of the study sample were enrolled in a point of service plan.
Table 2 shows average (unadjusted) adherence rates by race, education and income in each therapeutic
class. Average adherence was highest among patients taking antihypertensives (78.4 percent, oral
antidiabetic 74.7 percent, antihypertensive 75.3 percent), and among those taking medications in 2 or 3 of
the classes analyzed (35.7%). Further, there were significant differences by race/ethnicity, education and
income. Non-White, low educational attainment, and low household income were associated with lower
34
adherence rates in all three classes. For example, the average adherence rate among whites taking an oral
antidiabetic was 8.4 percentage points higher than Hispanics, which is equivalent to one less month (31
days) of medication use per year. The mean PDC of patients in the lowest income bracket (<$40K) was
9.4 percentage points lower than those in the highest income bracket (>$100K), equivalent to 34 less days
covered by medication in a year.
Table 3 shows how racial/ethnic gradients in adherence changed as we added additional sets of covariates.
The baseline model reports unadjusted differences for each racial/ethnic group relative to whites, similar
to Table 2. For example, the average PDC of Hispanics is 7.9 to 9.2 percentage points lower than whites.
Similarly, the average PDC of blacks is 7.5 to 8.7 percentage points lower than whites without
adjustment. Controlling for the full set of covariates reduces these differences by 30% to 40% across the
three drug classes. For example, the average PDC of Hispanics is 4.9 to 6.5 percentage points lower than
whites after adjustment, while the black-white differential is reduced to 4.8 to 5.7 percentage points.
Average adherence rates of Asians are only modestly lower than whites before adjustment, and remain
similar in absolute terms in the full model. The average PDC of Asians is 0.8 to 3.3 percentage points
lower than whites after adjustment.
Table 4 compares the adjusted odds of discontinuing therapy and having suboptimal adherence (PDC
<0.8) for each race/ethnicity relative to whites, by therapeutic class. Blacks and Hispanics are more likely
to discontinue an antihyperlipidemic (odds ratio of 1.19, 1.37 respectively), but not antihypertensive or
antidiabetic medications after adjustment. On the other hand, blacks and Hispanics actively using these
chronic medications are 43 to 66 percent more likely to have poor adherence than Whites, defined as a
PDC<.80.
35
Discussion
We found that controlling for socioeconomic characteristics reduced, but did not eliminate racial and
ethnic disparities in medication adherence. Average adherence rates of blacks and Hispanics were 4.8 to
6.5 percentage points lower than whites across three widely used drug classes, which translates to about
20-25 fewer days of medication use per year. We found smaller racial/ethnic differences in adherence to
medications for more symptomatic conditions such as antidiabetics, wherein the consequences of poor
adherence are more evident to the patient. While blacks and Hispanics take these medications less
consistently than whites, they were not at higher risk for discontinuing a medication, although the results
varied by therapeutic class. This suggests that reducing disparities in adherence should focus on
improving daily adherence rather than cessation of therapy.
Prior work has shown that higher adherence is associated with better management of chronic conditions
(A1C, LDL, and blood pressure), lower hospitalizations, and reduced mortality risk.
58-62
For example, Ho
et al found that nonadherence to antihypertensive and antihyperlipidemic medications was associated with
a 10 to 40 percent relative increase in risk of cardiovascular hospitalizations and a 50 to 80 percent higher
risk of mortality.
63
Thus, even modest improvements in the medication adherence of minorities has
potential to narrow observed differences in health across race/ethnicity.
Our findings are generally consistent with the previous literature, but smaller in magnitude.
34,35,45
This is
likely due to better controls for socioeconomic status, which has been shown to be strongly correlated
with better health and health behaviors.
64,65
We also hypothesized that racial/ethnic differences in
adherences may be driven by differential rates of discontinuation. Patients commonly discontinue
therapy, either at the direction of their provider, or more commonly, on their own due to side-effects,
36
perceived ineffectiveness or cost.
38,66,67
Prior work typically measures adherence over a pre-determined
period of time, such as a calendar year or twelve months after an initial fill date, underestimating overall
adherence rates and often categorizing those who have stopped taking a medication as poor adherers.
While this is an important methodological distinction, it did not explain differences in average adherence
by race/ethnicity.
Our study has several limitations. First, while pharmacy claims have been widely used to estimate
adherence, they do not measure actual pill-taking behavior. Second, our study sample consists of
younger, privately insured patients of above-average SES. Thus, our results may not generalize to lower
SES groups such as a Medicaid population. Third, we measure adherence among those actively using a
medication, excluding periods when patients have discontinued therapy for at least 6 months. We
conservatively used 180 days, but there is no standard definition of discontinuation or persistence to a
chronic medication. Using this definition, we found that about 1 in 5 patients discontinued their chronic
medication over an average follow-up of 2.5 years. Defining discontinuation based on a gap of 90 days or
more reduced the number of patients stopping therapy and lowered average adherence rates, but did not
substantively change adherence rates differentially across race/ethnicity. Lastly, our regression results do
not yield a causal explanation for non-adherence, rather they provide estimates of the effects of education,
income and race separately on adherence, which can guide potential interventions for moderating these
differences.
Public Health Implications
Our findings indicate that non-white patients were no more likely to discontinue a medication than
whites; rather, they took their chronic medications less consistently. While changing patient behavior is
difficult, prescription drug claims and electronic medical records allow for real-time interventions via text
message or low-cost reminder devices that could prove impactful for non-White patients who have gaps
37
in coverage. Alternatively, such data could be monitored to identify patients who did not refill a
medication in a timely fashion, eliciting a call from a pharmacist, nurse or health care provider to
understand the reason for nonadherence and intervene if appropriate. The combination of real-time
pharmacy claims and the near universal adoption of electronic devices could potentially prove more cost-
effective than previous approaches to improving adherence, although there is no conclusive evidence on
the long-term effectiveness of patient reminders.
68,69
38
Tables and Figures
39
Table 1 – Characteristics of Patients by Therapeutic Class
Therapeutic Class
Oral Antidiabetic
(N=56,720)
Antihypertensive
(N=156,468)
Antihyperlipidemic
(N=144,673)
Follow-up Period
a
, mean, y 2.4 2.5 2.5
Discontinued, % 24 26 24
Age, mean (SD), y 52.0 (9.4) 52.9 (8.5) 54.2 (7.5)
Male, % 51 57 60
Race, %
White 67 74 79
Asian 4 3 3
Black 15 14 10
Hispanic 13 9 8
Education, %
College Degree 12 15 18
Some College 52 53 54
High School and Less 36 32 28
Annual Household Income, %
>$100K 28 44 49
$75K-$99K 18 18 18
$60K-$74K 13 12 11
$50K-$59K 9 8 7
$40K-$49K 8 7 6
<$40K 14 11 9
Insurance Plan Type, %
Point of Service (POS) 74 74 75
Exclusive Provider Organization (EPO) 15 14 13
Health Maintenance Organization (HMO) 9 9 9
Others (PPO, Indenity and others) 2 3 3
Charlson Comorbidity Index, mean (SD) 2.3 (2.0) 1.4 (1.8) 1.3 (1.8)
40
Note. Percentages of patients in each group of race/ethnicity and SES variables (education and annual household
income) are calculated based on patients with known race/ethnicity and SES information. The percentages of
patients with unknown race/ethnicity and SES information are reported in supplement table A1.
a
Follow-up period for a therapeutic class is defined as the period from the first fill (in either 2011 or 2013) until Dec.
31
st
, 2013.
41
Table 2 – Average Adherence Rates (PDC) by Race, Education and Household Income
Therapeutic Class
Oral Antidiabetic
(N=56,720)
Antihypertensive
(N=156,468)
Antihyperlipidemic
(N=144,673)
All, mean (SD) 74.7 (24.3) 78.4 (22.2) 75.3 (22.6)
By Race
White 76.9 80.2 77.0
Asian 74.3 77.5 72.9
Black 69.4 72.5 68.4
Hispanic 68.5 72.3 67.9
By Education
College Degree 78.6 81.9 78.4
Some College 75.4 79.1 75.9
High School and Less 72.2 75.8 72.1
By Household Income
HH Income > $100K 78.4 81.4 77.7
$75K-$99K 75.7 79.0 75.5
$60K-$74K 74.2 77.6 74.2
$50K-$59K 73.1 76.2 72.9
$40K-$49K 72.1 75.3 71.7
<$40K 69.0 73.2 69.4
42
Table 3. Contribution of Factors Towards Explaining the Racial Gradient in Medication Adherence
Oral Antidiabetic Antihypertensive Antihyperlipidemic
Average PDC for White Patients, %
76.9 80.2 77.0
Covariates in the Model Differences in Average PDC between Hispanics and Whites
b
, %
Baseline
a
-8.4 -7.9 -9.2
(1) Demographics (Gender, Age, Geographic Area) -6.6 -6.6 -7.8
(2) = (1) + Comorbidities (Charlson Index, Other Drugs Taken) -6.0 -6.2 -7.5
(3) = (2) + Out-of-Pocket (OOP) Cost -6.0 -6.2 -7.6
(4) = (3) + Education -5.4 -5.6 -7.0
(5) = (4) + Income -4.9 -5.2 -6.5
Differences in Average PDC between Blacks and Whites
b
, %
Baseline
a
-7.5 -7.8 -8.7
(1) Demographics (Gender, Age, Geographic Area) -6.4 -6.6 -7.2
(2) = (1) + Comorbidities (Charlson Index, Other Drugs Taken) -6.0 -6.3 -7.1
(3) = (2) + OOP Cost -6.0 -6.2 -7.2
(4) = (3) + Education -5.5 -5.6 -6.4
(5) = (4) + Income -4.8 -5.0 -5.7
Differences in Average PDC between Asians and Whites
b
, %
Baseline
a
-2.6 -2.8 -4.2
(1) Demographics (Gender, Age, Geographic Area) -0.7 -2.1 -3.1
(2) = (1) + Comorbidities (Charlson Index, Other Drugs Taken) 0.1 -1.7 -2.7
43
Note. This table shows how the racial differences in adherence evolved as we added in more controls into the model. In each step, the controls included all the
controls from previous steps.
a
Baseline gap in PDC are the difference in average (unadjusted) in PDC between racial groups. The averages for different racial groups were reported in Table 2.
b
Differences in average PDC are calculated as the differences relative to Whites. A negative value indicates the group has lower average PDC than Whites.
(3) = (2) + OOP Cost 0.1 -1.7 -2.8
(4) = (3) + Education -0.9 -2.3 -3.4
(5) = (4) + Income -0.8 -2.2 -3.3
44
Table 4. Adjusted Odds Ratio of Discontinuation and Nonadherence (PDC <0.8) by Race/Ethnicity
% of Patients
That Discontinued
Adjusted Odds Ratio
Of Discontinuation
% of Nonadherent
Patients (PDC < 0.8)
Adjusted Odds Ratio
Of Being Nonadherent
Oral Antidiabetic
White 23.0 1 (Reference) 43.1 1 (Reference)
Asian 22.1 0.80 (0.72-0.89) 48.2 1.10 (1.01-1.20)
Black 24.8 0.93 (0.88-0.99) 56.7 1.43 (1.36-1.51)
Hispanic 26.5 1.02 (0.96-1.09) 58.2 1.44 (1.36-1.52)
Antihypertensive
White 24.6 1 (Reference) 36.1 1 (Reference)
Asian 24.4 0.98 (0.91-1.06) 41.2 1.18 (1.11-1.27)
Black 29.0 1.05 (1.01-1.08) 51.8 1.52 (1.47-1.57)
Hispanic 28.9 1.10 (1.05-1.15) 51.3 1.50 (1.44-1.56)
Antihyperlipidemic
White 22.1 1 (Reference) 43.1 1 (Reference)
Asian 25.2 1.13 (1.06-1.22) 51.9 1.34 (1.26-1.42)
Black 29.2 1.19 (1.14-1.24) 58.4 1.50 (1.44-1.56)
Hispanic 31.5 1.37 (1.31-1.43) 60.4 1.66 (1.60-1.73)
Note. Patients are considered to have discontinued the medications if they are not uncovered by any medication in the last 180 days of the follow-up period.
Adjusted odds ratios of discontinuation and being nonadherence are both calculated based on our multivariate logistic regression models, adjusting for
demographics, comorbidites, SES and out-of-pocket cost.
45
Chapter 3 – The Effects of Extended-Release
Formulations on Medication Adherence, A1C Control and
Hospital Outcomes Among Patients Treated with Type 2
Diabetes Therapies
46
Text
Introduction
Adherence to oral antidiabetic therapies is essential to achieve glycemic control and improved long-run
health outcomes among Type 2 diabetes patients.
70,71
Yet, poor adherence to chronic oral antidiabetic
therapies is endemic.
66,71,72
It has been estimated that suboptimal adherence to chronic medications cost
the US over $100 billion dollars annually.
72
Finding an effective solution to addressing long-run suboptimal adherence behavior, however, has proved
challenging.
66,73
Past studies have suggested the use of extended-release (XR) formulations as potential
strategies to improve adherence, as they reduce dosing frequencies and provide other pharmacokinetic
benefits.
66,74-76
While the effects of XR formulations on adherence have been widely studied, their impacts
on long-run hospital outcomes remain inconclusive.
77
Identifying the effects of formulations on long-run
adherence and hospital outcomes could inform designing cost-effective interventions to improve patients’
adherence and long-run health outcomes.
In this project, we used a large-scale pharmacy and medical claims datasets from a US-based insurance
provider to estimate the effects of prescribing XR formulations on long-run (> 1 year) adherence and
hospital outcomes. We identified a group of patients that were prescribed either XR or non-XR
formulation of metformin or glipizide, two of the most widely prescribed oral antidiabetic agents,
78
for at
least a year from 2011 to 2012. We measured their medication adherence, changes in A1C levels, and
hospital outcomes. Multivariate regression analyses were conducted to study the association between
formulation and these health behavior and outcomes. To address patients’ selection into different
formulations, we also built an endogenous binary variable model with an instrumental variable to estimate
the casual effects of using XR formulations on health behavior and outcomes.
47
Methods
Data
This study uses a 25% sample of De-identified Clinformatics® Data Mart (OptumInsight, Eden Prairie,
MN) from 2011 to 2013.
79
We used member enrollment, pharmacy, lab test, and hospitalization claims
data to identify patients treated with Type 2 diabetes therapies, and estimate their health behavior and
outcomes. Member enrollment files contain detailed information on patients’ enrollment period, age,
gender and plan type. Pharmacy claims include detailed information on medication, fill date, days of
supply, and out-of-pocket cost. Lab test claims include test date, test type, and test results. Only lab
results ordered through OPTUM and its affiliates were available. Hospitalization claims files include
admission date, length of stay, five diagnosis codes, standard cost, and out-of-pocket cost. Pharmacy
claims were used to quantify patients’ adherence to their medications, lab test for glycemic control
(changes in A1C), and hospitalization claims for hospital outcomes.
Sample selection
Patients were selected into the sample, if they were between 18 to 64 in age, continuously enrolled in the
same pharmacy and medical benefit plans from 2011 to 2013, and had at least two fills (at least 30 days
apart) of either formulation of metformin or glipizide from 2011 to 2012. To ensure at least a year of
follow-up period in measuring adherence, we required the first prescription to be observed in the year
2011. Patients who used both formulations were excluded from our sample (3.6% of patients were
dropped). To satisfy one of the key identification assumption in our casual model, we also restricted our
analysis to patients who come from “large” groups with at least 100 members (details to be explained in
the statistical analysis session).
Metformin and glipizide were included among the plethora of oral antidiabetic medications in our study,
since they had both formulations available throughout the observation period, intended for long-term use,
48
and had at least 100 unique users in either formulation group. Patients that prescribed both formulations
were excluded from our sample.
Outcomes measures
Three types of health behavior and outcomes in this study: adherence, glycemic control, and count of
diabetes-related hospital admissions. The chart below provides an overview of the measurement time
periods for each outcome.
Medication adherence is measured on a drug and formulation level (e.g. Metformin XR) by the
medication possession ratio (MPR), from the first fill date until 12/31/2012.
51
We used changes in A1C level between the first and last available A1C readings from lab results files to
assess patients’ glycemic control. Changes in A1C level was calculated as the difference between the first
and last available A1C readings, for patients with at least two A1C lab test claims. Following Menzin et
al. (2011), changes in A1C were calculated only for patients whose first A1C test took place within 6
months of the first observed fill date for a drug, and there must be more than 12 months between the test
dates of the first and last tests.
80
49
We examined counts of diabetes-related hospitalization outcomes in the year 2013. Hospitalizations were
considered diabetes-related based on a diagnosis, in any of the 5 diagnosis codes (ICD-9CM), for 1 of the
16 selected complications of diabetes identified in Menzin et al. 2010.
80
Statistical Models
We conducted two types of statistical analyses to investigate the relationship between XR use and
outcomes of interests. The first was a set of multivariate regression models that adjusted for a
comprehensive set of patient and plan characteristics. To address patients’ selection into different kinds of
formulations based on unobserved characteristics, we also built an endogenous binary variable model
with an instrumental variable, which only affected patients’ selection into formulations but not directly on
hospital outcomes.
Multivariate Regression Models
In the multivariate regression models, we used linear specifications for adherence and changes in A1C
models, and Poisson regression for count of hospital admission. All models controlled for XR status,
patients’ demographics characteristics (race, gender, age, and geographic area of residence),
socioeconomic status (patient-level education, annual household income), health status (other drugs taken,
Charlson index, and whether any hospital stays before), month of the first fill, and type of medication
(dummy variable indicating whether it is metformin or glipizide). For hospitalization models, we also
added plan-level average out-of-pocket cost of hospitalization, calculated based on hospitalization claims
of year 2011 and 2012, as measure of plan generosity towards hospitalization. Changes in A1C models
included the baseline (first) A1C reading.
From the output of multivariate regression models, we examined whether dummy variable for XR
remained statistically significant after controlling for the wide range of covariates. Based on the results,
we calculated the adjusted rates of adherence, A1C control and hospital outcomes.
50
Causal Models - Endogenous binary variable model and Instrumental variable (IV)
Since patients were unlikely to be randomly selected into different formulations, differences in health
outcomes between the two observed patient populations was resulted from the effects of XR treatment,
and the differences in patient characteristics that affected their selection into different formulations and
subsequent health outcomes. Multivariate regression analyses reduced the effects of selection into
formulations, but we were unlikely to have controlled for all the characteristics that affected the selection
and health outcomes. If XR-users were inherently sicker, measured by unobserved health status, a
comparison between XR and non-XR users will underestimate the true effects of XR at the entire patient
population level.
We used an endogenous binary variable model (EBVM) to model the effects of unobserved factors on
selection into different formulation, and related health outcomes. An important assumption of the model
is that the unobserved factors affect both the probability of patients’ getting XR formulations, and the
health outcomes. These effects are captured in the residual terms in the equations that determines the
probability of getting XR, and the health outcomes. EBVM assumes a bivariate normal correlation
structure between the error terms.
The model can be formally presented as the following.
𝑙𝑙𝑙𝑙𝑙𝑙 𝑗𝑗 𝑑𝑑 𝑙𝑙 𝑑𝑑𝑑𝑑 𝑙𝑙 𝑙𝑙 𝑎𝑎 𝑑𝑑 𝑖𝑖 𝑑𝑑 𝑑𝑑 𝑖𝑖𝑖𝑖 𝑖𝑖 𝑑𝑑 𝑖𝑖 𝑎𝑎𝑙𝑙
𝑦𝑦 𝑗𝑗 − 𝑑𝑑 𝑖𝑖𝑙𝑙 𝑜𝑜 𝑑𝑑 𝑜𝑜 𝑙𝑙 𝑖𝑖 𝑎𝑎 𝑣𝑣 𝑖𝑖𝑎𝑎 𝑣𝑣 𝑙𝑙 𝑙𝑙
𝑋𝑋 𝑗𝑗 − 𝑖𝑖𝑙𝑙 𝑜𝑜𝑙𝑙 𝑑𝑑 𝑣𝑣 𝑑𝑑 𝑜𝑜 𝑜𝑜𝑑𝑑 𝑖𝑖 𝑎𝑎𝑣𝑣 𝑖𝑖𝑎𝑎𝑙𝑙 𝑙𝑙 𝑐𝑐
𝑋𝑋𝑋𝑋
𝑗𝑗 − 𝑤𝑤 ℎ 𝑙𝑙𝑙𝑙 ℎ 𝑙𝑙𝑣𝑣 𝑋𝑋𝑋𝑋 𝑜𝑜𝑑𝑑 𝑣𝑣𝑜𝑜 𝑖𝑖 𝑙𝑙𝑎𝑎
𝑦𝑦 𝑗𝑗 = 𝑋𝑋 𝑗𝑗 𝛽𝛽 + 𝛿𝛿 𝑋𝑋𝑋𝑋
𝑗𝑗 + 𝜀𝜀 𝑗𝑗
𝑋𝑋𝑋𝑋
𝑗𝑗 ∗
= 𝑋𝑋 𝑗𝑗 𝛾𝛾 + 𝜃𝜃 𝐼𝐼𝐼𝐼
𝑗𝑗 + 𝜐𝜐 𝑗𝑗
𝑋𝑋𝑋𝑋 = �
1, 𝑖𝑖 𝑜𝑜 𝑋𝑋𝑋𝑋
𝑗𝑗 ∗
> 0
0, 𝑂𝑂 𝑙𝑙 ℎ 𝑙𝑙𝑣𝑣𝑤𝑤𝑖𝑖 𝑐𝑐𝑙𝑙
� 𝜀𝜀 𝑗𝑗 , 𝜐𝜐 𝑗𝑗 � ~ 𝐵𝐵 𝑖𝑖𝑖𝑖 𝑎𝑎𝑣𝑣 𝑖𝑖𝑎𝑎𝑙𝑙 𝑙𝑙 𝑁𝑁 𝑑𝑑𝑣𝑣 𝑜𝑜 𝑎𝑎 𝑙𝑙 {(0,0), �
𝜎𝜎 2
𝜌𝜌 𝜎𝜎 𝜌𝜌 𝜎𝜎 1
�
51
The EBVM is estimated using maximum likelihood.
81
To improve identification, we also develop a plan-level relative XR cost index, which plausibly only
affects patients’ decision to use XR formulations, but not directly on health outcomes. The relative XR
cost index is a weighted average of the differences in out-of-pocket cost between XR and non-XR
formulations across the 21 most prescribed medications with both formulations in the claims database.
These 21 drugs had both formulations available from 2011 to 2013, at least 100 users in either
formulation group.
IV
i
= � w
j
(copay
i, j
XR
−
2 1
j = 1
copay
i, j
N on − XR
)
(i – group number, j – molecule/drug, 𝑤𝑤 𝑗𝑗 - determined by total days of supply)
In theory, for an IV to be effective, two assumptions must be satisfied:
(1) (correlation) the IV is effective in predicting the formulation choice
(2) (exclusion) the IV does not directly predict health outcomes, or in individual health status of
patients should have little or no influence over the generosity in coverage of XR formulations.
Cost has been shown to be a concern in utilization of healthcare services.
38
The relative cost index
provided us with a measure/proxy for the generosity of the plans towards XR formulations. As we used
the out-of-pocket cost of 21 different kinds of medications, the influence of the health status of diabetes
patients on plan generosity should be greatly reduced. To further satisfy the exclusion assumption, we
also further selected our sample to include only patients from groups with at least 100 users, which
reduces the impact of individual health status on plan generosity.
Results
Table 1 summarizes the key characteristics of patients selected into our sample by XR status. 23,409
patients were included in our sample. 27% of the patients were users of XR formulations, and 85% used
52
Metformin. XR users were younger (52.3 vs 53.1 years old) and had higher socioeconomic status,
measured by education attainment and annual household income, than non-XR users. 13.2% of XR users
were college graduates, compared to 10.8% for non-XR users. 38.6% of XR users had an annual
household income of more than 100K USD, compared to 10.8% for non-XR users. In the baseline, XR
users also had better health status, measured by average A1C at the baseline, and number of other
medications taken.
Table 2 summarizes the key outcome measures by XR status. Compared with non-XR users, XR users
had higher average adherence rates (78.7% vs. 75.6%, p<0.01). Among the subset of patients with
changes in A1C measures, XR and non-XR users had similar reductions in A1C readings (XR Users: -
0.15%, non-XR users: - 0.19%, p>0.1). In the year 2013, XR-users had a lower probability of being
admitted into hospitals, measured by both all cause and diabetes-related hospitalizations. The per capita
count of diabetes-related hospitalization for XR users is -0.025 lower than that of non-XR users (XR-
users: 0.043, non-XR users: 0.064, p<0.01)
Table 3a & 3b summarize the key findings from multivariate regression analyses. After controlling for
covariates, the use of XR was significantly associated with lower diabetes-related hospitalization, higher
adherence, and greater reduction in A1C (all with p<0.01). On an annual basis, XR users had 0.018 (95%
CI: 0.009 to 0.028) fewer cases of hospitalization compared with non-XR users. The average adjusted
adherence rates of XR users were 2.6 percentage points higher than those of non-XR users (95% CI: 2.0
to 3.3). Among the 5,430 patients with changes in A1C measures, the use of XR was associated with 0.14
percentage points more reductions in A1C (95% CI: 0.05 to 0.23).
Table 6a & 6b summarize the results from our causal models estimates. The average relative cost index of
XR formulations is lower for existing XR users (Table 4). The XR index does not directly predict the
outcomes of interests (count of hospitalizations), indicating potential for satisfying the exclusion
restrictions (Table 5). In all structural estimations, we found that the instrumental variable, the generosity
53
index, negatively affects the probability of being prescribed XR formulations. Use of XR lowers the per
capita count of diabetes-related hospitalization by 0.163 annually (95 CI: -0.209 to -0.118), improves
adherence by 7 percentage points (95% CI: 5.5 to 8.8), and further lower A1C by 1.5 percentage points
(95% CI: 1.1 to 1.9).
The model diagnostic statistics also provide us with valuable information about the validities of the
models (Table 6a & 6b). The chi-square statistics confirm that the correlation coefficient between the
error terms are statistically different from zero, confirming our model choice. The estimates output of the
correlation coefficients between the error terms of the selection into XR and count of diabetes-related
hospitalization equation is 0.603, which indicates that the unobserved factors that increases the
probability of patients’ being prescribed XR formulations also increases the count of diabetes-related
hospitalization.
Discussion
In this longitudinal investigation of the effects of the prescribing XR formulations among patients treated
with metformin or glipizide therapies, our analysis show that the use of XR formulations could potentially
lead to improved adherence (up to 7 percentage points in adherence rates, measured by MPR), 1.5
percentage points of in reduction in A1C, and a per capita reduction of 0.163 cases of diabetes-related
hospitalization annually.
Based on our model estimates, among the 6.9 million patients of the age range of 18 to 64 treated with
Type 2 diabetes in the United States were taking extended-release formulations, diabetes-related
hospitalization would reduce by at least 96,600, which translates into savings of at least $1.8 billion US
dollars annually (Table 7). The change from non-XR to XR formulations are estimated to cost an extra
$232 million US dollars. Extended-release formulations could be a cost-effective strategy to reduce
nonadherence and social cost of Type-2 diabetes.
54
Our findings of improved adherence and the use of XR formulation on adherence is consistent with the
findings from previous literature, but smaller in scale.
77,82,83
The improve adherence translates into greater
reduction in A1C among XR-users and lower all-cause and diabetes-related hospitalization.
66,70,84
But it is
worth noting that the reduction in diabetes-related hospitalization could also be attributed to other
pharmacokinetic benefits of XR formulations besides improved adherence.
Our causal model results indicate that higher out-of-pocket cost of XR formulations could be a barrier to
prescribing XR formulations among patients treated with Type-2 diabetes therapies. XR-users in our
sample came from plans with more generous coverage of XR formulations, and the generosity index had
a significantly negative impact on the probability of getting XR formulations. Expanding coverage of XR
formulations, and lowering patient out-of-pocket cost of XR formulations could be an effective strategy to
improve adherence, and long-run health outcomes.
The correlation coefficients from the causal models also provide us with additional insights into the
unobserved selection mechanism into different formulations. Patients whose unobserved health status led
them to higher risk of more diabetes-related hospitalizations were also more likely to be prescribed XR
formulations. This indicates that XR users in our dataset were patients with worse unobserved health
status. This could explain the discrepancy in estimates of the effects of XR formulations on
hospitalization between multivariate and the causal models. Since the multivariate models did not control
for the unobserved health status, the effects of XR on hospitalizations were biased downwards, as XR
users were sicker. By correcting for the unobserved health status in the casual model, the reductions in
diabetes-related hospitalization were greater in the causal models than the multivariate ones.
Our study has several limitations. The study population are privately insured and of high SES, thus they
might not generalize to the entire US population. While claims-based analysis offers an objective way to
measure adherence, it might not reflect the actual adherence. We constructed the generosity index to
proxy for the generosity of the insurance plans towards XR formulations, but it is an imperfect measure.
55
Endogenous formulation of patient groups, and the discrepancy between observed copay and actual
benefit terms could invalidate our identification strategies. Nontheless, this measure provides us with a
novel strategy to identify the casual impacts of formulations on health behavior and outcomes. We did not
separately identify how much of the reductions in hospitalizations is due to improved adherence, versus
other pharmacokinetic channels.
56
Tables and Figures
57
Table 1 – Characteristics of Patients by XR Status
User Type All (N=23,409) Non-XR (N=16,990) XR (N=6,419)
Male, % 50.8 52.3 46.8
Age, mean, y 52.9 53.1 52.3
Metformin, % 85.3 90.3 72.2
Race, %
White 63.2 62.2 65.6
Education, %
College Graduate 11.5 10.8 13.2
Annual Household Income, %
>$100K 35.2 33.9 38.6
Health Status
Other Drugs Taken 6.7 6.8 6.6
Charlson Comorbidity Index, mean 2.1 2.1 2.0
Any Hospital Stay Before, % 14.3 14.7 13.1
All group differences are statistically significant (p<0.01).
58
Table 2 – Summary Statistics of Outcomes by XR Status
Patient Type
All
(N=23,409)
Non-XR
(N=16,990)
XR
(N=6,419)
Adherence
MPR, mean, % 76.4 75.6 78.7
Hospital Outcomes in 2013
Any Hospitalization, % 8.4 8.7 7.6
Any Diabetes-related Hospitalization, % 4.2 4.5 3.3
Count of Diabetes-related Hospitalization per capita, mean 0.059 0.064 0.043
A1C Results
Count of Patients with A1C results 5,432 3,954 1,479
First A1C Reading, % 7.6 7.7 7.4
Last A1C Reading, % 7.5 7.5 7.3
Changes between first and last A1C reading
a
, %
-0.18 -0.19 -0.15
a – while XR users have less reduction in A1C, the difference is not statistically significant
59
Table 3a – Use of XR was associated with lower hospitalizations
Outcome Count of All-Cause Hosp. Count of Diab.-related Hosp.
Models Difference in Means Poisson Difference in Means Poisson
N 23,409
Coefficients
XR Dummy - -0.194*** - -0.343***
Predictions (with 95% CI)
XR – Non-XR -0.030 [-0.044, -0.015] -0.023 [-0.038, -0.008] -0.021 [-0.030, -0.012] -0.018 [-0.028, -0.009]
XR 0.104 [0.092, 0.115] 0.108 [0.096, 0.120] 0.043 [0.036, 0.050] 0.045 [0.038, 0.052]
Non-XR 0.133 [0.125, 0.142] 0.131 [0.123, 0.139] 0.064 [0.059, 0.070] 0.063 [0.058, 0.069]
Model Diagnostics
Pseudo R-Square - 0.106 - 0.131
60
Table 3b – Use of XR was associated with better adherence, and greater reduction in A1C
Outcome MPR Changes in A1C, %
Models Difference in Means Linear Difference in Means Linear
N 23,409 5,430
Coefficients
XR Dummy - 0.026*** - -0.14***
Predictions (with 95% CI)
XR – Non-XR 0.031 [0.024, 0.038] 0.026 [0.020, 0.033] 0.04 [-0.06, 0.15] -0.14 [-0.23, -0.05]
XR 0.787 [0.781, 0.793] 0.784 [0.778, 0.789] -0.15 [-0.24, -0.06] -0.28 [-0.36, -0.21]
Non-XR 0.756 [0.752, 0.759] 0.757 [0.754, 0.761] -0.19 [-0.25, -0.14] -0.14 [-0.19, -0.10]
Model Diagnostics
R-Square - 0.103 - 0.284
61
Table 4 – Summary statistics of IV by XR Status
User Type Min Mean Max SD
Non-XR -16.04 7.78 30.55 4.28
XR -2.59 7.58 30.55 3.96
All -16.04 7.72 30.55 4.20
62
Table 5 – First stage IV exclusion assumption check
Model Probit Poisson Poisson
Dependent variable Use of XR All-cause Hosp. Diab. Hosp.
Statistics coef/p coef/p coef/p
Relative Cost Index -0.007*** -0.003 0.001
0.00 0.59 0.88
* All models controlled for the same set of covariates as in the multivariate regression models
63
Table 6a – XR led to lower hospitalizations
Outcome Count of All-cause Hosp. Count of Diab.-related Hosp.
N 23,409
Coefficients
Plan-level Relative Cost of XR -0.007*** -0.007***
XR Dummy -1.358*** -2.260***
Model Diagnostics
Rho 0.443*** 0.603***
Chi-square statistics 29.36*** 86.54***
Predictions (with 95% CI)
XR – Non-XR -0.162 [-0.218, -0.107] -0.163 [-0.209, -0.118]
If everyone uses XR 0.056 [0.044, 0.068] 0.019 [0.015, 0.223]
If everyone uses Non-XR 0.218 [0.174, 0.263] 0.183 [0.139, 0.226]
64
Table 6b – Use of XR improved adherence, and led to greater reduction in A1C.
Outcome MPR Changes in A1C, %
N 23,409 5,430
Coefficients
Plan-level Relative Cost of XR -0.008*** -0.022***
XR Dummy 0.070*** -1.502***
Model Diagnostics
Rho -0.117*** 0.513***
Chi-square statistics 30.03*** 36.68***
Predictions (with 95% CI)
XR – Non-XR 0.070 [0.051, 0.088] -1.502 [-1.906, -1.098]
If everyone uses XR 0.815 [0.802, 0.829] -1.275 [-1.578, -0.972]
If everyone uses Non-XR 0.745 [0.739, 0.751] 0.227 [0.111, 0.343]
65
Table 7 – Social Gain Calculation from Prescribing XR formulations
Social Gain From Prescribing XR Formulations
Models Poisson Structural with IV (EBVM)
Diabetes-related Hospitalization
Status Quo (27% XR users) 407,100 407,100
If everyone uses XR 310,500 131,100
Improvement 96,600 276,000
Potential Savings* 1.8 Billion 5.2 Billion
Potential savings = reduction in hospitalization * average cost per admission ($18,666)
66
REFERENCES
1. National Health Expenditure Data Highlights. Center for Medicare & Medicaid Services
Report. 2015.
2. Ma J, Ward EM, Siegel RL, Jemal A. TEmporal trends in mortality in the united states,
1969-2013. JAMA. 2015;314(16):1731-1739.
3. Underlying Causes of Death 1999-2015. 2016. http://wonder.cdc.gov/ucd-icd10.html
4. Mozaffarian D, Benjamin EJ, Go AS, et al. Heart Disease and Stroke Statistics—2016
Update. American Heart Association. 2016;133(4):e38-e360.
5. Ni H, Xu J. Recent Trends in Heart Failure-related Mortality: United States, 2000–2014.
NCHS Data Brief. December 2015;No. 231.
6. Bibbins-Domingo K, Pletcher MJ, Lin F, et al. Racial Differences in Incident Heart
Failure among Young Adults. New England Journal of Medicine. 2009;360(12):1179-
1190.
7. Huffman MD, Berry JD, Ning H, et al. Lifetime Risk for Heart Failure Among White and
Black AmericansCardiovascular Lifetime Risk Pooling Project. Journal of the American
College of Cardiology. 2013;61(14):1510-1517.
8. East MA, Peterson ED, Shaw LK, Gattis WA, O'Connor CM. Racial differences in the
outcomes of patients with diastolic heart failure. American Heart Journal.
2004;148(1):151-156.
9. Durstenfeld MS, Ogedegbe O, Katz SD, Park H, Blecker S. Racial and Ethnic
Differences in Heart Failure Readmissions and Mortality in a Large Municipal
Healthcare System. JACC: Heart Failure. 2016;4(11):885-893.
10. Fonarow GC, Hernandez AF, Solomon SD, Yancy CW. Potential mortality reduction
with optimal implementation of angiotensin receptor neprilysin inhibitor therapy in heart
failure. JAMA Cardiology. 2016;1(6):714-717.
11. Yancy CW, Jessup M, Bozkurt B, et al. 2013 ACCF/AHA Guideline for the Management
of Heart Failure. A Report of the American College of Cardiology Foundation/American
Heart Association Task Force on Practice Guidelines. 2013;128(16):e240-e327.
12. Sacubitril/valsartan (entresto) for heart failure. JAMA. 2015;314(7):722-723.
13. Goldman DP, Shang B, Bhattacharya J, et al. Consequences of health trends and medical
innovation for the future elderly. Health Affairs. 2005;24 Suppl 2:W5R5-17.
14. Goldman DP, Cutler D, Rowe JW, et al. Substantial health and economic returns from
delayed aging may warrant a new focus for medical research. Health Affairs.
2013;32(10):1698-1705.
15. Goldman DP, Orszag PR. The Growing Gap in Life Expectancy: Using the Future
Elderly Model to Estimate Implications for Social Security and Medicare. American
Economic Review: AEA Papers and Proceedings. 2014;104(5):230-233.
16. Gaudette E, Goldman DP, Messali A, Sood N. Do Statins Reduce the Health and Health
Care Costs of Obesity? PharmacoEconomics. 2015.
17. Goldman DP, Zheng Y, Girosi F, et al. The benefits of risk factor prevention in
Americans aged 51 years and older. American Journal of Public Health.
2009;99(11):2096-2101.
67
18. Goldman DP, Leaf DE, Sullivan J, Tysinger B, Xie Z. Innovation in Heart Failure
Treatment: Life Expectancy, Disability, and Health Disparities - Technical Appendix.
USC Schaeffer Center Working Paper. 2017.
19. Dolan P. Modeling valuations for EuroQol health states. Medical Care.
1997;35(11):1095-1108.
20. Shaw JW, Johnson JA, Coons SJ. US valuation of the EQ-5D health states: development
and testing of the D1 valuation model. Medical Care. 2005;43(3):203-220.
21. Gold M, Siegel J, Russell L, Weinstein M. Cost-effectiveness in Health and Medicine.
New York, NY: Oxford University Press; 1996.
22. Heidenreich PA, Albert NM, Allen LA, et al. Forecasting the Impact of Heart Failure in
the United States. A Policy Statement From the American Heart Association. 2013.
23. Heidenreich PA, Trogdon JG, Khavjou OA, et al. Forecasting the Future of
Cardiovascular Disease in the United States. A Policy Statement From the American
Heart Association. 2011;123(8):933-944.
24. Huffman MD, Berry JD, Ning H, et al. Lifetime Risk for Heart Failure Among White and
Black Americans: Cardiovascular Lifetime Risk Pooling Project. Journal of the American
College of Cardiology. 2013;61(14):1510-1517.
25. Wong CY, Chaudhry SI, Desai MM, Krumholz HM. Trends in Comorbidity, Disability,
and Polypharmacy in Heart Failure. The American Journal of Medicine.124(2):136-143.
26. McMurray JJV, Packer M, Desai AS, et al. Angiotensin–Neprilysin Inhibition versus
Enalapril in Heart Failure. New England Journal of Medicine. 2014;371(11):993-1004.
27. Fala L. Entresto (Sacubitril/Valsartan): First-in-Class Angiotensin Receptor Neprilysin
Inhibitor FDA Approved for Heart Failure. Am Health Drug Benefits. 2016;9(Spec
Feature):78-82.
28. Quality AfHRa. National Healthcare Disparities Report 2007. Feburary, 2008.
29. Étienne Gaudette BT, Alwyn Cassil and Dana Goldman. Health and Health Care of
Medicare Beneficiaries in 2030. USC Schaeffer Center Working Paper. 2015.
30. Holmes HM, Luo R, Hanlon JT, Elting LS, Suarez-Almazor M, Goodwin JS. Ethnic
Disparities in Adherence to Antihypertensive Medications in Medicare Part D
Beneficiaries. Journal of the American Geriatrics Society. 2012;60(7):1298-1303.
31. Lauffenburger JC, Robinson JG, Oramasionwu C, Fang G. Racial/Ethnic and Gender
Gaps in the Use and Adherence of Evidence-Based Preventive Therapies among Elderly
Medicare Part D Beneficiaries after Acute Myocardial Infarction. Circulation.
2014;129(7):754-763.
32. Zhang Y, Baik SH. Race/Ethnicity, Disability, and Medication Adherence Among
Medicare Beneficiaries with Heart Failure. Journal of General Internal Medicine.
2014;29(4):602-607.
33. Hussein M, Waters TM, Chang CF, Bailey JE, Brown LM, Solomon DK. Impact of
Medicare Part D on Racial Disparities in Adherence to Cardiovascular Medications
Among the Elderly. Medical Care Research and Review. 2015;73(4):410-436.
34. Benner JS, Glynn RJ, Mogun H, Neumann PJ, Weinstein MC, Avorn J. Long-term
persistence in use of statin therapy in elderly patients. JAMA. 2002;288(4):455-461.
35. Trinacty CM, Adams AS, Soumerai SB, et al. Racial differences in long-term adherence
to oral antidiabetic drug therapy: a longitudinal cohort study. BMC Health Services
Research. 2009;9(1):24.
68
36. Lewey J, Shrank WH, Avorn J, Liu J, Choudhry NK. Medication Adherence and
Healthcare Disparities: Impact of Statin Co-Payment Reduction. The American journal of
managed care. 2015;21(10):696.
37. Choudhry NK, Bykov K, Shrank WH, et al. Eliminating medication copayments reduces
disparities in cardiovascular care. Health affairs. 2014;33(5):863-870.
38. Goldman DP, Joyce GF, Zheng Y. Prescription drug cost sharing: Associations with
medication and medical utilization and spending and health. JAMA. 2007;298(1):61-69.
39. Alsabbagh MHDW, Lemstra M, Eurich D, et al. Socioeconomic Status and
Nonadherence to Antihypertensive Drugs: A Systematic Review and Meta-Analysis.
Value in Health.17(2):288-296.
40. Haynes RB, McKibbon KA, Kanani R. Systematic review of randomised trials of
interventions to assist patients to follow prescriptions for medications. The Lancet.
1996;348(9024):383-386.
41. Heisler M, Faul JD, Hayward RA, Langa KM, Blaum C, Weir D. Mechanisms for racial
and ethnic disparities in glycemic control in middle-aged and older Americans in the
health and retirement study. Arch Intern Med. 2007;167.
42. Adams AS, Trinacty CM, Zhang F, et al. Medication Adherence and Racial Differences
in A1C Control. Diabetes Care. 2008;31(5):916.
43. Lewey J, Shrank WH, Bowry ADK, Kilabuk E, Brennan TA, Choudhry NK. Gender and
racial disparities in adherence to statin therapy: A meta-analysis. American Heart
Journal. 2013;165(5):665-678.e661.
44. Saha S, Freeman M, Toure J, Tippens KM, Weeks C, Ibrahim S. Racial and Ethnic
Disparities in the VA Health Care System: A Systematic Review. Journal of General
Internal Medicine. 2008;23(5):654-671.
45. Adams AS, Uratsu C, Dyer W, et al. Health system factors and antihypertensive
adherence in a racially and ethnically diverse cohort of new users. JAMA internal
medicine. 2013;173(1):54-61.
46. Gerber BS, Cho YI, Arozullah AM, Lee S-YD. Racial Differences in Medication
Adherence: A Cross-Sectional Study of Medicare Enrollees. The American journal of
geriatric pharmacotherapy. 2010;8(2):136-145.
47. Adams AS, Soumerai SB, Lomas J, Ross-Degnan D. Evidence of self-report bias in
assessing adherence to guidelines. International Journal for Quality in Health Care.
1999;11(3):187-192.
48. Stirratt MJ, Dunbar-Jacob J, Crane HM, et al. Self-report measures of medication
adherence behavior: recommendations on optimal use. Translational Behavioral
Medicine. 2015;5(4):470-482.
49. Control CoD. Chronic Diseaes Prevention and Health Promotion. In:2016.
50. Wells TS, Ozminkowski RJ, Hawkins K, Bhattarai GR, Armstrong DG. Leveraging big
data in population health management. Big Data Analytics. 2016;1(1):1.
51. Sattler ELP, Lee JS, Perri M. Medication (Re)fill Adherence Measures Derived from
Pharmacy Claims Data in Older Americans: A Review of the Literature. Drugs & Aging.
2013;30(6):383-399.
52. Capoccia K, Odegard PS, Letassy N. Medication Adherence With Diabetes Medication.
The Diabetes Educator. 2015;42(1):34-71.
69
53. Raebel MA, Schmittdiel J, Karter AJ, Konieczny JL, Steiner JF. Standardizing
terminology and definitions of medication adherence and persistence in research
employing electronic databases. Medical care. 2013;51(8 0 3):S11.
54. Gu Q, Zeng F, Patel BV, Tripoli LC. Part D coverage gap and adherence to diabetes
medications. The American journal of managed care. 2009;16(12):911-918.
55. Peacock E, Krousel-Wood M. Adherence to Antihypertensive Therapy. Medical Clinics
of North America. 2017;101(1):229-245.
56. Nau DP. Proportion of days covered (PDC) as a preferred method of measuring
medication adherence.
57. Smith JP. Nature and causes of trends in male diabetes prevalence, undiagnosed diabetes,
and the socioeconomic status health gradient. Proceedings of the National Academy of
Sciences of the United States of America. 2007;104(33):13225-13231.
58. Bramley TJ, Nightengale BS, Frech-Tamas F, Gerbino PP. Relationship of Blood
Pressure Control to Adherence With Antihypertensive Monotherapy in 13 Managed Care
Organizations. Journal of Managed Care Pharmacy. 2006;12(3):239-245.
59. Ho P, Rumsfeld JS, Masoudi FA, et al. Effect of medication nonadherence on
hospitalization and mortality among patients with diabetes mellitus. Archives of Internal
Medicine. 2006;166(17):1836-1841.
60. Rasmussen JN, Chong A, Alter DA. Relationship between adherence to evidence-based
pharmacotherapy and long-term mortality after acute myocardial infarction. JAMA.
2007;297(2):177-186.
61. Spertus JA, Kettelkamp R, Vance C, et al. Prevalence, predictors, and outcomes of
premature discontinuation of thienopyridine therapy after drug-eluting stent placement.
Circulation. 2006;113(24):2803-2809.
62. Ho PM, Magid DJ, Shetterly SM, et al. Medication nonadherence is associated with a
broad range of adverse outcomes in patients with coronary artery disease. American
Heart Journal.155(4):772-779.
63. Ho PM, Magid DJ, Shetterly SM, et al. Medication nonadherence is associated with a
broad range of adverse outcomes in patients with coronary artery disease. American
Heart Journal. 2008;155(4):772-779.
64. Pampel FC, Krueger PM, Denney JT. Socioeconomic Disparities in Health Behaviors.
Annual review of sociology. 2010;36:349-370.
65. Braveman PA, Cubbin C, Egerter S, Williams DR, Pamuk E. Socioeconomic Disparities
in Health in the United States: What the Patterns Tell Us. American Journal of Public
Health. 2010;100(Suppl 1):S186-S196.
66. Brown MT, Bussell JK. Medication Adherence: WHO Cares? Mayo Clinic Proceedings.
2011;86(4):304-314.
67. Egan M, Philipson TJ. Health Care Adherence and Personalized Medicine. National
Bureau of Economic Research Working Paper Series. 2014;No. 20330.
68. Choudhry NK, Krumme AA, Ercole PM, et al. Effect of reminder devices on medication
adherence: The remind randomized clinical trial. JAMA Internal Medicine.
2017;177(5):624-631.
69. Thakkar J, Kurup R, Laba T, et al. Mobile telephone text messaging for medication
adherence in chronic disease: A meta-analysis. JAMA Internal Medicine.
2016;176(3):340-349.
70
70. Khunti K, Seidu S, Kunutsor S, Davies M. Association Between Adherence to
Pharmacotherapy and Outcomes in Type 2 Diabetes: A Meta-analysis. Diabetes Care.
2017;40(11):1588.
71. García-Pérez L-E, Álvarez M, Dilla T, Gil-Guillén V, Orozco-Beltrán D. Adherence to
Therapies in Patients with Type 2 Diabetes. Diabetes Therapy. 2013;4(2):175-194.
72. Osterberg L, Blaschke T. Adherence to Medication. New England Journal of Medicine.
2005;353(5):487-497.
73. McDonald HP, Garg AX, Haynes R. Interventions to enhance patient adherence to
medication prescriptions: Scientific review. JAMA. 2002;288(22):2868-2879.
74. Choudhry NK, Fischer MA, Avorn J, et al. The implications of therapeutic complexity on
adherence to cardiovascular medications. Archives of Internal Medicine.
2011;171(9):814-822.
75. Coleman CI, Limone B, Sobieraj DM, et al. Dosing Frequency and Medication
Adherence in Chronic Disease. Journal of Managed Care Pharmacy. 2012;18(7):527-
539.
76. Weeda ER, Coleman CI, McHorney CA, Crivera C, Schein JR, Sobieraj DM. Impact of
once- or twice-daily dosing frequency on adherence to chronic cardiovascular disease
medications: A meta-regression analysis. International Journal of Cardiology.
2016;216(Supplement C):104-109.
77. Wang L, Sun X, Du L, et al. Effects and patient compliance of sustained-release versus
immediate-release glipizides in patients with type 2 diabetes mellitus: a systematic
review and meta-analysis. Journal of Evidence-Based Medicine. 2011;4(4):232-241.
78. Inzucchi SE. Is It Time to Change the Type 2 Diabetes Treatment Paradigm? No!
Metformin Should Remain the Foundation Therapy for Type 2 Diabetes. Diabetes Care.
2017;40(8):1128-1132.
79. Van Nuys K, Joyce G, Ribero R, Goldman DP. Frequency and magnitude of co-payments
exceeding prescription drug costs. JAMA. 2018;319(10):1045-1047.
80. Menzin J, Korn JR, Cohen J, et al. Relationship between glycemic control and diabetes-
related hospital costs in patients with type 1 or type 2 diabetes mellitus. Journal of
Managed Care Pharmacy. 2010;16(4):264-275.
81. Terza JV. Estimating count data models with endogenous switching: Sample selection
and endogenous treatment effects. Journal of econometrics. 1998;84(1):129-154.
82. A. DL, D. MA, R. PE. Adherence in patients transferred from immediate release
metformin to a sustained release formulation: a population‐based study. Diabetes,
Obesity and Metabolism. 2009;11(4):338-342.
83. Ingersoll KS, Cohen J. The impact of medication regimen factors on adherence to chronic
treatment: a review of literature. Journal of behavioral medicine. 2008;31(3):213-224.
84. Schwartz S, Fonseca V, Berner B, Cramer M, Chiang Y-K, Lewin A. Efficacy,
Tolerability, and Safety of a Novel Once-Daily Extended-Release Metformin in Patients
With Type 2 Diabetes. Diabetes Care. 2006;29(4):759-764.
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays in health economics and provider behavior
PDF
Understanding primary nonadherence to medications and its associated healthcare outcomes: a retrospective analysis of electronic medical records in an integrated healthcare setting
PDF
Three essays on estimating the effects of government programs and policies on health care among disadvantaged population
PDF
The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins
PDF
Essays on health economics
PDF
Investigating racial and ethnic disparities in patient experiences with care and health services use following colorectal cancer diagnosis among older adults with comorbid chronic conditions
PDF
Three essays on health & aging
PDF
Essays on health insurance programs and policies
PDF
Essays in health economics
PDF
Essays on development and health economics
PDF
The impact of treatment decisions and adherence on outcomes in small hereditary disease populations
PDF
Economic, clinical, and behavioral outcomes from medical and pharmaceutical treatments
PDF
Long-term impacts of childhood adversity on health and human capital
PDF
Essays in health economics: evidence from Medicare
PDF
Value in health in the era of vertical integration
PDF
Racial/ethnic variation in care preferences and care outcomes among United States hospice enrollees
PDF
The impact of Patient-Centered Medical Home on a managed Medicaid plan
PDF
Essays in opioid use and abuse
PDF
Essays on health and aging with focus on the spillover of human capital
PDF
Essays in pharmaceutical and health economics
Asset Metadata
Creator
Xie, Zhiwen ""Richard""
(author)
Core Title
Three essays in health economics
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
07/31/2018
Defense Date
05/09/2018
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
extended-release medications,Health policy,heart failure,medication adherence,OAI-PMH Harvest,racial disparities,social cost,socioeconomic disparities
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Goldman, Dana (
committee chair
), Joyce, Geoffrey (
committee member
), Nugent, Jeffrey (
committee member
), Romley, John (
committee member
)
Creator Email
xizhwe186@gmail.com,zhiwenxi@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c89-44213
Unique identifier
UC11670849
Identifier
etd-XieZhiwenR-6583.pdf (filename),usctheses-c89-44213 (legacy record id)
Legacy Identifier
etd-XieZhiwenR-6583.pdf
Dmrecord
44213
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Xie, Zhiwen ""Richard""
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
extended-release medications
heart failure
medication adherence
racial disparities
social cost
socioeconomic disparities