IMPACT OF MEDICARE PART D COVERAGE GAP ON MEDICARE BENEFICIARIES WITH COPD: ADHERENCE, HEALTHCARE RESOURCE

USE, AND COST

by

YANNI FAN YU

LARRY R. HERALD, COMMITTEE CHAIR

MEREDITH L. KILGORE HAIYAN QU

MIDGE N. RAY

A DISSERTATION

Submitted to the graduate faculty of The University of Alabama at Birmingham,

in partial fulfillment of the requirements for the degree of Doctor of Science

BIRMINGHAM, ALABAMA

2015

Copyright by Yanni Fan Yu

2015

iii

IMPACT OF MEDICARE PART D COVERAGE GAP ON MEDICARE

BENEFICIARIES WITH COPD: ADHERENCE, HEALTHCARE RESOURCE USE,

AND COST

YANNI FAN YU

ADMINISTRATION – HEALTH SERVICES

ABSTRACT

Medicare Part D provides prescription drug coverage for beneficiaries to support

their pharmacological treatment; however, the complex deductible structure within

benefit plans creates a major coverage gap and unexpected consequences. Some evidence

has demonstrated reduced adherence resulting from the coverage gap; however, little

research has evaluated the effect on healthcare resource use (HRU) and cost, and no

studies have been conducted for beneficiaries with chronic obstructive pulmonary disease

(COPD). This study examined the impact of the coverage gap on medication adherence

as well as healthcare resource use and medical cost among beneficiaries with COPD.

Claims data based on a 5% random sample of Medicare beneficiaries were used in

this retrospective cohort study. For each year from 2007 to 2010, beneficiaries diagnosed

with COPD were assigned to either an exposure cohort if they were at risk of the

coverage gap or a control cohort if they were not. Exposure and control cohorts were

matched using a high-dimensional propensity scores. Adherence was defined as no less

iv

than 80% of the proportion of days covered (PDC) by long-acting bronchodilators

(LABDs). HRU included medical encounters occurring in all care settings. Cost included

non-drug cost paid by Medicare. All outcomes were estimated at the calendar year level.

Multivariable logistic and generalized linear model (GLM) regressions controlling for

unbalanced covariates post-matching were applied with generalized estimating equation

technique to correct for potential correlation between repeated observations of the same

beneficiary.

The final exposure and control cohorts each included 4,147 patient-year

observations. The results showed that the coverage gap was associated with lower

adherence. Both positive and negative associations with HRU were found, but no

significant difference in cost was observed between two cohorts.

This is the first study assessing the effect of the coverage gap on patients with

COPD. The findings provide support for phasing out the coverage gap by 2020. More

generally, the findings highlight opportunities to design benefit offerings that can

improve beneficiaries’ access to healthcare in ways that can impact healthcare quality and

utilization.

Keywords: Medicare Part D, coverage gap, medication adherence, healthcare resource use, cost, Chronic Obstructive Pulmonary Disease

v

ACKNOWLEDGMENTS

I would like to thank my committee chairman, Dr. Larry Hearld, and committee

members Dr. Meredith Kilgore, Dr. Haiyan Qu, and Dr. Midge Ray for their guidance

and comments provided for my dissertation. Larry has been greatly patient and helpful to

guide me through the dissertation journey, edit my dissertation drafts meticulously, and

offer his advice tirelessly. I am deeply grateful for his valuable mentorship. I am very

thankful to Dr. Kilgore for providing me with access to Medicare claims data and helping

facilitate the technical set-up on my computer. His in-depth knowledge of Medicare data

and Part D programs has been tremendously helpful in my dissertation research. I greatly

appreciate that Haiyan and Midge gave their time, thoughts, and assistance throughout

the process; their comments have greatly improved my dissertation. I truly hold precious

the time spent with my whole committee, and the inspiration and learning obtained from

all of you.

Dr. Kilgore also connected me to his colleagues in the School of Public Health --

Dr. Huifeng Yun and Dr. Haichang Xin to obtain information. I want to thank Hui and

Haichang for their assistance. I would also like to thank Robert, Matthew, and Marcie

Battles for their IT expertise and their help with all of the technical problems related to

remote access, software, and hardware of my computer.

I would like to acknowledge the leadership provided by Dr. Robert S. Hernandez

in leading the program, the excellent operation led by Ms. Leandra Y. Celaya, and the

vi

enormous support provided by Dr. M. Elizabeth Hendrix. Also, I want to thank my

classmates for their company, interaction, and friendship for the past years. Because of all

of you, this adventure has been more enjoyable and memorable.

My deepest gratitude to my family, without whom, none of this would have been

possible. I want to thank Andrew, my husband, and my adorable daughters Esther and

Enya for their love, support, understanding, and patience, which makes my endeavor

more meaningful. You are treasures in my life. I want to thank my parents for their

timeless love and selfless support. You keep encouraging me to go further and make me

always believe the power and the value of education. Also my thanks to my parents-in-

law for their immeasurable assistance that freed me up from housework and enabled me

to devote more time to my research. Thanks to all of you, I am really fortunate to be able

to pursue something I want to pursue, to taste the joy of achieving challenging goals, and

to carry over your endless love throughout my whole life.

vii

TABLE OF CONTENTS

ABSTRACT .............................................................................................................. iii

ACKNOWLEDGMENTS ..........................................................................................v

LIST OF TABLES .................................................................................................... ix

LIST OF FIGURES .................................................................................................. xi

LIST OF ABBREVIATIONS .................................................................................. xii

CHAPTER ONE: INTRODUCTION .........................................................................1

Background ......................................................................................................................1

Problem Statement ...........................................................................................................5

Research Questions ........................................................................................................11

CHAPTER TWO: LITERATURE REVIEW ...........................................................14

Impact of Medicare Part D .............................................................................................14

Impact of Medicare Part D Coverage Gap .....................................................................21

Theoretical Framework ..................................................................................................30

CHAPTER THREE: METHODS .............................................................................38

Study Design ..................................................................................................................38

Data Source ....................................................................................................................41

Sample Selection ............................................................................................................43

viii

Outcome Measures .........................................................................................................48

Variables.........................................................................................................................51

Statistical Analysis .........................................................................................................62

CHAPTER FOUR: RESULTS .................................................................................67

Sample Size ....................................................................................................................68

Demographic and Baseline Characteristics ....................................................................72

Descriptive Statistics of Outcome Variables ..................................................................81

Regression Analysis and Hypotheses Testing ................................................................88

CHAPTER FIVE: DISCUSSION ...........................................................................103

Review of Findings and Comparison with Existing Evidence .....................................103

Strengths and Limitations.............................................................................................109

Implications ..................................................................................................................114

Conclusion ....................................................................................................................121

REFERENCES .......................................................................................................122

APPENDIX .............................................................................................................133

INSTITUTIONAL REVIEW BOARD Documentation .............................................133

ix

LIST OF TABLES

Table 1. Maintenance medications used for COPD ..................................................45

Table 2: ICD-9-CM diagnosis codes for CCI conditions .........................................55

Table 3. ICD-9-CM codes for select relevant comorbidities ....................................56

Table 4. Procedure codes for supplemental oxygen therapy ....................................57

Table 5. Oral corticosteroid ......................................................................................57

Table 6. A list of variables for primary and subgroup analyses ...............................59

Table 7. Sample size of study cohorts and subgroups. .............................................72

Table 8. Patient demographic and baseline characteristics of study cohorts before

and after matching.....................................................................................................75

Table 9. Demographic and baseline characteristics of subgroups of the exposure

cohort before matching. ............................................................................................79

Table 10. Adherence to LABDs in the matched control and exposure cohorts ........81

Table 11. Annual HRU in the matched control and exposure cohorts. ....................82

Table 12. Annual all-cause medical cost for the matched control and exposure

cohorts .......................................................................................................................83

Table 13. Adherence to LABDs in the mid-gap and late-gap subgroups in the

matched exposure cohort ..........................................................................................84

x

Table 13a. Quarterly adherence for the late-gap subgroup. ......................................85

Table 14. Monthly HRU of the mid-gap and the late-gap subgroups in the matched

exposure cohort .........................................................................................................86

Table 15. Monthly all-cause medical cost for the mid-gap and the late-gap

subgroups in the matched exposure cohort ...............................................................87

Table 16. Conditional logistic regression on adherence to LABDs ..........................90

Table 16a. Unadjusted and adjusted adherence ........................................................90

Table 17. GLM regression on annual number of all-cause outpatient visits ...........93

Table 17a. Unadjusted and adjusted number of all-cause outpatient visits ..............94

Table 18. GLM regression on annual number of all-cause ER visits ......................95

Table 18a. Unadjusted and adjusted number of all-cause ER visits .........................96

Table 19. GLM regression on annual number of all-cause inpatient visits. .............97

Table 19a. Unadjusted and adjusted number of all-cause inpatient visits ................98

Table 20. GLM regression on annual all-cause medical cost .................................100

Table 20a. Unadjusted and adjusted annual all-cause medical cost .......................101

Table 21. Summary for regression analysis and hypotheses testing .......................102

xi

LIST OF FIGURES

Figure 1. Average annual prevalence of COPD among adults by age group and

gender in the United States (2007-2009). ...................................................................3

Figure 2. Rational choice theory model. ...................................................................33

Figure 3. Subgroups of the exposure cohort. ............................................................47

Figure 4. Patient selection flow chart.. .....................................................................69

xii

LIST OF ABBREVIATIONS

ACE angiotensin-converting-enzyme

AIDS Acquired immunodeficiency syndrome

AMI acute myocardial infarction

ARB angiotensin II receptor blocker

ATS American Thoracic Society

CBO Congressional Budget Office

CCI Charlson Comorbidity Index

CHF congestive heart failure

CI confidence interval

CMS Centers for Medicare & Medicaid Services

COPD Chronic obstructive pulmonary disease

CRN cost-related non-adherence

DA descriptive analysis

DDD defined daily dose

DME durable medical equipment

ER emergency room

ERS European Respiratory Society

ESRD end-stage-renal-diseases

FD&C Federal Food, Drug and Cosmetic

FDA Food and Drug Administration

xiii

FFS Fee for Service

GEE generalized estimating equation

GERD gastroesophageal reflux disease

GLM generalized linear model

GOLD Global initiative for chronic Obstructive Lung Disease

HbA1c hemoglobin A1c, i.e., glycated hemoglobin

HCPCS Healthcare Common Procedure Coding System

HDPS high-dimensional propensity score

HD-PSM high-dimensional propensity score matching

HF heart failure

HMO health maintenance organization

HRU healthcare resource use

ICD-9-CM International Classification of Diseases, Ninth Edition, Clinical

Modification

IRR incidence rate ratio

LABD long-acting bronchodilator

LIS low income subsidy

MA multivariable analysis

MA-PD Medicare Advantage plan with prescription drug coverage

MMA Medicare Modernization Act

MPR medication possession ratio

MRA Medication Refill Adherence

NDC national drug code

xiv

OOP out-of-pocket

OR odds ratio

PDC proportion of days covered

PDE Part D Event

PDP Part D plans

PPIs proton-pump inhibitors

PSM propensity score matching

RR relative ratio

SABD short-acting bronchodilator

SD standard deviation

SE standard error

TrOOP true out-of-pocket

US United States

WHO World Health Organization

1

CHAPTER ONE

INTRODUCTION

Background

Chronic Obstructive Pulmonary Disease

Chronic obstructive pulmonary disease (COPD) is a progressive lung

disease characterized by airway narrowing or airflow obstruction, resulting in

breathing difficulties, reduced exercise capacity and physical limitation (GOLD,

2014). COPD can deteriorate over time, leading to respiratory worsening and

serious disability that require hospitalization. Many people suffer for years and

die prematurely from COPD or its complications (GOLD, 2014). COPD has

become the fourth leading cause of death in the world (WHO, 2008). Chronic

lower respiratory disease, which primarily includes COPD has become the third

leading cause of death in the United States (Hoyert & Xu, 2012) .

There is no cure for COPD and no existing medication has been shown to

modify the long-term decline in lung function; however, treatments and lifestyle

changes may help to slow the progression of COPD and lessen symptoms

(GOLD, 2014). For example, smoking cessation has been found to be effective

in improving symptoms for patients who smoke (Anthonisen et al., 1994; Baillie,

Mattick, Hall, & Webster, 1994). Appropriate pharmacologic therapy can help to

relieve symptoms and reduce the frequency and severity of exacerbations;

pulmonary rehabilitation can be used for patients with shortness of breath while

walking, and oxygen therapy is often prescribed to patients with severe COPD

(GOLD, 2014; Stoller, Panos, Krachman, Doherty, & Make, 2010).

2

Although as many as 24 million Americans may have COPD, only 12-15

million have been diagnosed (American-Lung-Association, 2014; NIH, 2008),

accounting for about 5% of adults in the United States (Akinbami & Liu, 2011).

COPD occurs more often in older age groups and in females. According to the

CDC report, almost half of all COPD patients are aged 65 years or older (Centers

for Disease Control and Prevention, 2012). As shown in Figure 1, COPD

prevalence was highest among men aged 75–84 years (11.2%), followed by

women aged 65–74 years (10.4%), and women aged 75–84 years (9.7%). Access

to appropriate therapies to manage COPD, especially pharmacologic treatment,

used to be challenging due to the lack of coverage for prescription medications

among the elderly population. With the increasing size of the elderly population

and the rising prevalence of chronic diseases, it has become important for

healthcare policy development to improving access to health services and

assisting treatment needs for this population.

3

Figure 1. Average annual prevalence of COPD among adults by age group and gender in the United States (2007-2009).

Note: Graphed based on data from CDC/NCHS, Health Data Interactive

and National Health Interview Survey

Medicare and Medicare Part D

Medicare is a national social insurance program in the United States (US).

Established in 1965 and administered by the federal government, Medicare

provides access to health insurance for Americans aged 65 years or older who

have worked and paid taxes for a certain number of years and for younger people

with disabilities or certain illnesses such as end stage renal disease or

amyotrophic lateral sclerosis (Centers for Medicare & Medicaid, 2013). The

Medicare program is composed of four parts as described in detail below. Part A

helps cover inpatient care in hospitals, skilled nursing facilities (not custodial or

long-term care), hospice care, and some home health. Part B covers physician

2.0 2.0

3.9

6.4

8.3

11.2

7.2

3.0

4.1

7.5

8.7

10.49.7

7.8

0.0

2.0

4.0

6.0

8.0

10.0

12.0

18-24 25-44 45-54 55-64 65-74 75-84 >=85

Age groups in years

Male

Female

Percent

4

services and outpatient care as well as other medical services not covered by Part

A, such as occupational therapy.

For Part C, with the passage of the 1997 Balanced Budget Act, Medicare

beneficiaries have been given the option to enroll in capitated health insurance

(Part C plans) instead of through the original fee for service (FFS) Medicare

payment system (Centers for Medicare & Medicaid). Initially known as

“Medicare+Choice”, most of the Part C plans have been re-branded as

“Medicare Advantage” plans since the Medicare Modernization Act (MMA) of

2003.

Medicare Part D, also known as the Medicare prescription drug benefit,

went into effect on January 1, 2006 as a result of the passage of MMA. It is a

federal program to subsidize the cost of prescription drugs and prescription drug

insurance premiums for Medicare beneficiaries. Anyone with Medicare Part A or

B is eligible for Part D. If they do not have drug coverage under their retirement

insurance plan, beneficiaries must enroll in a stand-alone Prescription Drug Plan

(PDP) or Medicare Advantage plan with prescription drug coverage (MA-PD)

when they are first eligible, or pay a penalty if they choose to join later. These

plans are approved and regulated by the Medicare program but designed and

administered by private health insurance companies and pharmacy benefit

managers. Plans choose which drugs or drug classes to cover and decide how

much of the drug cost to reimburse. (Centers for Medicare & Medicaid, 2013).

The Medicare program provides different levels of coverage for elderly

patients with COPD depending on the type of care and treatment they need. For

5

example, Medicare Part A provides some coverage if patients were hospitalized

due to exacerbation; Part B helps pay for smoking cessation counseling or

pulmonary rehabilitation if COPD is moderate to severe, and oxygen therapy

(equipment, supplies, and oxygen) (Medicare-Made-Clear, 2014). Since 2006,

pharmacologic treatments such as bronchodilators and inhaled steroids are

covered by Medicare Part D, with patients responsible for deductibles and

copayments. However, as noted earlier, the amount of patients’ out-of-pocket

(OOP) expenses and the specific medications included in the coverage vary

across different Part D plans.

Problem Statement

Burden of COPD and Challenges in COPD Management

COPD has a significant health impact and may generate a cycle of

physical, social, and psychosocial consequences (GOLD, 2014). In the early

stages of COPD, dyspnea may occur during daily activities such as walking and

climbing stairs, forcing individuals to limit their daily tasks at home and avoid

social situations involving these activities. Increasing inactivity reduces patients’

fitness level and leads to worsening shortness of breath, further immobility and

lack of fitness. Thus, patients with COPD are significantly less active than

healthy individuals (Pitta et al., 2005), and they are 11 times more likely to

report fair or poor health, 10 times more likely to report depression, and 5.5

times more likely to report poor sleep (Eisner, Yelin, Trupin, & Blanc, 2002).

Comorbidities are common in patients with COPD. Significantly higher rates of

both respiratory and non-respiratory comorbidities were reported in individuals

with COPD when compared to those without COPD (Darkow, Kadlubek, Shah,

6

Phillips, & Marton, 2007). The common comorbid conditions include: asthma

and other lung disease, hypertension, cardiovascular conditions such as

myocardial infarction, angina and heart failure, diabetes, depression,

osteoporosis, respiratory infection, bone fractures, sleep disorders, excessive

weight loss and muscle wasting, and glaucoma (ATS/ERS, 2004; GOLD, 2014).

The economic burden associated with COPD is also substantial. The total

cost to the United States for COPD was projected to be close to $50 billion in

2010, including $29.5 billion in direct healthcare expenditures: $13.2 billion in

hospital care, $5.5 billion in physician services, and $5.8 billion in prescription

drugs (American-Lung-Association, 2014; NHLBI, 2009). Indirect costs account

for over $20 billion of the expenditures, including $8 billion in indirect

morbidity cost and $12.4 billion in indirect mortality cost (American-Lung-

Association, 2014). COPD expenditures exceed the costs associated with heart

failure and is over two-fold higher than the cost for asthma (NHLBI, 2007). In

terms of healthcare resource use, COPD is associated with 636,000 hospital

admissions and more than 15 million physician office visits each year in the

United States (NHLBI, 2007).

COPD is more prevalent in older adults, with approximately two-thirds

of COPD patients aged 50 years or older (American-Lung-Association, 2013)

and 50% aged 65 years or older (Centers for Disease Control and Prevention,

2012); therefore, management of COPD has become one of the priorities for

CMS. However, managing COPD in the Medicare population is exceptionally

challenging due to the following reasons:

7

• Complex clinical management of COPD. The diagnosis of

COPD is not straightforward due to the early physical

symptoms of COPD common to many conditions. Ongoing

monitoring and responsive disease management is required as

the disease progresses and needs to be patient-specific, guided

by severity of disease, risk of exacerbation, drug availability,

and patient’s response (GOLD, 2014).

• The presence of other comorbid conditions in elderly patients

with COPD. Co-existing conditions such as asthma,

pneumonia, or congestive heart failure add more challenges to

COPD management. Managing the care of patients with

comorbid conditions incurred about 1.4 to 2.1 times higher

expenditures when compared to managing the care of

similarly aged Medicare beneficiaries with COPD who did

not have comorbidities (Blanchette, Gutierrez, Ory, Chang, &

Akazawa, 2008; Grasso, Weller, Shaffer, Diette, & Anderson,

1998).

Furthermore, managing elderly COPD patients is expensive and requires

a large amount of healthcare resources. The estimated total 12-month

expenditures in Medicare FFS beneficiaries with COPD were $8,482 on average

in 1992, about 2.4 times higher than the mean expenditures in those without

COPD ($3,511) (Grasso et al., 1998). The average 12-month total medical

8

spending in community-dwelling Medicare beneficiaries with COPD was as

high as $21,488, 10 years later (B. C. Stuart et al., 2007).

Medication Adherence in COPD

COPD is a progressive illness with worsening symptoms, therefore

patients need to be prescribed appropriate therapies and have ongoing

assessment to manage their disease and improve their health status (Make, Dutro,

Paulose-Ram, Marton, & Mapel, 2012). Pharmacotherapy is a cornerstone of

COPD management, and maintenance medications are effective in controlling

symptoms, maintaining lung function, and preventing COPD exacerbations

(ATS/ERS, 2004; GOLD, 2014).

Despite the availability of evidence-based guidelines and advances in the

management of COPD, research suggests that about 66% of commercially

insured patients with COPD and 71% of Medicare beneficiaries with COPD

were not prescribed any maintenance therapy (Make et al., 2012). Although

COPD medication adherence approached 80% in clinical trials (Vestbo et al.,

2009), observed adherence levels in real-world settings ranged from 10% to 60%,

and only half of all patients continuously used their prescribed medications for

one year (Breekveldt-Postma, Koerselman, Erkens, Lammers, & Herings, 2007;

Charles et al., 2010; Rand, 2005; Restrepo et al., 2008; Simoni-Wastila et al.,

2012; Toy et al., 2011).

Research has revealed a series of factors associated with COPD

medication non-adherence, including age and disease duration (Osterberg &

Blaschke, 2005), patients’ belief in their health and medication effectiveness

(Khdour, Hawwa, Kidney, Smyth, & McElnay, 2012); severity of COPD

9

(Cecere et al., 2012; Khdour et al., 2012); smoking status (Cecere et al., 2012;

Khdour et al., 2012); presence of comorbidities, especially depression (Khdour

et al., 2012; Yohannes, Baldwin, & Connolly, 2006); medication regimen

complexity (Cecere et al., 2012); and perception of provider skill (Cecere et al.,

2012). In addition to clinical and psychosocial variables (Charles et al., 2010),

financial factors are also important to treatment decision making. Cost sharing,

copayment, availability of generic versions of medications, and level of

insurance benefit coverage are found to be independent risk factors for

medication non-adherence among patients with chronic conditions (Dusetzina,

Winn, Abel, Huskamp, & Keating, 2014; Gibson et al., 2006; Gu, Zeng, Patel, &

Tripoli, 2010; Ho, Bryson, & Rumsfeld, 2009; E. Kim et al., 2010; Shrank et al.,

2006; WHO, 2003).

Coverage Gap in Medicare Part D

Although Medicare Part D provides prescription drug coverage to

support elderly patients with COPD for their pharmacological treatment, the

deductible structure for Part D is a complex design with multiple payment tiers

for Medicare and beneficiaries that creates a coverage gap for beneficiaries. The

deductible structure in a standard Part D plan can be summarized as:

1) First, there is a deductible (e.g., $295 in 2009) that the

beneficiary pays out-of-pocket before an insurer makes any

payments;

2) Second, once the initial deductible is met, subsequent

payments are split between the beneficiary (25%) and the

10

Medicare Part D plan (75%) up to a threshold (e.g., $896.25

in 2009);

3) Third, once this threshold is reached, a second deductible (e.g.,

$896.25 to $4,350.25 in 2009) is required for and paid solely

by beneficiaries up to a second threshold (e.g., $4,350.25 in

2009);

4) Finally, once the second threshold is met (e.g., $4,350.25 in

2009), additional payments are once again split between the

beneficiary (5%) and Medicare (95%), which is known as

catastrophic coverage.

The second deductible incurred in the third payment tier is a coverage

gap known as a “donut hole” because beneficiaries have to pay full cost of their

prescription drugs before they enter the final payment tier (catastrophic

coverage). The deductible amount thresholds vary by year. The coverage gap

was included in the Part D benefit design because the cost of providing

continuous coverage for prescription drugs with no gap would have exceeded the

budgetary limit imposed by the MMA (Jack Hoadley et al., 2007).

About 1.5 million beneficiaries reached the coverage gap in 2006 when a

large number of Part D enrollees had less than full year benefit (IMS, 2007), and

over 3 million of the 24 million Part D enrollees were estimated to reach the gap

in 2007 when most enrollees had full-year coverage (Hoadley et al., 2007).

Many beneficiaries with moderate to high drug expenses, especially those with

chronic illness such as COPD, face breaks in coverage that result in high out-of-

11

pocket cost for prescription drugs. Under these circumstances, beneficiaries may

adjust their demand for drug therapies. For example, beneficiaries may

discontinue the effective brand-name drugs they were taking and switch to a

cheaper but less effective treatment after they hit the coverage gap.

Research Questions

The Part D coverage gap raises concerns that enrollees who reach the

threshold might forgo needed medication because they face the full cost of their

prescriptions they during the gap. Multiple studies have assessed the effect of the

coverage gap on medication adherence in elderly patients (Polinski et al., 2011)

and in those with different chronic diseases including cardiovascular disease

(Jung, Feldman, & McBean, 2014; Polinski et al., 2012), diabetes (Gu et al.,

2010; Zeng, Patel, & Brunetti, 2013), mental illness (Fung et al., 2013), heart

failure (HF) (Nair et al., 2011), hypertension (Nair et al., 2011), and acute

myocardial infarction (B. Stuart et al., 2013). Findings have shown that the use

of brand-name medications was significantly decreased for most conditions, but

adherence to antidepressants or HF drugs did not change or was only slightly

reduced. However, no published studies have been found in the existing

literature, which evaluate the effect of the Part D coverage gap on medication

utilization among Medicare beneficiaries with COPD. In addition, the research

related to the effect of Part D coverage gap on resource use and cost is very

limited.

This study assessed the impact of Part D coverage gap among

beneficiaries diagnosed with COPD on three types of outcomes: (1) adherence to

COPD maintenance medications, (2) healthcare resource use, and (3) medical

12

cost. The following research questions guided the study: compared to the

Medicare beneficiaries with COPD who were not exposed to the Part D coverage

gap,

1) Is medication adherence lower for Medicare beneficiaries

with COPD who reached the Part D coverage gap?

2) Is total healthcare resource use (HRU) higher for Medicare

beneficiaries with COPD who reached the Part D coverage

gap?

3) Is the total medical cost (non-drug) higher for Medicare

beneficiaries with COPD who reached the Part D coverage

gap?

The findings from this study contributed to the literature in several ways

and have a number of policy implications. By examining the impact of Part D

coverage gap on medication adherence, resource use, and medical cost among

patients with COPD, research on the effect of the coverage gap is expanded to

another prevalent and serious chronic condition and to other important outcomes

beyond adherence, which have not been widely studied in the current literature.

Information derived from this study can help researchers and policy makers

better understand issues related to Medicare Part D and its coverage gap so as to

(1) further address potential issues related to Medicare Part D benefit design; (2)

provide relevant evidence for policy makers to form legislation or make reforms

to support the potential phase-out of Part D coverage gap by 2020; and (3)

inform clinicians, insurers, and public health administrators regarding the

13

association between cost-sharing and patient behavior, especially medication

adherence. A good balance between cost and health outcomes should be taken

into account in future benefit design to maximize the treatment benefit among

patients with chronic diseases such as CODP.

14

CHAPTER TWO

LITERATURE REVIEW

This chapter reviews the extant literature on the impact of Medicare Part

D and the coverage gap on adherence, healthcare resource use, and cost,

followed by a description of the theoretical framework related to the research

questions, concluding with the development of research hypotheses based on the

relevant empirical and theoretical literature. In the literature review section,

research on the impact of Medicare Part D is reviewed first, followed by a

review of the research on the impact of Medicare Part D coverage gap. Both

sections are organized around the three types of outcomes: adherence, healthcare

resource use, and cost. For adherence, the evidence is summarized for the

general Medicare population and subgroups with different chronic conditions

separately, when such research was available. Similarly, research related to

healthcare resource use and cost is summarized separately for drug-related and

non-drug related (medical or total) services when such studies were available.

Impact of Medicare Part D

Since 2006, when Part D was added to Medicare program to provide

outpatient prescription drug coverage, Medicare Part D has been claimed as a

success story for both beneficiaries and tax payers. Below are some highlights

that have been discussed:

• Part D spending is about 45% lower than the original 2004-

2013 projections. The 10-year projected cost has been reduced

15

by $100 billion in each of the last three years (Congressional

Budget Office, 2011, 2012).

• Average beneficiary Part D premiums are only 50% of the

projected amount. The average monthly premium was $30 in

2013 versus the original forecast of $61 (Centers for Medicare

& Medicaid, 2011, 2012).

• Other medical spending declined significantly after Part D.

Research showed that the implementation of Part D was

associated with a $1,200 decrease per beneficiary in annual

non-drug medical cost in both 2006 and 2007 (McWilliams,

Zaslavsky, & Huskamp, 2011).

• With Part D expanded coverage for seniors, beneficiaries are

highly satisfied with the program. In 2011, 90% of

beneficiaries had comprehensive drug coverage, and most of

them said their coverage worked well. Medicare and Medicaid

dually-eligible beneficiaries exhibited the highest satisfaction

(KRCresearch, 2013).

A large amount of research has been conducted to evaluate the impact of

Medicare Part D after its implementation. This research has mostly been

concentrated in the following areas: drug utilization; cost, including out-of-

pocket (OOP) cost or non-drug/other medical cost; healthcare resource use such

as hospitalizations, disparity in healthcare access, healthcare practice in nursing

homes; and medication adherence by patients with various chronic diseases

16

including hypertension, depression, diabetes and osteoporosis. Considering the

large number of published studies and the relevance to the research questions

presented in this proposal, the literature review on this topic focused on

quantitative analysis of the impact of Medicare Part D on medication adherence,

healthcare resource use, and cost in Medicare population and is organized by

these three outcomes of interest.

Impact on Medication Adherence

Medication non-adherence has been an issue reported by older

Americans owing to unaffordable OOP drug cost (Piette, Heisler, & Wagner,

2004; Safran et al., 2005; Safran et al., 2002). The implementation of Medicare

Part D was intended to address medication non-adherence by increasing access

to medication. Several studies have found that self-reported cost-related non-

adherence (CRN) among Medicare beneficiaries significantly decreased after

Part D implementation and the reduction was even larger in 2007 than in 2006,

with odds ratios ranging from 0.58 to 0.85 (Kennedy, Maciejewski, Liu, &

Blodgett, 2011; Madden, Graves, Ross-Degnan, Briesacher, & Soumerai, 2009;

Madden et al., 2008). One exception to this finding was that the significant

change in CRN was not observed in beneficiaries with fair-to-poor health

(Kennedy et al., 2011; Madden et al., 2008). The effect was consistent among

subgroups defined by disability status and the number of morbidities.

Beneficiaries with Different Chronic Diseases

Diabetes, hyperlipidemia, and hypertension are common chronic

conditions occurring in elderly people. Zhang et al. (2010) evaluated adherence

over six months among beneficiaries with diabetes, hyperlipidemia and/or

17

hypertension by comparing their medication possession ratio (MPR) based on

the level of their prior drug coverage. The authors found that Part D improved

MPRs by about 13 percentage points for hyperlipidemia and hypertension and

18 percentage points for diabetes among those without prior prescription drug

coverage. Improvement was lower, however, in groups with limited prior

coverage. It was also noted that even with the Part D benefit, about 50-80% of

these beneficiaries still did not attain good adherence (i.e., MPR≥80%) (Zhang,

Lave, Donohue, et al., 2010).

Diabetes. Employer-based retiree drug benefits have long been viewed as

the gold standard of drug coverage for Medicare beneficiaries. One study

examined anti-diabetic agent adherence for Medicare members who enrolled in

Medicare PDP compared to adherence for those who enrolled in retiree plans to

assess the effect of Part D on adherence of anti-diabetics (B. Stuart et al., 2011).

Similar adherence was observed across the two groups of patients, suggesting

that Part D benefit helps Medicare beneficiaries with diabetes to be comparably

adherent with their therapies to those with retiree drug benefits.

Depression. The beneficial effect of Part D coverage on medication

adherence was not consistent for beneficiaries with depression. The self-reported

CRN was less affected or not found to decline among respondents with

depressive symptoms (Zivin, Madden, Graves, Zhang, & Soumerai, 2009) or

depression (Kennedy et al., 2011). However, one study found the CRN

significantly decreased for groups of beneficiaries with no coverage, coverage

from Medigap plans, or coverage from Medicare HMO (ORs between 0.4 and

18

0.6) (Safran et al., 2010). The effect of the Part D program on medication

adherence among beneficiaries with depression was further evaluated in another

investigation using Medicare Advantage claims data (Donohue et al., 2011). The

study measured adherence by using MPR over a six month period after an index

antidepressant prescription, stratified by levels of prior drug coverage, and

showed that the odds of being adherent (MPR>=80%) was significantly higher

after Part D among the groups whose coverage improved with Part D (OR=1.86

for no prior coverage group; OR=1.74 for prior $150 cap group; OR=1.19 for

prior $350 cap group).

In summary, Medicare Part D has shown a positive effect on improving

medication adherence in general and in various disease populations, though the

effect is not consistent in the depressive population.

Impact on Healthcare Resource Use

The literature related to the impact of Part D on healthcare resource use is

limited and has shown mixed results. Afendulis et al. (2011) used hospital data

from 2005 to 2007 and linked it with state-level drug coverage data for 23 states

to compare changes in the probability of hospitalization before and after Part D

implementation (Afendulis & Chernew, 2011). The authors reported that Part D

reduced the overall rate of hospitalization by 4.1% and about 42,000

hospitalizations were avoided annually for eight ambulatory care sensitive

conditions (i.e., short-term complication of diabetes, uncontrolled diabetes,

COPD, CHF, angina, asthma, stroke, and AMI). Their calculation was

extrapolated to all 50 states and the District of Columbia in the United States,

with the assumption that the same relationship between drug coverage rate

19

changes and the trend in the ACSC hospitalization rate found in the 23-state

sample also prevailed in other states. This calculation indicated that the annual

number of hospitalizations avoided would have been approximately 76,000 per

year (Afendulis, He, Zaslavsky, & Chernew, 2011). In contrast, Liu and

colleagues did not find a statistically significant decrease in emergency room use

and hospitalizations as a result of the Part D coverage during the first year of

implementation (Liu et al., 2011).

Impact on Cost

Out-of-Pocket (OOP) Cost for Drugs

One intention of implementing Part D is to increase beneficiaries’ access

to prescription drugs by decreasing their OOP cost. A number of studies have

investigated the effect of Part D on OOP spending using longitudinal claims data

or other applicable data sources. Multiple studies reported a range of 13%-18%

decrease in patients’ OOP cost ($143 to $148 reduction per year) (Briesacher et

al., 2011; Ketcham & Simon, 2008; Polinski, Kilabuk, Schneeweiss, Brennan, &

Shrank, 2010; Yin et al., 2008; Zhang, Lave, Newhouse, & Donohue, 2010).

One study estimated the OOP cost from 2003 to 2006 among non-

institutionalized Medicare beneficiaries who enrolled in Part D and stratified the

results by prior drug coverage status (no coverage, with coverage from Medicare

HMO, with coverage from Medigap, or employer-sponsored coverage). The

study found significantly lower odds of spending more than $100 or $300 per

month on prescription drugs after Part D in all groups except for the group with

prior employer-sponsored coverage (Safran et al., 2010). However, mixed

findings were reported for different subgroups of Medicare beneficiaries. For

20

example, the cost reduction was not significant for elderly dual-eligible

beneficiaries (Polinski et al., 2010).

Furthermore, Basu, Yin, and Alexander compared beneficiaries’ OOP

expense and total expenditures in the first 18 months after the Part D

implementation between a “treatment” group (65-78 years old patients with dual

eligibility on 1/1/2005) and a “control” group (60-63 years old patients with

Medicaid coverage on 1/1/2005). They found no significant changes in the dual-

eligibles’ OOP cost or total monthly expenditures (Basu, Yin, & Alexander,

2010).

Non-Drug Cost

The research related to non-drug cost for traditional fee-for-service

Medicare members produced inconsistent results (Ingber, Greenwald, Freeman,

& Healy, 2010). One study compared the non-drug medical cost in those with

limited prior coverage and those with generous prior coverage and found that

total non-drug medical cost decreased after 1/1/2006 by an average of $306 per

quarter for beneficiaries with limited prior drug coverage relative to those with

generous prior coverage, mostly attributable to changes in spending on inpatient

(-$204 per quarter) and skilled nursing facility care (-$586 per quarter)

(McWilliams et al., 2011). The findings suggested that Part D was associated

with significant reductions in non-drug medical cost for Medicare beneficiaries

with limited prior drug coverage. Another study estimated the savings from Part

D for beneficiaries with CHF and reported the non-drug cost were $1,827 lower

per beneficiary per year (Dall, Blanchard, Gallo, & Semilla, 2013). The

magnitude was even greater among the previously uninsured population ($2,050

21

per year), but smaller for patients with limited ($773) or moderate ($465)

coverage prior to Part D enrollment. The authors estimated a cost reduction of

$2.9 billion savings in medical expenses, or $2.6 billion if accounting for the

increase in medication spending (Dall et al., 2013).

The potential of Part D to reduce non-drug cost may be limited to certain

disease populations. For example, cost savings on non-drug services was not

observed for the Medicare population with arthritis, where a non-significant

decrease in medical cost was observed (Cheng & Rascati, 2012).

Impact of Medicare Part D Coverage Gap

Despite the beneficial impact of Medicare Part D discussed earlier, the

coverage gap in Medicare Part D is a realistic challenge faced by Medicare

beneficiaries. During the first year of Medicare Part D implementation, large

variations in the number of beneficiaries who fell in the coverage gap for that

year were reported, depending on which plan beneficiaries were enrolled in: 6%-

58.8% for MA-PD plan enrollees (Ettner et al., 2010; Karaca, Streeter, Barton, &

Nguyen, 2008; M. H. Kim, Lin, & Kreilick, 2009; Raebel, Delate, Ellis, &

Bayliss, 2008; Schmittdiel et al., 2009; Schneeweiss, Patrick, et al., 2009; Zhang,

Donohue, Newhouse, & Lave, 2009), 43% for non-Medicaid beneficiaries in

PDPs (Karaca et al., 2008), and 40% for beneficiaries with employer-sponsored

coverage (Zhang et al., 2009). In 2007, a more consistent range of beneficiaries

(18.5%-26%) was reported to reach the Part D coverage gap (J Hoadley,

Hargrave, Cubanski, & Neuman, 2008; Pedan, Lu, & Varasteh, 2009).

Although concerns about potential negative effects of the coverage gap

on Medicare beneficiaries’ adherence and health outcomes were brought up at

22

the beginning of the program, Medicare still incorporated the Part D coverage

gap because the budget to fund the Part D program would have been insufficient

otherwise and policy makers wanted to use this as a means to contain healthcare

expenditures and lower the financial burden of Medicare through cost sharing

with beneficiaries (Rosenthal, 2004).

Impact on Medication Adherence

Medicare Part D with drug coverage is intended to improve access to

prescription drugs by helping patients mitigate the effects of OOP drug cost.

However, the coverage gap can work at cross-purposes with this goal. The U.S.

Department of Health and Human Services estimates that over a quarter of Part

D participants stop filling their prescribed drugs when they hit the coverage gap

(Claffey, 2010).

Adherence to Different Drug Classes

A 2008 report published by the Kaiser Family Foundation was one of the

first studies to examine the impact of the coverage gap on drug use/adherence.

Using national patient-level retail pharmacy claims data for Part D enrollees

from IMS Health and focusing on Part D enrollees taking one or more drugs in

the eight drug classes for common chronic conditions including Alzheimer’s

disease, high cholesterol, depression, diabetes, gastroesophageal reflux disease,

heart failure, hypertension, and osteoporosis, the researchers found that 15%

stopped taking one or more medications, 5% switched to an alternative drug in

that class, and 1% reduced their medication use after reaching the coverage gap

(J Hoadley et al., 2008).

23

Later, using Defined Daily Dose (DDD), Schneeweiss and colleagues

(2009) studied adherence to drugs in four essential medication classes:

clopidogrel, drugs used to treat asymptomatic conditions (statins), drugs used to

treat symptomatic conditions [proton-pump inhibitors (PPIs)], and less

expensive drugs (warfarin) and reported significant decreases ranging from 4.8

percentage points for statins to 6.3 percentage points for warfarin among

beneficiaries reaching the coverage gap (Schneeweiss, Patrick, et al., 2009).

Raebel et al. (2008) compared medication adherence measured by

Medication Refill Adherence (MRA) for six drug classes (statins,

antidepressants, anti-diabetic agents, anti-hypertensives, beta-blockers, and

diuretics) between Medicare members who reached the coverage gap and age-

and gender-matched members who did not reach the coverage gap. The study

found that adherence to chronic medications declined over time from 2005 to

2006 in both groups, but the reduction was greater for beneficiaries who reached

the coverage gap (Raebel et al., 2008). For the group reaching the coverage gap,

the decline in adherence from 2005 to 2006 was statistically significant for all

drug classes except for anti-diabetic agents and beta-blockers, and the decline

was smallest for diabetes drugs (3.4%) and largest for diuretics (8.3%).

Polinski and colleagues (2011) studied community-dwelling FFS

Medicare beneficiaries with prescription drug coverage through PDPs or a

retiree drug plan in 2006 or 2007 regarding changes in their use of

cardiovascular and hypoglycemic drugs (Polinski et al., 2011). They established

an “early Part D” cohort (2005-2006) and an “Established Part D” cohort (2006-

24

2007). Within each cohort, the “exposed” group (coverage gap) was matched

with an “unexposed” group (no coverage gap). Compared to unexposed patients,

exposed patients in both “early Part D” and “established Part D” cohorts had

significantly higher likelihood to discontinue a drug.

Adherence in Beneficiaries with Diabetes

Reduction in adherence related to the coverage gap for beneficiaries with

diabetes has been reported in several studies. Compared to those without the

coverage gap, the odds of adherence among patients with the coverage gap

decreased from 17% to almost 40% (ORs between 0.62 and 0.83). Two

additional studies found that MPR for antidiabetics decreased by 10.3% (Zhang,

Baik, & Lave, 2013), or 2.9 - 3.3 percentage points among patients with the

coverage gap with the biggest impact on the differential rates of starting or

stopping for several expensive drug classes (Joyce, Zissimopoulos, & Goldman,

2013).

Adherence in Beneficiaries with Heart Failure

Compared to the diabetic population, a smaller scale of effect on

adherence was observed in patients with heart failure. After beneficiaries entered

the coverage gap, a reduction of 2.5%-3.6% in adherence was reported by two

studies based on the same database (Baik et al., 2012; Zhang et al., 2013).

Adherence in Beneficiaries with Other Cardiovascular Diseases

Based on a 5% random sample of Medicare claims data, Li and

colleagues (2012) used a quasi-experimental design to determine the magnitude

of the effect of coverage gap on the usage of antihypertensive and lipid lowering

medications among beneficiaries with hypertension and hyperlipidemia (Li,

25

McElligott, Bergquist, Schwartz, & Doshi, 2012). Comparing across four groups

[with low income subsidy (LIS), with generic drug coverage, with generic and

brand-name drug coverage, and no gap coverage], the authors found that the

limited coverage was associated with an increased risk for medication non-

adherence (measured by PDC; OR=1.60 for antihypertensive; OR=1.59 for lipid

lowering drugs) as well as non-persistence (indicated if no supply of medication

for ≥ 30 days; OR=1.38 for anti-hypertensives; OR=1.35 for lipid lowering

drugs). In addition, the generic drug coverage during the coverage period did not

mitigate these effects, and the gap effect did not occur to medications which treat

symptoms such as pain relievers.

Another study based on a 5% random sample of Medicare claims data

assessed the effect of Part D benefit phases on adherence (measured by PDC)

with evidence-based medications (statin, clopidogrel, beta-blocker, and ACE

inhibitor/ARBs) among Medicare beneficiaries following acute myocardial

infarction. Benefit phases were defined as initial coverage phase, coverage gap

phase, and catastrophic coverage phase. For non-LIS enrollees, entering the

coverage phase was associated with significant reductions in mean PDC for all

four drug classes (-7.8% for statins, -7.0% for clopidogrel, -5.9% for beta-

blockers, and -5.1% for ACE inhibitors/ARBs); however, no significant changes

in adherence were observed when beneficiaries transitioned from the coverage

gap to the catastrophic coverage gap (B. Stuart et al., 2013).

Adherence in Beneficiaries with Depression

In several observational studies using claims data, the reported effect of

Part D coverage gap on adherence in the population with depression was

26

insignificant, either no change (Baik et al., 2012) or with modest effect (6.9%-12%

reduction without significance compared to pre-gap) (Zhang, Baik, Zhou,

Reynolds, & Lave, 2012). However, findings from another study based on

qualitative and semi-structured interviews with non-dually eligible Medicare

beneficiaries who had a mental illness suggested that adherence was negatively

impacted by the benefit structure (Bakk, Woodward, & Dunkle, 2014).

Overall, Medicare Part D coverage gap was found to be associated with

reduced medication adherence, although the association in the population with

depression or mental disorder was not conclusive.

Impact on Resource Use

Only one study was identified that examined the impact of the coverage

gap on healthcare resource use. Raebel et al. (2008) found that, compared to age-

and gender-matched members who did not have the coverage gap, those who

reached the coverage gap had 85% higher risk of being hospitalized [incidence

rate ratio (IRR)=1.85] and 60% higher risk of using emergency room services

(IRR=1.60) (Raebel et al., 2008).

Impact on Cost

The effect of the coverage gap on cost has not been widely studied and

most of the published research has focused on OOP cost. Limited data have been

reported regarding the impact of the coverage gap on other types of cost such as

non-drug cost.

OOP Cost

Multiple studies have shown that beneficiaries with the coverage gap had

higher OOP cost than those without (Sun & Lee, 2007). However, when OOP

27

costs were compared between the pre- and post-gap periods, Sun and Lee

reported that costs fell 28% from $2,441 to $1,757 due to the reduced days of

therapy by 16% (Sun & Lee, 2007). Similarly, for Medicare beneficiaries with

diabetes or heart failure, an overall decrease in monthly medication spending

after they entered the gap was observed, primarily due to the decreased use of

brand-name drugs (Zhang et al., 2013). The group without gap coverage had

their monthly pharmacy spending reduced by $73.15, on average, of which

$66.65 was for brand-name drugs. Although the coverage gap does disrupt the

use of prescription drugs among seniors with diabetes, the declines in usage are

modest and concentrated on higher cost, brand-name medications.

Non-Drug Cost

Only one study was identified regarding the effect of the coverage gap on

non-drug cost. Based on a 5% random sample of Medicare data, Zhang et al.

(2012) found that there were no significant increases in non-drug medical cost

after beneficiaries entered the coverage gap among those who were diagnosed

with depression and continuously enrolled in stand-alone Part D plans in 2007

(Zhang et al., 2012). In sum, consumption of brand prescription drugs was

impacted more than generic drugs as a result of the coverage gap and the effect

on non-drug cost was not conclusive.

Brief Review of Research Design/Methods

All of the reviewed research consisted of observational studies.

Difference-in-Difference or pre-post comparison was the most common design

used to reduce the potential selection bias between beneficiaries with and

without the coverage gap. Adjustment for confounders were incorporated in

28

most of the studies and propensity score matching technique was often used

when an “exposure” group (having coverage gap) and a “control” group (no

coverage gap) were defined. Stratifications in the “exposure” group were often

based on benefit design related to the coverage gap, i.e., coverage gap without

any additional benefit, coverage gap with additional generic drug benefit, or

coverage gap with additional benefit for both generic and brand-name drugs.

Assessment of Literature

The preceding review of the literature on the effects of Medicare Part D

indicates that Part D is associated with lower OOP cost, especially for long-term

medications; however, differences across subgroups of beneficiaries have been

observed. Studies on non-drug medical cost and resource uses were limited and

mixed results were reported. Some recent studies demonstrated savings and

reduction in non-drug medical cost and services attributable to Part D, but also

showed differences across subgroups.

As shown by the review, the coverage gap has posed a significant

challenge to beneficiaries in terms of their medication adherence and cost-

sharing burden. Although some research reported an offset effect on the OOP

cost when utilization of brand-name drugs decreased as a result of the coverage

gap, suboptimal health outcomes in beneficiaries and unexpected economic

burden to Medicare may still result from the coverage gap.

The review also highlights several gaps in the extant literature, including:

1) The majority of the research was based on the enrollees in

MA-PDs. Limited evidence was available on the impact of

Part D coverage gap among PDP enrollees. The beneficiaries

29

in PDP may have different characteristics compared to the

beneficiaries in MA-PDs, and they can experience different

benefit designs. So they may respond to the Part D and the

Part D coverage gap quite differently. Investigating the effect

of Part D coverage gap on the impacted beneficiaries in

various health insurance systems will be helpful to gain better

understanding regarding the Part D program.

2) Research on the impact of Part D or Part D coverage gap has

been devoted to heterogeneous subgroups of Medicare

beneficiaries defined by chronic diseases. However, the

current research has been limited to subgroups with diabetes,

depression, heart failure, and myocardial infarction. Given the

debilitating disease progression and the substantial burden of

COPD, as well as the high prevalence of COPD in the

Medicare population, assessing the impact of Part D coverage

gap on the COPD population would be useful to future

management of COPD in Medicare.

3) Study outcomes in the assessment of the impact of Part D

coverage gap are typically medication utilization (e.g.,

percentage of usage), adherence (PDC, MPR, MRA), and

OOP cost. There is limited evidence regarding the impact on

resource use and non-drug cost or total cost. Assessing the

impact of Part D coverage gap on overall resource use or non-

30

drug cost beyond adherence may help to build a holistic

picture regarding the consequences of the Part D coverage gap

from both clinical and economic perspectives, and may

provide stronger evidentiary support for refining policy or

developing strategies to solve identified problems.

Theoretical Framework

Theory Overview

Medication adherence is basically a patient’s choice regarding whether

he/she will take medicine, and how frequently or for how long he/she will take

the medicine. A patient, as a self-interested individual, usually makes such a

choice based on his/her preferences to maximize gain and to minimize loss.

Similarly, when a patient makes a decision to go to a physician’s office or an

emergency room, or when a patient decides how to spend money on healthcare,

he/she makes a preferred choice and takes rational actions to optimize his/her

benefit. One of the theories describing this behavior is rational choice theory,

also known as choice theory or rational action theory. Rational choice theory

provides a framework for understanding and modelling social and economic

behavior (Blume & Easley, 2008; Sen, 2008).

Rational choice theory was pioneered by sociologist Goerge Homas and

was developed further by other theorists such as Blau, Coleman, and Cook

during the 1960s and 1970s (Scott, 2000). This theory evolved to a formal

mathematical model of rational choice, and even became a basis of a Marxist

theory of class and exploitation. The theory not only has a strong economic

orientation, such as quantitative measurement of cost or profit and utility

31

motivated analysis, but also has a robust linkage to social and sociological

factors (Calvert, 1994).

Rational choice theory provides an approach to assessing decision-

making based on empirical evidence, understanding choices, and rationalizing

the inferences and conclusions. It consists of systematic evaluation of choice

options through an analysis of the various consequences of the choices,

including validity, rationality, value, and risk. Intuitively speaking, as people are

often motivated by the potential of making a profit or impacted by financial

factors such as cost, they tend to calculate the possible cost and benefits of any

action before deciding what to do.

Rational choice theory is built upon one central assumption, which is that

all social phenomena can be explained in terms of the individual actions that lead

to the phenomena. Another key element is the belief that all actions are

fundamentally “rational” or rationally motivated, even though some may appear

to be irrational, such as discontinuing effective therapy and seeking care in an

emergency room when the disease relapses. However, rational choice theory has

been criticized for its overemphasis on individualistic actions and inadequate

explanations of collective actions and social norms such as altruism, reciprocity

and trust, and situations where collective and non-individual benefits are pursued

(e.g., not-for-profit charity organizations or groups of volunteers) (Scott, 2000).

The application of rational choice theory is inherently a multilevel

enterprise and the theory is often used as an assumption of the behavior of

individuals in microeconomic models of human decision-making (Coleman,

32

1990; Hechter & Kanazawa, 1997; Hedström & Stern, 2008; Lohmann, 2008).

As depicted by Figure 2, at the lower level (micro level), the rational choice

model contains assumptions about individual cognitive capacities and values as

indicated by a lower case of x, and actions or behaviors taken by individuals are

indicated by a lower case of y. At the higher level (macro level), the rational

choice model includes specifications of different scenarios, where the upper case

of X is the current scenario for individuals at Time 1 and the upper case of Y is

the new scenario resulting from the actions of individuals at Time 2. In this

model, relationship 1 indicates that individual cognitive capacity or values (x)

are shaped by current scenario (X); relationship 2 describes how a person subject

to a given scenario at Time 1 will behave at Time 2 (y) based on his or her

values or beliefs (x); and relationship 3 denotes the change of the scenario (Y) as

a result of these behaviors (y). For example, the belief in the importance of

taking medications (x) is associated with how much an individual needs to pay

for medication and other healthcare resources (X). An individual may switch

from taking medication to using physician services (y) if he or she thinks

physician services are more affordable (x). As a result of these behaviors (y),

medication adherence is reduced and physician office visits are increased (Y).

33

Figure 2. Rational choice theory model.

Note: Adapted based on Figure 1 from Hechter & Kanazawa, Annu Rev Sociol (23):191-214.

Application of the Rational Choice Theory

Patients usually have a choice of whether to follow, modify or reject a

prescribed treatment. The choice is influenced, however, by attributes of the

treatments and other contextual factors such as new health policy or changes in

existing policies. More specifically, non-adherence may be explained rationally

by the following reasons:

• A patient’s belief that the treatment is not working;

• Side effects that negatively affect patient’s quality of life;

• Practical barriers to the treatment such as high cost; and

• Patients wanting to check if the illness is still there when they

stop taking medications.

Time 1 Time 2

Cognitive capacities,

values, beliefs

Actions, behaviors, Choices

Current scenario New scenario

34

When rational choice theory is applied to assess the impact of Medicare

Part D coverage gap on medication adherence, the consumption of goods and

services (i.e., prescription drugs and healthcare services in this case) is largely

driven by cost (i.e., out-of-pocket cost in this case). Medicare patients are

“rationally” motivated by their health needs and goals when they decide to take

their prescription drugs. When patients are aware of the Part D coverage gap,

they may make a rough prediction of their healthcare needs based on their

current health status, calculate their drug expenses at the beginning of a year, and

take “rational” actions to adjust their medication-taking and care-seeking

behavior accordingly. If taking medications is more affordable than healthcare

received in other settings, beneficiaries are more likely to comply with their

therapy to maintain their health status instead of using other resources.

However, in the situation where taking medications becomes more

expensive than seeking care in a physician office, emergency room (ER), or

hospital, beneficiaries may skip or forgo their prescription drugs or switch to less

costly generic drugs. Alternatively, they may leverage their Part A or B benefit

to use alternative resources more frequently when they cannot control symptoms

or deteriorate after discontinuing medications. In the end, beneficiaries will most

likely make the choice that they think can give them the greatest value or best

outcome possible. This thought process may result in several different scenarios

related to the coverage gap, including:

1) If Medicare beneficiaries are relatively healthy, they do not

need to worry about the coverage gap because the probability

35

of reaching the coverage gap is very small. In this case, their

medication taking behavior is not expected to have a

noticeable change as a result of the introduction of the

coverage gap, and their overall healthcare resource use and

cost are not expected to fluctuate as well.

2) If beneficiaries are marginally sick, they may carefully

arrange or organize their medication needs so that they may

manage to enter the coverage gap later during the year. Their

consumption of medications or alternative healthcare

resources can be marginally impacted by the coverage gap,

for example, they may potentially skip one or two

prescriptions or stop medications for a short period of time,

and seek care from healthcare providers when needed.

3) If beneficiaries have suboptimal health status, they are more

likely to enter the coverage gap and enter the coverage gap

earlier during the year. Their medication-taking behavior is

expected to shift dramatically after they reach the coverage

gap in comparison to before reaching the coverage gap. They

may not refill their medications by schedule or even stop

taking their medication completely because they have to

assume full cost of the prescription drugs. In this case,

beneficiaries may greatly increase the use of other healthcare

resources to achieve their health goals.

36

4) If beneficiaries are very sick, they may quickly enter and exit

the coverage gap and then enter the catastrophic coverage. As

these patients may want to maximize their benefit from the

catastrophic coverage, the coverage gap may exert a very

different effect on their medication-taking and care-seeking

behavior. They may use medication as much as possible at the

beginning of year so that they can be covered by the

catastrophic benefit as soon as possible. Substitute usage of

other healthcare resources may be less aggressive among

these patients.

As summarized in the literature review section, the Part D coverage gap

has resulted in significant reduced medication adherence. Although direct

evidence of the effect on healthcare resource use and cost is limited, a

relationship can be conjectured based on past research that has demonstrated

non-adherence to prescribed medications to be associated with “poor health

outcomes for patients, missed opportunities for therapeutic gain, and increased

health care cost often associated with a worsening of the condition being treated”

(Clifford & Coyne, 2014, p. 650). Therefore, the following hypotheses were

tested in this study:

1) Among Medicare beneficiaries with COPD, the coverage gap

will be associated with lower medication adherence to COPD

long-term maintenance therapies.

37

2) Among Medicare beneficiaries with COPD, the coverage gap

will be associated with higher consumption of healthcare

resources (non-drug).

3) Among Medicare beneficiaries with COPD, the coverage gap

will be associated with higher all-cause medical cost (from

payer’s perspective).

38

CHAPTER THREE

METHODS

This chapter describes the methods used in the study. It consists of study

design, data sources, sample selection, outcome measures, and data analysis.

Study Design

This was a retrospective cohort study using longitudinal observational

data. Details on the study cohort definitions are specified in the Study Cohorts

section. The primary analysis was focused on adherence to COPD long-term

maintenance therapies, all-cause medical healthcare resource use (HRU), and all-

cause medical cost during a calendar year between 2007 and 2010 among the

Part D beneficiaries with COPD who were at risk of and reached the coverage

gap (“exposure” cohort) as compared to those with COPD who were not at risk

of the coverage gap. Those who were dually eligible or had other drug benefits

that covered both brand and generic drugs during the gap (“control” cohort) were

not subject to the coverage gap, and thus were not included in the study.

In addition to the primary analysis, subgroups were constructed within

the exposure cohort based on when they reached the coverage gap. Specifically,

beneficiaries were assigned to the late-gap group if they reached the coverage

gap in November or later, the mid-gap group if between March and November,

or the early-gap group if earlier than March (detailed definitions are described in

the Study Cohort section). Outcomes prior to the coverage gap (pre-gap) and

during the coverage gap (in-gap) were compared between the subgroups. As the

39

focus of this study is the impact of Part D coverage gap, the period after COPD

patients reached the catastrophic threshold is considered out of the study scope

and was not included in this study.

High-Dimensional Propensity Score Analysis and Matching

In an observational study, selection bias is an important issue when

comparing groups and exposure is not randomly assigned. Propensity score

matching (PSM) was used in this study to generate comparable exposure and

control cohorts with balanced demographic and clinical characteristics. A

propensity score for an individual is the conditional probability of receiving

exposure given the observed factors or characteristics prior to exposure (Little &

Rubin, 2000). Therefore, individuals with similar propensity scores will tend to

have similar levels of the covariates, thereby removing or minimizing the bias

due to the covariates (Little & Rubin, 2000). Generally, a logistic regression is

constructed to estimate the propensity of subjects getting treatment, with the

dependent variable being the treatment received (dichotomous variable with 1

and 0) (D'Agostino, 1998).

Following the calculation of propensity scores, selection bias can be

accounted for in multiple ways: stratification, adjustment in a regression analysis,

and matching (D'Agostino, 1998). Unlike regression adjustment that is made

during the calculation, matching removes or minimizes the bias before

estimating the effect. Generally, matching is employed when there is a relatively

large sample size and a sufficient number of confounders that are available or

can be created for matching. After matching the observations in the two groups

on their propensity scores, the significance of differences in outcomes between

40

the two matched groups will be analyzed using techniques for non-independent

samples or matched pairs.

In a traditional approach to generating propensity scores, a number of

relevant confounders or covariates are defined based on available data and then

specified in the logistic model. Model specification is primarily guided by

knowledge related to exposure and the study population characteristics. When

using longitudinal healthcare claims data, the covariates typically include

demographics (e.g., age, gender), history of major medical conditions, overall

comorbidity scores, prior medication use, or history of healthcare resource use

(e.g., ER, hospitalization, physician office visits) over a given period of time

before exposure initiation (Charlson, Pompei, Ales, & MacKenzie, 1987; Gagne,

Glynn, Avorn, Levin, & Schneeweiss, 2011; Romano, Roos, & Jollis, 1993;

Schneeweiss et al., 2001). However, these covariates are not always adequate to

specify the degree of certainty/uncertainty in the causality between exposure and

outcome.

Although claims data provide rich information about patients and health

services, other important attributes are unavailable (e.g., laboratory results,

functional status, smoking status, over-the-counter medication) (Brookhart,

Sturmer, Glynn, Rassen, & Schneeweiss, 2010). In addition, empirically

identifying appropriate proxies for patient health status out of a large number of

variables in claims data is a significant challenge. In this context, Schneeweiss

and colleagues developed an automated algorithm for healthcare claims data to

set up proxies by assessing diagnosis codes, procedure codes, and prescribed

41

medication codes (Schneeweiss, Rassen, et al., 2009). These empirically

identified variables can be used alone or in conjunction with investigator-

selected variables to estimate a propensity score. This method has been labeled

high-dimensional propensity score (HDPS) analysis.

In this study, the dependent variable for the logistic regression used in the

HDPS analysis was the membership of the exposure or control cohort. The

propensity score indicates the probability of being in the exposure cohort. The

initial exposure and control cohorts were matched 1:1 based on the propensity

score using Greedy matching method (Parsons, 2009).

Data Source

A random sample of 5% of Medicare beneficiaries was used for this

study. The Medicare administrative claims database is a comprehensive data

source covering all beneficiaries who were enrolled in Medicare, capturing

information on demographic characteristics, enrollment, prescription drug events

(PDE), medical encounters in inpatient and outpatient settings, and health

services incurred in other facilities such as hospice or skilled nursing home. The

5% random sample used for this study included data from the Beneficiary

Summary files from 2005 to 2010, Part A and Part B claims from 2005 to 2010,

Part D Event (PDE) Data, and the Plan Characteristics files from 2006 to 2010.

This study was approved by the UAB Institutional Review Board and by the

CMS Privacy Board.

The Beneficiary Summary File provides demographic and enrollment

information about beneficiaries. Starting in 2006, this file also includes Part D

enrollment information. The Part A data file includes information on inpatient

42

hospital stays, including length of stay, diagnosis-related group, department-

specific charges, and up to 10 individual diagnosis and procedure codes. Part B

claims data include claims submitted by physicians and other healthcare

providers and facilities for the services covered by Medicare Part B. Each claim

contains ICD-9-CM (International Classification of Disease, Ninth Edition,

Clinical Modification) diagnosis and procedure codes, date and place of services,

demographic information of beneficiaries, and a physician identification number.

Data from outpatient hospitals, skilled nursing homes, hospice care facilities, and

durable medical equipment (DME) are also included in Part B claims data. All

data files can be linked with the denominator file that provides demographic

information of all the beneficiaries entitled to Medicare, including state and

county codes, zip code, date of birth, date of death, gender, race, and age.

The Part D Event data and the Drug and Plan Characteristics files contain

elements that provide information on beneficiary demographics, plan

characteristics, prescription fill date, drug characteristics (e.g., national drug

code [NDC] number, days of supply, quantity supplied, and fill number), and

cost and payment information (e.g., dispensing fee, patient paid amount, Part D

paid amount). The drug characteristics file can be used to determine the generic

equivalence of different medications (i.e., brand-name, generic name, strength

and dosage form). The plan characteristics file can be used to determine: (1)

whether a particular plan was PDP or MA-PD, (2) whether it offered coverage

for some or all drugs during the coverage gap, and (3) type of cost-sharing

43

strategies (e.g., deductible) used before beneficiaries reach the coverage gap. All

data files can be linked via the unique de-identified ID number.

Sample Selection

General Inclusion and Exclusion Criteria

Because the coverage gap thresholds varied by calendar year, patient

selection and outcome measures were employed at a yearly level. Considering

that many Medicare beneficiaries did not have a full year benefit in 2006, data

files from 2007 to 2010 were used for this study. Beneficiaries who met all of

the following inclusion criteria were selected to form a general patient pool:

• Had “of age” listed as the reason for Medicare eligibility, i.e.,

age is greater than 65 years as of six months prior to January

1 of a calendar year;

• Had a full year eligibility during a respective calendar year

and six months of continuous eligibility prior to January 1 of

the respective calendar year (baseline period);

• Had at least two outpatient claims with a diagnosis of COPD

(ICD-9-CM diagnosis codes: 491.xx, 492.xx, 494.xx, 496.xx)

on different dates OR at least one emergency room (ER) or

inpatient claim with COPD as the primary diagnosis during a

respective calendar year;

• Had at least two prescriptions of long-term maintenance

therapy for COPD (long-acting bronchodilator, LABD) filled

on different dates during a respective calendar year (Table 1).

44

Exclusion criteria:

• Were enrolled with a Medicare Advantage plan in any month

during a respective calendar year;

• Had a diagnosis of asthma (ICD-9-CM diagnosis code:

493.xx) during a respective calendar year, because some of

the LABD medications are also indicated for asthma;

• Had a diagnoses of cancer (ICD-9-CM diagnosis codes:

140.xx-239.xx) during a respective calendar year, because

patients with cancer may have different medication utilization

and spending patterns compared to other Medicare

beneficiaries;

• Had a disability or end-stage-renal-disease (ESRD, ICD-9-

CM diagnosis codes: 585.5x, 585.6x) during a respective

calendar year, because their benefits can differ substantially

from other Medicare beneficiaries.

45

Table 1

Maintenance Medications Used for COPD

Class Medication Generic Names Medication Brand-names (Patent Expiration Time)

LABD Medications

• Arformoterol

• Formoterol

• Indacaterol

• Salmeterol

• Tiotropium

• Budesonide+formoterol

• Fluticasone+salmeterol

• Mometasone+formoterol

• Brovana

• Foradil

• Onbrez/Arcapta

• Serevent

• Spiriva

• Symbicort

• Advair

• Seretide

SABD Medications

• Albuterol+ Ipratropium

• Ipratropium

• Levalbuterol

• Metaproterenol

• Pirbuterol

• Albuterol or Salbutamol

• Combivent

• Atrovent

• Xopenes

• Alupent

• Maxair

• Accuneb, Ventolin, Preventil, Asthalin, Asthavent, ProAir, Airomir, AZMASOL, Ventosol, Asmol,Vospire

Note: LABD – long-acting bronchodilator, SABD – short-acting bronchodilator.

Study Cohorts

Beneficiaries who met the selection criteria were divided into two study

cohorts: a “control” cohort that included beneficiaries who were not subject to or

at risk of the coverage gap and an “exposure” cohort that included beneficiaries

who were subject to the coverage gap. The definitions of these two cohorts are as

follows:

46

Control cohort

If beneficiaries fell into one of the following categories, they were

assigned to the control cohort:

• Had Medicare-Medicaid dual eligibility for the whole year;

• Qualified for Part D low-income-subsidies (LIS), i.e.,

received LIS for at least one month before and after they

entered the coverage gap; or

• Had additional benefits covering brand and generic drugs

during the gap.

Because beneficiaries in the control cohort were not exposed to the

coverage gap, even when their pharmacy spending exceeded the coverage gap

threshold, their drug coverage remained intact throughout the whole calendar

year.

Exposure cohort and subgroups in exposure cohort

If beneficiaries did not have dual eligibility or low-income-subsidies or

full benefit to help with the coverage gap during a calendar year, they were

assigned to the exposure cohort. It should be noted that beneficiaries who were

exposed to but did not reach the coverage gap in a respective year were

identified as “no-reaching gap subgroup.” Though characteristics of this group

were described in subgroup analysis, this subgroup was not included in the main

analysis. This exclusion was based on the assumption that beneficiaries who are

relatively healthy were much less likely to reach the coverage gap; therefore,

their medication-taking behavior was not expected to noticeably change as a

result of presence of the coverage gap.

47

Three additional subgroups were identified for subgroup analysis.

Beneficiaries who reached the coverage gap between March 1 and October 31

were identified as the “mid-gap subgroup”. Beneficiaries who reached the

coverage gap on and after November 1 were identified as the “late-gap

subgroup”. Finally, patients who reached the coverage gap before March 1 were

identified as the “early-gap subgroup”. Similar to the “no-reaching gap

subgroup”, the early-gap subgroup was included in the subgroup analysis but not

in the main analysis. This exclusion was based on the assumption that

beneficiaries who are very sick may have reached the coverage gap early and

wanted to maximize their medication usage during the gap period to enter the

catastrophic phase sooner. This group of patients was anticipated to be small and

to respond to the coverage gap differently than other subgroups. Figure 3 depicts

the subgroup designation within the exposure cohort.

Figure 3. Subgroups of the exposure cohort.

For the subgroup analysis, the day when a beneficiary reached the

coverage gap (i.e., his/her total drug spending reached the coverage gap

threshold in a calendar year) was defined as the gap date. The period prior to the

Jan 1 Mar 1 Nov 1 Dec 31

Late-gap subgroup:

Reach the coverage gap after 10/31

Mar

Mid-gap subgroup: Reach the coverage gap on 3/1--10/31

Early-gap subgroup: Reach the coverage gap before 3/1

Exposure cohort

No-reaching gap subgroup Not reach the coverage gap

48

gap date in a calendar year was defined as a pre-gap period; the period from the

index date until the last day of the coverage gap when a beneficiary reached the

catastrophic coverage or death or the end of a respective calendar year,

whichever was earlier, was defined as the in-gap period.

Outcome Measures

Primary Analysis

Adherence

In past research, methods to evaluate medication adherence included pill

counting, patient diaries, self-reporting, and use of claims data (Hess, Raebel,

Conner, & Malone, 2006). In this study, prescription claims data were used and

medication adherence was assessed at a drug class level, i.e., medications in the

LABD category were treated as one drug class. With retrospective pharmacy

claims data, multiple measures have been used in the literature to measure

adherence, such as the proportion of days covered (PDC), medication possession

ratio (MPR), and gaps in filling prescriptions (Hess et al., 2006). In this study,

adherence was measured by PDC.

Yearly PDC was defined as the proportion of days covered by LABD

relative to the treatment period during a calendar year. The treatment period was

calculated as the duration from the fill date of the first LABD prescription until

the end of the year. The formula to calculate the yearly PDC is specified below:

Yearly PDC = 100 x

Treatment period in days

Number of days covered by LABD prescriptions within a calendar year

49

PDC equal to or greater than 80% was regarded as good adherence. A

dichotomous variable for adherence was constructed as 1 if PDC ≥ 80%, and 0

otherwise.

Healthcare resource use (HRU)

All-cause HRU was measured as the number of visits in different service

settings, including outpatient (including physician office and other outpatient

encounters), emergency room (ER), and inpatient hospitals.

These variables were constructed for each calendar year from 2007 to

2010. In addition, considering COPD patients may switch from LABD to SABD

because SABD cost is lower, the utilization of SABD was assessed as well.

Specifically, the analysis included the following HRU outcomes:

• Number of outpatient visits per year

• Number of ER visits per year

• Number of inpatient visits per year

• Number of days supplied for SABD prescriptions per year

Cost

All-cause medical cost was estimated from the payer’s perspective, so

beneficiaries’ copays and deductibles were not included in the calculation.

Specifically, all-cause medical cost was defined as the cost related to medical

services offered in all settings including outpatient, ER, and inpatient. Like HRU

outcome variables, the cost variable was also constructed separately for each

calendar year from 2007 to 2010.

50

Subgroup Analysis

Adherence

For the subgroup analysis, PDC was estimated for the pre-gap and the in-

gap periods separately. The pre-gap treatment period was calculated as the

number of days between the fill dates of the first LABD prescription to the gap

date. The in-gap treatment period was calculated as the number of days between

the fill dates of the first LABD prescription after reaching the gap until the end

of gap, or end of a calendar year, or death, whichever occurred earlier. The

formula to calculate the pre-gap and in-gap PDC is specified below:

Pre-Gap PDC = 100 x

In-Gap PDC = 100 x

Healthcare resource use (HRU)

The number of visits in outpatient, ER, and inpatient settings and the

number of days supplied for SABD prescriptions were calculated on a monthly

basis separately for the pre-gap and the in-gap periods. Specifically, the

following variables were included for the HRU assessment:

• Monthly number of outpatient visits

• Monthly number of ER visits

• Monthly number of inpatient visits

• Monthly number of days supplied for SABD prescriptions

Number of days covered by LABD prescriptions within the pre-gap period

Pre-gap treatment period in days

Number of days covered by LABD prescriptions within the in-gap period

In-gap treatment period in days

51

Cost

Finally, all-cause medical cost was calculated on a monthly basis

separately for the pre-gap and the in-gap periods.

Variables

The following variables were considered as independent variables for the

analysis.

Variables for Primary Analysis

For the primary analysis, the key independent variable of interest was a

dichotomous indicator of membership in the exposure or the control cohort

(1=exposure cohort, 0=control cohort).

For the PSM, 300 variables were generated as proxies based on ICD-9-

CM diagnosis codes and CPT procedure codes from different dimensions

(outpatient, ER, and inpatient) using the HDPS technique. In addition, the

following variables were also included in the PSM:

• Age: Age was calculated as of six months before January 1 of

each calendar year.

• Gender: Male or female (1=female, 0=male, with male as

reference group).

• Ethnicity: Based on the categories provided in the data,

ethnicity was defined as Caucasian or other (1=Caucasian,

0=other).

• Region: Geographic locations of patient residence were

determined from residence state documented in the eligibility

52

file and were grouped based on the U.S. census classification

(i.e., Northeast, Midwest, South, and West regions).

• Charlson Comorbidity Index (CCI) in the baseline period:

CCI was assessed using medical claims six months prior to

the beginning of a respective calendar year (baseline period)

based on the primary or secondary ICD-9-CM diagnosis

codes listed on claims. The CCI adapted by Deyo and

described in Table 2 is the most commonly used index in

health outcome studies (D'Hoore, Bouckaert, & Tilquin, 1996;

Deyo, Cherkin, & Ciol, 1992; Schneeweiss & Maclure, 2000).

To achieve an aggregate score, the CCI assigns a weight

ranging from 1 to 6 (higher indicating greater disease severity)

to separate conditions defined by ICD-9-CM codes that are

associated with medical services provided to a patient.

• Presence of select relevant comorbidities in the baseline

period: Presence of a comorbidity was defined as having or

not having the condition based on ICD-9-CM diagnosis code

(binary, 1=having the condition, 0=not having the condition).

These comorbid conditions include hyperlipidemia,

hypertensive disease, heart disease, osteoporosis, depression,

diseases of the musculoskeletal system and connective tissue.

Table 3 includes details on all of the conditions considered in

the analysis.

53

• Number of unique drug prescriptions filled in the baseline

period. The national drug code (NDC) is a unique 10-digit, 3-

segment numeric identifier assigned to each medication listed

with the Federal Food, Drug and Cosmetic (FD&C) Act

enforced by US Food and Drug Administration (FDA). The

CMS created an 11-digit NDC derivative which includes the

labeler, product or package code segments of NDC with

leading zeroes wherever they are needed to result in a fixed

length of 5-4-2 configuration. The first segment of a code

identifies the drug manufacturer, the second segment

identifies the specific product, and the third segment identifies

the package size. The first nine digits of the CMS NDC

derivative recorded in the Part D data file were used to

identify the unique drug dispensed to beneficiaries. Therefore,

the number of unique drugs filled for that year was defined as

the number of different unique 9-digit CMS NDC derivatives

recorded during that year.

• Number of all-cause ER visits in the baseline period

• Number of all-cause inpatient visits in the baseline period

• Any COPD diagnosis in the baseline period (1= at least one

ICD-9 code for COPD, 0=no ICD-9 code for COPD).

Beneficiaries with no COPD diagnosis in the baseline period

were considered as incident COPD patients.

54

• Any LABD prescription in the baseline period (1= at least

one LABD prescription, 0=no LABD prescription).

Beneficiaries without LABD prescription in the baseline

period were considered as LABD new users.

• Any supplemental oxygen therapy in the baseline period.

The consensus guidelines recommend adding supplemental

oxygen therapy for patients with very severe COPD (GOLD,

2014); therefore, it can serve as a proxy of the severity of

COPD. Supplemental oxygen therapy was identified by the

ICD-9 procedure codes or Healthcare Common Procedure

Coding System (HCPCS) codes listed in Table 4.

• Any use of oral corticosteroid in the baseline period (1= at

least one oral corticosteroid, 0=no oral corticosteroid). Oral

corticosteroids are recommended by consensus guidelines for

patients during exacerbations of COPD (GOLD, 2014);

therefore, it can also work as a proxy of the higher severity of

COPD if patients receive systemic corticosteroid (Table 5).

55

Table 2

ICD-9-CM Diagnosis Codes for CCI Conditions

Charlson Comorbid Conditions ICD-9-CM Diagnosis

Codes

Myocardial infarction 410.x, 412.x

Congestive heart failure 428.x

Peripheral vascular disease 443.9, 441.x, 785.4, V43.4

Cerebrovascular disease 430-438.x

Chronic pulmonary disease 490-496.x, 500-505.x, 506.4

Dementia 290.x

Paralysis 342.x, 344.1x

Diabetes 250.0x – 250.3x, 250.7x

Diabetes with sequela 250.4x – 250.6x, 250.8x, 250.9x

Moderate or severe renal disease 582.x, 583.x, 585.x, 586.x, 588.x

Mild liver disease/Various cirrhosis 571.x

Moderate or severe liver disease 572.x, 456.x

Ulcer disease 531-534.x

Rheumatologic disease 710.x, 714.x, 725.x

AIDS 042.x – 044.x

Any tumor 140-195.x

Metastatic solid tumor 196-199.x

Note: Created based on Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index for use with ICD-9-CM administrative databases. J Clin Epidemiol. 1992;45(6):613-619

56

Table 3

ICD-9-CM Codes for Select Relevant Comorbidities

Chronic conditions or Disease classes ICD-9-CM Diagnosis

Codes

Asthma 493.xx

Diabetes 250.xx

Hypertensive disease 401.xx-405.xx

Heart disease 410.xx-429.xx

Cerebrovascular disease 430.xx-438.xx

Depressive disorder 296.2x, 296.3x, 300.4, 311

Anxiety 293.84, 300.0x, 300.21, 300.22 300.23, 300.29, 300.3x, 300.5x, 300.89, 300.9x, 308.xx 309.81, 313.0x, 313.1x, 313.21, 313.22, 313.3x, 313.82, 313.83

Diseases of the musculoskeletal system and connective tissue

710.xx-719.xx

Deficiency anemia 281.xx, 285.xx

Lipid disorder 272.0x, 272.1x, 272.2x, 272.3x, 272.4x

Osteoporosis 733.0x, V17.81

Osteoarthritis 715.xx V13.4 (arthritis)

GERD 530.81, 530.10, 530.11, 530.12, 530.19

Sleep apnea 780.51, 780.53, 780.57, 327.20, 327.21, 327.23, 327.27, 327.29

Obesity

278.xx, V77.8, V85.2-V85.5

57

Table 4

Procedure Codes for Supplemental Oxygen Therapy

Supplemental Oxygen Therapy Description

ICD-9 procedure code: V46.2 Machine dependent supplemental oxygen

HCPCS codes:

E0431 Compressed-oxygen systems

E1390, E1391 Oxygen concentrator

E1392 Portable oxygen concentrator

Table 5

Oral Corticosteroid

Brand-name Generic Name

Diprolene, Betaderm, Betnovate, Diprosone Betamethasone

Decadron, Maxidex, Ozurdex, Baycadron Dexamethasone

Cortenema, Solu-cortef, Cortef, Cortifoam Hydrocortisone

Medrol Dosepak, Solu-Medrol, Medrol, MethylPREDNISolone

Methylprednisolone

Deltasone, Sterapred, Rayos, Sterapred DS Prednisolone

Deltasone, Sterapred, Rayos, Sterapred DS Prednisone

Kenalog-40, Aristocort, Azmacort, Kenalog-10

Triamcinolone

Cortone Acetate Cortisone

Depo-Dilar Paramethasone

The distribution of these characteristics was presented before and after

the matching process. Variables with a statistically significant difference

between exposure and control cohorts after propensity score matching were

included in a separate multivariable model to further adjust for differences. This

multivariable model included the following covariates:

58

• Unbalanced covariates after the PSM

• Cohort membership, i.e., in the exposure cohort or the control

cohort

Variables for Subgroup Analysis

The same variables described in primary analysis were included in the

descriptive portion of the subgroup analysis. Table 6 below summarizes the

variables included in the primary analysis and subgroup analysis, respectively,

and the definition of each variable.

Table 6

A List of Variables for Primary and Subgroup Analyses

Variable Definition Variable Type

Primary Analysis

(PSM/DA/MA)

Subgroup Analysis

(DA only)

Outcome Variable

Proportion of days covered (PDC)

The proportion of days covered by LABD relative to the treatment period during a calendar year

Continuous, ranging between 0 and 1

DA, MA DA

Adherence PDC is equal to or greater than 80% Binary, 1 indicating adherent, and 0 otherwise

DA, MA DA

All-cause outpatient visits

Number of all-cause outpatient visits occurred in a calendar year

Count DA, MA DA

All-cause ER visits Number of all-cause ER visits occurred in a calendar year

Count DA, MA DA

All-cause inpatient visits Number of all-cause inpatient visits occurred in a calendar year

Count DA, MA DA

All-cause medical cost Cost related to medical services offered in all settings including outpatient, ER, and inpatient in a calendar year, paid by Medicare

Continuous DA, MA DA

Independent Variables

59

Variable Definition Variable Type

Primary Analysis

(PSM/DA/MA)

Subgroup Analysis

(DA only)

Age Age in years as of 6 months before January 1st of each calendar year

Continuous PSM, DA, MA DA

Gender Male or female Binary, 1 indicating female, and 0 male

PSM, DA, MA DA

Ethnicity Caucasian or other ethnicities Binary, 1 indicating Caucasian, and 0 otherwise

PSM, DA, MA DA

Region Grouped state codes based on the U.S. census classification: NorthEast, MidWest, South, or West

Categorical: NorthEast, MidWest, South, or West

PSM, DA, MA DA

Baseline CCI score An aggregate score based on a weight ranging from 1 to 6 assigned to 17 different conditions identified by ICD-9-CM codes

Continuous PSM, DA, MA DA

Presence of select comorbidities in the baseline period

Comorbidities considered relevant to COPD population and defined by ICD-9-CM diagnosis codes listed on medical claims

Binary, 1 indicating presence of the disease, and 0 otherwise

PSM, DA, MA DA

Number of unique drugs filled in the baseline period

The number of different unique 9-digit CMS NDC derivatives recorded during a calendar year

Continuous PSM, DA, MA DA

60

Variable Definition Variable Type

Primary Analysis

(PSM/DA/MA)

Subgroup Analysis

(DA only)

Any COPD diagnosis in the baseline period

Having medical claims with COPD diagnosis code in the baseline period.

Binary, 1 indicating prevalent COPD patients, and 0 incident COPD patients

PSM, DA, MA DA

Any LABD prescription in the baseline period

Receiving LABD prescription in the baseline period.

Binary, 1 indicating receiving LABD prescription before, and 0 LABD new users

PSM, DA, MA DA

Any supplemental oxygen therapy in the baseline period

Receiving supplemental oxygen therapy in the baseline period, identified by ICD-9-CM procedure code or Healthcare Common Procedure Coding System (HCPCS) codes

Binary, 1 indicating received oxygen therapy, and 0 otherwise

PSM, DA, MA DA

Any oral corticosteroid use in the baseline period

Receiving oral corticosteroid in the baseline period.

Binary, 1 indicating received oral corticosteroid, and 0 otherwise

PSM, DA, MA DA

Number of all-cause ER visits in the baseline period

Number of all-cause ER visits occurred in the baseline period

Count PSM, DA, MA DA

61

Variable Definition Variable Type

Primary Analysis

(PSM/DA/MA)

Subgroup Analysis

(DA only)

Number of all-cause inpatient visits in the baseline period

Number of all-cause inpatient visits occurred in the baseline period

Count PSM, DA, MA DA

Membership of the exposure or the control cohort

Patients categorized into the exposure (with the coverage gap) or the control cohort (without the coverage gap)

Binary, 1 indicating exposure cohort, and 0 control cohort

PSM, DA, MA --

Membership of the mid-gap or the late-gap subgroup

Patients who reached the Part D gap between 4/1 and 10/31 were assigned to the mid-gap subgroup; patients who reached the gap after 10/31 were assigned to the late-gap subgroup.

Categorical: mid-gap, and late-gap

MA DA

Note: PSM – propensity score matching, DA – descriptive analysis, MA – multivariable analysis

62

63

Statistical Analysis

The following sections describe the data analysis and statistical methods

applied in this study.

Descriptive Analysis

A descriptive analysis was conducted to summarize patient demographic

and clinical characteristics, HRU, and cost for study cohorts. Means and

standard deviations (SD) were reported for continuous variables and frequency

distributions with percentages reported for categorical variables.

Primary analysis

Patient demographic and clinical characteristics of the exposure and the

control cohorts were summarized before and after the PSM. Yearly adherence,

HRU, and cost were reported for the two matched cohorts after the PSM. Before

the PSM, the Student’s t-test was used to detect differences between the

exposure and the control cohorts for continuous variables (e.g., age, CCI score,

medical encounters, cost), and the Chi-square test for categorical variables,

including demographics (e.g., gender, ethnicity) and comorbidities (e.g., diabetes,

hypertensive disease). After the PSM, McNemar’s test was used for categorical

variables and the paired t-test for continuous variables.

Subgroup analysis

Patient demographic and clinical characteristics were summarized for all

subgroups of the exposure cohort. To assess differences between the subgroups,

the Student’s t-test was used for continuous variables (e.g., age, CCI score), and

the Chi-square test for categorical variables including demographics (e.g.,

gender, ethnicity) and comorbidities (e.g., diabetes, hypertensive disease). To

64

assess differences in monthly HRU and cost before and after reaching the

coverage gap for the mid-gap and the late-gap subgroups in the matched

exposure cohort, the McNemar’s test was used for categorical variables and the

paired t-test for continuous variables.

Multivariable Analysis

The following section describes the multivariable analysis performed for

the outcome variables of adherence, HRU, and cost in the primary analysis. No

multivariable analysis was conducted for the subgroup analysis.

Logistic Regression for HDPS Analysis and Matching

A logistic regression model was employed to generate the propensity

score with membership in the exposure cohort or the control cohort as the

dependent variable. The independent variables included the empirically

identified covariates based on HDPS analysis and other independent variables

specified in the variable section. In HDPS analysis, the diagnosis code and the

procedure codes in outpatient, ER, and inpatient settings were specified as data

dimensions (i.e., six dimensions) and the 200 most prevalent codes in each data

dimension were used in the analysis after code recurrence assessment. Within

each data dimension, the possible amount of confounding was calculated for

each variable based on a multiplicative model and all variables were sorted in

descending order (details specified in (Schneeweiss, Rassen, et al., 2009).

Next, the top 300 variables were selected, as research shows that the

HDPS algorithm might have reached or been very close to its full potential to

adjust for confounding effect with approximately 300 empirically selected

covariates (J. A. Rassen, Glynn, Brookhart, & Schneeweiss, 2011). A SAS

65

Macro for HDPS algorithm developed by Rassen and colleagues was adapted for

this analysis (J. Rassen, Doherty, Huang, & Schneeweiss, 2013).

After the propensity score was generated, matching was conducted at 1:1

between the exposure cohort and the control cohort using the Greedy 5�1 digit

technique (Parsons, 2009). With this technique, propensity scores were arranged

in descending order and then observations were attempted to be matched on the

first five digits of the score. If all cases were not matched, then a four digit

match was attempted. This process was repeated until matches were attempted

on the first digit of the propensity score. This process maximized the number of

matched pairs while minimizing errors. Observations that could not be matched

using this technique were excluded.

Generalized linear models (GLMs)

Because statistical significance still existed between cohorts for some

covariates after the PSM, multivariable regression analysis was performed to

adjust for potential residual confounding effects when comparing the matched

exposure cohort and the matched control cohort.

Adherence. A conditional logistic regression model was constructed with

adherence (1= PDC ≥80%), 0 = PDC<80%) as the dependent variable. The

independent variables included unbalanced covariates after the PSM and the

variable indicating membership in the exposure or the control cohort.

HRU. Three Generalized Linear Models (GLMs) with negative binomial

distribution and log link function were employed with the number of outpatient,

ER, or inpatient visits as the dependent variables. Unlike Poisson models,

66

negative binomial regression models do not require the assumption of equality

between the conditional mean and variance, and can particularly correct for

overdispersion, i.e., the variance is greater than the conditional mean (Hilbe,

2011; Osgood, 2000; Paternoster & Brame, 1997). The independent variables

included unbalanced covariates after the PSM and the variable indicating

membership in the exposure or the control cohort.

Cost. A GLM regression with gamma distribution and log link function

was employed with all-cause medical cost as the dependent variables. The

independent variables included unbalanced covariates after the PSM and the

variable indicating membership in the exposure or the control cohort.

Generalized Estimating Equation (GEE)

In this study, over 50% of patients had observations for two or more

years from 2007 to 2010. When patients have multiple observations over time,

there may be correlation between repeated measures. GEE technique was applied

in the multivariable models to correct for the correlation between repeated

observations of a patient. (Hardin & Hilbe, 2003; Liang & Zegger, 1986). GEE

is an extension of the quasi-likelihood approach used to analyze longitudinal and

other correlated data (Burton, Gurrin, & Sly, 1998; Diggle, Liang, & Zeger,

1994; Wedderburn, 1974).

All analyses were performed using SAS® 9.2 (SAS Institute Inc., Cary,

NC, USA). P-values less than 0.05 were considered to be statistically significant.

67

CHAPTER FOUR

RESULTS

This chapter presents the results from the analyses conducted to examine

the study hypotheses. The results are reported in four parts: (1) sample size of

study cohorts and subgroups after application of inclusion and exclusion criteria;

(2) demographic and baseline characteristics of the study sample before and after

PSM; (3) descriptive statistics of the outcome variables; and (4) the findings for

multivariable regression analysis and hypothesis testing associated with the

outcome variables.

As described in Chapter 3, two sets of analyses were conducted – a

primary analysis and a subgroup analysis. Considering that beneficiaries in the

exposure cohort could enter into the Part D coverage gap at different time points

during a calendar year, subgroups were created to explore if the timing of

reaching the Part D gap led to different outcomes. In the primary analysis, the

analyses were conducted at the cohort level for the exposure cohort and the

control cohort. In the subgroup analysis, the analyses were conducted at the

subgroup level defined within the exposure cohort, including no-reaching gap,

early-gap, mid-gap, and late-gap subgroups depending on whether and when

beneficiaries reached the Part D coverage gap during a calendar year. The

primary analysis included both descriptive and multivariable analyses; while the

subgroup analysis only included a descriptive analysis as the comparison

68

between the pre-gap and in-gap periods within the exposure cohort is not a main

objective of this research.

Sample Size

Application of the patient selection criteria resulted in 5,366; 5,650;

5,991; and 6,268 unique beneficiaries diagnosed with COPD and treated with

LABD for the years 2007-2010, respectively (Figure 4). Each year nearly 20%

of those beneficiaries enrolled with Part D benefit were not subject to the

coverage gap (i.e., assigned into the control cohort) and the remaining

beneficiaries were at risk of the coverage gap (i.e., assigned into the overall

exposure cohort).

69

Figure 4. Patient selection flow chart.

The overall exposure cohort was further categorized into different

subgroups based on whether and when they reached the Part D coverage gap

(Table 7). Between 33% and 44% of the exposure cohort (33% in 2007 and 44%

in 2010) did not reach the coverage gap and were not included in the final

70

exposure cohort to implement matching. Among those who reached the coverage

gap, only 1.5% or fewer (n=118 for four years) reached the coverage gap earlier

than March 1 in the year (i.e., early-gap subgroup) and most of them (88%-95%)

entered the catastrophic phase. The early-gap subgroup was not included in the

final exposure cohort to implement matching as well considering that this group

of patients may have very different medication utilization patterns because they

were motivated to maximize their medication usage to enter the catastrophic

phase sooner after they reach the coverage gap. Over 80% of the beneficiaries

who reached the coverage gap reached the gap between March 1 and October 31

in the year (i.e., mid-gap subgroup), and over 20% of them entered the

catastrophic phase before the year end. About 17%-19% of the beneficiaries who

reached the Part D gap reached the gap on or after November 1 in the year (i.e.,

late-gap subgroup), and only one patient entered the catastrophic phase before

the year end.

As reported in Table 7, from 2007 to 2010, the final control cohort

contained 1,011; 1,012; 1,145; and 1,176 beneficiaries and the final exposure

cohort (the mid-gap + the late-gap subgroups) contained 2,786; 2,746; 2,721;

and 2,751 beneficiaries. For the final control cohort, over 7% of the patients had

observations across four years, about 16% across three years, and 33% across

two years; for the final exposure cohort, only about 1% of the patients had

observations across four years, 6% across three years, and 21% across two years.

Combined across all years, there were 4,344 patient-year observations in the

control cohort and 11,004 patient-year observations in the exposure cohort

71

before implementation of propensity score matching (PSM). After the 1:1

matching, both matched control and exposure cohorts included 4,147 patient-

year observations, which was the final study sample for hypothesis testing.

72

Table 7

Sample Size of Study Cohorts and Subgroups

Before matching

Year

2007

Year

2008

Year

2009

Year

2010 Total

Enrolled with Part D: Not exposed to

Part D coverage gap 1011 1012 1145 1176 4344

Enrolled with Part D: Exposed to

Part D coverage gap 4191 4487 4675 4920 18273

No-reaching gap subgroup 1372 1700 1930 2149 7151 Reached the gap 2819 2787 2745 2771 11122

1) Early-gap subgroup 33 41 24 20 118 Entered the catastrophic phase 29 39 22 19 109

2) Mid-gap subgroup 2305 2271 2203 2231 9010 Entered the catastrophic

phase 511 504 483 469 1967 3) Late-gap subgroup 481 475 518 520 1994

Entered the catastrophic phase 1 0 0 0 1 Final exposure cohort (mid-gap+late-gap) before matching 2786 2746 2721 2751 11004 Final control cohort before matching 1011 1012 1145 1176 4344

After matching

Year

2007

Year

2008

Year

2009

Year

2010 Total

Matched exposure cohort 987 970 1087 1103 4147

1) Mid-gap subgroup 821 804 896 912 3433 Entered the catastrophic phase 216 227 228 248 919

2) Late-gap subgroup 166 166 191 191 714 Entered the catastrophic phase 0 0 0 0 0

Matched control cohort 987 970 1087 1103 4147

Note: Early-gap subgroup: entering the coverage gap before March 1, mid-gap subgroup:entering the coverage gap between March 1 and October 31, late-gap subgroup: entering the coverage gap on and after November 1.

Demographic and Baseline Characteristics

The mean age of the patients was 77.4 years (SD=7.6), the majority were

female, and over 90% were Caucasians. The study sample was heavily

concentrated in the South, followed by the Midwest and Northeast. Beneficiaries

in the West were underrepresented in the study. The mean CCI score was about

73

2 and among the selected comorbidities, the most prevalent one in the baseline

period was hypertensive disease (over 65%), followed by heart disease (over

50%) and hyperlipidemia (over 45%). Over 80% of the study sample was

prevalent COPD patients, over 70% of the patients had used a LABD and a large

proportion of them received oxygen therapy or oral corticosteroids (about 30%)

in the baseline period. Beneficiaries had substantial medication burden in the

baseline period, with an average of over 10 different classes of medications, and

had 0.4 ER visits and 0.3 inpatient visits in the baseline six month period.

Detailed results are reported in Table 8.

Primary Analysis: Control vs. Exposure Cohort Before and After Matching

As shown in Table 8, before matching, the control cohort and the

exposure cohort were significantly different in almost all of the demographic and

baseline characteristics except for several baseline comorbidities. The control

cohort was a little older, included more female beneficiaries and fewer

Caucasians, contained more residents in the Midwest and fewer in the West, and

had a higher CCI score. Among selected baseline comorbidities, the control

cohort had higher prevalence than the exposure cohort for all diseases except for

hyperlipidemia and sleep disorder (lower prevalence), and asthma and

osteoporosis (similar prevalence).

A higher percentage of patients in the exposure cohort had a diagnosis of

COPD in the baseline period, used LABD or oxygen therapy but had fewer

classes of medication in the baseline period than the control cohort. In addition,

the exposure cohort used fewer ER and inpatient services in the baseline period.

Details are presented in Table 8.

74

After implementing PSM, the control and the exposure cohorts were

generally balanced in demographic and baseline characteristics. Statistical

differences were only observed in the prevalence of several baseline

comorbidities including hyperlipidemia, anemia, depression, anxiety, GERD,

sleep disorder, disease of the musculoskeletal system and connective tissue, and

hypertensive disease. However, compared to results before matching, the

differences became smaller as indicated in Table 8. In addition, post-match

standardized difference was also calculated to assess whether the distribution of

the variables describing patients’ demographic and baseline characteristics was

similar between the matched cohorts, i.e., to diagnose the balance of matching

(Austin, 2009). The standardized difference was generally small with most of the

absolute values less than 10% except several baseline comorbidity variables,

indicative of acceptable balance between the matched cohorts (Normand et al.,

2001). Detailed results are presented in Table 8.

Table 8

Patient Demographic and Baseline Characteristics of Study Cohorts Before and After Matching

Before matching After matching

Control

(n=4344) Exposure (n=11004)

P-value Control

(n=4147) Exposure (n=4147)

P-value Std Diff ƚ

in %

Age (mean, SD) 77.41 (7.64) 76.59 (7.22) <.0001 77.38 (7.64) 77.31 (7.46) .9081 -0.93 Female (n, %) 3114 (71.69) 7187 (65.31) <.0001 2963 (71.45) 2933 (70.73) .4675 -1.60 Caucasian (n, %) 4008 (92.27) 10744 (97.64) <.0001 3925 (94.65) 3912 (94.33) .5316 -1.37 Region (n, %) <.0001 .1259

NorthEast 1121 (25.81) 2485 (22.58) 1064 (25.66) 1025 (24.72) -2.17 MidWest 1363 (31.38) 2583 (23.47) 1291 (31.13) 1311 (31.61) 1.04 West 226 (5.2) 1473 (13.39) 226 (5.45) 284 (6.85) 5.82 South 1632 (37.57) 4457 (40.5) 1564 (37.71) 1525 (36.77) -1.95

Deyo-Charlson comorbidity index in the baseline period* (mean, SD)

2.22 (1.69) 1.87 (1.44) <.0001 2.16 (1.63) 2.17 (1.65) .975 0.61

Other comorbidities in the baseline period (n, %)

Asthma 429 (9.88) 1079 (9.81) .8953 405 (9.77) 457 (11.02) .0613 4.11 Hyperlipidemia 1970 (45.35) 5223 (47.46) <.0001 1881 (45.36) 2294 (55.32) <.0001 20.02 Heart disease 2480 (57.09) 5633 (51.19) <.0001 2331 (56.21) 2353 (56.74) .6261 1.07 Deficiency anemia 1035 (23.83) 1793 (16.29) <.0001 952 (22.96) 828 (19.97) .0009 -7.29 Depression 768 (17.68) 1119 (10.17) <.0001 717 (17.29) 526 (12.68) < .0001 -12.93 Anxiety 462 (10.64) 770 (7) <.0001 443 (10.68) 353 (8.51) .0027 -7.37 Osteoporosis 584 (13.44) 1533 (13.93) .4302 552 (13.31) 608 (14.66) .0763 3.89 Osteoarthritis 998 (22.97) 2076 (18.87) <.0001 928 (22.38) 918 (22.14) .7918 -0.58 GERD 869 (20) 1554 (14.12) <.0001 819 (19.75) 683 (16.47) .0001 -8.52 Sleep disorder 236 (5.43) 881 (8.01) <.0001 217 (5.23) 367 (8.85) < .0001 14.17 Diseases of the musculoskeletal

system and connective tissue 1739 (40.03) 3570 (32.44) <.0001 1624 (39.16) 1532 (36.94) .0375 -4.57

75

Hypertensive disease 2998 (69.01) 7170 (65.16) <.0001 2828 (68.19) 2921 (70.44) .0268 4.86 Obesity 182 (4.19) 378 (3.44) .0247 159 (3.83) 177 (4.27) .3161 2.20

Prevalent COPD diagnosis in the baseline period (n, %)

3625 (83.45) 9370 (85.15) .0084 3461 (83.46) 3491 (84.18) .3711 1.96

Prescribed with LABDs in the baseline period (n, %)

3204 (73.76) 8339 (75.78) .0089 3073 (74.1) 3023 (72.9) .2135 -2.73

Prescribed with oral corticosteroid in the baseline period (n, %)

1237 (28.48) 3037 (27.6) .2749 1175 (28.33) 1173 (28.29) .9611 -0.11

Order of oxygen therapy in the baseline period (n, %)

1273 (29.30) 3811 (34.63) <.0001 1245 (30.02) 1252 (30.19) .8669 0.37

Number of unique prescription drugs in the baseline period (mean, SD)

11.95 (7.25) 10.02 (5.63) <.0001 11.59 (6.85) 11.69 (6.35) .0933 1.51

Number of all-cause ER visits in the baseline period (mean, SD)

0.43 (0.87) 0.33 (0.78) <.0001 0.42 (0.84) 0.44 (0.98) .5431 2.19

Number of all-cause inpatient visits in the baseline period (mean, SD)

0.33 (0.72) 0.27 (0.66) <.0001 0.32 (0.71) 0.34 (0.76) .9636 2.72

Note: ƚ Std Diff=Standardized Difference.

*The baseline period was defined as six months prior to the start of a calendar year.

76

77

Subgroup Analysis: Subgroups in Exposure Cohort Before Matching

Among the 18,273 patient-year observations from 2007 to 2010 in the

overall exposure cohort (before cohort matching), there were 7,151 observations

in the no-reaching gap subgroup, 118 observations in the early-gap subgroup,

9,010 observations in the mid-gap group, and 1,994 observations in the late-gap

group. Although the early-gap and the no-reaching gap subgroups were not

included in the final exposure cohort to implement matching, they were included

in the descriptive analysis.

As described in Table 9, among the four subgroups of the overall

exposure cohort (before cohort matching), the early-gap subgroup was, on

average, the youngest (mean age=74.4 years) and the mid-gap subgroup was, on

average, the oldest (mean age=76.7 years). The no-reaching subgroup had the

lowest percentage of female beneficiaries (about 60%). Across all subgroups,

over 60% of the beneficiaries were females and a majority of the subgroups

resided in the South. The early-gap subgroup appeared to have the worst

comorbidity burden with a mean CCI score of 2.8 (SD=2.1) while the no-

reaching gap subgroup had lowest mean CCI score of 1.5 (SD=1.3). Similarly,

the early-gap subgroup had the highest prevalence of baseline comorbidities

compared to other subgroups, while the no-reaching gap subgroup had the

lowest prevalence. Similar patterns were observed for medication use and ER or

inpatient services in the baseline period.

When comparing the mid-gap and the late-gap subgroups in the overall

exposure cohort (before cohort matching), the mid-gap subgroup was older, had

a higher mean CCI score (1.93 vs. 1.58, P< .0001), had a higher prevalence of

78

comorbidities except anxiety and obesity, and was more likely to use LABD

(77.48% vs. 68.10%, P< .0001), oral corticosteroids (28.52% vs. 23.42%,

P< .0001), or oxygen therapy (35.66% vs. 29.99%, P< .0001) in the baseline

period (Table 9). The mid-gap subgroup also used more classes of medications

(10.54 vs. 7.64, P< .0001) and ER/inpatient services (0.34 vs. 0.28, P=.0004 for

ER; 0.28 vs. 0.21, P< .0001 for inpatient visits) in the baseline period than the

late-gap subgroup.

Table 9

Demographic and Baseline Characteristics of Subgroups of the Exposure Cohort Before Matching

Early-gap

subgroup (n=118) No reaching gap

subgroup (n=7151) Mid-gap subgroup

(n=9010) Late-gap subgroup

(n=1994)

P-value (mid- vs. late-gap)

Age (mean, SD) 74.42 (6.31) 75.86 (6.88) 76.75 (7.25) 75.86 (7.03) < .0001

Female (n, %) 77 (65.25) 4327 (60.51) 5912 (65.62) 1275 (63.94) .1553

Caucasian (n, %) 115 (97.46) 6892 (96.38) 8796 (97.62) 1948 (97.69) .8560

Region (n, %)

.0019

NorthEast 28 (23.73) 1263 (17.66) 2102 (23.33) 383 (19.21)

MidWest 23 (19.49) 1820 (25.45) 2060 (22.86) 523 (26.23)

West 19 (16.1) 1094 (15.3) 1198 (13.3) 275 (13.79)

South 48 (40.68) 2971 (41.55) 3645 (40.46) 812 (40.72)

Deyo-Charlson comorbidity index in the baseline period* (mean, SD)

2.75 (2.08) 1.48 (1.25) 1.93 (1.46) 1.58 (1.28) < .0001

Other comorbidities in the baseline period (n, %)

Asthma 12 (10.17) 552 (7.72) 895 (9.93) 184 (9.23) .3376 Hyperlipidemia 65 (55.08) 3157 (44.15) 4803 (53.31) 968 (48.55) .0001 Heart disease 77 (65.25) 2853 (39.9) 4786 (53.12) 847 (42.48) < .0001 Deficiency anemia 30 (25.42) 866 (12.11) 1543 (17.13) 250 (12.54) < .0001 Depression 32 (27.12) 449 (6.28) 952 (10.57) 167 (8.38) .0034 Anxiety 10 (8.47) 368 (5.15) 626 (6.95) 144 (7.22) .6645 Osteoporosis 19 (16.1) 797 (11.15) 1324 (14.69) 209 (10.48) < .0001 Osteoarthritis 32 (27.12) 1110 (15.52) 1755 (19.48) 321 (16.1) .0005 GERD 27 (22.88) 813 (11.37) 1325 (14.71) 229 (11.48) .0002

79

Sleep disorder 21 (17.8) 451 (6.31) 747 (8.29) 134 (6.72) .0194 Diseases of the musculoskeletal

system and connective tissue 55 (46.61) 1923 (26.89) 3018 (33.5) 552 (27.68) < .0001

Hypertensive disease 80 (67.8) 4048 (56.61) 6005 (66.65) 1165 (58.43) < .0001 Obesity 7 (5.93) 186 (2.6) 321 (3.56) 57 (2.86) .1183

Prevalent COPD diagnosis in the baseline period (n, %)

111 (94.07) 5493 (76.81) 7738 (85.88) 1632 (81.85) < .0001

Prescribed with LABDs in the baseline period (n, %)

94 (79.66) 4490 (62.79) 6981 (77.48) 1358 (68.10) < .0001

Prescribed with oral corticosteroid in the baseline period (n, %)

41 (34.75) 1478 (20.67) 2570 (28.52) 467 (23.42) < .0001

Order of oxygen therapy in the baseline period (n, %)

53 (44.92) 1950 (27.27) 3213 (35.66) 598 (29.99) < .0001

Number of unique prescription drugs in the baseline period (mean, SD)

17.97 (8.03) 6.75 (4.66) 10.54 (5.68) 7.64 (4.71) < .0001

Number of all-cause ER visits in the baseline period (mean, SD)

0.43 (0.83) 0.18 (0.52) 0.34 (0.83) 0.28 (0.77) .0004

Number of all-cause inpatient visits in the baseline period (mean, SD)

0.4 (0.85) 0.18 (0.52) 0.28 (0.68) 0.21 (0.56) < .0001

Note: *The baseline period was defined as six months prior to the start of a calendar year. SD=standard deviation.

80

81

Descriptive Statistics of Outcome Variables

Primary Analysis: Matched Control and Exposure Cohorts

LABD Adherence

After cohort matching, the mean annual PDC in the matched control

cohort was 0.70 (SD=0.25), while the mean annual PDC in the matched

exposure cohort was 0.69 (SD=0.24; Table 10). Given that adherence is defined

as a PDC equal to or greater than 0.8, 45.7% of observations in the matched

control cohort showed as adherent, while 42% of the observations of the

matched exposure cohort reflected adherence.

Table 10

Adherence to LABDs in the Matched Control and Exposure Cohorts

Matched control cohort

(n=4147) Matched exposure cohort

(n=4147)

PDC (mean, SD) 0.70 (0.25) 0.69 (0.24) Adherent (n, %) 1894 (45.70%) 1742 (42.01%)

Note: PDC≥0.80 was defined as being adherent. SD=standard deviation.

HRU

As reported in Table 11, the mean number of annual outpatient visits was

higher in the matched control cohort than that in the matched exposure cohort

(27.19 vs. 26.31). On average, the matched control cohort was prescribed with

more days of SABD than the matched exposure cohort (89.27 vs. 71.33 days).

On the other hand, the matched control and exposure cohorts had similar

numbers of ER and inpatient visits.

82

Table 11

Annual HRU in the Matched Control and Exposure Cohorts

Matched control cohort

(n=4147)

Matched exposure cohort

(n=4147)

Number of all-cause outpatient visits (mean, SD) 27.19 (21.98) 26.31 (18.39) Number of all-cause ER visits (mean, SD) 0.85 (1.39) 0.83 (1.47) Number of all-cause inpatient visits (mean, SD) 0.66 (1.18) 0.66 (1.19) Number of days supplied for SABD prescriptions (mean, SD)

89.27 (131.91) 71.33 (110.56)

Note: SD=standard deviation

All-Cause Medical Cost

Table 12 summarizes the mean annual all-cause medical cost for the

matched control and exposure cohorts. The matched control and the matched

exposure cohorts had a comparable mean all-cause medical cost. Specifically, on

average, Medicare paid about $13,871 and $13,396 per patient per year for the

medical services for the control and the exposure cohort beneficiaries,

respectively. Considering the high degree of skewness of the cost, the log-

transformed cost was also calculated for two cohorts. The average all-cause

medical cost was similar between the matched control and exposure cohorts,

8.61 and 8.66, respectively.

83

Table 12

Annual All-Cause Medical Cost for the Matched Control and Exposure Cohorts

Matched control cohort

(n=4147) Matched exposure cohort

(n=4147)

Mean (SD) Median (q1, q3) Mean (SD) Median (q1, q3)

All-cause medical cost

13871.07 (20110.86)

5175.48 (2044.48, 17189.43)

13396.26 (19861.42)

5573.61 (2304.81, 15811.42)

Log (all-cause medical cost) 8.61 (1.49) 8.55 (7.62, 9.75) 8.66 (1.37) 8.63 (7.49, 9.67) Note: SD=standard deviation. q1=25%tile, q3=75%tile

Subgroup Analysis: Pre-Gap vs. In-Gap period for Mid-gap and Late-Gap

Subgroups in the Matched Exposure Cohort

LABD Adherence

The pre-gap treatment period was calculated as the duration from the fill

date of the first LABD prescription to the gap date; the in-gap treatment period

was calculated as the duration from the fill date of the first LABD prescription

after reaching the gap until the end of gap or calendar year or death, whichever

occurred first. In the matched exposure cohort, the mean pre-gap treatment

period was 153 and 253 days within a calendar year for the mid- and late-gap

subgroups, respectively. The mean in-gap treatment period was 134 and 32 days

within a calendar year for the mid-gap and the late-gap subgroups, respectively.

In the mid-gap subgroup, pre-gap PDC and in-gap PDC were 0.50 and 0.46,

respectively. Likewise, a higher adherence rate was observed in the pre-gap

period (about 20%) than in the in-gap period (about 16%). In the late-gap

subgroup, PDC and adherence rate during the in-gap period were both

84

significantly higher than the rates during the pre-gap period. However, this result

needs to be interpreted with caution as the exposure time of the late-gap

subgroup for receiving LABDs was much shorter than that of the mid-gap

groups due to the definition of the subgroups. That is, as patients who entered

the coverage gap on and after November 1 of the year were assigned to the late-

gap subgroup, the late-gap subgroup could only have a maximum of two months

of exposure time by the end of a calendar year. Table 13 presents detailed results.

Table 13

Adherence to LABDs in the Mid-Gap and Late-Gap Subgroups in the Matched

Exposure Cohort

Pre-gap period In-gap period

Mid-gap subgroup (n=3433)

Treatment duration in days (mean, SD) 152.60 (68.37) 134.30 (56.88) PDC (mean, SD) 0.49 (0.29) 0.46 (0.49) Adherent (n, %) 672 (19.57%) 555 (16.17%)

Late-gap subgroup (n=714)

Treatment duration in days (mean, SD) 252.68 (85.86) 31.62 (17.79) PDC (mean, SD) 0.29 (0.22) 0.88 (0.18) Adherent* (n, %) 32 (4.48%) 310 (43.42%)

Note: * Adherent was defined if PDC≥0.80. SD=standard deviation.

Additional analysis was conducted to assess adherence in different

quarters of the pre-gap period for the late-gap subgroup. It appeared that patients

were much less adherent earlier during the year, and became more adherent

when they got closer to the date when they hit the coverage gap. This trend was

quite consistent across the year from 2007 to 2010 (Table 13a).

85

Table 13a

Quarterly Pre-Gap Adherence for the Late-Gap Subgroup

Quarter 1 prior to the coverage gap

start date

Quarter 2 prior to the coverage gap start

date

Quarter 3 prior to the coverage gap start

date

2007

% being adherent* 14.03% 5.37% 0.00% 2008

% being adherent 11.26% 2.51% 0.00% 2009 % being adherent 13.88% 7.51% 0.00% 2010 % being adherent 12.20% 5.66% 0.00%

Note: * PDC≥0.80 is defined as being adherent.

HRU

In the matched exposure cohort, the mean duration of the period from

January 1 until reaching the gap (pre-gap period) was 189 and 332 days within a

calendar year for the mid-gap and late-gap subgroups, respectively. The mean

duration of the in-gap period was 152 and 33 days within a calendar year for the

mid-gap or the late-gap subgroups, respectively. In the mid-gap subgroup,

patients had similar monthly HRU in all settings and similar length of SABD

therapy for the pre-gap and the in-gap periods. Higher average monthly SABD

days in the in-gap period as compared to the pre-gap period was observed,

although this result should be interpreted with caution for the reasons described

earlier – by definition, the duration of the in-gap period for the late-gap subgroup

was much shorter compared to the pre-gap period, so the length of exposure to

SABD therapy was shorter for the in-gap period than the pre-gap period

(denominator). In addition, almost 50% of SABDs prescriptions filled by

86

beneficiaries had 25 or 30 days of supply, and about 25% ranged from 15 to 20

days or 90 days –15 days (6%), 20 days (5%) or 90 days (4%). Because of these

differences, beneficiaries in the in-gap period and pre-gap period could

potentially accumulate a similar number of days of supply (numerator) and the

in-gap period could end up with having a much higher number of monthly

SABD days than the pre-gap period. In terms of resource use, the late-gap

subgroup had higher monthly number of outpatient, ER, and inpatient visits in

the pre-gap period than in the in-gap period. In addition, rates of HRU of the

late-gap subgroup were lower than rates of the mid-gap subgroup for both pre-

gap and in-gap periods. Table 14 presents detailed results.

Table 14

Monthly HRU of the Mid-Gap and the Late-Gap Subgroups in the Matched

Exposure Cohort

Pre-gap period

(mean, SD) In-gap period (mean, SD)

Mid-gap subgroup (n=3433)

Length of period in days 189.19 (64) 151.86 (50.18) Monthly number of all-cause outpatient visits 2.26 (1.75) 2.20 (1.79) Monthly number of all-cause ER visits 0.17 (0.13) 0.16 (0.22) Monthly number of all-cause inpatient visits 0.16 (0.17) 0.14 (0.11) Monthly number of days supplied for SABD prescriptions 13.43 (11.27) 14.29 (11.35)

Late-gap subgroup (n=714)

Length of period in days 332.5 (18.15) 31.62 (17.79) Monthly number of all-cause outpatient visits 1.90 (1.42) 1.56 (2.35) Monthly number of all-cause ER visits 0.16 (0.13) 0.10 (0.35) Monthly number of all-cause inpatient visits 0.14 (0.11) 0.09 (0.29) Monthly number of days supplied for SABD prescriptions 9.51 (9.24) 50.53 (101.03)

All-Cause Medical Cost

In the matched exposure cohort, the mid-gap subgroup had higher

monthly expenses per patient for medical services in the pre-gap period

87

compared to the in-gap period ($1,135 vs. $1,085, P=0.0236; respectively).

Similarly, the late-gap subgroup had higher monthly expenses per patient for

medical services in the pre-gap period compared to the in-gap period ($987.71 vs.

$338.70, respectively). In addition, the average medical cost of the late-gap

subgroup was lower than that of the mid-gap subgroup for both pre-gap and in-

gap periods. Table 15 presents detailed results.

Table 15

Monthly All-Cause Medical Cost for the Mid-Gap and the Late-Gap Subgroups

in the Matched Exposure Cohort

Pre-gap period In-gap period

Mean (SD) Median (q1, q3) Mean (SD) Median (q1, q3)

Mid-gap subgroup (n=3433)

Monthly all-cause medical cost

1135.23 (2089.17)

345.76 (141.90, 1106.62)

1085.87 (2185.31)

313.58 (146.23, 880.25)

Late-gap subgroup (n=714)

Monthly all-cause medical cost

987.71 (1673.54)

338.70 (135.27, 1094.99)

575.25 (2057.77)

112.44 (0.00, 331.90)

Note: SD=standard deviation.

88

Regression Analysis and Hypotheses Testing

Below are the three hypotheses which were presented in Chapter 2:

4) Hypothesis 1: Among Medicare beneficiaries with COPD, the

coverage gap will be associated with lower medication

adherence to COPD long-term maintenance therapies.

5) Hypothesis 2: Among Medicare beneficiaries with COPD, the

coverage gap will be associated with higher consumption of

healthcare resources (non-drug).

6) Hypothesis 3: Among Medicare beneficiaries with COPD, the

coverage gap will be associated with higher medical cost

(from payer’s perspective).

Multiple regression models were constructed to test these hypotheses.

The outcome variables for the models included LABD adherence, annual

number of all-cause outpatient visits, annual number of all-cause ER visits,

annual number of all-cause inpatient visits, and annual all-cause medical cost. In

all models, age, gender, cohort membership (exposure vs. control), and the

remaining unbalanced variables after implementing PSM were included as

independent variables.

Regression Analyses for Hypothesis 1 (LABD Adherence)

The first hypothesis pertains to the association between reaching the

coverage gap and LABD adherence in patients with COPD. A conditional

logistic regression model was constructed to test this hypothesis, including a

binary variable with 1 indicating being adherent and 0 otherwise as the outcome

variable. The regression results are summarized in Table 16.

89

Unadjusted results showed the matched exposure cohort had lower

adherence rate than the matched control cohort. After adjustment for sample

selection bias (via PSM) and controlling for age, gender and the unbalanced

covariates after the PSM in the regression model, beneficiaries who reached the

coverage gap were associated with lower odds of being adherent compared to

beneficiaries who did not reach the coverage gap. Specifically, beneficiaries in

the mid-gap subgroup (i.e., reached the coverage gap earlier than October) had

about 7% lower odds but without statistical significance (Odds ratio [OR]=0.931,

95% confidence interval [CI]: 0.846, 1.024); while beneficiaries in the late-gap

subgroup had nearly 40% lower odds (OR=0.603, 95% CI: 0.493, 0.738) to be

adherent. In addition, hyperlipidemia, depression, and disease of the

musculoskeletal system and connective tissues were found to be associated with

lower likelihood of being adherent (Table 16). The adjusted adherences were

estimated from the regression model and reported in Table 16a, and the results

were slightly higher than unadjusted estimations.

90

Table 16

Conditional Logistic Regression on Adherence to LABDs

Variable OR 95% CI P-value

Mid-gap exposure subgroup vs. control cohort 0.931 0.846 1.024 .1415

Late-gap exposure subgroup vs. control cohort 0.603 0.493 0.738 < .0001 Age in years 1.005 0.998 1.013 .1809 Female (vs. Male) 1.104 0.967 1.261 .1425 Hyperlipidemia 0.869 0.774 0.975 .0172 Deficiency anemia 1.034 0.899 1.189 .6374 Depression 0.848 0.721 0.997 .0463 Anxiety 0.974 0.798 1.188 .7927 GERD 1.057 0.913 1.224 .4568 Sleep disorder 0.981 0.793 1.215 .8622 Diseases of the musculoskeletal system and connective tissue

0.809 0.719 0.911 .0004

Hypertensive disease 0.970 0.852 1.105 .6502

Note: OR= odds ratio, CI=confidence interval.

Table 16a

Unadjusted and Adjusted Adherence

Unadjusted Outcomes Adjusted Outcomes Matched control

Matched exposure

Matched control

95% CI Matched exposure

95% CI

Adherence* 45.70% 42.01% 48.6% 48.1%, 49.2% 43.4%

42.8%, 43.9%

Note: * Adherence is defined if PDC≥0.80. CI=confidence interval.

91

Regression Analyses for Hypothesis 2 (All-Cause HRU)

The second hypothesis was related to the association between reaching

the Part D gap and all-cause HRU for patients with COPD. Three separate GLM

models with a negative binomial distribution and log link function were

constructed for the three HRU outcomes: all-cause outpatient visits, all-cause ER

visits, and all-cause inpatient visits. Because the study sample covered multiple

years and more than half of the patients had observations in multiple calendar

years, GEE technique was applied in the models to correct for the potential

correlation between repeated measures within one patient.

Annual Number of All-Cause Outpatient Visits

Unadjusted results showed the matched exposure cohort had lower mean

number of outpatient visits than the matched control cohort. After adjustment for

sample selection bias (via PSM) and controlling for age, gender, and the

unbalanced covariates after the PSM in the regression model, cohort membership

was shown as a significant predictor of the number of outpatient visits; further,

the timing of reaching the coverage gap had a different effect on the number of

outpatient visits. Specifically, beneficiaries reaching the gap earlier were

associated with almost 4% higher number of outpatient visits than the control

cohort (relative ratio [RR]=3.78%, β=0.0371, standard error [SE]=0.016,

P= .0201);while beneficiaries reaching the gap later in the year were associated

with over 5% lower number of outpatient visits than the control cohort with

marginal statistical significance (RR= -5.46%, β= -0.056, SE=0.029, P= .0545).

All other independent variables were significantly associated with the outcome

variable. Specifically, one additional year in age was associated with about 1.5%

92

higher number of outpatient visits (relative ratio [RR]=1.48%, β=0.015, P< .001),

while female patients had over 5% fewer outpatient visits than male patients

(RR= -5.64%, β= -0.058, P= .0011). All of the unbalanced baseline

comorbidities were associated with a higher number of outpatient visits, with the

highest difference in diseases of the musculoskeletal system and connective

tissue (RR=34.04%, β=0.293, P< .0001) and the lowest difference in

hyperlipidemia (RR=6.82%, β=0.066, P< .0001). Detailed regression results are

presented in Table 17. The adjusted annual number of all-cause outpatient visits

were estimated from the regression model and reported in Table 17a, and the

results were similar to unadjusted estimations.

93

Table 17

Negative Binomial Regression on Annual Number of All-Cause Outpatient Visits

Variable Coefficient (β) SE 95% CI

P-value RR

Intercept 1.765 0.082 1.604 1.926 < .0001 -- Exposure: mid-gap subgroup vs. control cohort

0.037 0.016 0.006 0.068 .0201 3.78%

Exposure: late-gap subgroup vs. control cohort

-0.056 0.029 -0.114 0.001 .0550 -5.46%

Age in years 0.015 0.001 0.013 0.017 < .0001 1.48% Female (vs. Male) -0.058 0.018 -0.093 -0.023 .0011 -5.64% Hyperlipidemia 0.066 0.016 0.034 0.098 < .0001 6.82% Deficiency anemia 0.280 0.018 0.244 0.315 < .0001 32.25% Depression 0.266 0.021 0.224 0.307 < .0001 30.41% Anxiety 0.074 0.026 0.022 0.125 .0051 7.66% GERD 0.140 0.019 0.102 0.178 < .0001 14.99% Sleep disorder 0.200 0.028 0.144 0.254 < .0001 22.01% Diseases of the musculoskeletal system and connective tissue

0.293 0.016 0.262 0.324 < .0001 34.04%

Hypertensive disease 0.110 0.018 0.073 0.146 < .0001 11.57%

Note: SE – standard error, CI – confidence interval, RR – relative ratio.

94

Table 17a

Unadjusted and Adjusted Number of All-Cause Outpatient Visits

Unadjusted Outcomes Adjusted Outcomes Matched control

Matched exposure

Matched control

95% CI Matched exposure

95% CI

Number of all-cause outpatient visit per year

27.19 26.31 26.87 26.57, 27.17

26.80 26.52, 27.08

Note: CI – confidence interval.

Annual Number of All-Cause ER Visits

Unadjusted results showed the matched exposure cohort had a similar

mean number of ER visits as the matched control cohort. After adjusting for

sample selection bias (via PSM) and controlling for age, gender, and the

unbalanced covariates after the PSM, the mid-gap subgroup was not a significant

predictor of the number of ER visits (β=0.016, SE=0.039, P= .6728), and the

late-gap subgroup was associated with about 13% lower number of ER visits

(RR= -13.23%, β= -0.142, SE=0.072, P= .0496). One additional year in age was

associated with 1.33% higher number of ER visits (RR=1.33% β=0.013,

P< .0001), while gender was not a significant factor. Several baseline

comorbidities including deficiency anemia, depression, anxiety, GERD, disease

of musculoskeletal system and connective tissue, and hypertensive disease

showed a significant positive relationship with the number of ER visits, with the

highest difference for GERD (RR=32.47%, β=0.281, P< .0001) followed by

diseases of the musculoskeletal system and connective tissue (RR=32.35%,

β=0.280, P< .0001). Hyperlipidemia and sleep disorder were not significantly

associated with the number of ER visits. Detailed results are presented in Table

18. The adjusted annual number of all-cause ER visits was estimated from the

95

regression model and reported in Table 18a, and the results were similar to

unadjusted estimations.

Table 18

Negative Binomial Regression on Annual Number of All-Cause ER Visits

Variable Coefficient (β) SE 95% CI

P-value RR

Intercept -1.548 0.192 -1.925 -1.172 < .0001 -- Exposure: mid-gap subgroup vs. control cohort

0.017 0.039 -0.060 0.093 .6728 1.66%

Exposure: late-gap subgroup vs. control cohort

-0.142 0.072 -0.284 -0.0002 .0496 -13.23%

Age in years 0.013 0.003 0.008 0.018 < .0001 1.33% Female (vs. Male) -0.071 0.044 -0.158 0.015 .1073 -6.86% Hyperlipidemia 0.009 0.038 -0.066 0.084 .8133 0.90% Deficiency anemia 0.272 0.044 0.185 0.359 < .0001 31.21% Depression 0.164 0.050 0.066 0.262 .0010 17.79% Anxiety 0.144 0.061 0.024 0.264 .0189 15.45% GERD 0.281 0.051 0.182 0.380 < .0001 32.47% Sleep disorder 0.016 0.068 -0.118 0.150 .8156 1.61% Diseases of the musculoskeletal system and connective tissue

0.280 0.039 0.204 0.357 < .0001 32.35%

Hypertensive disease 0.136 0.043 0.052 0.219 .0014 14.53%

Note: SE – standard error, CI – confidence interval, RR – relative ratio.

Table 18a

Unadjusted and Adjusted Number of All-Cause ER Visits

Unadjusted Outcomes Adjusted Outcomes Matched control

Matched exposure

Matched control

95% CI

Matched exposure

95% CI

Number of all-cause ER visit per year

0.85 0.83 0.86 0.85, 0.87

0.82 0.81, 0.83

Note: CI – confidence interval.

Annual Number of All-Cause Inpatient Visits

Unadjusted results showed the matched exposure cohort had a similar

mean number of inpatient visits as the matched control cohort. After adjusting

96

for sample selection bias (via PSM) and controlling for age, gender, and the

unbalanced covariates after the PSM, cohort membership and the time of

reaching the coverage gap were not significant predictors of the number of

inpatient visits (β=0.030, SE=0.042, P= .4683 for the mid-gap subgroup; β= -

0.081, SE=0.077, P= .0932 for the late-gap subgroup). One additional year in

age was associated with 1.12% higher number of inpatient visits (RR=1.12%,

β=0.011, P< .0001), and female patients were associated with over 14% fewer

inpatient visits (RR=14.68%, β= -0.159, P= .0004). Baseline comorbidities,

including deficiency anemia, depression, GERD, sleep disorder, disease of

musculoskeletal system and connective tissue, and hypertensive disease, were

significantly and positively associated with the number of inpatient visits, with

highest difference for deficiency anemia (RR=38.31%, β=0.324, P<0.0001),

followed by GERD (RR=29.24%, β=0.257, P< .0001). Hyperlipidemia and

anxiety were not significantly associated with number of inpatient visits.

Detailed regression results are presented in Table 19. The adjusted annual

number of all-cause inpatient visits were estimated from the regression model

and reported in Table 19a, and the results were similar to unadjusted estimations.

97

Table 19

Negative Binomial Regression on Annual Number of All-Cause Inpatient Visits

Variable Coefficient (β) SE 95% CI

P-value RR

Intercept -1.528 0.199 -1.918 -1.139 < .0001 -- Exposure: mid-gap subgroup vs. control cohort

0.030 0.042 -0.051 0.111 .4683 3.06%

Exposure: late-gap subgroup vs. control cohort

-0.081 0.077 -0.231 0.070 .2932 -7.73%

Age in years 0.011 0.003 0.006 0.016 < .0001 1.12% Female (vs. Male) -0.159 0.045 -0.247 -0.071 .0004 -14.68% Hyperlipidemia -0.010 0.041 -0.091 0.071 .8104 -0.99% Deficiency anemia 0.324 0.045 0.236 0.412 < .0001 38.31% Depression 0.130 0.055 0.022 0.239 .0184 13.93% Anxiety 0.056 0.064 -0.070 0.182 .3847 5.75% GERD 0.257 0.048 0.163 0.350 < .0001 29.24% Sleep disorder 0.166 0.082 0.005 0.327 .0436 18.02% Diseases of the musculoskeletal system and connective tissue

0.240 0.041 0.160 0.320 < .0001 27.11%

Hypertensive disease 0.107 0.046 0.017 0.196 .0194 11.26%

Note: SE – standard error, CI – confidence interval, RR – relative ratio.

98

Table 19a

Unadjusted and Adjusted Number of All-Cause Inpatient Visits

Unadjusted Outcomes Adjusted Outcomes Matched control

Matched exposure

Matched control

95% CI

Matched exposure

95% CI

Number of all-cause inpatient visit per year

0.66 0.66 0.67 0.66, 0.67

0.66 0.65, 0.66

Note: CI – confidence interval.

Regression Analyses for Hypothesis 3 (Annual All-Cause Medical Cost)

The third hypothesis was related to the association between reaching the

Part D gap and annual all-cause medical cost in beneficiaries with COPD. A

GLM model with a Gamma distribution and log link function was constructed

for the cost outcome - annual all-cause medical cost. Once again, given that the

study sample included data spanning multiple years and more than half of the

beneficiaries had observations in multiple calendar years, GEE technique was

applied in the models to correct for the potential correlation between repeated

measures within one beneficiary.

Unadjusted results showed the matched exposure cohort had similar

mean annual medical cost as the matched control cohort. After adjusting for

sample selection bias (via PSM) and controlling for age, gender, and the

unbalanced covariates after the PSM in the regression model, cohort membership

and the timing of reaching the coverage gap were not significantly associated

with medical cost (β=0.004, SE=0.034, P= .8974 for the mid-gap subgroup; β= -

0.069, SE=0.068, P= .3049 for the late-gap subgroup). Regarding other

covariates included in the model, each additional year in age was associated with

1.72% higher medical cost (RR=1.72%, β=0.017, P< .0001) and female patients

99

were associated with nearly 13% lower medical cost than male patients (RR= -

12.47%, β= -0.133, P= .0003). All baseline comorbidities except hyperlipidemia

and anxiety were positively and significantly associated with medical cost, with

highest difference for deficiency anemia (RR=43.02%, β=0.358, P< .0001),

followed by disease of musculoskeletal system and connective tissue

(RR=35.59%, β=0.305, P< .0001). Detailed regression results are presented in

Table 20. The adjusted annual all-cause medical cost was estimated from the

regression model and reported in Table 20a, and the results were similar to

unadjusted estimations.

100

Table 20

GLM Regression on Annual All-Cause Medical Cost

Variable Coefficient (β) SE 95% CI

P-value RR

Intercept 7.877 0.167 7.549 8.204 < .0001 -- Mid-gap exposure subgroup vs. control cohort

0.004 0.034 -0.061 0.070 .8974 0.43%

Late-gap exposure subgroup vs. control cohort

-0.069 0.068 -0.202 0.063 .3049 -6.70%

Age in years 0.017 0.002 0.013 0.021 < .0001 1.72% Female (vs. Male) -0.133 0.037 -0.206 -0.060 .0003 -12.47% Hyperlipidemia 0.008 0.033 -0.0570 0.074 .8020 0.83% Deficiency anemia 0.358 0.037 0.286 0.430 < .0001 43.02% Depression 0.221 0.044 0.136 0.306 < .0001 24.73% Anxiety 0.060 0.052 -0.042 0.160 .2504 6.11% GERD 0.217 0.040 0.138 0.295 < .0001 24.20% Sleep disorder 0.269 0.062 0.148 0.391 < .0001 30.92% Diseases of the musculoskeletal system and connective tissue

0.305 0.0326 0.241 0.368 < .0001 35.59%

Hypertensive disease 0.096 0.0380 0.021 0.170 .0117 10.03%

Note: SE – standard error, CI – confidence interval, RR – relative ratio.

101

Table 20a

Unadjusted and Adjusted Annual All-Cause Medical Cost

Unadjusted Outcomes Adjusted Outcomes Matched control

Matched exposure

Matched control

95% CI Matched exposure

95% CI

Annual all-cause medical cost

13871.07 13396.26 13970.91

13798.01, 14143.81

13428.30 13269.54, 13587.06

Note: CI – confidence interval.

Summary of Findings for Regression Analysis and Hypothesis Testing

Table 21 summarizes the hypothesized relationships, as compared to the

observed relationship, between the outcome variables and exposure status (i.e.,

reaching the Part D coverage gap as well as the timing of reaching the gap). The

direction of the relationships is presented as positive (+) or negative (-). Positive

“+” indicates that reaching the Part D gap was associated with greater likelihood

of being adherent, higher HRU, or higher cost; while negative “-” indicates that

reaching the coverage gap was associated with lower likelihood of being

adherent, lower HRU, or lower cost. For the column of “Statistical significance”,

“x” indicates having statistical significance and “√” indicates not having

statistical significance. Generally, hypothesis 1 (LABD adherence) was

supported by the findings of this study, though statistical significance was not

found for the mid-gap subgroup (i.e., reaching the coverage gap earlier than

October). The evidence from this analysis did not support the hypothesized

positive relationship between reaching the coverage gap and HRU or medical

cost, except that a positive association was detected between reaching the

coverage gap and the number of outpatient visits if beneficiaries reached the

coverage gap earlier than November. On the other hand, if beneficiaries reached

102

the coverage gap after October, they were observed to be associated with a lower

number of ER or inpatient visits. These results are discussed in detail in Chapter

5.

Table 21

Summary for Regression Analysis and Hypotheses Testing

Outcome variable

Hypothesized relationship with the Part

D gap vs. Control cohort Finding

s Statistical

significance

Hypothesis 1 LABD adherence Negative (-) Exposure cohort:

Mid-gap subgroup - x

Exposure cohort: Late-gap subgroup

- �

Hypothesis 2 All-cause outpatient visits

Positive (+) Exposure cohort: Mid-gap subgroup

+ �

Exposure cohort: Late-gap subgroup

- x

All-cause ER visits

Positive (+) Exposure cohort: Mid-gap subgroup

+ x

Exposure cohort: Late-gap subgroup

- �

All-cause inpatient visits

Positive (+) Exposure cohort: Mid-gap subgroup

+ x

Exposure cohort: Late-gap subgroup

- x

Hypothesis 3 All-cause medical cost

Positive (+) Exposure cohort: Mid-gap subgroup

+ x

Positive (+) Exposure cohort: Late-gap subgroup

- x

Note: x – without statistical significance, � – with statistical significance

103

CHAPTER FIVE

DISCUSSION

The objective of this study was to assess the impact of the Part D

coverage gap on medication adherence, healthcare resource use, and total

medical cost among Medicare beneficiaries with COPD by addressing the

following research questions:

1) Is medication adherence to LABD lower for Medicare patients

with COPD who reached the Part D coverage gap?

2) Is total healthcare resource use (HRU) in the outpatient, ER and

inpatient setting higher for Medicare patients with COPD who

reached the Part D coverage gap?

3) Is total all-cause medical cost (non-drug) higher for Medicare

patients with COPD who reached the Part D coverage gap?

This chapter provides a summary and discussion of the key findings

related to these questions and presents their implications for future research,

management, and policy. First, the results related to each hypothesis are

summarized and explained. Then, the strengths and limitations of the study are

described. Finally, the implications of the study’s findings for researchers,

healthcare administrators, and policy makers are discussed.

Review of Findings and Comparison with Existing Evidence

Research Question/Hypothesis 1: LABD Adherence

Question 1: Is medication adherence to LABD lower for Medicare

patients with COPD who reached the Part D coverage gap?

104

Prior to multivariable regression analysis, the matched exposure cohort

was shown to have statistically lower yearly PDC for LABD therapies than the

matched control cohort. Compared to the control cohort, 3.7% fewer (or relative

difference of 8%) of the patients in the exposure cohort were adherent. The

regression analysis that controlled for age, gender, cohort membership, and

unbalanced characteristics found that patients in the exposure cohort were

associated with lower odds of being adherent than patients in the control cohort.

Higher magnitude of effect was observed in beneficiaries who reached the

coverage gap earlier. Collectively, these findings provide support for the

hypothesis that reaching the Part D coverage gap negatively affects medication

adherence among Medicare patients with COPD.

One explanation for this finding is offered by rational choice theory,

which suggests that individuals are motivated to make “rational” choices in order

to maximize their benefits. When Medicare beneficiaries enter the coverage gap,

they bear a higher economic burden to obtain their medications. Under these

circumstances, when assessing the value and benefit of their medications, if they

believe that stopping or skipping brand drugs or using cheaper generic drugs will

offer more benefit than paying higher out-of-pocket cost, they are more likely to

choose non-adherence to more expensive brand-name drugs.

This finding is consistent with most of the existing evidence that shows

the Part D coverage gap is associated with reduced medication adherence. For

example, Fung et al. (2010) found that the odds of adherence among diabetic

patients with the Part D coverage gap decreased by 17% compared to those

105

patients without the Part D coverage gap (Fung et al., 2010). Likewise, Stuart

and colleagues (2013) found that the PDC was 7.8% lower for statins, 7.0%

lower for clopidogrel, or 5.9% lower for beta-blocker for beneficiaries exposed

to the coverage gap compared with those not exposed.

Additionally, in the subgroup analysis of the exposure cohort in this

study, the mid-gap subgroup (i.e., entering the coverage gap between March 1

and October 31) had a 6% lower mean PDC after reaching the coverage gap

compared to their adherence prior to reaching the gap. Comparatively, a

reduction of 2.5%-3.6% in MPR was reported in beneficiaries with heart failure

after they entered the coverage gap (Baik et al., 2012; Zhang et al., 2013).

Research Question/Hypothesis 2: All-Cause HRU

Question 2: Is total healthcare resource use (HRU) in the outpatient, ER,

and inpatient setting higher for Medicare patients with COPD who reached the

Part D coverage gap?

The hypothesis that Medicare beneficiaries who were exposed to the

coverage gap will experience higher consumption of healthcare resources (non-

drug) was partially supported by the analysis. The descriptive analysis showed

that the annual number of all-cause HRU was similar between the exposure and

the control cohorts, except for the number of outpatient visits and days supplied

for SABD (lower for the exposure cohort). After controlling for post-PSM

differences in the multivariable analysis, different directions of relationship

between cohorts and the annual number of all-cause HRU were detected based

on the timing of reaching the gap. Generally, a positive relationship was

observed in beneficiaries who reached the gap between March and October with

106

significance found for the outpatient setting only, and a negative relationship

was observed in beneficiaries who reached the gap after October with

significance found in the ER and inpatient settings.

This hypothesis was based on a belief that beneficiaries may seek care at

a physician’s office, ER, or hospital more often after they were exposed to the

coverage gap because they would reduce their use of medications when their

out-of-pocket cost increased. If beneficiaries hit the coverage gap earlier in the

year (i.e., mid-gap subgroup), they were more likely to follow this pattern

because there were a couple of months or longer when they had to pay full price

of COPD prescription drugs before the start of next coverage cycle (i.e., a new

calendar year), especially in the outpatient setting where health services were

relatively inexpensive when they had other Medicare health benefit coverage;

but less dramatic shift was observed in the ER or inpatient settings for this mid-

gap subgroup. Also, this does not appear to be the case for the beneficiaries who

reached the coverage gap after October (late-gap subgroup), and in fact opposite

results were shown that this group of beneficiaries used less HRU especially in

the ER and the inpatient settings. There are several potential explanations for the

lack of a substitution effect between LABDs and medical services. First, COPD

is symptomatic; as the GOLD guidelines specify, maintenance therapy with

LABDs is usually needed to control symptoms and prevent exacerbation when

COPD is at the moderate level or higher (GOLD, 2014), and there were few

convenient alternatives (such as generic LABDs) available to help slow disease

progression for the period of 2007-2010. Second, the out-of-pocket (OOP) cost

107

of medical services can be comparable to OOP cost of brand LABDs when

COPD is out of control, which may present an irrational choice for beneficiaries.

For example, it was estimated that healthcare expenses for seniors averaged

$2,714 per day for inpatient stays and $651 for ER visits (Machlin, 2009). Given

the OOP amount paid by Medicare beneficiaries accounts for approximately 15%

of the total expense (Machlin, 2009), beneficiaries pay about $407 per day for a

hospital stay and $98 for an ER visit. According to general Medicare benefit

policy, the hospital inpatient deductible for beneficiaries is $1,206 per benefit

period with additional coinsurance if the length of stay exceeds 60 days

(Medicare, 2015). In comparison, on average, the full price of 30-day supply of

brand LABDs ranges from about $150 to $350 (GoodRx, 2015). Thus, patients

might be less motivated to completely give up maintaining their lung function

and controlling their symptoms via medication as other options could be more

costly. Another possibility is that these patients did not have severe COPD and

had relatively stable health status, so they did not consume excess healthcare

resources over the year. In addition, there is possibly a lagged effect of coverage

gap exposure on utilization. That is, in this study only HRU in the current year

was estimated for each calendar year and it may take a longer time to see the

effect on health resource utilization.

Notably, the findings of this study are not consistent with the findings

reported by Raebel et al. (2008) who found that beneficiaries who reached the

coverage gap had 85% higher risk of being hospitalized and 60% higher risk of

using emergency room services as compared to those without the coverage gap

108

(Raebel et al., 2008). One potential reason for the difference relates to the study

population. The study by Raebel et al. focused on Medicare beneficiaries

covered by a commercial integrated healthcare delivery network (Kaiser

Permanente Colorado, KPCO) without focusing on a specific chronic condition

in defining the study cohort. Also, their study examined these relationships

shortly after the Part D coverage gap was implemented (2005-2006), and our

study examined the associations for multiple years after the implementation from

2007 to 2010.

Research Question/Hypothesis 3: All-Cause Medical Cost

Question 3: Is total all-cause medical cost (non-drug) higher for

Medicare patients with COPD who reached the Part D coverage gap?

The study found that the mean annual all-cause medical cost was similar

between the exposure and the control cohorts. Thus, the hypothesis that exposure

to the Part D coverage gap would be associated with higher all-cause medical

cost was not supported. The rationale for this hypothesis was that cost, as a

monetary manifestation of HRU, would increase as beneficiaries increased other

types of utilization as a substitute for prescription drugs. However, as the

analysis indicated, HRU in the ER and inpatient settings, which usually was the

driver of overall medical cost, did not significantly increase following exposure

to the Part D coverage gap.

This finding is consistent with the study by Zhang and colleagues (2012)

who reported no significant increases in non-drug medical spending in

beneficiaries with depression who reached the coverage gap as compared to

those who were not exposed to the coverage gap (Zhang et al., 2012).

109

Strengths and Limitations

Strengths

This is the first study to evaluate the impact of the Part D coverage gap

using the longitudinal national Medicare claims data for beneficiaries with

COPD. Other studies have analyzed the general Medicare population and sub-

populations with other conditions (e.g., diabetes, cardiovascular diseases, mental

illness). To date, however, no research has been conducted in COPD population

in regards to Medicare Part D coverage gap. Chronic lower respiratory disease,

which primarily includes COPD has become the third leading cause of death in

the United States (Hoyert & Xu, 2012). The total annual cost (direct and indirect)

of treating and managing COPD in the United States was estimated to be close to

$50 billon (American-Lung-Association, 2014; NHLBI, 2009). Approximately

12 to 15 million adults in the United States are diagnosed with COPD

(American-Lung-Association, 2014; NIH, 2008), and almost half of them were

aged 65 years or older (Centers for Disease Control and Prevention, 2012).

Consequently, effective management of COPD has become one of the priorities

for Medicare. Thus, this study provided information regarding the impact of the

Part D coverage gap in an important Medicare population. In addition, this study

added to the limited evidence base on the impact of the coverage gap on HRU

and cost.

This is the first study to explore the effect of timing of hitting the

coverage gap on outcomes. The published articles in the existing literature

usually assessed the impact of coverage gap from the perspective of with vs.

without or before vs. after; no other studies have investigated the difference

110

possibly related to when beneficiaries hit the gap. This study is also one of the

few studies evaluating outcomes related to the coverage gap that included

Medicare beneficiaries who were not exposed to the coverage gap as a control

group. Only two other studies used a similarly defined control group focused on

assessing the effect of reaching the coverage gap on drug utilization (Polinski et

al., 2010; Polinski et al., 2012). Instead, most published studies used pre-post

design and Difference-in-Difference methods where the beneficiaries who

reached the coverage gap served as a control group for themselves. The pre-post

design or Difference-in-Difference method does not require balanced baseline

characteristics between comparative groups because only one group is included

in the analysis, so selection bias is minimized by design. When chronic and

progressive diseases such as cardiovascular disease, diabetes, or COPD are

assessed in studies, a patient’s disease may change differently in the post-period

compared to the pre-period, especially if pre-period and post-period are not short.

Consequently, the effect of a policy on the outcomes may be contaminated by a

changing disease state. It can be challenging to adjust for this change in the pre-

post design. In contrast, in a matched design (such as the one used in this study)

the adjustment of baseline disease status is a part of the exposure-control design

to make the two groups more comparable and have similar disease states. As a

result, the effect of disease progression may be less of a concern

The methodologies used in the analysis also provide several strengths for

this study. First, the high-dimensional propensity score matching (HD-PSM) was

adopted to further mitigate possible selection biases and adjust for the observed

111

confounding effect between the exposure and the control cohorts. In an

observational study, when the exposure is not randomly assigned, propensity

score matching (PSM) is one of the often-used approaches to help reduce

selection bias based on predefined demographics, comorbidities, historical

medication use or healthcare resource use before the exposure initiation. The

HD-PSM method extends beyond traditional PSM by maximizing the usage of

information provided by claims data. Matching two cohorts with the propensity

score generated in this way is able to estimate effect closer to randomized trials

compared to the traditional PSM with typical covariates only (Schneeweiss,

Rassen, et al., 2009). This is the first study applying HD-PSM method in the

assessment of the impact of Part D coverage gap. Second, four years of data

were used in this study and outcomes were defined at the year level to account

for different Part D thresholds for each year. Consequently, beneficiaries with

enrollment for multiple years can have repeated observations. GEE technique

was used in the assessment to correct for the potential correlation between the

repeated observations from the same beneficiary. Third, multivariable regression

models that controlled for unbalanced characteristics post-matching adjusted for

residual confounding effects.

Limitations

This study has several limitations. First, medical and pharmacy claims

data used in this analysis were primarily used for administrative purposes to

obtain reimbursement, therefore, there is potential for coding errors that may

cause diagnostic and procedural misclassification. At least two COPD diagnosis

codes in an outpatient setting and at least two LABD prescriptions were required

112

in the patient identification stage to help reduce potential misclassifications. In

addition, the pharmacy claims only provided information on whether

prescriptions were filled by patients and there was no confirmation on whether

patients actually took the medication. Therefore, adherence was measured based

on the assumption that patients took the drugs after they filled the prescriptions.

Second, the study is subject to the limitations of retrospective

observational studies. As it was not an experimental design and did not

randomly assign exposure, the findings can only be interpreted as association

and no causality can be concluded. Although multiple strategies were applied to

minimize selection bias, they could only account for observed covariates and

were unable to control for unobserved factors, for example, patient’s lung

function, beneficiaries’ functional status, or beneficiaries’ behavior (e.g., self-

selection of high premium plan to avoid or reduce the burden produced by the

coverage gap).

Third, the methods adopted in this study have their own limitations.

Though matching based on HD-PS method can improve performance due to its

ability to adjust for additional confounding effects omitted in the traditional

PSM, HD-PS analysis was still empirical in nature and heavily relied on how

data were collected and recorded in the database. For example, the performance

of HD-PS analysis using administrative claims data might be different from

using electronic medical records data. Furthermore, because HD-PS analysis is

capable of generating hundreds of covariates, selection and prioritization of

appropriate covariates are usually based on the magnitude of the association

113

between each covariate and the study outcome. When the study outcome is rare,

the prioritization rule may miss potentially important covariates. (Schneeweiss,

Rassen, et al., 2009).

Fourth, the GLM regression provides a broad statistical approach to

analyzing various outcomes or predictors with different distributions. It is more

generally applicable than the traditional Ordinary Least-Squares regression;

however, it is still based on the normality assumption. The cost outcome is

usually highly skewed and outliers are commonly observed. Although the log-

link function and gamma distribution were applied to address the skewness, it is

possible that the transformed distribution in the GLM was still not perfectly

normal.

Fifth, the study cohort was composed of beneficiaries from the Medicare

FFS program, so the results might not be generalizable to the Medicare

population enrolled with Medicare Advantage plans. Similarly, the study period

ended in 2010 due to data availability and it was not clear if the impact of the

coverage gap identified in this study remained after 2010. Studies in the

managed Medicare population or using more recent data may provide more

insight.

Finally, the cost estimation in this analysis was only focused on medical

cost. Medicare does not directly pay drug claims submitted by the Part D plans,

but rather pays the part D plans in the way of subsidized premiums for their

provided Part D coverage. Future studies utilizing data including drug cost

114

would help to determine if and how much the Part D coverage gap can produce

cost offsets between prescription drugs and medical services for Medicare.

Implications

This study assessed the impact of the Medicare Part D coverage gap on

adherence, healthcare resource use, and medical cost among the beneficiaries

with COPD and has important research, managerial, and policy implications.

The following section describes these implications in greater detail.

Implications for Future Research

Like most published studies that have examined the Part D coverage gap

and medication adherence, a patient’s adherence was defined at the drug class

level (e.g., LABDs for COPD in this study; antidiabetics, statin or beta-blockers

or ACE inhibitors for MI, antihypertensive drugs, or lipid-lowering drugs for

other studies). Discontinuation of a particular LABD or switch between different

LABDs were not defined and not regarded as non-adherence in this study. MPR

and PDC are the most commonly used measures, but they do not capture actual

medication taking behaviors. This might be one reason for less consistent

directional association with adherence observed for the population with mental

illness compared to those with other chronic conditions in the literature,

considering individuals with mental illness often have more irregular patterns in

medication utilization. More refined adherence measurements that capture more

detailed drug-taking behavior is an area for future research on the impact of Part

D coverage gap on adherence. This research could include topics such as

discontinuation and restart, brand switch due to different formulary status, drug

augmentation or titration, and delay in drug dispensing. A lagged effect of

115

adherence level in one calendar year on the adherence level in the following year

could be another area to explore.

In this study, longitudinal data across four years were used to assess the

overall impact of the Part D coverage gap. Several things, however, are worth

noting with respect to establishing the coverage gap. First, different coverage

gap thresholds are established each calendar year with an increasing trend.

Second, different Part D drug plans can implement their own threshold for the

coverage gap based on their respective benefit designs. Future studies may want

to explore whether the level of the threshold leads to different adherence or

health service utilization. Likewise, the elasticity of the threshold may be an

interesting topic for future research, i.e., evaluating how sensitive beneficiaries

are to the threshold (e.g., how much medication adherence or care seeking

behavior is changed in response to different threshold levels), or what level of

threshold will generate cost-offsets or even cost-savings for insurers.

This study focused on the comparison of beneficiaries who were exposed

to and reached the coverage gap with those who were not exposed to the gap

because of financial assistance or other provided benefit or coverage. The

subgroup in the exposure cohort that was exposed to but did not reach the gap

was excluded from the analysis because they were assumed to be relatively

healthy and their drug utilization may not be impacted by the existence of the

coverage gap. This subgroup accounted for 30-40% of the overall exposure

cohort. Further research can be conducted to investigate their pattern of drug use

and health service utilization and related cost to confirm this assumption.

116

Another subgroup that was exposed to and reached the gap prior to March 1 was

excluded as well based on the assumption that they may have maximized their

drug utilization after reaching the gap in order to enter the catastrophic coverage

sooner. This subgroup accounted for less than 1% of the overall exposure cohort,

but it would be interesting to further investigate their drug and resource

utilization and related cost to test this assumption.

The defined outcomes (adherence, resource use, and cost) were not

assessed for the catastrophic phase as it was beyond the study scope of this

research. In this analysis, about 10% of the overall exposure cohort entered the

catastrophic phase during a respective calendar year. Whether and how drug

utilization and resource use in the catastrophic phase differs from the in-gap and

pre-gap periods are interesting questions to be taken up in future research.

Given the observed different effects on outcomes in the mid-gap and the

late-gap subgroups as compared to the control cohort in this study, researchers

might consider further assessing the impact of timing of reaching the gap in

greater detail using more advanced methodology such as time-varying analysis

to ascertain its association with outcomes.

Although no consistent significant association was observed with HRU

or cost in this study, researchers might want to explore the impact of the Part D

coverage gap on other types of outcomes such as clinical outcomes (e.g., disease

exacerbation, adverse event or episode, blood pressure control, or lung function

maintenance) or patient-reported outcomes (e.g., quality of life, or patient

satisfaction) so that more information on the impact of the coverage gap can be

117

provided to healthcare administrators and policymakers to better inform their

decision making.

Robust analytic techniques have been applied to this study to mitigate the

potential selection bias and confounding effects. For future studies, different

design methods can be explored based on the selected data source to further

validate the findings of this analysis. Researchers may even consider a

randomized trial with a small to moderate scale to make causal inferences.

Implications for Management and Policy

The findings from this study also provide insights for decision-makers

and administrators in healthcare. Medicare Part D is intended to lower

medication expenditures for Medicare beneficiaries. However, its complex cost-

sharing design creates gaps in coverage. Previous research (Baik et al., 2012;

Fung et al., 2010; Gu et al., 2010; Joyce et al., 2013; Zhang et al., 2013) and this

analysis have shown the gap negatively impacts medication adherence. Decline

in adherence indicates a disruption in medication treatment, which suggests that

insurance benefit with a gap in coverage possibly brought about negative

unintended consequences while improving beneficiaries’ access to healthcare.

Cycling in and out of a coverage gap may be disruptive to beneficiaries because

they have to make changes to their care plans in order to lower their risk of

falling into the gap and minimize the potential consequences of entering the gap.

Beneficiaries would have to give different behavioral responses at different

phases related to the coverage gap in order to mitigate the potential short-term

health effect. For example, prior to entering the coverage gap, beneficiaries

exposed to the gap might be cautious of their drug utilization so they can delay

118

the entry into the gap. After entering the gap, beneficiaries may choose not to

spend or reduce the prescription drug expense if it is close to year end, but may

choose to maximize their drug utilization to enter the catastrophic coverage

sooner if it is still early in the year. If beneficiaries are in the catastrophic phase,

they may be less concerned due to the catastrophic coverage. As the coverage

gap is initiated for each calendar year, beneficiaries without financial assistance

or subsidies may adjust their behavior and drug usage in anticipation of the gap

based on the experience of the previous year. This type of behavioral adjustment

can be stressful to the elderly beneficiaries and bring about certain long-term

consequences on beneficiaries’ health.

The coverage gap is planned to close out by 2020. By 2020, beneficiaries

will pay 25% of the total cost for covered brand-name and generic drugs during

the gap. Although the cost of closing the coverage gap may present a serious

challenge to policy makers in the current fiscal climate, it is expected that the

coverage gap closure will benefit beneficiaries. However, one study suggested

that phasing out the coverage gap under healthcare reform may still increase

drug cost if the drug has low clinical value (Li et al., 2012). The question of how

to redesign the drug benefit post gap close-out is what policy-makers and benefit

managers need to think through. A stable benefit with higher coinsurance can

make the process less complicated and reduce the financial risk associated with

pharmaceutical expenditures for healthcare payers. Also, more nuanced cost-

sharing insurance policies such as value-based insurance design may be needed

to provide incentives to encourage beneficiaries to use high-value medications

119

and discourage them from overusing medications with low marginal benefit

(Fendrick, Smith, Chernew, & Shah, 2001). With that, policy makers can

consider Part D cost-sharing approaches and utilization management tools to

promote appropriate use of covered products and maximize the clinical and

economic value for beneficiaries as well as insurers.

On the other hand, other research showed that the elderly were likely to

reduce the use of essential medications due to drug copayment (Artz, Hadsall, &

Schondelmeyer, 2002; Ellis et al., 2004; Federman, Adams, Ross-Degnan,

Soumerai, & Ayanian, 2001; Piette, Wagner, Potter, & Schillinger, 2004;

Tamblyn et al., 2001). A study evaluated the “first-dollar coverage” (i.e., no cost

sharing) of ACE inhibitors for Medicare beneficiaries with diabetes and reported

that the program helped to extend patient life and reduce Medicare program cost,

suggesting that full Medicare coverage of essential medications or high-value

drugs can be a cost-effective option for Medicare (Rosen et al., 2005).

These discussions suggest that benefit redesign not be a “one-size-fits-all”

strategy. Healthcare administrators need to work with multi-disciplinary expert

teams covering medical, economics, policy, and other areas and take both

clinical and economic values into consideration to establish robust coverage

policies that can be customized for beneficiaries with different profiles. Such

efforts, however, present nontrivial challenges for administrators and policy

makers in the form of substantial time, resources, and administrative efforts

required to design “personalized” coverage for a diverse population.

120

Prior to the close-out of the Part D coverage gap, healthcare

administrators and health plans should make efforts to help beneficiaries

transition through the gap smoothly and minimize the risk of experiencing high

out-of-pocket cost and preventable adverse outcomes from medication non-

adherence. Health plans can take more proactive approaches to raise awareness

of the coverage gap among beneficiaries and physicians, such as mailing or

calling members to alert them of their proximity to the gap long before they

reach it. Health plans can also provide beneficiaries with personalized

information on cost-saving options that may help delay their entry into the gap.

Finally, health plans may consider educating beneficiaries of the importance of

adherence and develop strategies to help them remain compliant with their

medication regimens and ultimately reduce the use of expensive medical

encounters and decrease cost for both beneficiaries and the health plan.

Health plans may also want to consider working with healthcare

providers on strategies to improve medication adherence. Patients directly

interact with and usually trust healthcare providers for their health issues.

Therefore, patients receiving education from healthcare providers regarding the

importance of medication adherence and optimal treatment regimen may be

more receptive than the same information received from health plans.

Healthcare cost is increasing rapidly and the projected net expenditure

for the Part D program from 2009 to 2018 is estimated to be $727.3 billion

(Roumie, 2012). Healthcare payers and decision-makers as well as beneficiaries

may be more receptive to policy changes if they promote appropriate drug

121

utilization and achieve a consistent approach for Part D plan formularies, and

subsequently result in cost savings to Medicare, less cost-related non-adherence

and lower economic burden to beneficiaries, and ultimately improvement in the

health of Medicare beneficiaries.

Conclusion

The purpose of this study was to investigate association of the Medicare

Part D coverage gap relative to medication adherence, healthcare resource use,

and cost among Medicare beneficiaries with COPD. Findings from this analysis

revealed that reaching Part D coverage gap lowered the odds of medication

adherence in COPD. However, the association with healthcare resource use was

mixed and the association with cost was inconclusive.

There is little information available regarding the effect of Part D

coverage gap on health resource use or cost outcomes and no published studies

on the Part D coverage gap and beneficiaries with COPD. This study provides

insights to fill this evidence gap. Building on these findings, additional research

related to other important illness or other meaningful outcomes will be pertinent

for increasing the knowledge base in this area, improving current benefit design,

developing future benefit structures, and optimizing the quality of health policy

decisions in order to help Medicare provide the best healthcare for its members

with the most cost-effective outcomes.

122

REFERENCES

Afendulis, C. C., & Chernew, M. E. (2011). State-level impacts of Medicare Part D. Am J Manag Care, 17 Suppl 12, S.

Afendulis, C. C., He, Y., Zaslavsky, A. M., & Chernew, M. E. (2011). The impact of Medicare Part D on hospitalization rates. Health Serv Res,

46(4), 1022-1038. doi: 10.1111/j.1475-6773.2011.01244.x Akinbami, L. J., & Liu, X. (2011). Chronic obstructive pulmonary disease

among adults aged 18 and over in the United States, 1998-2009. NCHS

Data Brief(63), 1-8. American-Lung-Association. (2013). Trends in chronic obstructive bronchitis

and emphysema. Lung.org: Epidemiology & statistics unit, research and Health Education Division.

American-Lung-Association. (2014). Chronic obstructive pulmonary disease (COPD) fact sheet. Retrieved from http://www.lung.org/lung-disease/copd/resources/facts-figures/COPD-Fact-Sheet.html#Sources

Anthonisen, N. R., Connett, J. E., Kiley, J. P., Altose, M. D., Bailey, W. C., Buist, A. S., . . . et al. (1994). Effects of smoking intervention and the use of an inhaled anticholinergic bronchodilator on the rate of decline of FEV1. The Lung Health Study. Jama, 272(19), 1497-1505.

Artz, M. B., Hadsall, R. S., & Schondelmeyer, S. W. (2002). Impact of generosity level of outpatient prescription drug coverage on prescription drug events and expenditure among older persons. Am J Public Health,

92(8), 1257-1263. ATS/ERS. (2004). Standards for the diagnosis and management of patients with

COPD (Internet). Version 1.2. New York: American Thoracic Society. Austin, P. C. (2009). Balance diagnostics for comparing the distribution of

baseline covariates between treatment groups in propensity-score matched samples. Stat Med, 28(25), 3083-3107. doi: 10.1002/sim.3697

Baik, S. H., Rollman, B. L., Reynolds, C. F., 3rd, Lave, J. R., Smith, K. J., & Zhang, Y. (2012). The effect of the US Medicare Part D coverage gaps on medication use among patients with depression and heart failure. J

Ment Health Policy Econ, 15(3), 105-118. Baillie, A. J., Mattick, R. P., Hall, W., & Webster, P. (1994). Meta-analytic

review of the efficacy of smoking cessation interventions. Drug Alcohol

Rev, 13(2), 157-170. doi: 10.1080/09595239400185231 Bakk, L., Woodward, A. T., & Dunkle, R. E. (2014). The Medicare Part D

coverage gap: implications for non-dually eligible older adults with a mental illness. J Gerontol Soc Work, 57(1), 37-51. doi: 10.1080/01634372.2013.854857

Basu, A., Yin, W., & Alexander, G. C. (2010). Impact of Medicare Part D on Medicare-Medicaid dual-eligible beneficiaries' prescription utilization and expenditures. Health Serv Res, 45(1), 133-151. doi: 10.1111/j.1475-6773.2009.01065.x

Blanchette, C. M., Gutierrez, B., Ory, C., Chang, E., & Akazawa, M. (2008). Economic burden in direct costs of concomitant chronic obstructive

123

pulmonary disease and asthma in a Medicare Advantage population. J

Manag Care Pharm, 14(2), 176-185. Breekveldt-Postma, N. S., Koerselman, J., Erkens, J. A., Lammers, J. W., &

Herings, R. M. (2007). Enhanced persistence with tiotropium compared with other respiratory drugs in COPD. Respir Med, 101(7), 1398-1405. doi: 10.1016/j.rmed.2007.01.025

Briesacher, B. A., Zhao, Y., Madden, J. M., Zhang, F., Adams, A. S., Tjia, J., . . . Soumerai, S. B. (2011). Medicare part D and changes in prescription drug use and cost burden: national estimates for the Medicare population, 2000 to 2007. Med Care, 49(9), 834-841. doi: 10.1097/MLR.0b013e3182162afb

Brookhart, M. A., Sturmer, T., Glynn, R. J., Rassen, J., & Schneeweiss, S. (2010). Confounding control in healthcare database research: challenges and potential approaches. Med Care, 48(6 Suppl), S114-120. doi: 10.1097/MLR.0b013e3181dbebe3

Burton, P., Gurrin, L., & Sly, P. (1998). Extending the simple linear regression model to account for correlated responses: an introduction to generalized estimating equations and multi-level mixed modelling. Stat Med, 17(11), 1261-1291.

Calvert, R. (1994). Identify, expression, and rational-choice theory. In Michael Martin & L. McIntyre (Eds.), Readings in the Philosophy of the Social

Sciences (pp. 569-596). Cambridge: MIT press. Cecere, L. M., Slatore, C. G., Uman, J. E., Evans, L. E., Udris, E. M., Bryson, C.

L., & Au, D. H. (2012). Adherence to long-acting inhaled therapies among patients with chronic obstructive pulmonary disease (COPD). Copd, 9(3), 251-258. doi: 10.3109/15412555.2011.650241

Centers for Disease Control and Prevention, C. (2012). Chronic obstructive pulmonary disease among adults--United States, 2011. Morbidity and

mortality weekly report (Vol. 61, pp. 938-943). cdc.gov: Centers for Disease Control and Prevention

Centers for Medicare & Medicaid, C. Medicare Advantage Plans. Retrieved from http://www.medicare.gov/sign-up-change-plans/medicare-health-plans/medicare-advantage-plans/medicare-advantage-plans.html

Centers for Medicare & Medicaid, C. (2011). Medicare prescription drug premiums will not increase, more seniors receiving free preventive care, discounts in the donut hole [Press release]

Centers for Medicare & Medicaid, C. (2012). Medicare Prescription Drug Premiums to Remain Steady for Third Straight Year [Press release]

Centers for Medicare & Medicaid, C. (2013, August 7, 2013). Medicare Program - General Information. Retrieved from http://www.cms.gov/Medicare/Medicare-General-Information/MedicareGenInfo/index.html

Charles, M. S., Blanchette, C. M., Silver, H., Lavallee, D., Dalal, A. A., & Mapel, D. (2010). Adherence to controller therapy for chronic obstructive pulmonary disease: a review. Curr Med Res Opin, 26(10), 2421-2429. doi: 10.1185/03007995.2010.516284

124

Charlson, M. E., Pompei, P., Ales, K. L., & MacKenzie, C. R. (1987). A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis, 40(5), 373-383.

Cheng, L. I., & Rascati, K. L. (2012). Impact of Medicare Part D for Medicare-age adults with arthritis: prescription use, prescription expenditures, and medical spending from 2005 to 2008. Arthritis Care Res (Hoboken),

64(9), 1423-1429. doi: 10.1002/acr.21696 Claffey, J. (2010, August 14, 2010). Medicare 'Doughnut Hole' Checks in the

Mail. Foster's Daily Democrat. Clifford, S., & Coyne, K. S. (2014). What is the value of medication adherence?

J Manag Care Pharm, 20(7), 650-651. Coleman, J. (1990). Foundations of social theory. Cambridge MA: Harvard

University. Congressional Budget Office, C. (2011). Preliminary Analysis of the President’s

Budget for 2012. Washington D.C.: U.S. Congress. Congressional Budget Office, C. (2012). Updated Budget Projections: Fiscal

Years 2012 to 2022 (pp. 9): U.S. Congress. D'Agostino, R. B., Jr. (1998). Propensity score methods for bias reduction in the

comparison of a treatment to a non-randomized control group. Stat Med,

17(19), 2265-2281. D'Hoore, W., Bouckaert, A., & Tilquin, C. (1996). Practical considerations on

the use of the Charlson comorbidity index with administrative data bases. J Clin Epidemiol, 49(12), 1429-1433.

Dall, T. M., Blanchard, T. D., Gallo, P. D., & Semilla, A. P. (2013). The economic impact of Medicare Part D on congestive heart failure. Am J

Manag Care, 19(6 Suppl), s97-100. Darkow, T., Kadlubek, P. J., Shah, H., Phillips, A. L., & Marton, J. P. (2007). A

retrospective analysis of disability and its related costs among employees with chronic obstructive pulmonary disease. J Occup Environ Med,

49(1), 22-30. doi: 10.1097/JOM.0b013e31802db55f Deyo, R. A., Cherkin, D. C., & Ciol, M. A. (1992). Adapting a clinical

comorbidity index for use with ICD-9-CM administrative databases. J

Clin Epidemiol, 45(6), 613-619. Diggle, P., Liang, K., & Zeger, S. (1994). Analysis of longitudinal data. Oxford,

UK: Oxford University Press. Donohue, J. M., Zhang, Y., Aiju, M., Perera, S., Lave, J. R., Hanlon, J. T., &

Reynolds, C. F., 3rd. (2011). Impact of Medicare Part D on antidepressant treatment, medication choice, and adherence among older adults with depression. Am J Geriatr Psychiatry, 19(12), 989-997. doi: 10.1097/JGP.0b013e3182051a9b

Dusetzina, S. B., Winn, A. N., Abel, G. A., Huskamp, H. A., & Keating, N. L. (2014). Cost sharing and adherence to tyrosine kinase inhibitors for patients with chronic myeloid leukemia. J Clin Oncol, 32(4), 306-311. doi: 10.1200/jco.2013.52.9123

125

Eisner, M. D., Yelin, E. H., Trupin, L., & Blanc, P. D. (2002). The influence of chronic respiratory conditions on health status and work disability. Am J

Public Health, 92(9), 1506-1513. Ellis, J. J., Erickson, S. R., Stevenson, J. G., Bernstein, S. J., Stiles, R. A., &

Fendrick, A. M. (2004). Suboptimal statin adherence and discontinuation in primary and secondary prevention populations. J Gen Intern Med,

19(6), 638-645. doi: 10.1111/j.1525-1497.2004.30516.x Ettner, S. L., Steers, N., Duru, O. K., Turk, N., Quiter, E., Schmittdiel, J., &

Mangione, C. M. (2010). Entering and exiting the Medicare part D coverage gap: role of comorbidities and demographics. J Gen Intern

Med, 25(6), 568-574. doi: 10.1007/s11606-010-1300-6 Federman, A. D., Adams, A. S., Ross-Degnan, D., Soumerai, S. B., & Ayanian,

J. Z. (2001). Supplemental insurance and use of effective cardiovascular drugs among elderly medicare beneficiaries with coronary heart disease. Jama, 286(14), 1732-1739.

Fendrick, A. M., Smith, D. G., Chernew, M. E., & Shah, S. N. (2001). A benefit-based copay for prescription drugs: patient contribution based on total benefits, not drug acquisition cost. Am J Manag Care, 7(9), 861-867.

Fung, V., Mangione, C. M., Huang, J., Turk, N., Quiter, E. S., Schmittdiel, J. A., & Hsu, J. (2010). Falling into the coverage gap: Part D drug costs and adherence for Medicare Advantage prescription drug plan beneficiaries with diabetes. Health Serv Res, 45(2), 355-375. doi: 10.1111/j.1475-6773.2009.01071.x

Fung, V., Price, M., Busch, A. B., Landrum, M. B., Fireman, B., Nierenberg, A., . . . Hsu, J. (2013). Adverse clinical events among medicare beneficiaries using antipsychotic drugs: linking health insurance benefits and clinical needs. Med Care, 51(7), 614-621. doi: 10.1097/MLR.0b013e31829019c5

Gagne, J. J., Glynn, R. J., Avorn, J., Levin, R., & Schneeweiss, S. (2011). A combined comorbidity score predicted mortality in elderly patients better than existing scores. J Clin Epidemiol, 64(7), 749-759. doi: 10.1016/j.jclinepi.2010.10.004

Gibson, T. B., Mark, T. L., Axelsen, K., Baser, O., Rublee, D. A., & McGuigan, K. A. (2006). Impact of statin copayments on adherence and medical care utilization and expenditures. Am J Manag Care, 12 Spec no., Sp11-19.

GOLD. (2014). Global Strategy for Diagnosis, Management, and Prevention of COPD Global Initiative for Chronic Obstructive Lung Disease (GOLD): GOLD committee.

GoodRx. (2015). Prices for Popular Long-Acting Beta Agonists. Retrieved from http://www.goodrx.com/long-acting-beta-agonists

Grasso, M. E., Weller, W. E., Shaffer, T. J., Diette, G. B., & Anderson, G. F. (1998). Capitation, managed care, and chronic obstructive pulmonary disease. Am J Respir Crit Care Med, 158(1), 133-138. doi: 10.1164/ajrccm.158.1.9710041

Gu, Q., Zeng, F., Patel, B. V., & Tripoli, L. C. (2010). Part D coverage gap and adherence to diabetes medications. Am J Manag Care, 16(12), 911-918.

126

Hardin, J., & Hilbe, J. (2003). Generalized Estimating Equations. London, UK: Chapman and Hall/CRC.

Hechter, M., & Kanazawa, S. (1997). Sociological rational choice theory. Annu

Rev: Sociol, 23, 24. Hedström, P., & Stern, C. (2008). Rational choice and sociology. In S. N.

Durlauf & L. E. Blume (Eds.), The New Palgrave Dictionary of

Economics (2nd ed.): Palgrave Macmillan. Hess, L. M., Raebel, M. A., Conner, D. A., & Malone, D. C. (2006).

Measurement of adherence in pharmacy administrative databases: a proposal for standard definitions and preferred measures. Ann

Pharmacother, 40(7-8), 1280-1288. doi: 10.1345/aph.1H018 Hilbe, J. M. (2011). Negative binomial regression (second ed.). Cambridge, UK:

Cambridge University Press. Ho, P. M., Bryson, C. L., & Rumsfeld, J. S. (2009). Medication adherence: its

importance in cardiovascular outcomes. Circulation, 119(23), 3028-3035. doi: 10.1161/circulationaha.108.768986

Hoadley, J., Hargrave, E., Cubanski, J., & Neuman, T. (2008). The Medicare Part D coverage gap: costs and consequences in 2007. kpp.org: Kaiser Famiy Foundation.

Hoadley, J., Thompson, J., Hargrave, E., Merrell, K., Cubanski, J., & Neuman, T. (2007). Medicare Part D 2008 Data Spotlight: The Coverage Gap. kff.org: Kaiser Family Foundation.

Hoyert, D. L., & Xu, J. (2012). Deaths: Preliminary Data for 2011 (N. c. f. h. statistics, Trans.) National Vital Statistics Reports (Vol. 61, pp. 1-51). cdc.gov: Department of Health and Human Services, Centers for Disease Control and Prevention.

IMS. (2007). Medicare Part D: The First Year IMS Health. Ingber, M., Greenwald, L., Freeman, S., & Healy, D. (2010). Medicare Part D

program evaluation: analysis of the impact of Medicare Part D on the FFS program. CMS.gov: CMS.

Joyce, G. F., Zissimopoulos, J., & Goldman, D. P. (2013). Digesting the doughnut hole. J Health Econ, 32(6), 1345-1355. doi: 10.1016/j.jhealeco.2013.04.007

Jung, K., Feldman, R., & McBean, A. M. (2014). Nonlinear pricing in drug benefits and medication use: the case of statin compliance in Medicare Part D. Health Serv Res, 49(3), 910-928. doi: 10.1111/1475-6773.12145

Karaca, Z., Streeter, S., Barton, V., & Nguyen, K. (2008). The impact of Medicare Part D on beneficiaries with Type 2 Diabetes/drug utilization and out-of-pocket costs. Avalerehealth.net: Avalere Health LLC.

Kennedy, J. J., Maciejewski, M., Liu, D., & Blodgett, E. (2011). Cost-related nonadherence in the Medicare program: the impact of Part D. Med Care,

49(5), 522-526. doi: 10.1097/MLR.0b013e318210443d Ketcham, J. D., & Simon, K. I. (2008). Medicare Part D's effects on elderly

patients' drug costs and utilization. Am J Manag Care, 14(11 Suppl), Sp14-22.

127

Khdour, M. R., Hawwa, A. F., Kidney, J. C., Smyth, B. M., & McElnay, J. C. (2012). Potential risk factors for medication non-adherence in patients with chronic obstructive pulmonary disease (COPD). Eur J Clin

Pharmacol, 68(10), 1365-1373. doi: 10.1007/s00228-012-1279-5 Kim, E., Gupta, S., Bolge, S., Chen, C. C., Whitehead, R., & Bates, J. A. (2010).

Adherence and outcomes associated with copayment burden in schizophrenia: a cross-sectional survey. J Med Econ, 13(2), 185-192. doi: 10.3111/13696991003723023

Kim, M. H., Lin, J., & Kreilick, C. (2009). National assessment of Medicare prescription plan coverage gaps among patients with atrial fibrillation in the US. Adv Ther, 26(8), 784-794. doi: 10.1007/s12325-009-0054-1

KRCresearch. (2013). Seniors’ Opinions About Medicare Prescription Drug Coverage: Eighth Year Update, September 2013 (pp. 51). Medicaretoday.org: KRCresearch.

Lawrence E. Blume, & Easley, D. (2008). Rationality. In S. N. Durlauf & L. E. Blume (Eds.), The New Palgrave Dictionary of Economics (2nd ed.): Palgrave Macmillan.

Li, P., McElligott, S., Bergquist, H., Schwartz, J. S., & Doshi, J. A. (2012). Effect of the Medicare Part D coverage gap on medication use among patients with hypertension and hyperlipidemia. Ann Intern Med, 156(11), 776-784, w-263, w-264, w-265, w-266, w-267, w-268, w-269. doi: 10.7326/0003-4819-156-11-201206050-00004

Liang, K.-Y., & Zegger, S. L. (1986). Longitudinal data analysis using generalized linear models. Biometrika, 73(1), 10.

Little, R. J., & Rubin, D. B. (2000). Causal effects in clinical and epidemiological studies via potential outcomes: concepts and analytical approaches. Annu Rev Public Health, 21, 121-145. doi: 10.1146/annurev.publhealth.21.1.121

Liu, F. X., Alexander, G. C., Crawford, S. Y., Pickard, A. S., Hedeker, D., & Walton, S. M. (2011). The impact of Medicare Part D on out-of-pocket costs for prescription drugs, medication utilization, health resource utilization, and preference-based health utility. Health Serv Res, 46(4), 1104-1123. doi: 10.1111/j.1475-6773.2011.01273.x

Lohmann, S. (2008). Rational choice and political science. In S. N. Durlauf & L. E. Blume (Eds.), The New Palgrave Dictionary of Economics (2nd ed.): Palgrave Macmillan.

Machlin, S. R. (2009). Trends in health care expenditures for the elderly age 65 and over: 2006 versus 1996. Statistical brief (Vol. 256). ahrq.gov: Agency for Healthcare Research and Quality.

Madden, J. M., Graves, A. J., Ross-Degnan, D., Briesacher, B. A., & Soumerai, S. B. (2009). Cost-related medication nonadherence after implementation of Medicare Part D, 2006-2007. Jama, 302(16), 1755-1756. doi: 10.1001/jama.2009.1516

Madden, J. M., Graves, A. J., Zhang, F., Adams, A. S., Briesacher, B. A., Ross-Degnan, D., . . . Soumerai, S. B. (2008). Cost-related medication nonadherence and spending on basic needs following implementation of

128

Medicare Part D. Jama, 299(16), 1922-1928. doi: 10.1001/jama.299.16.1922

Make, B., Dutro, M. P., Paulose-Ram, R., Marton, J. P., & Mapel, D. W. (2012). Undertreatment of COPD: a retrospective analysis of US managed care and Medicare patients. Int J Chron Obstruct Pulmon Dis, 7, 1-9. doi: 10.2147/copd.s27032

McWilliams, J. M., Zaslavsky, A. M., & Huskamp, H. A. (2011). Implementation of Medicare Part D and nondrug medical spending for elderly adults with limited prior drug coverage. Jama, 306(4), 402-409. doi: 10.1001/jama.2011.1026

Medicare-Made-Clear. (2014). COPD and Medicare. Retrieved from http://blog.medicaremadeclear.com/blog/bid/178249/COPD-and-Medicare

Medicare. (2015). Medicare 2015 costs at a glance. Retrieved from http://www.medicare.gov/your-medicare-costs/costs-at-a-glance/costs-at-glance.html

Nair, K. V., Frech-Tamas, F., Jan, S., Wolfe, P., Allen, R. R., & Saseen, J. J. (2011). Comparing pre-gap and gap behaviors for Medicare beneficiaries in a Medicare managed care plan. J Health Care Finance, 38(2), 38-53.

NHLBI. (2007). Morbidity and Mortality: 2007 Chart Book on Cardiovascular, lung, and blood disease. Bethesda, MD: National Institutes of Heart, Lung, and Blood (NHLBI).

NHLBI. (2009). Morbidity and Mortality: 2009 Chart Book on Cardiovascular, lung, and blood disease. Bethesda, MD: National Institutes of Heart, Lung, and Blood (NHLBI).

NIH. (2008). Fact Book Fiscal Year 2007. Bethesda, MD: National Heart, Lung, and Blood Institute.

Normand, S. T., Landrum, M. B., Guadagnoli, E., Ayanian, J. Z., Ryan, T. J., Cleary, P. D., & McNeil, B. J. (2001). Validating recommendations for coronary angiography following acute myocardial infarction in the elderly: a matched analysis using propensity scores. J Clin Epidemiol,

54(4), 387-398. Osgood, D. (2000). Poisson‐based Regression Analysis of Aggregate Crime

Rates. Journal of Quantitative Criminology, 16, 24. Osterberg, L., & Blaschke, T. (2005). Adherence to medication. N Engl J Med,

353(5), 487-497. doi: 10.1056/NEJMra050100 Parsons, L. S. (2009). Reducing Bias in a Propensity Score Matched-Pair

Sample Using Greedy Matching Techniques. Paper presented at the SAS User Group International (SUGI), Seatle WA. http://www2.sas.com/proceedings/sugi26/p214-26.pdf

Paternoster, R., & Brame, R. (1997). Multiple routes to delinquency? A test of developmental and general theories of crime. . Criminology, 35, 40.

Pedan, A., Lu, J., & Varasteh, L. T. (2009). Assessment of drug consumption patterns for Medicare Part D patients. Am J Manag Care, 15(5), 323-327.

129

Piette, J. D., Heisler, M., & Wagner, T. H. (2004). Cost-related medication underuse among chronically ill adults: the treatments people forgo, how often, and who is at risk. Am J Public Health, 94(10), 1782-1787.

Piette, J. D., Wagner, T. H., Potter, M. B., & Schillinger, D. (2004). Health insurance status, cost-related medication underuse, and outcomes among diabetes patients in three systems of care. Med Care, 42(2), 102-109. doi: 10.1097/01.mlr.0000108742.26446.17

Pitta, F., Troosters, T., Spruit, M. A., Probst, V. S., Decramer, M., & Gosselink, R. (2005). Characteristics of physical activities in daily life in chronic obstructive pulmonary disease. Am J Respir Crit Care Med, 171(9), 972-977. doi: 10.1164/rccm.200407-855OC

Polinski, J. M., Kilabuk, E., Schneeweiss, S., Brennan, T., & Shrank, W. H. (2010). Changes in drug use and out-of-pocket costs associated with Medicare Part D implementation: a systematic review. J Am Geriatr Soc,

58(9), 1764-1779. doi: 10.1111/j.1532-5415.2010.03025.x Polinski, J. M., Shrank, W. H., Glynn, R. J., Huskamp, H. A., Roebuck, M. C.,

& Schneeweiss, S. (2012). Beneficiaries with cardiovascular disease and thePart D coverage gap. Circ Cardiovasc Qual Outcomes, 5(3), 387-395. doi: 10.1161/circoutcomes.111.964866

Polinski, J. M., Shrank, W. H., Huskamp, H. A., Glynn, R. J., Liberman, J. N., & Schneeweiss, S. (2011). Changes in drug utilization during a gap in insurance coverage: an examination of the medicare Part D coverage gap. PLoS Med, 8(8), e1001075. doi: 10.1371/journal.pmed.1001075

Raebel, M. A., Delate, T., Ellis, J. L., & Bayliss, E. A. (2008). Effects of reaching the drug benefit threshold on Medicare members' healthcare utilization during the first year of Medicare Part D. Med Care, 46(10), 1116-1122. doi: 10.1097/MLR.0b013e318185cddd

Rand, C. (2005). Patient adherence with COPD therapy. Eur Respir Rev, 14(96), 5.

Rassen, J., Doherty, M., Huang, W., & Schneeweiss, S. (2013). Pharmacoepidemiology Toolbox (Version 2.4.15). Boston, MA. Retrieved from http://www.drugepi.org/dope-downloads/

Rassen, J. A., Glynn, R. J., Brookhart, M. A., & Schneeweiss, S. (2011). Covariate selection in high-dimensional propensity score analyses of treatment effects in small samples. Am J Epidemiol, 173(12), 1404-1413. doi: 10.1093/aje/kwr001

Restrepo, R. D., Alvarez, M. T., Wittnebel, L. D., Sorenson, H., Wettstein, R., Vines, D. L., . . . Wilkins, R. L. (2008). Medication adherence issues in patients treated for COPD. Int J Chron Obstruct Pulmon Dis, 3(3), 371-384.

Romano, P. S., Roos, L. L., & Jollis, J. G. (1993). Adapting a clinical comorbidity index for use with ICD-9-CM administrative data: differing perspectives. J Clin Epidemiol, 46(10), 1075-1079; discussion 1081-1090.

Rosen, A. B., Hamel, M. B., Weinstein, M. C., Cutler, D. M., Fendrick, A. M., & Vijan, S. (2005). Cost-effectiveness of full medicare coverage of

130

angiotensin-converting enzyme inhibitors for beneficiaries with diabetes. Ann Intern Med, 143(2), 89-99.

Rosenthal, M. B. (2004). Doughnut-hole economics. Health Aff (Millwood),

23(6), 129-135. doi: 10.1377/hlthaff.23.6.129 Roumie, C. L. (2012). The doughnut hole: it's about medication adherence. Ann

Intern Med, 156(11), 834-835. doi: 10.7326/0003-4819-156-11-201206050-00010

Safran, D. G., Neuman, P., Schoen, C., Kitchman, M. S., Wilson, I. B., Cooper, B., . . . Rogers, W. H. (2005). Prescription drug coverage and seniors: findings from a 2003 national survey. Health Aff (Millwood), Suppl Web

Exclusives, W5-152-w155-166. doi: 10.1377/hlthaff.w5.152 Safran, D. G., Neuman, P., Schoen, C., Montgomery, J. E., Li, W., Wilson, I. B.,

. . . Rogers, W. H. (2002). Prescription drug coverage and seniors: how well are states closing the gap? Health Aff (Millwood), Suppl Web

Exclusives, W253-268. Safran, D. G., Strollo, M. K., Guterman, S., Li, A., Rogers, W. H., & Neuman,

P. (2010). Prescription coverage, use and spending before and after Part D implementation: a national longitudinal panel study. J Gen Intern Med,

25(1), 10-17. doi: 10.1007/s11606-009-1134-2 Schmittdiel, J. A., Ettner, S. L., Fung, V., Huang, J., Turk, N., Quiter, E. S., . . .

Mangione, C. M. (2009). Medicare Part D coverage gap and diabetes beneficiaries. Am J Manag Care, 15(3), 189-193.

Schneeweiss, S., & Maclure, M. (2000). Use of comorbidity scores for control of confounding in studies using administrative databases. Int J Epidemiol,

29(5), 891-898. Schneeweiss, S., Patrick, A. R., Pedan, A., Varasteh, L., Levin, R., Liu, N., &

Shrank, W. H. (2009). The effect of Medicare Part D coverage on drug use and cost sharing among seniors without prior drug benefits. Health

Aff (Millwood), 28(2), w305-316. doi: 10.1377/hlthaff.28.2.w305 Schneeweiss, S., Rassen, J. A., Glynn, R. J., Avorn, J., Mogun, H., & Brookhart,

M. A. (2009). High-dimensional propensity score adjustment in studies of treatment effects using health care claims data. Epidemiology, 20(4), 512-522. doi: 10.1097/EDE.0b013e3181a663cc

Schneeweiss, S., Seeger, J. D., Maclure, M., Wang, P. S., Avorn, J., & Glynn, R. J. (2001). Performance of comorbidity scores to control for confounding in epidemiologic studies using claims data. Am J Epidemiol, 154(9), 854-864.

Scott, J. (2000). Understanding contemporary Society: theories of the present. Sage: Belknap Press.

Sen, A. (2008). Rational behaviour. In S. N. Durlauf & L. E. Blume (Eds.), The

New Palgrave Dictionary of Economics (2nd ed.): Palgrave Macmillan. Shrank, W. H., Hoang, T., Ettner, S. L., Glassman, P. A., Nair, K., DeLapp, D., .

. . Asch, S. M. (2006). The implications of choice: prescribing generic or preferred pharmaceuticals improves medication adherence for chronic conditions. Arch Intern Med, 166(3), 332-337. doi: 10.1001/archinte.166.3.332

131

Simoni-Wastila, L., Wei, Y. J., Qian, J., Zuckerman, I. H., Stuart, B., Shaffer, T., . . . Bryant-Comstock, L. (2012). Association of chronic obstructive pulmonary disease maintenance medication adherence with all-cause hospitalization and spending in a Medicare population. Am J Geriatr

Pharmacother, 10(3), 201-210. doi: 10.1016/j.amjopharm.2012.04.002 Stoller, J. K., Panos, R. J., Krachman, S., Doherty, D. E., & Make, B. (2010).

Oxygen therapy for patients with COPD: current evidence and the long-term oxygen treatment trial. Chest, 138(1), 179-187. doi: 10.1378/chest.09-2555

Stuart, B., Davidoff, A., Erten, M., Gottlieb, S. S., Dai, M., Shaffer, T., . . . Shenolikar, R. (2013). How Medicare Part D benefit phases affect adherence with evidence-based medications following acute myocardial infarction. Health Serv Res, 48(6 Pt 1), 1960-1977. doi: 10.1111/1475-6773.12073

Stuart, B., Simoni-Wastila, L., Yin, X., Davidoff, A., Zuckerman, I. H., & Doshi, J. (2011). Medication use and adherence among elderly Medicare beneficiaries with diabetes enrolled in Part D and retiree health plans. Med Care, 49(5), 511-515. doi: 10.1097/MLR.0b013e31820bf885

Stuart, B. C., Simoni-Wastila, L., Zuckerman, L., Doshi, J. A., Shea, D., Shaffer, T., & Lao, L. (2007). Medication use by age and disabled Medicare beneficiaries across the spectrum of morbidity: A Charbook. Baltimore, MD: The Peter Lamy Center on Drug Therapy and Aging.

Sun, S. X., & Lee, K. Y. (2007). The Medicare Part D doughnut hole: effect on pharmacy utilization. Manag Care Interface, 20(9), 51-55, 59.

Tamblyn, R., Laprise, R., Hanley, J. A., Abrahamowicz, M., Scott, S., Mayo, N., . . . Mallet, L. (2001). Adverse events associated with prescription drug cost-sharing among poor and elderly persons. Jama, 285(4), 421-429.

Toy, E. L., Beaulieu, N. U., McHale, J. M., Welland, T. R., Plauschinat, C. A., Swensen, A., & Duh, M. S. (2011). Treatment of COPD: relationships between daily dosing frequency, adherence, resource use, and costs. Respir Med, 105(3), 435-441. doi: 10.1016/j.rmed.2010.09.006

Vestbo, J., Anderson, J. A., Calverley, P. M., Celli, B., Ferguson, G. T., Jenkins, C., . . . Jones, P. W. (2009). Adherence to inhaled therapy, mortality and hospital admission in COPD. Thorax, 64(11), 939-943. doi: 10.1136/thx.2009.113662

Wedderburn, R. (1974). Quasi-likelihood functions, generalized linear models, and the Gauss-Newton method. Biometrika, 61(3), 9.

WHO. (2003). Adherence to long-term therapy: evidence for action. In E. Sabate (Ed.). who.int: World Health Organization.

WHO. (2008). COPD predicted to be third leading cause of death in 2030 World

health statistics 2008. Geneva: World Health Organization. Yin, W., Basu, A., Zhang, J. X., Rabbani, A., Meltzer, D. O., & Alexander, G.

C. (2008). The effect of the Medicare Part D prescription benefit on drug utilization and expenditures. Ann Intern Med, 148(3), 169-177.

132

Yohannes, A. M., Baldwin, R. C., & Connolly, M. J. (2006). Depression and anxiety in elderly patients with chronic obstructive pulmonary disease. Age Ageing, 35(5), 457-459. doi: 10.1093/ageing/afl011

Zeng, F., Patel, B. V., & Brunetti, L. (2013). Effects of coverage gap reform on adherence to diabetes medications. Am J Manag Care, 19(4), 308-316.

Zhang, Y., Baik, S. H., & Lave, J. R. (2013). Effects of Medicare Part D coverage gap on medication adherence. Am J Manag Care, 19(6), e214-224.

Zhang, Y., Baik, S. H., Zhou, L., Reynolds, C. F., & Lave, J. R. (2012). Effects of Medicare Part D coverage gap on medication and medical treatment among elderly beneficiaries with depression. Arch Gen Psychiatry, 69(7), 672-679. doi: 10.1001/archgenpsychiatry.2011.1402

Zhang, Y., Donohue, J. M., Newhouse, J. P., & Lave, J. R. (2009). The effects of the coverage gap on drug spending: a closer look at Medicare Part D. Health Aff (Millwood), 28(2), w317-325. doi: 10.1377/hlthaff.28.2.w317

Zhang, Y., Lave, J. R., Donohue, J. M., Fischer, M. A., Chernew, M. E., & Newhouse, J. P. (2010). The impact of Medicare Part D on medication adherence among older adults enrolled in Medicare-Advantage products. Med Care, 48(5), 409-417. doi: 10.1097/MLR.0b013e3181d68978

Zhang, Y., Lave, J. R., Newhouse, J. P., & Donohue, J. M. (2010). How the Medicare Part D drug benefit changed the distribution of out-of-pocket pharmacy spending among older beneficiaries. J Gerontol B Psychol Sci

Soc Sci, 65(4), 502-507. doi: 10.1093/geronb/gbp111 Zivin, K., Madden, J. M., Graves, A. J., Zhang, F., & Soumerai, S. B. (2009).

Cost-related medication nonadherence among beneficiaries with depression following Medicare Part D. Am J Geriatr Psychiatry, 17(12), 1068-1076. doi: 10.1097/JGP.0b013e3181b972d1

133

APPENDIX

INSTITUTE REVIEW BOARD Documentation

134

135