GENDER AND RACIAL DISPARITIES IN COST-RELATED MEDICATION
NONADHERENCE: THE EFFECT OF MEDICARE PART D AND FACTORS IMPACTING
COMPLIANCE AMONG OLDER WOMEN AND BLACK AMERICANS
By
Louanne Bakk

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR IN PHILOSOPHY
Social Work
2012

ABSTRACT
GENDER AND RACIAL DISPARITIES IN COST-RELATED MEDICATION
NONADHERENCE: THE EFFECT OF MEDICARE PART D AND FACTORS IMPACTING
COMPLIANCE AMONG OLDER WOMEN AND BLACK AMERICANS
By
Louanne Bakk
The purpose of this dissertation is to better understand how cost-related medication
nonadherence (CRN) has been impacted by the implementation of Medicare Part D. CRN can
include not filling, stopping, or skipping doses of a prescription because of cost (Gellad, Haas, &
Safran, 2007; Safran et al., 2005). Research demonstrating the impact of Medicare Part D on
CRN is extremely limited. Studies have primarily investigated the effect of the benefit utilizing
pharmaceutical claims and health care plan data (Chen et al., 2008; Joyce, Goldman, Vogt, Sun,
& Jena, 2009; Lichtenberg & Sun, 2007; Schneeweiss et al., 2009; Yin et al., 2008) and do not
differentiate beneficiaries by demographics or health status. While gender and racial disparities
in noncompliance prior to Medicare Part D clearly existed (Gellad, et al., 2007; Heisler, Wagner,
& Piette, 2005; Klein, Turvey, & Wallace, 2004; Piette, Heisler, & Wagner, 2004a; Soumerai et
al., 2006; Zivin, Ratliff, Heisler, Langa, & Piette, 2010), studies have not examined the
prevalence of racial and gender disparities in CRN since the implementation of Medicare Part D.
This dissertation is comprised of three discrete empirical manuscripts, with introductory
and concluding essays. Cumulative advantage/disadvantage theory was used as a foundation for
understanding the effectiveness of Medicare Part D. Longitudinal and cross-sectional analyses
were conducted using data from the Health and Retirement Study (HRS) and the Prescription
Drug Study (PDS), a subsample of the HRS, to examine racial and gender differences in CRN
before and after the implementation of Medicare Part D and factors associated with the benefit

that can potentially impact adherence. In particular, this study addressed the following questions:
1) To what extent do racial and gender disparities in CRN exist since the implementation of
Medicare Part D? 2) Do the Medicare Part D Low-Income Subsidy (LIS), coverage gap, and
restrictions directly and indirectly affect the relationship between race, gender, and CRN? and 3)
How do factors associated with cumulative advantage/disadvantage theory, specifically
socioeconomic and health status, directly and indirectly affect the relationship between race,
gender, and CRN? The results suggest that older Black Americans and females are more likely to
report CRN before and after Medicare Part D than older Whites and males. Applying for the LIS
increases the risk of CRN and mediates gender differences. Racial disparities in CRN appear to
be driven by having a Medicare Part D plan with a gap in coverage. Poorer health and lower
annual income increases the likelihood of CRN, even after controlling for LIS status and
Medicare Part D’s cost containment provisions. Further, experiencing the coverage gap and
restrictions have a direct effect on nonadherence due to cost. The findings provide important
insights into Medicare Part D’s effectiveness in eliminating racial and gender differentials in
CRN. Implications for practice, policy, education, and future research are discussed.

Copyright by
LOUANNE BAKK
2012

ACKNOWLEDGEMENTS
I feel enormously blessed and grateful for the people in my life that provided
encouragement and support through this long and at times, seemingly-ever-ending process. I
would first like to extend my warmest thanks and appreciation to the members of my dissertation
committee. My committee chair, Amanda T. Woodward, was extremely kind, helpful, patient,
and supportive to me. Her advice and insights were invaluable during my doctoral studies. Rena
Harold was helpful to me in my development as a both researcher and teacher. I always
appreciated her words of encouragement. Clifford Broman’s expertise and feedback were
helpful in this process, and will continue to benefit me as a move towards publication. A special
thanks goes to Ruth Dunkle who I have known since my Master’s program at the University of
Michigan. Ruth has been a fantastic mentor, and has always seemed to find time for me despite
her busy schedule.
I also extend gratitude to my Hartford “buddies,” Tammy and Jean. Without these two
wonderful ladies, this would have been a very lonely and difficult journey. They were always
there to listen to me, and their friendship has meant more to me than they will ever know.
My friends and family were always supportive, kind, and inspiring. I am especially
grateful my sister Janet, and to my friends Elizabeth, Miriam, and Yule. Also, thanks go to Ernie
and Treasure, my feline companions, who faithfully sat with me every day as I wrote.
And finally, words cannot express the gratitude, appreciation, and I love I feel for my
husband, Terry. Terry has been a tremendous support to me through this entire process, and has
been there for me. His encouragement, love, and presence made this hard work possible. I am so
grateful that we can share this wonderful accomplishment together.

v

TABLE OF CONTENTS
LIST OF TABLES ......................................................................................................................... ix
LIST OF FIGURES .........................................................................................................................x
CHAPTER ONE
INTRODUCTION ...........................................................................................................................1
Cumulative Advantage/Disadvantage Theory .....................................................................2
Cumulative Advantage/Disadvantage, Race, Gender, and Medicare Part D ...........4
Overview and Structure of Dissertation ...............................................................................8
Significance of the Study ...................................................................................................10
CHAPTER TWO
RACIAL AND GENDER DISPARITIES IN MEDICATION NONADHERENCE: A
PRE/POST MEDICARE PART D COMPARISON .....................................................................13
Abstract ..............................................................................................................................13
Background ........................................................................................................................15
Literature Review...............................................................................................................17
Cumulative Advantage/Disadvantage and Medicare Part D .................................17
Income and education ................................................................................18
Health status ...............................................................................................19
Medication Access and Adherence ........................................................................20
Pre-Medicare Part D ..................................................................................20
Post-Medicare Part D .................................................................................20
CRN and the Coverage Gap ...........................................................22
Methodology ....................................................................................................................25
Sample....................................................................................................................25
Measures ................................................................................................................27
Dependent variable ....................................................................................27
Independent variables ................................................................................29
Analytic Strategy ...................................................................................................31
Results
....................................................................................................................34
Description of Sample at Baseline .........................................................................34
Bivariate Results ....................................................................................................34
Mixed-Effect Logistic Regression Results ............................................................40
Discussion
....................................................................................................................44
CHAPTER THREE
COST-RELATED MEDICATION NONADHERENCE UNDER MEDICARE PART D AND
THE LOW-INCOME SUBSIDY: A RACIAL AND GENDER COMPARISON ......................56
Abstract
....................................................................................................................56
Background ....................................................................................................................58
Literature Review...............................................................................................................61
Cumulative Advantage/Disadvantage and Medicare Part D .................................61

vi

Income and education ................................................................................63
Health status ...............................................................................................63
Low-Income Subsidy .................................................................................64
Medication Access and Adherence ........................................................................64
Pre-Medicare Part D ..................................................................................64
Post-Medicare Part D .................................................................................66
CRN and the Low-Income Subsidy ...............................................68
Methodology ....................................................................................................................71
Sample....................................................................................................................71
Measures ................................................................................................................74
Dependent variable ....................................................................................74
Independent variables ................................................................................75
Analytic Strategy ...................................................................................................78
Results
....................................................................................................................81
Description of Sample at Baseline .........................................................................81
Bivariate Results ....................................................................................................81
Mixed-Effect Logistic Regression Results ............................................................87
Discussion
....................................................................................................................91
CHAPTER FOUR
MEDICARE PART D COVERAGE GAP AND RESTRICTIONS: RACE, GENDER, AND
COST-RELATED MEDICATION NONADHERENCE............................................................104
Abstract ............................................................................................................................104
Background ..................................................................................................................106
Literature Review.............................................................................................................108
Cumulative Advantage/Disadvantage and Medicare Part D ...............................108
Income, education, and prescription insurance coverage ........................109
Health status .............................................................................................110
Medication Access and Adherence Under Medicare Part D ...............................111
Medicare Part D Cost-Containment Provisions and Adherence ..........................113
CRN and the Coverage Gap .....................................................................115
Coverage restrictions ...............................................................................116
Methodology ..................................................................................................................118
Sample..................................................................................................................118
Measures ..............................................................................................................120
Dependent variable ..................................................................................120
Independent variables ..............................................................................121
Analytic Strategy .................................................................................................124
Results
..................................................................................................................126
Description of Sample at Baseline .......................................................................126
Bivariate Results ..................................................................................................126
Logistic Regression Results .................................................................................132
Discussion
..................................................................................................................136
CHAPTER FIVE
CONCLUSIONS

..................................................................................................................145

vii

Overview
..................................................................................................................145
Summary of Findings.......................................................................................................146
Cumulative Advantage/Disadvantage Theory and CRN .....................................148
Implications for Social Work Practice, Policy, and Education ........................................151
Social Work Practice............................................................................................151
Social Work Policy ..............................................................................................156
Social Work Education ........................................................................................159
Limitations and Future Research .....................................................................................160
Conclusions ..................................................................................................................165
BIBLIOGRAPHY

..................................................................................................................166

viii

LIST OF TABLES
Table 2.1.

Description of Sample of Individuals Age 65 and Over Reporting Medication
Adherence/Nonadherence in 2005 .........................................................................35

Table 2.2.

Bivariate Relationships Between Independent Variables and Medication
Adherence/Nonadherence, 2005 and 2007 ............................................................37

Table 2.3.

Mixed-Effects Logistic Regression Results of Cost-Related Medication
Nonadherence ........................................................................................................41

Table 3.1.

Description of Sample of Individuals Age 65 and Over Reporting Medication
Adherence/Nonadherence in 2005 and Enrolling in Medicare Part D ..................82

Table 3.2.

Bivariate Relationships Between Independent Variables and Medication
Adherence/Nonadherence, 2005 and 2007 ............................................................84

Table 3.3.

Mixed-Effects Logistic Regression Results of Cost-Related Medication
Nonadherence ........................................................................................................88

Table 4.1.

Description of Sample of Individuals Age 65 and Over Reporting Medication
Adherence/Nonadherence Under Medicare Part D or MA Plan in 2007.............127

Table 4.2.

Bivariate Relationships Between Independent Variables and Medication
Adherence/Nonadherence Under Medicare Part D or MA Plan in 2007.............130

Table 4.3.

Logistic Regression Results of Cost-Related Medication Nonadherence Under
Medicare Part D or MA Plan in 2007: Race and Gender ....................................132

Table 4.4.

Logistic Regression Results of Cost-Related Medication Nonadherence Under
Medicare Part D or MA Plan in 2007 ..................................................................133

ix

LIST OF FIGURES
Figure 1.1. Theoretical Model ........................................................................................................7
Figure 3.1: Medicare Drug Benefit Subsidies for Low-Income Beneficiaries, 2010 ...................61

x

CHAPTER ONE
INTRODUCTION
The Medicare Prescription Drug, Improvement, and Modernization Act of 2003 (MMA)
is considered the largest benefit expansion in the history of the Medicare program. The MMA
made substantial changes to the structure of Medicare and provided prescription drug coverage
for beneficiaries under Part D. Medicare Part D was crafted as a response to older adults’ need
for prescription drug insurance. While overall the MMA helped facilitate the purchase of needed
medications, core stipulations of the policy can potentially prevent access to medications because
out-of-pocket expenditure requirements can be substantial. Provisions contained in the Act allow
Medicare Part D plans to utilize coverage restrictions that include limiting the number and/or
type of medications that are paid for by the plan and requiring increased cost-sharing for certain
drugs. Further, Medicare Part D contains a benefit threshold or coverage gap. Entering the gap
means that prescriptions previously covered by the prescription drug plan become the sole
responsibility of the Medicare beneficiary after reaching a drug specified spending limit. There is
concern that these cost-containment provisions can make it difficult for more vulnerable
populations to purchase their medications, particularly older women and Black Americans. This,
in turn, could impact their ability to adhere to their medication regimen.
The purpose of this dissertation is to better understand how Medicare Part D affected
racial and gender disparities in cost-related medication nonadherence (CRN) and explain factors
associated with the benefit that can potentially impact adherence. This chapter briefly outlines
cumulative advantage/disadvantage theory, the theoretical perspective used in this dissertation. It
then discusses the socioeconomic and health implications of cumulative disadvantage for older
women and Black Americans and how adherence can be impacted under the Medicare Part D

1

benefit. The structure of the dissertation and main objectives of each chapter will be explained
followed by a discussion of the significance of the three studies.
Cumulative Advantage/Disadvantage Theory
Cumulative advantage/disadvantage theory can be traced to Robert K. Merton’s (1968a)
introduction of the “Matthew Effect.” This term was developed by Merton, a sociologist, and
named after a passage in the Gospel of Matthew 25:29 (Merton, 1968a). The framework was
developed to provide a basis for understanding inequality, or the skewed system of recognition
and productivity in science (O'Rand, 2003). Merton (1968b) primarily focused on advantage, and
asserted that the scientists who are recognized for exceptional performance early in their careers
produce returns that grow or accumulate over time. They continue to attract additional resources
and rewards such as computers, research assistants, and guidance from colleagues, which
provide them with the opportunity to perform research and receive additional recognition
(Anderson, Long, & Krauze, 1982; Cole & Cole, 1973; Ross & Wu, 1996). The status of
prominent scientists is constantly enhanced, and the system disproportionately advantages those
that already maintain a privileged position. Their work is more highly valued than that of
scientists who are relatively unknown, and the effects accumulate over time and involve
differentiation, or increasing inequality due to the accumulation of resources (Merton, 1968a,
1968b).
Merton’s ideas have been integrated into the field of aging. Cumulative
advantage/disadvantage theory has increasingly been used as a paradigm to describe
stratification and growing trajectories over time (Dannefer, 1987; Farkas, 2003; Kail, Quadagno,
& Keene, 2009; A. E. Wilson, Shuey, & Elder, 2007). The concept is based on the accrual of
advantages and disadvantages over a lifespan, and considers how systematic disadvantages

2

impact later life. It is suggested that advantages and disadvantages not only accrue throughout
the life course but are magnified and can have a feedback effect (Crystal & Shea, 1990a; Ferraro
& Kelley-Moore, 2003). In other words, “the more disadvantages individuals experience, the
more likely they are to accrue subsequent and greater disadvantages” (Kail, et al., 2009, p. 557).
Because of cumulative advantages and disadvantages throughout the life course, heterogeneity
increases with advancing age and inequalities are most pronounced in later life (Crystal & Shea,
1990a, 1990b; Dannefer, 1987; Ross & Wu, 1996). For example, individuals from economically
disadvantaged families tend to have fewer years of education. This creates differences in
employment opportunities, which can lead to lower wages, unemployment, less advantageous
working conditions, and fewer or no accumulated assets, health, or pension benefits. These
disadvantages continue to accrue over the life span (Dannefer, 2003). Lower income due to
limited employment and educational opportunities is often compounded because of poorer health
status (O'Rand & Hamil-Luker, 2005). In this way, economic and health inequalities accumulate
and become magnified in later life (Dannefer, 2003; Sambamoorthi, Shea, & Crystal, 2003).
When considering health status in later life, socioeconomic status is strongly related to
level of functional impairment (Clark & Maddox, 1992) and is a primary determinant of
disparities in health (Williams, 2002; A. E. Wilson, et al., 2007). Consistent with cumulative
advantage/disadvantage theory, studies have found that health trajectories continue to diverge
across the life course (Clark, 1997; Dannefer, 2003; Hatch, 2005; Kelley-Moore & Ferraro,
2004b). Individuals with a lower level of education often have more functional limitations due to
accumulated disadvantages. Level of functioning and the development of acute and chronic
health conditions are often dependent on accrued resources, particularly accumulated income and
financial assets (Crystal & Shea, 1990a).

3

Cumulative Advantage/Disadvantage, Race, Gender, and Medicare Part D
This dissertation focuses on racial and gender disparities in CRN under Medicare Part D.
Older women and Black Americans are significantly impacted by cumulative disadvantage
which can subsequently impact their ability to purchase prescribed medications and maintain
adherence with a medication regimen under the MMA. Cost-containment provisions within the
policy allow Medicare Part D plans to enforce restrictions by refusing to pay for certain
medications or limiting the number, type, and dollar amount of medications that are covered.
Additionally, substantial cost sharing is required once a beneficiary reaches the benefit threshold,
or enters the coverage gap. These provisions, in turn, can increase out-of-pocket prescription
drug expenditures and can disproportionately impact females and Black Americans because of
lower SES (Crystal & Shea, 1990a; Finkle, Hartmann, & Lee, 2007; Wright & Devine, 1994)
and increased chronic conditions (Clark, 1997; Kelley-Moore & Ferraro, 2004a) due to
accumulated disadvantages. Higher expenditures resulting from the Medicare Part D coverage
gap and restrictions can subsequently impact their ability to purchase medications and maintain
compliance (Wei, Akincigil, Crystal, & Sambamoorthi, 2006).
The ability to afford pharmaceuticals and maintain adherence to a prescribed medication
regimen are largely dependent on retirement earnings. Income in later life is essentially based on
pre-retirement experiences, and racial and gender inequities continue to increase with advancing
age (Crystal & Shea, 1990a; Finkle, et al., 2007; Wright & Devine, 1994). Wages for females
and Black Americans have consistently lagged behind those received by White males, and
women and Black Americans are more likely to have been segregated into positions without
retirement or pension benefits (Gonyea & Hooyman, 2005). Since Social Security is based on a
40-year work history, older women and Black Americans are typically subject to lower benefits

4

in later life than White males due to lower wages and fewer years in the labor force (Olson,
1994) and thus have less income available to purchase prescribed medications. Income
inequalities are even more pronounced for Black Americans because of racial discrimination and
the unequal educational, social, and economic opportunities throughout the life course (Calasanti
& Slevin, 2001; Kail, et al., 2009; Quinn, 1996). Furthermore, while poverty in old age is twice
as likely for women as for older men, it is three times higher for Black Americans than Whites
(Finkle, et al., 2007; Gonyea & Hooyman, 2005; Wright & Devine, 1994).
Older women and Blacks Americans’ lower income is typically exacerbated by a higher
prevalence of chronic health conditions. Racial and gender inequities in morbidity exist
throughout the life course and health trajectories continue to diverge with increasing age (Clark,
1997; Kelley-Moore & Ferraro, 2004a). Women are more likely to report functional limitations,
a higher number of chronic conditions, and rate their health as poor or fair than their male
counterparts (Murtagh & Hubert, 2004). While the gender gap in disability is relatively small at
age 65, it widens considerably into older age (M. J. Gibson & Brangan, 1998). Of all racial
groups, health status is poorest for older Black Americans (Collins, Estes, & Bradsher, 2001) and
compared to Whites, older Black Americans report higher rates of hypertension, diabetes,
depression, and difficulty with physical and social functioning (Schoenbaum & Waidmann,
1997; Schoenborn & Heyman, 2009) and have a higher risk of developing cancer, cardiovascular
disease, and stroke (Manton & Stallard, 1997).
As the result of increased morbidity, females and Black Americans are more reliant on
pharmaceuticals (Goulding, 2005). However, their ability to afford prescribed medications is
often compromised because of lower SES due to disadvantages throughout the life course.
Compared to White males, older women and Black Americans spend a higher proportion of their

5

income on pharmaceuticals (Rogowski, Lillard, & Kington, 1997; Sambamoorthi, et al., 2003). It
is expected that they will continue to experience a disproportionately heavier financial burden
under Medicare Part D and will be adversely impacted by the coverage gap and restrictions (Wei,
et al., 2006).
While racial (Bambauer et al., 2007; Gellad, et al., 2007; Heisler, et al., 2005; Klein, et
al., 2004; Piette, Heisler, & Wagner, 2004c; Roth, Esserman, Ivey, & Weinberger, 2009; Roth &
Ivey, 2005; Soumerai, et al., 2006; Steinman, Sands, & Covinsky, 2001; Zivin, et al., 2010) and
gender (Heisler, et al., 2005; Piette, et al., 2004c; Zivin, et al., 2010) disparities in CRN clearly
existed before Medicare Part D, studies have not examined whether differentials in CRN exist
since the implementation of the MMA. This dissertation addresses this gap in research. It builds
on previous studies on racial and gender disparities in CRN by considering how Medicare Part D
has impacted compliance, and explores factors associated with the benefit that can potentially
affect adherence. Based on existing literature and using cumulative advantage/disadvantage
theory as a framework, the specific questions to be explored by this dissertation in three distinct
papers are:
1.

To what extent do racial and gender disparities in CRN exist since the
implementation of Medicare Part D?

2.

How do the Medicare Part D coverage gap, coverage restrictions, and LowIncome Subsidy (LIS) directly and indirectly affect the relationship between
race, gender, and CRN?

3.

How do socioeconomic and health status directly and indirectly affect the
relationship between race, gender, and CRN since the implementation of
Medicare Part D?

6

Figure 1.1 is the theoretical model for this dissertation.
Socioeconomic
Status:
Income
Education
LIS Status

Race

Health Status:
Morbidities
Perceived health

Gender

Cost-Related Medication
Nonadherence

Cost-Sharing
Provisions:
Coverage Gap
Coverage
Restrictions
= direct effect
= indirect/mediating effect
Figure 1.1: Theoretical model
As illustrated in this model, it is expected that race and gender will have a direct effect on
CRN. The Medicare Part D cost-containment provisions, specifically the coverage gap and
restrictions, will have both a direct and indirect effect on CRN. That is, encountering costcontainment provisions will increase CRN and these provisions will mediate the effect of race
and gender on CRN if the coefficients decrease or the relationship becomes nonsignificant with
the addition of these variables into the models. Further, it is anticipated that CRN will be
associated with early disadvantage, measured by income, education, health status, and LIS status.
These factors, in turn, will indirectly affect racial and gender differences in CRN.

7

Overview and Structure of Dissertation
This dissertation is comprised of three discrete empirical papers, with introductory and
concluding essays. It uses data from the Health and Retirement Study (HRS) and the Prescription
Drug Study (PDS), a subsample of the HRS. The PDS was specifically designed to examine
prescription drug utilization as Medicare Part D was implemented. The outcome variable of
interest in all studies is CRN, which includes not filling, stopping, or skipping doses of a
prescription because of cost.
The objectives of the first paper (Chapter Two) are to examine 1) whether racial and
gender disparities in CRN have changed since the implementation of Medicare Part D; and 2)
whether socioeconomic status, health status, and prescription insurance coverage directly and
indirectly affect the relationship between race and gender and CRN over time. The analytic
sample for this longitudinal analysis consists of 3,171 respondents age 65 and older who were
Medicare-eligible, taking at least one prescribed medication, and responded to questions about
adherence or nonadherence in either 2005 or 2007. While gender and racial disparities in CRN
prior to the Medicare prescription drug benefit (Part D) exist, studies have not considered racial
and gender disparities in CRN since the policy’s enactment. Thus, this study allows for the
examination of whether and to what extent nonadherence due to cost has changed for older
women and Black Americans since the implementation of the Part D benefit, and how factors
associated with cumulative disadvantage affect CRN.
The second paper (Chapter Three) uses a similar methodology as Chapter Two. While the
study in Chapter Two included respondents who met study criteria regardless of source of
prescription drug coverage, the analytic sample for Chapter Three only includes the subset of
these respondents who enrolled in Medicare Part D (n=1,466). In addition to examining racial

8

and gender differences in CRN over time among enrollees, this paper examines the direct and
indirect effects of the Low-Income Subsidy (LIS). Examining disparities among enrollees can
help us understand how Medicare Part D enrollment affects CRN for older women and Black
Americans over time and the impact of the LIS on adherence, a key provision of the MMA.
The third paper (Chapter Four) focuses specifically on the impact of the Medicare Part D
coverage gap and restrictions on racial and gender differences in CRN. The objectives of the
fourth chapter are to determine: 1) to what extent race and gender impact the likelihood of CRN
under Medicare Part D plans; and 2) to what extent the Medicare Part D coverage gap and
coverage restrictions directly and indirectly affect the relationship between race, gender,
socioeconomic and health status and CRN. The 2006 wave of the HRS and data from the 2007
wave of the PDS were utilized. The analytic sample for this cross-sectional analysis consists of
1,353 respondents age 65 and older who were enrolled in Medicare Part D, taking at least one
prescribed medication, and responded to questions about adherence or nonadherence in 2007.
The results of this research can help inform Medicare policy by identifying whether older women
and Black Americans are disproportionately nonadherent under the Part D benefit, and how the
coverage gap and restrictions directly and indirectly impact CRN.
The final chapter links the previous chapters of the dissertation together, and summarizes
and interprets the findings. Conclusions are drawn in relation to racial and gender differences in
CRN and Medicare Part D, and implications for social work practice, policy, and education are
discussed. The chapter concludes by summarizing the limitations of the study and presenting
recommendations for future research.

9

Significance of the Study
This dissertation will address how Medicare Part D has impacted nonadherence,
particularly for older women and Black Americans. CRN with a prescribed medication regimen
can have negative health ramifications. Individuals who restrict medication use because of cost
have significantly worse health outcomes than those who are adherent, and in one study were
twice as likely to exhibit a deterioration in overall health status over time (Heisler et al., 2004).
Prior to Medicare Part D, individuals with coverage restrictions were twice as likely to enter a
nursing home, and more likely to use hospital and outpatient services than those without
restrictions (Soumerai, McLaughlin, Ross-Degnan, Casteris, & Bollini, 1994; Soumerai, RossDegnan, Avorn, McLaughlin, & Choodnovskiv, 1991). In a recent study involving mental health
patients, approximately one quarter of all patients who were unable to access medications due to
Medicare Part D coverage restrictions experienced an adverse clinical event, and in many cases,
required an emergency room visit (West et al., 2007). These findings suggest that nonadherence
not only affects the health status and well-being of older individuals, but results in an increase in
overall health care costs and utilization.
While the body of knowledge pertaining to how Medicare Part D has impacted
medication access and adherence is slowly growing, there are several gaps that limit our
understanding. This study will begin to address these limitations, and makes the following
contributions to our understanding of gender and racial disparities and our ability to promote
equitable access to prescription medications.
First, while gender and racial disparities in noncompliance prior to Medicare Part D
existed, studies have not examined the prevalence of racial and gender disparities in
nonadherence due to cost since the implementation of the MMA. This study is the first to

10

examine whether and to what extent the policy has affected racial and gender disparities in
nonadherence.
Second, previous research primarily utilized pharmaceutical claims data or membership
databases associated with Medicare drug plans in an attempt to understand how the Medicare
Part D coverage gap has impacted medication use. There are limitations with the methodology
utilized in these past studies. First, nonadherence was estimated to occur based on a reduced
number of pharmaceutical claims as opposed to being reported by beneficiaries. In other words,
studies speculated without clear evidence that lower prescription drug expenditures were
associated with noncompliance due to increased cost-sharing resulting from the coverage gap.
Second, findings were based on aggregate claims and did not differentiate between beneficiaries
by accounting for demographics or health status. It is therefore impossible to determine whether
older women and Black Americans are disproportionately experiencing CRN as the result of the
coverage gap. This study not only aims to understand whether racial and gender disparities in
CRN occur under Medicare Part D, but it is the first to investigate directly whether the coverage
gap and restrictions contribute to nonadherence among older females and Black Americans.
Third, the MMA stipulated that prescription drug cost-sharing assistance be provided to
beneficiaries that qualify for the LIS. This study is significant in that it is the first to examine
how the LIS directly and indirectly impacts nonadherence among older women and Black
Americans. This knowledge can inform policymakers about the impact of the LIS on reduced
medication nonadherence.
This dissertation has distinct implications for social work practice and policy, and aims to
provide information that can help improve the health and well-being of the aging population, in
particular, older women and Black Americans. It will enhance our understanding of whether the

11

Medicare Part D coverage gap and restrictions serve as barriers for cumulatively disadvantaged
populations and continue to perpetuate health and economic disparities. Understanding how the
Medicare prescription drug benefit affects adherence among older women and Black Americans
can help practitioners recognize that clients may need additional resources and assistance in
order to avoid nonadherence, which can potentially avert adverse health events and improve
well-being.
This topic is particularly relevant given provisions contained within the Patient Protection
and Affordable Care Act, or PPACA, of 2010. While the PPACA increased coverage to
beneficiaries reaching the coverage gap, the provision will not be fully implemented until 2019
(Kaiser Family Foundation, 2009c). For example, beneficiaries received a 50 percent discount
for some brand name medications in 2011, and this discount will gradually increase to 75 percent
by 2020. Affording medications can still be problematic for older women and Black Americans
because the legislation does nothing to control the increasing cost of medications or Medicare
drug plan premium and copayment amounts. Additionally, restrictions will continue to impede
access to prescribed medications and if current patterns continue, will be increasingly
problematic in the future (Donohue, Huskamp, & Zuvekas, 2009; Polinski, Mohr, & Johnson,
2009). This study can be utilized to inform policymakers on how differential life experiences can
impact medication compliance under the MMA. Findings will help substantiate the need to
account for race and gender when evaluating policy alternatives in order to promote more
equitable access to prescription medications and avoid disparate heath outcomes.

12

CHAPTER TWO
MANUSCRIPT ONE
ABSTRACT
Purpose: While gender and racial disparities in cost-related medication nonadherence (CRN)
existed prior to the Medicare prescription drug benefit (Part D), studies have not examined the
prevalence of racial and gender disparities in CRN since the policy’s enactment. The objectives
of this longitudinal study are to determine: 1) whether racial and gender disparities in CRN have
changed since the implementation of Medicare Part D; and 2) whether socioeconomic status,
health status, and prescription insurance coverage directly and indirectly effect the relationship
between race, gender, and CRN over time. Methods: This study utilizes secondary data from the
2004 and 2006 waves of the Health and Retirement Study (HRS) and data from the 2005 and
2007 waves of the Prescription Drug Study (PDS), a subsample of the HRS, to evaluate changes
in CRN before and after Medicare Part D. The analytic sample consisted of 3,171 respondents
age 65 and older who were Medicare-eligible and taking at least one prescribed medication. The
outcome variable of interest was CRN, which included not filling, stopping, or skipping doses of
a prescription because of cost. The main predictors were race (Black and White) and gender
(male and female). Covariates included demographics, socioeconomic status, health status, and
type of prescription drug coverage. The analysis was conducted using mixed-effects logistic
regression models. Results: Results indicated that older Black Americans and females were
significantly more likely to report CRN both before and after Medicare Part D compared to older
Whites and men. While the rate of CRN for females significantly decreased over time in
comparison to males, there was no change in the rate for Black Americans relative to Whites.
The introduction of variables related to socioeconomic status, health status, and prescription

13

insurance coverage had little mediating effect on these findings. Health status, annual income,
education, and prescription insurance coverage, however, had a direct effect on CRN over time.
Conclusions: Despite the implementation of Medicare Part D, racial and gender disparities in
CRN continued to persist. In comparison to older males, older females had greater improvement
over time. Older adults with poorer health, a lower annual income, or less generous or no
prescription drug coverage had a greater likelihood of CRN both before and after Medicare Part
D.

14

CHAPTER TWO
RACIAL AND GENDER DISPARITIES IN MEDICATION NONADHERENCE: A
PRE/POST MEDICARE PART D COMPARISON
Background
Prescription drugs are an essential part of medical treatment and play an increasingly
important role in health care. When Medicare was initially enacted in 1965, prescription drugs
were less central to health care. Today, older adults are more likely to suffer from chronic
conditions, such as depression, diabetes, high blood pressure, or arthritis, that are treated through
the use of pharmaceuticals (Medicare Payment Advisory Commission, 2000). As a result,
prescription drug use and expenditures by older adults have grown substantially. Between 1992
and 2000, the average number of prescriptions utilized by older adults increased 45% and it is
projected that trend will continue (Families USA, 2000).
Cost can be a formidable barrier in medication access. Retail prescription drug prices
have increased an average of 6.9% annually from 1990 to 2007, far outpacing the rise in inflation
(Kaiser Family Foundation, 2008). Older adults are disproportionately impacted by the
increasing costs of prescription medications because while they represent 13% of the general
population, they account for 34% of all medications that are dispensed and 42% of all
prescription drug expenditures (Families USA, 2000). As drug costs and out-of-pocket
medication expenses rise, adherence tends to fall (Mojtabai & Olfson, 2003; Safran, et al., 2005).
Older adults – particularly individuals with several chronic conditions – often become
nonadherent when they are unable to afford medications (Heisler, et al., 2004; Neuman et al.,
2007; Soumerai & Ross-Degnan, 1999).
The Medicare Prescription Drug, Improvement, and Modernization Act of 2003 (MMA)
was passed in an effort to increase access to prescription drugs for older adults with an emphasis

15

on the need to provide medications to more vulnerable populations, particularly poor and
chronically-ill older individuals (Bass, 2005). To respond to the need to provide older adults with
drug coverage, the MMA added the Part D benefit to Medicare (Madden et al., 2008; Mayes,
2005). Part D includes a number of cost-containment provisions such as cost-sharing
requirements, plan premiums and a time period, known as the coverage gap, when beneficiaries
are responsible for 100 percent of their drug costs. Cost-sharing provisions stipulate that the
beneficiary is required to pay a fixed amount for each prescription or copayment, a proportion of
the medication price or coinsurance, and/or an annual payment before the plan will provide
coverage or deductible. Plan premiums require that individuals pay a monthly amount to receive
coverage and averaged $40.72 in 2011, an increase of 57% since Medicare Part D was
implemented in 2006 (Hoadley, Summer, Hargrave, Cubanski, & Neuman, 2010). There is
concern that these cost-containment provisions can inhibit the use of medications for older
women and Black Americans. Compared to older men and Whites, older women and Black
Americans tend to have a greater number of chronic health problems (Clark, 1997; Kelley-Moore
& Ferraro, 2004a) and are, therefore, more reliant on prescription medications (Rogowski, et al.,
1997; Sambamoorthi, et al., 2003). This means that older women and Black Americans tend to
have higher copayments and are more likely to reach the coverage gap each year. Combined
with their overall lower socioeconomic status, therefore, older women and Blacks are less able to
afford out-of-pocket expenditure requirements (Wei, et al., 2006) and more likely to have
difficulties adhering to their medication regimen.
Previous research has shown the connection between cost-related medication
nonadherence (CRN) and cost-containment provisions similar to those present in Medicare Part
D plans (Adams, Soumerai, & Ross-Degnan, 2001; T. B. Gibson, Ozimkowski, & Goetzel, 2005;

16

Tseng, Brook, Keeler, Steers, & Mangione, 2004). CRN can include not filling, stopping, or
skipping doses of a prescription because of cost. Prior to the MMA, racial and gender disparities
in CRN clearly existed (Heisler, et al., 2005; Klein, et al., 2004). Medicare Part D may continue
to impede access to medication because the benefit’s cost-containment provisions can increase
out-of-pocket expenditures (Hsu et al., 2008). These provisions, in turn, can disproportionately
impact females and Black Americans who tend to be of lower socioeconomic status and are also
living with more chronic conditions both of which subsequently affect their ability to maintain
adherence (Wei, et al., 2006). To date, studies have not examined the prevalence of racial and
gender disparities in CRN since the implementation of the Medicare prescription drug benefit.
This longitudinal study uses a cumulative advantage/disadvantage framework to build upon prior
research by examining whether and to what extent racial and gender disparities in CRN have
changed since the implementation of Medicare Part D.
Literature Review
Cumulative Advantage/Disadvantage and Medicare Part D
Cumulative advantage/disadvantage theory is useful as a foundation for understanding
and evaluating the effectiveness of Medicare Part D. The theory has increasingly been used as a
paradigm in the field of aging to describe stratification and growing trajectories over time
(Dannefer, 1987; Farkas, 2003; Kail, et al., 2009; A. E. Wilson, et al., 2007). Systematic
disadvantages have a considerable impact on later life, and it is suggested that they not only
accrue throughout the life course but are magnified and can have a feedback effect (Crystal &
Shea, 1990a; Ferraro & Kelley-Moore, 2003). In other words, “the more disadvantages
individuals experience, the more likely they are to accrue subsequent and greater disadvantages”
(Kail, et al., 2009, p. 557). For example, individuals from economically disadvantaged families

17

are more likely to have fewer years of education. This creates differences in employment
opportunities, and subsequently places them at greater risk for incurring lower wages,
unemployment, and less advantageous working conditions. Lower income due to limited
employment and educational opportunities is often compounded because of poorer health status,
lesser or no health insurance, and greater health care expenditures (Sambamoorthi, et al., 2003).
Because of cumulative advantages and disadvantages throughout the life course, heterogeneity
increases with age, and inequalities are greatest in later life (Crystal & Shea, 1990a; Dannefer,
1987; Ross & Wu, 1996). For these reasons, the aged should not be stereotyped (Butler, 1985) or
treated as a homogeneous group (Neugarten, 1982) when deliberating social policy initiatives
intended to target older individuals. Inequities resulting from disadvantages throughout the life
course need to be considered in order to ensure equitable access to benefits.
This study will focus on racial and gender disparities in CRN before and after the
implementation of the Medicare Part D benefit. As the result of cumulative advantages and
disadvantages throughout the life course, it is anticipated that higher out-of-pocket expenditures
will be related to personal characteristics associated with early disadvantage, or income,
education, and health status. These factors, in turn, can potentially impact CRN under the
benefit.
Income and education. The ability to afford pharmaceuticals and maintain adherence to
a prescribed medication regimen are largely dependent on retirement earnings. Income in later
life is essentially based on pre-retirement experiences, and racial and gender inequities continue
to increase with advancing age (Crystal & Shea, 1990a). Wages for females and Black
Americans have consistently lagged behind those received by White males, and women and
Black Americans are more likely to have been segregated into positions without retirement or

18

pension benefits (Gonyea & Hooyman, 2005). Since Social Security is based on a 40-year work
history, older women and Black Americans are typically subject to lower benefits in later life
than White males due to lower wages and fewer years in the labor force (Olson, 1994) and thus
have less income available to purchase prescribed medications. Catastrophic changes such as
widowhood – which is more common in the lives of women (Moody, 2009) – can be particularly
difficult because older females and Black Americans are in a disadvantaged position. Women
and Black Americans not only enter old age poorer than men and Whites (Finkle, et al., 2007;
Gonyea & Hooyman, 2005; Wright & Devine, 1994) but can become poorer with age as a result
of widowhood (Minkler & Stone, 1985). Further, income inequalities are even more pronounced
for Black Americans because of racial discrimination and the resulting unequal educational and
economic opportunities throughout the life course (Kail, et al., 2009).
Health status. Lower income is typically exacerbated by a higher prevalence of chronic
health conditions among older women and Black Americans. Racial and gender inequities in
morbidity exist throughout the life course and health trajectories continue to diverge with
increasing age (Clark, 1997; Kelley-Moore & Ferraro, 2004a). Women are more likely to report
functional limitations, a higher number of chronic conditions, and rate their health as poor or fair
than their male counterparts (Murtagh & Hubert, 2004). Of all racial groups, health status is
poorest for older Black Americans (Collins, et al., 2001) and compared to Whites, older Black
Americans report higher rates of hypertension, diabetes, depression, and difficulty with physical
and social functioning (Schoenbaum & Waidmann, 1997; Schoenborn & Heyman, 2009).
As the result of increased morbidity, females and Black Americans are more reliant on
pharmaceuticals (Goulding, 2005). However, their ability to afford prescribed medications is
often compromised because of lower socioeconomic status. Compared to White males, older

19

women and Black Americans spend a higher proportion of their income on pharmaceuticals
(Rogowski, et al., 1997; Sambamoorthi, et al., 2003). For these reasons, it is expected that they
will continue to experience a disproportionately heavier financial burden under Medicare Part D
(Wei, et al., 2006). This, in turn, can subsequently affect their ability to purchase prescribed
medications and maintain adherence.
Medicare Access and Adherence
Pre-Medicare Part D. Prior to Medicare Part D, CRN was estimated to occur in one
fourth of all older adults (Safran, et al., 2005); those without prescription drug coverage or with
limited coverage were significantly more likely to be noncompliant because of cost (Blustein,
2000; Fillenbaum, Hanlon, Corder, Brock, & et al., 1993; Rogowski, et al., 1997). Nonadherence
due to cost was greater among older women than men (Heisler, et al., 2005; Piette, et al., 2004c;
Zivin, et al., 2010). Race was an even stronger predictor of CRN than gender. According to one
study, older Black Americans were 1.5 times more likely to report nonadherence to a medication
regimen than Whites (Gellad, et al., 2007), and numerous studies have found that Black
Americans were more likely than Whites to forgo or delay taking prescribed medications
because of cost (Bambauer, et al., 2007; Klein, et al., 2004; Roth, et al., 2009; Roth & Ivey,
2005; Soumerai, et al., 2006; Steinman, et al., 2001).
Post-Medicare Part D. When analyzing the MMA from certain economic perspectives,
the policy can appear highly successful. In 2008, 90 percent of Medicare’s 45 million
beneficiaries had prescription drug coverage (Joyce, et al., 2009) as compared to 76 percent in
2006 (Levy & Weir, 2009). Further, it is estimated that out-of-pocket spending on prescription
drugs decreased between 13 and 18.4 percent even as the number of medications used by older
adults increased from 5.9 to 12.8 percent since the implementation of the MMA (Chen, et al.,

20

2008; Joyce, et al., 2009; Lichtenberg & Sun, 2007; Schneeweiss, et al., 2009; Yin, et al., 2008).
However, these findings do not take into consideration the implications of the cost-containment
provisions – or cost-sharing requirements, plan premiums, or coverage gap – of the MMA, and
the impact of those provisions on cumulatively disadvantaged beneficiaries. In other words, there
was no differentiation of groups reported in these findings; beneficiaries were regarded as a
homogeneous group that gained access to a prescription drug benefit and lowered their
medication expenditures. It is important to recognize the potential implications of Medicare Part
D cost-sharing provisions on older females and Black Americans. For example, one implication
of the cost-sharing requirements is that access to medications is not in fact the same for all
beneficiaries. Those with lower income and increased morbidity, such as is often experienced by
older women and Black Americans, may be negatively and disproportionately influenced by
cost-sharing requirements both because of their higher medication expenses and their limited
income. This, in turn, can impact their ability to access medication and maintain adherence.
Thus, existing research on the impact of MMA does not recognize the differential consequences
of these provisions
Research examining changes in CRN before and after Medicare Part D is extremely
limited. One of the main reasons for this limitation is the restrictions placed on the availability of
data. The Centers for Medicare and Medicaid Services has historically been a primary source of
data for researchers. However, a provision contained in the MMA stipulated that private health
care plans were not required to release Medicare Part D claims data for nonpayment-related
purposes, which included research, public health purposes, and health care analysis (Crystal,
2008, November; Stuart, 2008). This stipulation largely accounts for the limitation with existing

21

research. Fortunately, this was modified in 2009, and data are currently in the process of being
released.
When considering studies conducted after the implementation of Medicare Part D, Safran
et al. (2009) reported that for older individuals with no or a meager drug benefit, enrollment
resulted in lower out-of-pocket prescription drug expenditures, higher medication utilization, and
overall lower CRN. However, low-income, chronically ill beneficiaries were significantly more
likely to report CRN and difficulty affording their medications. Similarly, Madden et al. (2008)
found that while there was a modest but significant decrease in overall CRN, no decrease was
observed among individuals in fair to poor health, with at least four morbidities, or having an
annual income less than $25,000. Although differences associated with race or gender were not
analyzed in either study, the findings shed some light on the experiences of these groups.
Because older women and Black Americans are more likely to be low income, rate their health
as poor or fair, and experience higher rates of chronic illness and health problems, these studies
suggest that the benefit may have little or no impact on CRN for these groups.
While previous studies have not specifically examined the effect of the Medicare drug
benefit on racial or gender disparities in CRN, research suggests ethnic disparities continue to
exist. One study conducted after the implementation of Medicare Part D, for example, found that
Hispanics were more likely to report CRN than non-Hispanics. Further, CRN was significantly
greater for non-Hispanic Blacks than non-Hispanic Whites. When examining gender, both
Hispanic and non-Hispanic females were more likely to report nonadherence due to cost than
males (Frankenfield, Wei, & Anderson, 2010).
CRN and the Coverage Gap. Other studies focused specifically on CRN resulting from
the Medicare Part D coverage gap. Provisions contained within the MMA stipulated that once a

22

beneficiary reaches the benefit threshold, known as the coverage gap, they are responsible for
100% of their prescription drug costs until spending approximately $4,550 out-of-pocket for
prescription medications. Overall, reaching the coverage gap resulted in lower drug consumption
(Pedan, Lu, & Varasten, 2009; Sun & Lee, 2007; Zhang, Donohue, Newhouse, & Lave, 2009)
and decreased adherence (Hsu, et al., 2008) because of increased out-of-pocket expenditure
requirements. Furthermore, Hsu et al. (2008) found that 36% percent of respondents
encountering the coverage gap stopped taking their medication, reduced the recommended
dosage, or switched to an alternate drug.
While research has not specifically examined the impact of the Medicare Part D coverage
gap on older women and Black Americans, beneficiaries with certain chronic illnesses – such as
depression and diabetes – have a greater risk of this cost-sharing provision because they spend
considerably more on medications and are, therefore, at greater risk of CRN (Hoadley, Hargrave,
Cubanski, & Neuman, 2008; Stuart, Simoni-Wastila, & Chauncey, 2005). Fifty-one percent of
older diabetics who reached the coverage gap or benefit threshold had a decline in out-of-pocket
spending on medications, suggesting that they may have become noncompliant (Hoadley, et al.,
2008). In comparing older adults with and without depressive symptoms, individuals with
depressive symptoms were more likely to report CRN and forego basic needs to pay for
medications following the implementation of Medicare Part D (Zivin, Madden, Graves, Zhang,
& Soumerai, 2009). Because older women and Black Americans experience higher rates of
diabetes (Gellad, Huskamp, Phillips, & Haas, 2006; Schoenborn & Heyman, 2009) and a greater
number of depressive symptoms (Skarupski, Mendes de Leon, Barnes, & Evans, 2009), they may
face a greater risk of CRN under Medicare Part D.

23

Overall, studies suggest that gender and racial disparities in CRN prior to Medicare Part
D existed. However, research demonstrating the impact of Medicare Part D on disparities in
nonadherence due to cost is limited. Studies have primarily investigated the effect of the
Medicare prescription drug benefit utilizing pharmaceutical claims and health care plan data and
do not differentiate beneficiaries by demographics. While a few studies have used nationally
representative datasets to examine CRN, racial and gender differences in change in medication
adherence before and after Medicare Part D have not been examined. Further, the one study
examining disparities in CRN after the implementation of the benefit was cross sectional as
opposed to longitudinal. Thus, research has not investigated whether racial and gender disparities
in CRN have changed since the implementation of the MMA; therefore, the policy’s
effectiveness in eliminating differentials is currently unknown.
This study will build on existing knowledge in a number of ways. First, it uses data from
a nationally representative longitudinal dataset specifically designed to examine prescription
drug utilization as Medicare Part D was implemented. This allows for the examination of gender
and racial differences in CRN both before and after Medicare Part D. Second, this study
examines changes in CRN over time for older women and Black Americans, thus expanding our
knowledge of how racial and gender disparities in CRN have increased or decreased since the
implementation of Medicare Part D. Third, the study examines whether and to what extent lower
socioeconomic status, poorer health status, and little or no prescription insurance coverage
directly and indirectly affect the relationship between race and gender and CRN over time,
specifically before and after the implementation of Medicare Part D. This furthers our
understanding of how factors associated with cumulative disadvantage can impact likelihood of
nonadherence due to cost.

24

Based on existing literature, there are four hypotheses in this study:
1)

As compared to older males, older females are more likely to report CRN over
time and less likely to report a decline in CRN over time.

2)

As compared to older Whites, older Black Americans are more likely to report
CRN over time and less likely to report a decline in CRN over time.

3)

Socioeconomic status, health status, and prescription insurance coverage will
mediate the effect of racial and gender differences in CRN over time.

4)

Lower socioeconomic status, poorer health status, and less generous or no
prescription insurance coverage will increase the likelihood of CRN over time.
Methodology

Sample
This study utilizes secondary data from the 2004 and 2006 waves of The University of
Michigan (U-M) Health and Retirement Study (HRS) as well as data from the 2005 and 2007
waves of the HRS Prescription Drug Study (PDS) to conduct a longitudinal analysis. Launched
in 1992, the HRS is a nationally representative population study of more than 20,000 Americans
over the age of 50 designed to assess the relationship between health and retirement. The core
HRS survey collects information pertaining to physical and mental health, insurance coverage,
financial status, family support systems, retirement situation, and work status (Juster & Suzman,
1995). The primary mode of data collection is by telephone; however, in situations where health
limitations would make the hour-plus telephone session difficult or where there is no telephone
in the household, the survey is conducted in the respondent’s home (HRS, 2011). Primarily
funded through the National Institute on Aging (NIA U01AG009740), the HRS is designed,
administered, and conducted by the Institute for Social Research (ISR) at the U-M. The overall

25

interview response rate for the core 2004 HRS was 86.2% and for the core 2006 HRS, the overall
response rate was 88.5% (Ofstedal, 2009, June).
The PDS is a subsample of the HRS, and is a two-wave mail survey conducted in 2005 and
2007. It was specifically designed to examine prescription drug utilization as Medicare Part D
was implemented. The baseline wave was administered pre-Medicare Part D in 2005 by the ISR,
and captured information pertaining to prescription drug use, coverage, and knowledge of the
Medicare prescription drug benefit. The second wave was conducted post-Medicare Part D in
2007, and captured similar information as well as data pertaining to the impact of the MMA costcontainment provisions such as the coverage gap and restrictions. The sample was drawn from
respondents who participated in the HRS in 2004. To be eligible for inclusion in the PDS,
respondents needed to be age 65 or older in 2007 (born in 1942 or later), or have Medicare or
Medicaid coverage at some time between 2002 and 2004. Because the Consumption and
Activities Mail Survey (CAMS) was fielded at the same time as the PDS, approximately 40% of
eligible participants were excluded because of their involvement in the CAMS. However, the
exclusion did not affect the representation of the PDS sample because CAMS participants were
randomly selected (J. Faul, Personal Communication, January 4, 2012). Of those who were
eligible for inclusion in the PDS, 4,684 individuals or 88.1% completed the PDS in 2005, and
3,536 persons or 74% responded to the survey in 2007 (HRS, 2011). Given that this is a
relatively new topic area, there are no reliability or validity measures associated with the PDS (J.
Faul, Personal Communication, February 3, 2012).
To be included in the analytic sample for this study, respondents were required to 1) be
either non-Hispanic White/Caucasian or non-Hispanic Black/African American; 2) be age 65 or
older in 2005 and therefore eligible for Medicare when the PDS was administered; 3) have

26

reported taking prescription medications in 2005 or 2007; and 4) have responded to questions
about prescription drug adherence in the 2005 or 2007 PDS. Of the 4,808 responding to the
2005 or 2007 PDS, 4,248 were either non-Hispanic White/Caucasian or non-Hispanic
Black/African American; 3,606 were age 65 and over; 3,323 reported taking prescription
medications; and 3,181 reported data on prescription drug adherence or nonadherence. Thus, a
total 3,181 respondents to the 2005 (n=2,863) or 2007 (n=2,301) PDS were included in the final
analytic sample.
The combination of HRS and PDS provides an excellent source for testing the proposed
hypotheses because indicators of cumulative advantage/disadvantage that could decrease or
increase older adults’ risk of CRN are available through these datasets. Variables from the 2005
and 2007 PDS included CRN, type of prescription drug insurance coverage, and Medicare Part D
enrollment status. All other variables were obtained from the 2004 and 2006 RAND HRS data
files. These data files contain a broad range of measures across HRS waves, and are developed
and maintained by the RAND Center for the Study of Aging, with support provided by the NIA
and Social Security Administration (RAND Center for the Study of Aging, 2010).
In regards to Institutional Review Board (IRB) approval, this study relied exclusively on
secondary data to complete all analysis. The HRS and PDS files contain de-identified individual
data that are available to the public. Therefore, the IRB at Michigan State University (MSU)
does not require human subjects review.
Measures
Dependent variable. A summary indicator of any CRN was constructed as the
dependent variable. In the 2005 and 2007 PDS, CRN was evaluated using questions about the
following three behaviors: 1) not filling a new prescription because of cost; 2) stopping taking a

27

medication because of cost; or 3) skipping doses of a medication to save money. If the
respondent indicated that they engaged in any of these behaviors, they were considered
nonadherent in this study. Earlier research typically used a single measure to assess CRN, and
focused solely on whether respondents failed to fill a prescription in the past year because of
cost. However, this measure did not assess for multiple ways CRN could occur. Thus, the
additional questions were developed to help monitor and identify changes in CRN, and have
demonstrated test-retest reliability (Pierre-Jacques et al., 2008) and construct validity (Soumerai,
et al., 2006). This composite variable approach was originally developed by Safran et al. (2003)
based on a series of validated survey questions concerning CRN to prescription regimens,
confirmed using factor analysis techniques, and used as a main outcome in subsequent research
examining drug nonadherence practices of older adults (Bambauer, et al., 2007; Gellad, et al.,
2007; Madden, et al., 2008; Safran, et al., 2005; Soumerai, et al., 2006; I. B. Wilson, Rogers,
Chang, & Safran, 2004; Zivin, et al., 2009). Because all three measures of CRN, specifically
filling, stopping, or skipping doses of a prescription because of cost, were highly correlated with
comorbidity, self-reported health status, and socioeconomic status (Bambauer, et al., 2007;
Safran, et al., 2005; Safran, et al., 2003; Soumerai, et al., 2006), it is important that they be
included in this study. Older females and Black Americans are more likely to have a lower
income and multiple health problems due to disadvantages throughout the life course and thus,
accounting for all behaviors can further understanding of CRN before and after the
implementation of Medicare Part D.
In this study, a dichotomous variable indicating any nonadherence or no nonadherence to
the three CRN behaviors in the PDS – not filling a new prescription because of cost, stopping
taking a medication because of cost, or skipping doses of a medication to save money – was

28

created for each time point, or for both the 2005 and 2007 measures. The CRN variable was
initially incorporated as a continuous variable indicating a sum total of the three behaviors.
However, results were significantly skewed because the majority of the sample were adherent to
their medication regimen and similar to previous research, the variable was dichotomized
(Bambauer, et al., 2007; Gellad, et al., 2007; Madden, et al., 2008; Safran, et al., 2005; Soumerai,
et al., 2006; I. B. Wilson, et al., 2004; Zivin, et al., 2009). Using this global, dichotomous
measure is common in existing literature, and will help provide a meaningful comparison to
previous research.
Independent variables. The main predictors of interest were race (Black and White) and
gender (female and male). A time variable was also created to evaluate CRN over time (2005
and 2007) using the “reshape” command in Stata version 11 and was used in all regression
analyses as a main predictor. The “reshape” command converted the data to a long format and
provided a single variable that was used to measure CRN at both time points. Using literature
pertaining to cumulative advantage/disadvantage as a guide, other independent variables
included socioeconomic characteristics, health status, and variables related to prescription drug
insurance coverage. Additionally, age and marital status were used as control variables.
Socioeconomic characteristic indicators included attained education and annual
household income. Disadvantages can accumulate throughout the life course as the result of
decreased educational attainment and lower income, and result in decreased ability to afford
prescription medications and maintain adherence. The level of attained education was a
continuous variable ranging from 0 to 17 or more years of education. Cumulatively
disadvantaged individuals tend to have lower levels of education. Subsequently, they are less
likely to have prescription drug coverage and the financial resources needed to purchase

29

prescribed medications. Utilizing this measure will help clarify how educational attainment can
impact CRN. Additionally, it can further our understanding of how education can indirectly
impact CRN among older women and Black Americans. Annual household income was a
continuous variable ranging from $0 to $2,761,657. Utilizing this variable can help understand
how income can impact CRN, and whether the poor and near-poor may be more likely to forego
medications because of cost.
Health status variables included number of chronic medical conditions and self-reported
health status. Increased morbidity is associated with cumulative disadvantage and increased
reliance on pharmaceuticals, particularly among older women and Black Americans. Thus, it is
important to understand how health status impacts CRN. First, a continuous variable indicating
the number of chronic medical conditions was used. Respondents were asked, “Has a doctor ever
told you that you have [name of chronic medical condition]” or “our records from your last
interview show that you have [name of chronic medical condition].” This question was asked for
a series of eight chronic medical conditions and included: high blood pressure, diabetes, cancer,
lung disease, heart disease, stroke, psychiatric problems, and arthritis. The second health-related
variable included self-reported health status. Respondents were asked, “Would you say your
health is excellent, very good, good, fair, or poor?” Responses ranged from 0 = “excellent” to 4 =
“poor” with a higher score indicating that the respondent perceived their health status as being
more poor.
Two prescription insurance coverage variables were included. Having less generous
prescription drug coverage is related to cumulative disadvantage, and can impact adherence.
Respondents were asked, “Which of the following best describes how your insurance works
when you buy prescription drugs?” Five levels of drug coverage response options were defined:

30

no coverage/pay full cost, insurance pays part of cost, prescription discount card, full
coverage/pay nothing, and some other type of coverage. In the 2007 PDS, respondents were also
asked, “Are you currently enrolled in a Medicare Part D plan or a Medicare HMO or Advantage
plan that helps pay for prescription drugs?” Possible responses response options included ‘yes,’
‘no,’ and ‘don’t know.’ Including the ‘don’t know’ option can help illustrate the confusion
associated with the Medicare Part D benefit (Heiss, McFadden, & Winter, 2006; Hsu, et al.,
2008). A substantial number of respondents chose this option.
Finally, other control variables included age and marital status. Age was a continuous
variable ranging from age 65 to 101. Marital status was a seven-category variable: married,
married with spouse absent, partnered, separated, divorced, widowed, and never married.
Analytic Strategy
Univariate descriptive statistics summarized all measures at baseline. Cross-tabulations
were conducted to determine the bivariate association between respondent characteristics and
CRN at baseline and in 2007. The Rao-Scott chi-square, a design-adjusted version of the Pearson
chi-square, was used to test the significance of the association (Rao & Scott, 1981). Mixedeffects logistic regression models were used for multivariate analysis to estimate racial and
gender differences in CRN over time, and to examine the direct and indirect effect of
socioeconomic and demographic characteristics, health status, and prescription insurance
coverage on the relationship between race and gender and CRN.
Mixed regression models are useful for this type of analysis for several reasons
(Blackwell, Mendes de Leon, & Miller, 2006; Rabe-Hesketh & Skrondal, 2008). First, they do
not require that respondents be measured on the same number of time points. In other traditional
methods such as repeated-measures ANOVA, cases are dropped if some measurements are

31

missing (Blackwell, et al., 2006). This is important because some participants did not respond to
questions pertaining to CRN in both surveys. In particular, 72.9% of respondents indicated
adherence or nonadherence in both the 2005 and 2007 PDS. Of the 27.1% not responding at both
waves, 51.4% completed the CRN questions in the 2005 PDS and 48.6% indicated adherence or
nonadherence in the 2007 PDS. In general, by allowing the inclusion of cases with missing data,
the sample size is larger and thus more closely approximates the general population (Ott &
Lngnecker, 2001).
Second, mixed regression models do not require that respondents complete the survey at
precisely the same time interval. This, again, is an essential point when considering the
administration of the PDS. The survey was mailed to respondents in October, and the field
period continued through March. It was completed at varying times over a six-month period.
Finally, mixed regression models are advantageous when analyzing repeated-measures data
because they provide an estimate of individual-specific change across time and also model
patterns of change between individuals. By modeling both fixed and random effects, there is the
ability to account for individual differences over time as well as deviations from the group trend
(Van Dongen, Olofsen, Dinges, & Maislin, 2004).
A series of six mixed regression models were run with the direct and indirect effects of
each set of variables on the relationship between race, gender, and CRN over time examined in
each model. Model 1 included only race and gender. Age and marital status variables were added
in Model 2 and in Model 3, socioeconomic characteristics were added. Health status variables
were added in Model 4. In Model 5, prescription insurance coverage variables were added.
Finally, year of adherence or nonadherence due to cost was interacted with race and gender in
Model 6 to compare racial and gender differences in change in CRN over time, controlling for all

32

other variables. A likelihood-ratio test was performed for each model to determine whether
including time as a fixed and random effect provided a better fit than the random-intercept only
model. A null model was first fit with the dependent variable only to determine the level of
between-person variance. Next, the fixed effect of time and a random effect associated with the
intercept of each respondent were added to the model. A third model then added a second
random effect that allowed each person to have a unique outcome at each time point. With the
exception of the null model, this process was repeated for Model 1 through Model 6. An alpha of
.05 was used as the cutoff for determining statistical significance. In every analysis, the
likelihood-ratio test was not significant. Thus, time was included only as a fixed effect in the
regression models.
All statistical analyses were performed in the Stata version 11.0 statistical package
(StataCorp). Stata’s survey commands were used in univariate statistics and bivariate tests of
association to account for the complex multistage clustered sampling design of the HRS when
computing standard errors. The PDS questionnaire weights (a product of the HRS sampling
weights) were applied to adjust for sample selection probability or oversampling of select
minority groups and for interviewee nonresponse in univariate and bivariate analysis only (HRS,
2011). Stata does not allow the incorporation of the survey command or application of
probability weights in mixed-effect logistic regression analyses. Thus, mixed regression models
did not control for the design effect and were not weighted. A consequence of not utilizing the
questionnaire weights or controlling for the design effect is the potential for inaccurate point
estimates and/or inaccurate standard errors, which in turn can impact the ability to infer results to
the general population. Despite this limitation, mixed-effect regression analyses can provide a
strong indication of racial and gender difference in CRN over time because unlike other methods

33

such as time-series ANOVA, cases are not dropped if there are missing values. Further, there
was no substantive difference in outcomes in bivariate models that were weighted and controlled
for the design effect as compared to multivariate models. Thus, the potential that the results will
inaccurately reflect the general population is minimal. Significance in multivariate analysis was
tested at the p<.05 level using a one-tailed test.
Results
Description of Sample at Baseline
A detailed description of the sample in 2005 is provided in Table 2.1. Of the 2,863
individuals responding to the question about medication adherence at baseline, 8.2% selfidentified as non-Hispanic African American and the majority (58%) were female. Respondents
ranged in age 65 to 101 with a mean age of 74.3 years, and over half (53.3%) were married. The
average household income was $50,614 (SD = $100,088) and the mean years of education was
12.42 (SD = 2.93). Respondent self-reports indicated that the majority felt their health was good
(35%), very good (27.1%), or fair (21.1%), followed by excellent (8.5%), and excellent (8.3%).
The average number of chronic health conditions was 2.38 (SD = 1.36). In regards to
prescription insurance coverage, most (59.7%) had insurance to pay for part of the cost of
medications, 15.3% paid the full cost of their prescriptions or had no coverage, 13.8% had a
prescription discount card, 6.7% paid nothing for their medications, and 4.5% listed “other” as
their source of coverage. No respondents were enrolled in the Medicare Part D benefit at baseline
because it was not yet implemented.
Bivariate Results
Bivariate analyses of medication adherence and nonadherence in 2005 and 2007 are
presented in Table 2.2. In regards to the main predictor variables, both race and gender appeared

34

TABLE 2.1
Description of Sample of Individuals Age 65 and Over Reporting Medication
Adherence/Nonadherence in 2005 (Baseline) (n=2,863)
Variable
Race
Black American
White
Gender
Female
Male
Age (range 65 to 101)
2
Marital status
Married
Married, spouse absent
Partnered
Separated
Divorced
Widowed
Never married
2
Education (range: 0 to 17)
Annual household income (range: $0 to $2,761,657)
Number of chronic medical conditions (range: 0 to 7)
2
Self-reported health
Excellent
Very good
Good
Fair
Poor
2
Prescription drug coverage
Pay full cost or no coverage
Insurance pays part of cost
Prescription discount card
Pay nothing
Other
Medicare Part D enrollment
No
Cost-related adherence or nonadherence
Adherent
Nonadherent
1
2

1

% / SD

422
2,441

8.2
91.8

1,713
1,150
74.38

58.0
42.0
7.38

1,579
32
52
23
222
886
68

53.3
1.0
1.5
1.0
8.0
32.5
2.7

12.42
50,614
2.38

2.93
100,088
1.36

241
749
989
637
244

8.5
27.1
35.0
21.1
8.3

504
1,497
436
198
117

15.3
59.7
13.8
6.7
4.5

2,863

100

2,390
473

85.2
14.8

Weighted %, unweighted N
Sample size varies slightly due to missing data within PDS or HRS

35

1

N /M

to be significantly related to CRN. In 2005 and 2007, a higher proportion of Black Americans
were nonadherent compared to Whites, and a greater proportion of females were nonadherent
compared to males.
In terms of control variables, age was significantly related to CRN. Nonadherent older
individuals reported a significantly lower age than adherent older adults before and after
Medicare Part D. Marital status was not significantly related to CRN.
In regards to socioeconomic status, educational attainment was significantly related to
nonadherence in 2005 and 2007. Respondents who reported CRN had significantly fewer years
of education. Further, nonadherent older individuals reported a significantly lower annual
income than adherent older adults before and after the implementation of Medicare Part D.
In regards to health status variables, respondents who reported CRN had a significantly
greater number of chronic conditions compared to those that maintained adherence in 2005 and
2007. A smaller proportion of respondents in excellent health reported nonadherence before and
after Medicare Part D than those in other categories with those in fair or poor health reporting the
most nonadherence. Finally, prescription insurance coverage was significantly related to CRN. A
much larger proportion of those reporting nonadherence in 2005 and 2007 utilized a prescription
discount card or had no coverage/paid the full cost of their medications as compared to those
who had insurance, paid nothing for medications, or had some other type of coverage. In 2007, a
higher proportion of those enrolled in Medicare Part D or unaware of whether they were enrolled
in this benefit reported CRN in 2007 compared to those who were not enrolled in Medicare Part
D at that time.
Bivariate analyses suggest that without controlling for other factors, older Black
Americans and women were more likely to report CRN than older Whites and men before and

36

TABLE 2.2
Bivariate Relationships Between Independent Variables and Medication Adherence/Nonadherence, 2005 and 2007
2005 (n=2,863)
2007 (n=2,301)
Adherence
Nonadherence
Adherence
Nonadherence
1

1

1

1

N

2

N

%

N

%

N

%

321
2,069

Race
Black American
White

%

77.3
85.9

101
372

22.7
14.1

221
1,719

75.0
86.8

73
288

25.0
13.2

84.3
88.0

232
129

15.7
12.0

11.95***

! , df=1

3

! , df=1
Gender
Female
Male

19.70***

1,382
1,008

2

2

Age (M±SD)

df=1, 51

82.6
88.8

3

2

Education (M±SD)

17.4
11.1

72.64 ± 6.98

34.57***

2

! , df=6

331
142

2

1,344
28
44
20
175
725
53
3

3

76.03 ± 6.87
3

74.32 ± 6.24

15.68***
86.8
90.9
86.8
86.8
81.4
83.7
81.4

235
4
8
3
47
161
15

13.2
9.1
13.2
13.2
18.6
16.3
18.6

1.77

12.48 ± 3.05
3

1,110
830
6.47*

74.68 ± 7.69

Marital status
Married
Married, spouse absent
Partnered
Separated
Divorced
Widowed
Never married

df=1, 51

3

3

14.21***

1,119
24
24
12
128
571
43

87.5
93.1
79.5
81.6
80.8
84.3
88.5

3

185
2
9
3
37
114
6

12.5
6.9
20.5
18.4
19.2
15.7
11.5

1.74
12.05 ± 3.09

5.19*

12.56 ± 2.85
3

12.23***

37

11.99 ± 2.82

TABLE 2.2
Bivariate Relationships Between Independent Variables and Medication Adherence/Nonadherence, 2005 and 2007
2005 (n=2,863)
2007 (n=2,301)
Adherence
Nonadherence
Adherence
Nonadherence
1

%

N
Annual household
2
income (M±SD)

53,391 ± 85,825
21.74***

df=1, 51
Number /chronic medical
2
conditions (M±SD)
df=1, 51
2
Self-reported health
Excellent
Very good
Good
Fair
Poor
2
! , df=4

3

2

Prescription drug coverage
Pay full cost or no
coverage
394
Insurance pays part of
cost
1,306
Prescription discount
card
306
Pay nothing
184
Other
109
2
3
! , df=4
20.07***

34,588 ± 52,804

1

%

N

48,113 ± 61,358
3

N

1

%

37,948 ± 50,520

7.91**

2.34 ± 1.35
3
13.61***
217
650
832
505
183
3
6.49***

1

%

N

2.61 ± 1.39

2.46 ± 1.35
3
24.91***

2.84 ± 1.38

90.1
88.5
85.5
81.7
77.5

24
99
157
132
61

9.9
11.5
14.5
18.3
22.5

165
558
701
348
147
3
7.88***

93.3
88.5
87.5
79.3
81.7

16
75
118
106
38

6.7
11.5
12.5
20.7
18.3

79.4

110

20.6

88

78.2

26

21.8

88.0

191

12.0

1,542

85.9

285

14.1

72.5
93.6
93.8

130
14
8

27.5
6.4
6.2

73
123
86
3
4.18**

76.2
90.2
92.9

27
14
7

23.8
9.8
7.1

38

TABLE 2.2
Bivariate Relationships Between Independent Variables and Medication Adherence/Nonadherence, 2005 and 2007
2005 (n=2,863)
2007 (n=2,301)
Adherence
Nonadherence
Adherence
Nonadherence
1
1
1
1
%
%
%
%
N
N
N
N
Medicare Part D enrollment
No
Yes
Don't know
2
! , df=2
1
2
3

2

2,390
-----

85.2
-----

473
-----

14.8
-----

788
968
129
3
10.54***

90.4
82.7
81.0

97
226
31

9.6
17.3
19.0

Weighted %, unweighted N's
Sample size varies slightly due to missing data within Prescription Drug Study (PDS) or Health and Retirement Study (HRS)

Test statistic is Rao-Scott chi-square for categorical variables and F means test for continuous variables
*Significant at p<.05; **Signficant at p<.01; ***Significant at p<.001

39

after the implementation of Medicare Part D as expected. Respondents who were younger, had
lower income, fewer years of education, or poorer health status appeared more likely to report
CRN over time. Having a prescription discount card or no coverage increased the likelihood of
CRN in 2005 and 2007. Finally, older individuals enrolled in Medicare Part D or who were
unaware of their enrollment status appeared more likely to report CRN in 2007.
Mixed-Effect Logistic Regression Results
Table 2.3 shows mixed-effects logistic regression models for CRN outcomes. Model 1
estimated the effects of race and gender on CRN over time, controlling for each other and time.
Compared to older Whites, Black Americans were over two and a half times more likely to
report CRN both before and after the implementation of Medicare Part D (OR=2.55). Females
were 1.82 times more likely than males to experience CRN. Further, CRN was not significantly
different in 2007 as compared to 2005.
The effects of race and gender on CRN over time remained significant with introduction
of age and marital status variables in Model 2, although the odds of older Black Americans
reporting CRN decreased 33% and for women, there was a 12% decline. Thus, the addition of
these variables appeared to have some indirect effect on the relationship between race, gender,
and CRN over time. Furthermore, both age and marital status had a direct effect on CRN in 2005
and 2007 when controlling for race, gender, and time. The odds of reporting CRN decreased as
age increased. Respondents who were older had a lower likelihood of nonadherence due to cost
over time (OR=.94). Additionally, respondents who were divorced (OR=1.47) or widowed
(OR=1.53) were more likely than those currently married to experience CRN over time.
The inclusion of socioeconomic characteristic variables in Model 3 appeared to mediate
some of the effect of race and gender on CRN over time, although the significant relationships

40

TABLE 2.3
Mixed-Effects Logistic Regression Results of Cost-Related Medication Nonadherence
Model 1
Model 2
Model 3
OR
OR
OR
Race
Black American
2.55***
2.22***
1.67**
White (reference group)
---Gender
Female
1.82***
1.70***
1.63***
Male (reference group)
--Year
2007
.88
.96
.98
2005 (reference group)
---Age (continuous)
.94***
.93***
Marital Status
Currently married (reference group)
--Married, spouse absent
.58
.50
Partnered
1.72
1.63
Separated
.75
.56
Divorced
1.47*
1.26
Widowed
1.53**
1.22
Never married
.94
.82
Education (continuous)
.93***
Annual household income (continuous)
.99***
Number of chronic medical conditions (continuous)
Self-reported health
Excellent (reference group)
Very good
Good
Fair
Poor

41

Model 4
OR

Model 5
OR

Model 6
OR

1.51*
--

1.53*
--

1.65
--

1.65***
--

1.42**
--

3.01***
--

.95
-.93***

.91
-.93***

1.26
-.93***

-.48
1.65
.51
1.19
1.16
0.79
.95*
.99***
1.17***

-.51
1.91
.76
1.33
1.23
.97
.95*
.99***
1.18***

-.52
1.90
.77
1.34
1.24
.97
.95*
.99***
1.18***

-1.32
1.48
2.25**
2.27**

-1.22
1.36
2.14**
2.20**

-1.24
1.37
2.18**
2.25**

TABLE 2.3
Mixed-Effects Logistic Regression Results of Cost-Related Medication Nonadherence
Model 1
Model 2
Model 3
Model 4
OR
OR
OR
OR
Prescription drug coverage
Pay full cost or no coverage (reference group)
Insurance pays part of cost
Prescription discount card
Pay nothing
Other
Medicare Part D enrollment
No (reference group)
Yes
Don't know
Year*Race
Year*Gender

Model 5
OR

Model 6
OR

-.35***
1.42*
.12***
.17***

-.35***
1.43*
.12***
.17***

-1.94***
1.79*

-2.01***
1.91*
.95
.59**

*Significant at p<.05; **Significant at p<.01; ***Significant at p<.001
Model 1: Race, Gender, and Year (n=3,181)
Model 2: Race, Gender, Year, Age, and Marital Status (n=3,177)
Model 3: Race, Gender, Year, Age, Marital Status, and Socioeconomic Status (n=3,172)
Model 4: Race, Gender, Year, Age, Marital Status, Socioeconomic, and Health Status (n=3,169)
Model 5: Race, Gender, Year, Age, Marital Status, Socioeconomic and Health Status, and Insurance Coverage (n=3,104)
Model 5: Interaction Effect of Race, Gender, and Year, Controlling for Age, Marital Status, Socioeconomic and Health Status, and
Insurance Coverage (n=3,104)

42

were maintained. The odds of older Black Americans experiencing CRN in 2005 and 2007
decreased 55% and for females, there was a 13% decline. Marital status was no longer a
significant predictor with the addition of these variables. Furthermore, both educational
attainment and annual household income had a direct effect on CRN before and after Medicare
Part D when controlling for race, gender, time, age, and marital status. CRN was significantly
less for respondents with greater number of years of education (OR=.93). Further, having a
greater annual income was associated with a lower likelihood of nonadherence due to cost
(OR=.99).
With the addition of health status variables in Model 4, the effects of race and gender on
CRN over time remained significant. Number of chronic medical conditions and self-reported
health appeared to have very little indirect effect on the relationship between race, gender, and
CRN as only a slight change in odds ratios was observed. Respondents with a greater number of
chronic medical conditions had an increased likelihood of CRN (OR=1.17). Further, as
compared to older individuals who reported their health as being excellent, odds of CRN were
greater for those who rated their health as fair (OR=2.25) or poor (OR=2.27).
The significant effects of race and gender on CRN before and after Medicare Part D were
maintained with addition of prescription insurance coverage variables in Model 5. The inclusion
of prescription drug coverage and Medicare Part D enrollment variables appeared to have some
mediating effect on the relationship between race and gender and CRN in 2005 and 2007.
Although the odds of older Black Americans reporting CRN increased slightly (2%), a 23%
decrease in CRN for women was observed. Furthermore, prescription drug coverage and
Medicare Part D enrollment had a direct effect on CRN over time when controlling for race,
gender, time, age, marital status, socioeconomic characteristics, and health status. As compared

43

to respondents that paid for the full cost of their medication(s) or had no prescription drug
coverage, individuals with insurance that covered part of the cost were 65% less likely (OR=.35)
and those with a prescription drug discount card were nearly one and a half times more likely
(OR=1.42) to report CRN before and after Medicare Part D. Furthermore, the odds ratios (.12
and .17, respectively) show that respondents who paid nothing for their medications were 88%
less likely to experience CRN and those who had some other type of coverage were 83% less
likely to report CRN (compared to respondents who paid for the full cost of their medications).
When considering Medicare Part D enrollment, the odds of CRN over time were nearly double
(OR=1.94) for individuals enrolled in this benefit than for non-Medicare Part D enrollees.
Further, respondents who did not know if they enrolled in Medicare Part D were 1.79 times more
likely to experience CRN before and after the implementation of the benefit.
Finally, year of adherence or nonadherence due to cost was interacted with race and
gender variables in Model 6 to compare racial and gender differences in change in CRN over
time, controlling for age, marital status, socioeconomic characteristics, health status, and
prescription insurance coverage. As compared to males, the rate of CRN for females
significantly decreased over time (OR=.59). For older Black Americans, the interaction was not
significant. Nonadherence due to cost, relative to Whites, did not change over time.
Discussion
The addition of the Part D drug benefit to Medicare was the largest expansion of the
program in over 40 years (Madden, et al., 2008). While Medicare Part D was designed to help
older persons gain access to needed medications, the extent to which the policy eliminated racial
and gender disparities in CRN was unknown. This is the first longitudinal study to examine
whether and to what extent nonadherence due to cost has changed for older females and Black

44

Americans since the implementation of the Medicare prescription drug benefit, and how factors
associated with cumulative disadvantage can directly and indirectly affect the likelihood of CRN
over time.
In this study, there was a significant disparity in CRN between older Black Americans
and Whites both before and after Medicare Part D, and these differences did not change over
time. Similarly, older females were significantly more likely than males to report CRN in both
2005 and 2007; however, women experienced a significantly greater decline in nonadherence
over time compared to men. Both racial and gender disparities remained significant after
controlling for age, marital status, socioeconomic characteristics, health status, and prescription
insurance coverage. Further, factors related to cumulative disadvantage, specifically lower
socioeconomic status, less education, poorer health status, and less generous or no prescription
insurance coverage, had a direct effect on CRN over time.
In terms of hypotheses organizing the present analyses, there are several findings in need
of discussion. The first hypothesis stated that compared to older males, older females are more
likely to report CRN over time and less likely to report changes in CRN over time. This was
partially supported. This study replicates previous findings showing that women were more
likely to experience CRN than men (Heisler, et al., 2005; Piette, et al., 2004c; Zivin, et al., 2010),
and expands this research by demonstrating that gender disparities existed before and after
Medicare Part D. This suggests that although a drug benefit became available under Medicare,
women continue to have more difficulty than men affording medications and maintaining
adherence. A potential reason why disparities persist is that Medicare Part D’s cost-sharing
provisions continue to present barriers to maintaining adherence.

45

The prediction that females would be less likely to have a decline in CRN over time,
however, was not supported. On the contrary, women were more likely to experience a decline in
CRN. Thus, it is suggested that Medicare Part D is more effective in reducing CRN among
women than men over time. One potential explanation is the availability of the Low-Income
Subsidy (LIS) that was established with the enactment of Medicare Part D. The LIS provides
financial assistance for Medicare Part D beneficiaries that have limited income and resources.
Those eligible receive assistance with costs associated with the Part D plan (monthly premiums,
annual deductible, medication copayments), and have a reduced coverage gap. Additional
analyses (not shown) demonstrated that in 2005 and 2007, a significantly higher proportion of
females had an annual household income in the lower two quartiles and therefore, they may be
closer to the financial guidelines established to receive LIS assistance. One study confirmed that
lower income individuals are more likely to participate in the LIS (Davidoff et al., 2010). Thus, it
is possible that women may have had a greater reduction in CRN over time than men because of
financial status and the availability of this assistance.
In terms of race, it was predicted that compared to older Whites, older Black Americans
would be more likely to report CRN over time and less likely to report a decrease in CRN over
time. This hypothesis was partially supported. This finding expands previous research illustrating
racial disparities in CRN (Gellad, et al., 2007; Klein, et al., 2004; Soumerai, et al., 2006;
Steinman, et al., 2001) by demonstrating that despite the presence of a drug benefit under
Medicare, older Black Americans were significantly more likely to report CRN both before and
after Medicare Part D. It is therefore suggested that cost continues to be a barrier to maintaining
adherence for older Black Americans.

46

When considering change over time, it appears that Medicare Part D may be more
effective in improving gender disparities in CRN than racial differentials. The rate of CRN for
females significantly decreased over time in comparison to males; however, there was no change
in the rate for Black Americans relative to Whites. A possible reason for this finding may be
Black Americans’ greater reliance on Medicaid for prescription drug coverage. Unlike those
who became eligible for LIS assistance with the enactment of Medicare Part D, Medicaid
recipients were automatically enrolled in Medicare Part D plans that continued to subsidize drug
costs. In other words, the assistance received through Medicaid was retained under Medicare
Part D plans. Additional analyses (not shown) revealed that in this study, old Black Americans
were significantly more likely to be enrolled in Medicaid in both 2005 and 2007. Despite the
presence of drug coverage under Medicaid, Black Americans have experienced problems
affording the cost-sharing requirements (Gellad, et al., 2007; Klein, et al., 2004). Thus, very little
change in CRN over time could be expected given previous coverage and cost-sharing
requirements that existed under Medicaid before Medicare Part D were essentially maintained.
The third hypothesis that socioeconomic status, health status, and prescription insurance
coverage would mediate the effect of racial and gender differences in CRN over time was not
supported. Although the odds of CRN for older Black Americans and females compared to older
Whites and males decreased with the addition of these variables, the effects were small and racial
and gender differences in CRN over time remained significant. In other words, disparities
continued even after accounting for factors associated with cumulative disadvantage.
It is difficult to assess the reasons why disparities persist after controlling for
socioeconomic status, health status, and prescription insurance coverage because of limitations
with existing research. There are, however, two potential explanations. First, while this study

47

controls for whether an individual actually enrolled in Medicare Part D, it could not control for
individual differences within Part D plans. There are over 40 Medicare Part D plans (Kaiser
Family Foundation, 2011b), and coverage restrictions and cost-sharing requirements vary
extensively within plans (Hoadley, Hargrave, Cubanski, & Neuman, 2006). Each plan’s coverage
policies can impact CRN. For example, antidepressants are often subject to policies designed to
reduce access, which can increase CRN (Huskamp, Stevenson, Donohue, Newhouse, & Keating,
2007). Because older women are more likely to have a greater number of depressive symptoms
(Skarupski, et al., 2009), their chances of incurring coverage restrictions or other costcontainment measures associated with Part D plans are greater. Similarly, cost-sharing
requirements are typically more for individuals with diabetes than most other chronic illnesses
under Medicare Part D plans (Ettner et al., 2010). Since older Black Americans are more likely
to be diabetic than Whites (Schoenbaum & Waidmann, 1997; Schoenborn & Heyman, 2009),
they have a greater likelihood of incurring cost sharing under Part D plans. Further, Black
Americans had a significantly greater number of medications in 2005 and 2007 as compared to
Whites (additional analysis, not shown). Thus, the variation in cost sharing may have a greater
impact on older Black Americans than Whites because their health status is typically poorer and
as a result, they often have higher prescription drug needs. This, in turn, can impact their ability
to purchase prescribed medications and maintain adherence.
A second explanation involves factors involving the physician-patient relationship and
trust. The physician-patient relationship has been found to influence CRN (I. B. Wilson, et al.,
2004). Higher out-of-pocket prescription drug expenditures have been associated with a greater
likelihood to forgo medications because of cost if trust in the physician is low (Piette, Heisler,
Krein, & Kerr, 2005). Further, communication regarding costs can influence CRN. Individuals

48

who have experienced problems paying for their medications are often not asked if they can
afford the cost (Piette, Heisler, & Wagner, 2004b). It is possible that for some older women and
Black Americans in this study, physician trust and/or communication were low and contributed
to CRN.
Finally, it was predicted that lower socioeconomic status, poorer health status, or little or
no prescription insurance coverage would increase the likelihood of CRN over time. When
considering socioeconomic status, this prediction was supported. Similar to previous crosssectional studies conducted before the implementation of Medicare Part D, having a greater
annual household income and more years of education decreased the likelihood of nonadherence
due to cost (Gellad, et al., 2007; Mojtabai & Olfson, 2003; Soumerai, et al., 2006; Steinman, et
al., 2001; Zivin, et al., 2010). This study expands these findings and suggests that despite the
implementation of Medicare Part D, older individuals with a lower income and less education are
especially vulnerable to CRN.
With regard to health status variables, poorer health directly affected CRN as predicted.
In this study, older persons with a greater number of chronic medical conditions were more
likely to report CRN before and after Medicare Part D. Additionally, nonadherence due to cost
was more likely for individuals reporting fair or poor health than those reporting excellent health.
This is consistent with previous findings that individuals in poor health with multiple
comorbidities have high and persistent CRN over time (Briesacher, Gurwitz, & Soumerai, 2007;
Madden, et al., 2008; Safran, et al., 2005). CRN can be caused by higher out-of-pocket
expenditures due to multiple medications (Frankenfield, et al., 2010; Zivin, et al., 2010), and
generosity of coverage can substantially reduce the risk of nonadherence for those with several
comorbidities (Soumerai, et al., 2006). While Medicare Part D may have alleviated some of the

49

financial burden for sicker beneficiaries, it is suggested that cost-sharing requirements under
Medicare Part D can still be problematic and thus, have little impact on CRN.
When considering prescription drug coverage variables, these findings support the
hypothesis that having no or less generous drug coverage will increase CRN over time. First, as
compared to older adults with no coverage or paying the full cost for their medications, having
partial coverage provided by insurance, paying nothing for medications, or having some “other”
source of coverage decreased the likelihood of CRN before and after Medicare Part D. This is
consistent with previous findings showing that older individuals with drug coverage were less
likely to report CRN than those without (Frankenfield, et al., 2010; Mojtabai & Olfson, 2003;
Steinman, et al., 2001), and expands research by demonstrating that these differences remained
despite the implementation of Medicare Part D. Additionally, CRN was significantly greater over
time between older individuals with a prescription discount card – which typically provides
minimal coverage – and those without coverage or paying the full cost of their medications. This
is a bit perplexing given that a discount card can provide some financial assistance with
medication costs. A potential reason for this finding is that those who acquire a discount card
may have greater difficulty with medication costs than those without this type of assistance. In
other words, they obtain the card because they are struggling with drug costs.
Second, when considering the impact of Medicare Part D on CRN, those enrolled in the
benefit or uncertain if they enrolled in the benefit were significantly more likely than those not
enrolled to report nonadherence due to cost over time. This finding may seem counterintuitive,
particularly because one might expect CRN to improve over time if enrolling in the benefit.
However, this study included all respondents, regardless of type of prescription drug coverage. In
other words, if an individual did not enroll in Medicare Part D, it may be because they had

50

coverage through an alternative source such as an employer-based plan. Research has shown
differential adherence to medications depending on the source of coverage. For example, CRN
was significantly greater for those enrolled in Medicare Part D as compared to individuals with
employer-based coverage (Safran, et al., 2009). These differences can be explained by out-ofpocket expenditures under Medicare Part D; as compared to employer-based plans, Part D
enrollees had higher cost-sharing requirements or a less generous benefit structure (Goedken,
Urmie, Farris, & Doucette, 2010; Safran, et al., 2009). Thus, it is suggested that older individuals
enrolled in Medicare Part D will continue to experience greater CRN because of cost sharing
under this program.
An additional finding worthy of note involves the age variable. It was somewhat
surprising that increased age decreased the likelihood of CRN over time, particularly because
research shows lower prescription drug coverage rates among older age groups (Kanavos &
Gemmill-Toyama, 2010). This finding, however, is consistent with cross-sectional studies using
HRS to examine CRN before the implementation of Medicare Part D. One potential reason
involved the sampling of the HRS and exclusion of institutionalized persons. Because risk of
institutionalization or nursing home placement increases with age, it was suggested that the older
age groups participating in the HRS are disproportionately healthy (Klein, et al., 2004). This
explanation, however, could not be confirmed through additional analysis. Therefore, future
research is needed to understand why younger persons were more likely to experience CRN than
older age groups.
Strengths of this study include the use of a large nationally representative sample of older
adults and its ability to examine multiple predictors simultaneously and over time. However,
there are a number of limitations that must be considered. In the future, a longer prepolicy and

51

postpolicy series would provide more clarity and permit a stronger assessment of Part D
enrollment. Other unmeasured factors may have influenced the results. For example, the
transition of dual eligibles from Medicaid to Medicare drug plans in 2006 was confusing for
many older adults (Kaiser Family Foundation, 2006) and resulted in difficulties obtaining
medications (West, et al., 2007). A significantly greater proportion of older Black Americans in
this study were dually eligible as compared to Whites, and thus, had a greater likelihood of being
impacted by the transition. As a result, CRN may have been overestimated because data
collection occurred shortly after the launch of Medicare Part D and at a time of initial confusion.
However, the PDS only contained two waves of data at the time this study was conducted. While
this research provides important evidence on the impact of the policy on racial and gender
disparities in CRN, the results should be considered early evidence until longer-term data
become available.
Second, socioeconomic and health status were measured in the 2004 and 2006 HRS, and
CRN was measured in the 2005 and 2007 PDS. Thus, there is the potential that income or certain
health conditions may have changed because these measures were not taken simultaneously or
when CRN was reported. While this seems less likely with income because older adults are
largely dependent on Social Security, there is greater potential for fluctuation in health status.
Depression, for example, can be transitory (Zivin, et al., 2010). Older females and Black
Americans who enrolled in Medicare Part D have a greater prevalence of depression as
compared to older males and Whites (Skarupski, et al., 2009). Thus, the indirect effect of health
status on the relationship between race, gender, and CRN may have been inaccurately
represented in the results.

52

Third, this study investigated nonadherence due to cost only and did not account for other
factors that can impact adherence. For example, an individuals’ concerns or beliefs regarding
treatment through the use of medications can certainly impact adherence (Aikens & Piette,
2009). Some medications may be viewed as more essential than others (Lau et al., 2008; Piette,
Heisler, & Wagner, 2006) and there could be racial and gender differences in whether a
prescription is perceived as necessary, particularly when faced with cost concerns. As compared
to older males and Whites, older females’ and Black Americans’ ability to afford out-of-pocket
expenditure requirements is typically less due to lower socioeconomic status (Wei, et al., 2006).
Thus, they may perceive the need for certain medications differently because of economic
concerns. Additionally, physician trust and communication can be a factor in medication
adherence (Piette, et al., 2005). It is possible that for some older females and Black Americans in
this study, physician trust and/or communication were low and contributed to CRN. While this
could account for some variation in CRN, these factors were not the focus of this research but
will be examined in future studies.
Fourth, older Black Americans had a significantly greater proportion of missing data
when asked about CRN in the 2007 PDS than older Whites. Individuals can be confused by the
Medicare Part D benefit because of its complexity (Davidoff, et al., 2010; Heiss, et al., 2006;
Hsu, et al., 2008). Fewer Black Americans may responded to the CRN question because as
stated previously, the survey was administered shortly after the implementation of Medicare Part
D – which was a time of confusion for many older persons. This, in turn, could bias the results
and underestimate or overestimate the racial effect of CRN over time. Further, this study relied
on self-reported survey measures. Self-reporting can be subject to social desirability bias
(Nederhof, 2006). Respondents may be reluctant to admit failure to adhere to a prescribed

53

medication regimen – particularly if it is due to cost factors. Because older women and Black
Americans are more likely to be economically disadvantaged, they may incur greater difficulty
affording the cost of prescribed medications – and, admitting failure to adhere because of
economic reasons. This bias could contribute to underestimating the prevalence of CRN.
Fifth, there are limitations with nonadherence measures contained in the PDS. The survey
did not ask respondents whether they took less than the prescribed dosage of a medication or if
they cut pills in half to make a medication last longer. The survey question asked if a dosage of a
medication was skipped to save money, and this measure could be a potential indicator of these
behaviors. However, failure to ask if these behaviors specifically occurred could underestimate
CRN.
Finally and as stated previously, this study utilized a composite measure to estimate CRN
as opposed to including all three behaviors as separate indicators: 1) not filling a new
prescription because of cost; 2) stopping taking a medication because of cost; or 3) skipping
doses of a medication to save money. While this is consistent with previous research and
therefore useful for cross-study comparison (Bambauer, et al., 2007; Gellad, et al., 2007;
Madden, et al., 2008; Safran, et al., 2005; Soumerai, et al., 2006; I. B. Wilson, et al., 2004; Zivin,
et al., 2009), measuring each behavior separately may provide important insights for social work
practice on racial and gender disparities in CRN. For example, one behavior could be more
prevalent than others among older women or Black Americans. By understanding which types or
types of nonadherence are more common in older women and Black Americans, interventions
that specifically target certain behaviors can be designed. Thus, future research will examine
each CRN item as an outcome.

54

Despite these limitations, the study makes a strong and significant contribution to our
understanding of whether and to what extent racial and gender disparities in CRN have changed
since the implementation of Medicare Part D. It also extends our understanding of the direct and
indirect effect of factors related to cumulative disadvantage on the relationship between CRN
and race and gender over time. Medicare Part D reduced out-of-pocket prescription drug
expenditures (Millett, Everett, Matheson, Bindman, & Mainous, 2010). This is an important
point, but the benefit may not be sufficient for all groups. The older adult population is not
homogeneous, and subgroup differences need to be considered. These results suggest that despite
the implementation of a drug benefit under Medicare, racial and gender disparities in CRN
continue to persist, and poorer health, being lower income, or having less generous or no
prescription drug coverage increases the likelihood of CRN. Findings can help substantiate the
need to account for race and gender when evaluating policy alternatives in order to promote
more equitable access to prescription medications and avoid disparate heath outcomes.

55

CHAPTER THREE
MANUSCRIPT TWO
ABSTRACT
Purpose: The viability of Medicare Part D and the Low-Income Subsidy (LIS) as a
means of reducing or eliminating racial and gender disparities in cost-related medication
nonadherence (CRN) is still uncertain. Using cumulative advantage/disadvantage theory as a
framework, the objectives of this longitudinal study were to determine: 1) to what extent racial
and gender disparities in CRN exist over time for older women and Black Americans enrolled in
Medicare Part D; and 2) to what extent the LIS directly and indirectly affects the relationship
between race, gender, socioeconomic and health status, and CRN before and after Medicare Part
D. Methods: To evaluate changes in CRN before and after Medicare Part D, secondary data from
the 2004 and 2006 waves of the Health and Retirement Study (HRS) and data from the 2005 and
2007 waves of the Prescription Drug Study (PDS), a subsample of the HRS, were utilized. The
analytic sample for this longitudinal analysis consisted of 1,466 respondents, age 65 and older,
who were enrolled in Medicare Part D, taking at least one prescribed medication, and responded
to questions about adherence or nonadherence in either 2005 or 2007. The outcome variable of
interest was CRN, which included not filling, stopping, or skipping doses of a prescription
because of cost. The main predictors were race (Black American and White), gender (male and
female) and LIS status (dual eligible/receiving LIS, applied for LIS, and no LIS). Covariates
included demographics, socioeconomic, and health status. The analysis was conducted using
mixed-effects logistic regression models. Results: Results indicated that racial and gender
disparities in CRN persisted after accounting for health and socioeconomic status. While LIS
status did not affect racial differentials, applying for the LIS mediated gender differences. Poorer

56

health, applying for the LIS, and having an annual income between $12,271 and $43,479
increased the likelihood of CRN over time. However, differences in between those earning
$12,721 to $23,363 and the highest income group were mediated by applying for the LIS.
Conclusions: Gender differences in CRN were substantially driven by whether an individual
applied for the LIS as were differentials between lower and higher income groups. However,
older Black Americans, the near-poor, and individuals in poor health continue to be
disproportionately nonadherent. These findings indicate the need to target outreach towards these
groups to raise awareness of the LIS and increase participation. Additionally, policymakers
should consider raising the LIS income requirements to 200% FPL.

57

CHAPTER THREE
COST-RELATED MEDICATION NONADHERNECE UNDER MEDICARE PART D
AND THE LOW-INCOME SUBSIDY: A RACIAL AND GENDER COMPARISON
Background
Prescription drug costs can present a significant barrier to older adults in need of
medications and, aside from health insurance premiums, represent the leading contributor to outof-pocket health care expenditures (Kaiser Family Foundation, 2010b). Retail prescription drug
prices have increased an average of 6.9% annually from 1990 to 2007, far outpacing the rise in
inflation (Kaiser Family Foundation, 2008). Older adults are disproportionately impacted by the
increasing costs of prescription medications because while they represent 13% of the general
population, they account for 34% of all medications that are dispensed and 42% of all
prescription drug expenditures (Families USA, 2000). As drug costs and out-of-pocket
medication expenses rise, adherence tends to fall (Mojtabai & Olfson, 2003; Safran, et al., 2005).
Older adults – particularly individuals with several chronic conditions – often become
nonadherent when they are unable to afford medications (Heisler, et al., 2004; Neuman, et al.,
2007; Soumerai & Ross-Degnan, 1999).
The Medicare Prescription Drug, Improvement, and Modernization Act of 2003 (MMA)
is the largest benefit expansion in the history of the program, and represents the most meaningful
health care policy in the past three decades because it reduces out-of-pocket health care costs. To
respond to the need to provide older adults with drug coverage, the MMA added the Part D
benefit to Medicare (Madden, et al., 2008; Mayes, 2005). Part D includes a number of costcontainment provisions such as cost-sharing requirements, plan premiums and a time period,
known as the coverage gap, when beneficiaries are responsible for 100 percent of their drug
costs. Cost-sharing provisions stipulate that the beneficiary is required to pay a fixed amount for

58

each prescription or copayment, a proportion of the medication price or coinsurance, and/or an
annual payment before the plan will provide coverage or deductible. Plan premiums require that
individuals pay a monthly amount to receive coverage and averaged $40.72 in 2011, an increase
of 57% since Medicare Part D was implemented in 2006 (Hoadley, et al., 2010). There is
concern that these cost-containment provisions can inhibit the use of medications for older
women and Black Americans. Compared to older men and Whites, older women and Black
Americans tend to have a greater number of chronic health problems (Clark, 1997; Kelley-Moore
& Ferraro, 2004a) and are, therefore, more reliant on prescription medications (Rogowski, et al.,
1997; Sambamoorthi, et al., 2003). This means that older women and Black Americans tend to
have higher copayments and are more likely to reach the coverage gap each year. Combined
with their overall lower socioeconomic status, therefore, older women and Blacks are less able to
afford out-of-pocket expenditure requirements (Wei, et al., 2006) and more likely to have
difficulties adhering to their medication regimen.
Previous research has shown the connection between cost-related medication
nonadherence (CRN) and cost-containment provisions similar to those present in Medicare Part
D plans (Adams, et al., 2001; T. B. Gibson, et al., 2005; Tseng, et al., 2004; Wang et al., 2010).
CRN can include not filling, stopping, or skipping doses of a prescription because of cost. Prior
to the MMA, racial and gender disparities in CRN clearly existed (Heisler, et al., 2005; Klein, et
al., 2004). Medicare Part D may continue to impede access to medications because the benefit’s
cost-containment provisions can increase out-of-pocket expenditures (Hsu, et al., 2008), and may
have a disproportionately greater impact on older females and Black Americans because of their
economic status, health status, and greater reliance on prescription medications (Wei, et al.,
2006).

59

Because out-of-pocket expenditure requirements can be a barrier to prescription drug
access, the Low-Income Subsidy (LIS) program (also known as “Extra Help”), was created
under the MMA. This program was intended to offset prescription drug costs for beneficiaries
with limited income and resources. There are two groups of beneficiaries that qualify for this
assistance. First, the MMA stipulated that individuals covered by both Medicare and Medicaid,
known as dual eligibles, automatically qualify and receive the LIS. Prior to Medicare Part D,
dual eligibles received prescription benefits through state Medicaid programs. The MMA
required that these beneficiaries enroll in Medicare Part D and the LIS to maintain this
assistance. States were no longer allowed to provide prescription drug coverage for dual
eligibles. Second, other low-income Medicare beneficiaries whose income and assets are below a
specified level can receive assistance; however, they are required to apply for the LIS. Coverage
is not automatic for this group.
The LIS reduces or eliminates out-of-pocket expenses associated with Medicare Part D,
including premiums, deductibles, copayments, coinsurance, and costs associated with the
coverage gap (Kaiser Family Foundation, 2009b). A summary of Medicare prescription drug
plan subsidies is provided in Figure 3.1. Though research pertaining to the efficacy of the LIS is
lacking, it is suggested that older women and Black Americans in particular will be impacted by
the LIS provision because of their lower-income status.
To date, studies have not examined the prevalence of racial and gender disparities in
CRN for beneficiaries that enrolled in Medicare Part D or the impact of the LIS on CRN. This
longitudinal study builds upon prior research by using a cumulative advantage/disadvantage
framework to examine to what extent racial and gender disparities in CRN have changed for
older women and Black Americans enrolled in Medicare Part D. Further, this study seeks to

60

Figure 3.1
Medicare Drug Benefit Subsidies
for Low-Income Beneficiaries, 2010
Low-Income Subsidy Level
Monthly
Annual
Premium
Deductible
Dual eligibles, or individuals
$0
$0
with Medicare and Medicaid:
Automatically enrolled and
receiving LIS

Copayments

$1.10-$2.50/generic
$3.30-$6.30/brandname;
no copayment after
total drug spending
reaches $6,440
Individuals with income <135%
$0
$0
$2.50/generic
of poverty and resources
$6.30/brand-name;
<$8,100/individual or
no copayment after
$12,910/couple:
total drug spending
Application for LIS required
reaches $6,440
Individuals with income 135% to
Sliding
$63
15% of total costs
150% of poverty and resources
scale up to
up to $6,440;
<$12,510/individual or
$31.94
$2.50/generic
$25,010/couple:
$6.30/brand-name
Application for LIS required
thereafter
SOURCE: Kaiser Family Foundation summary of Medicare drug benefit Low-Income
Subsidies in 2010.
NOTE: 2009 poverty level is $10,830/individual and $14,570/couple
understand to what extent the LIS directly and indirectly affects the relationship between race,
gender, socioeconomic and health status, and CRN over time. Examining disparities among
enrollees can help us understand how Medicare Part D enrollment has impacted CRN for older
women and Black Americans, and the impact of the LIS, a key and costly provision of the
MMA.
Literature Review
Cumulative Advantage/Disadvantage and Medicare Part D
When deliberating social policy initiatives intended to target older individuals, the aged
should not be stereotyped (Butler, 1985) or treated as a homogeneous group (Neugarten, 1982).
Inequities resulting from disadvantages throughout the life course need to be considered in order

61

to ensure equitable access to benefits. Cumulative advantage/disadvantage theory is useful as a
foundation for understanding and evaluating the effectiveness of Medicare Part D. The theory
has increasingly been used as a paradigm in the field of aging to describe stratification and
growing trajectories over time (Dannefer, 1987; Farkas, 2003; Kail, et al., 2009; A. E. Wilson, et
al., 2007). Systematic disadvantages have a considerable impact on later life, and it is suggested
that they not only accrue throughout the life course but are magnified and can have a feedback
effect (Crystal & Shea, 1990a; Ferraro & Kelley-Moore, 2003). In other words, “the more
disadvantages individuals experience, the more likely they are to accrue subsequent and greater
disadvantages” (Kail, et al., 2009, p. 557). For example, individuals from economically
disadvantaged families are more likely to have fewer years of education. This creates differences
in employment opportunities, and subsequently places them at greater risk for incurring lower
wages, unemployment, and less advantageous working conditions. Lower income due to limited
employment and educational opportunities is often compounded because of poorer health status,
lesser or no health insurance, and greater health care expenditures (Sambamoorthi, et al., 2003).
Because of cumulative advantages and disadvantages throughout the life course, heterogeneity
increases with age, and inequalities are greatest in later life (Crystal & Shea, 1990a; Dannefer,
1987; Ross & Wu, 1996).
This study will focus on racial and gender disparities in CRN under Medicare Part D. As
the result of cumulative advantages and disadvantages throughout the life course, it is anticipated
that higher out-of-pocket expenditures will be related to personal characteristics associated with
early disadvantage, or income, education, and health status. These factors, in turn, can
potentially impact CRN under the benefit.

62

Income and education. The ability to afford pharmaceuticals and maintain adherence to
a prescribed medication regimen are largely dependent on retirement earnings. Income in later
life is essentially based on pre-retirement experiences, and racial and gender inequities continue
to increase with advancing age (Crystal & Shea, 1990a). Wages for females and Black
Americans have consistently lagged behind those received by White males, and women and
Black Americans are more likely to have been segregated into positions without retirement,
pension, or health care benefits (Gonyea & Hooyman, 2005). Since Social Security is based on a
40-year work history, older women and Black Americans are typically subject to lower benefits
in later life than White males due to lower wages and fewer years in the labor force (Olson,
1994) and thus have less income available to purchase prescribed medications. Further,
catastrophic changes such as widowhood – which is more common in the lives of women
(Moody, 2009) – can be particularly difficult because older females and Black Americans are in
a disadvantaged position. Women and Black Americans not only enter old age poorer than men
and Whites but can become poorer with age as a result of widowhood (Minkler & Stone, 1985).
Health status. Lower income is typically accompanied by a higher prevalence of chronic
health conditions among older women and Black Americans. Racial and gender inequities in
morbidity exist throughout the life course and health trajectories continue to diverge with
increasing age (Clark, 1997; Kelley-Moore & Ferraro, 2004a). Women are more likely to report
functional limitations, a higher number of chronic conditions, and rate their health as poor or fair
than their male counterparts (Murtagh & Hubert, 2004). As the result of increased morbidity,
females and Black Americans are more reliant on pharmaceuticals (Goulding, 2005). Yet, their
ability to afford prescribed medications is often compromised because of lower socioeconomic
status. Compared to White males, older women and Black Americans spend a higher proportion

63

of their income on pharmaceuticals (Rogowski, et al., 1997; Sambamoorthi, et al., 2003). For
these reasons, it is expected that they will continue to experience a disproportionately heavier
financial burden under Medicare Part D (Wei, et al., 2006). This, in turn, can subsequently affect
their ability to purchase prescribed medications and maintain adherence.
Low-Income Subsidy. Cost-sharing assistance can be considerable for persons who
qualify for the LIS (Kaiser Family Foundation, 2009b). The LIS represents a significant policy
initiative because overall cost burden associated with prescribed medications can be greater for
older females and Black Americans due to disadvantages throughout the life course. Dual
eligibles and those applying for the LIS are typically lower-income individuals with multiple
chronic illnesses (Davidoff, et al., 2010). Thus, it is possible that older women and Black
Americans may be affected by the LIS provision more than older men and Whites because of
their financial and health status. Receiving this subsidy substantially reduces or eliminates outof-pocket expenses associated with Medicare Part D, and can potentially impact nonadherence
due to cost.
Medication Access and Adherence
Pre-Medicare Part D. Prior to Medicare Part D, CRN was estimated to occur in one
fourth of all older adults (Safran, et al., 2005); those without prescription drug coverage or with
limited coverage were significantly more likely to be noncompliant because of cost (Blustein,
2000; Fillenbaum, et al., 1993; Rogowski, et al., 1997). Nonadherence due to cost was greater
among older women than men (Heisler, et al., 2005; Piette, et al., 2004c; Zivin, et al., 2010).
Race was an even stronger predictor of CRN than gender. According to one study, older Black
Americans were 1.5 times more likely to report nonadherence to a medication regimen than
Whites (Gellad, et al., 2007), and numerous studies have found that Black Americans were more

64

likely than Whites to forgo or delay taking prescribed medications because of cost (Bambauer, et
al., 2007; Klein, et al., 2004; Roth, et al., 2009; Roth & Ivey, 2005; Soumerai, et al., 2006;
Steinman, et al., 2001). While Medicare Part D is expected to increase access to medications, it
is questionable whether the benefit will eliminate racial and gender disparities in CRN. Older
women and Black Americans are more likely to be reliant on prescription medications yet have a
decreased ability to afford out-of-pocket expenditure requirements due to lower socioeconomic
status. This, in turn, can affect their ability to maintain adherence.
In regards to Medicare and Medicaid recipients, CRN was less for dual eligibles than
non-dual eligibles before Medicare Part D (Gellad, et al., 2007; Soumerai, et al., 2006). Despite
lower nonadherence among dual eligibles as compared to non-dual eligibles, studies examining
CRN within the dual eligible population found that one fourth reported nonadherence (Safran, et
al., 2005) and nearly half had difficulties affording their medications (Saver, Doescher, Jackson,
& Fishman, 2004). It was suggested that dual eligibles were particularly sensitive to copayment
requirements under Medicaid plans because income and asset guidelines for coverage were
extremely low (Klein, et al., 2004; Soumerai, Avorn, Ross-Degnan, & Gortmaker, 1987).
Medicaid eligibility is dependent on having an annual income at or near the Federal Poverty
Level and extremely limited resources such as savings. As a result, affording medications can be
difficult even when copayment requirements are minimal. Individuals qualifying for coverage
typically have little disposable income left each month after paying for necessities such as food,
housing, transportation, or other health care costs – which in turn can impact their ability to
purchase prescribed medications (Soumerai, et al., 1987).
Studies conducted before the implementation of Medicare Part D found that older women
and Black Americans had higher odds of CRN than older men and Whites after controlling for

65

the presence of Medicaid coverage, or dual eligibility status (Bambauer, et al., 2007; Soumerai,
et al., 2006). Thus, it appears that receiving prescription benefits through state Medicaid
programs did little to mediate racial and gender differences in nonadherence due to cost. It is
possible that any copayment requirements posed difficulties older women and Black Americans
– regardless of whether or not they were minimal. It is unclear how CRN will be impacted for
dually eligible older women and Black Americans transitioning from Medicaid drug plans to the
LIS and Medicare Part D program. There was considerable variation in Medicaid copayment
requirements across states (Safran, et al., 2003). For this reason, copayment obligations were
expected to be higher under the LIS and Medicare Part D than Medicaid for some beneficiaries,
and lower for others (Guyer & Schneider, 2004).
Post-Medicare Part D. When analyzing the MMA from certain economic perspectives,
the policy can appear highly successful. In 2008, 90 percent of Medicare’s 45 million
beneficiaries had prescription drug coverage (Joyce, et al., 2009) as compared to 76 percent in
2006 (Levy & Weir, 2009). Further, it is estimated that out-of-pocket spending on prescription
drugs decreased between 13 and 18.4 percent even as the number of medications used by older
adults increased from 5.9 to 12.8 percent since the implementation of the MMA (Chen, et al.,
2008; Joyce, et al., 2009; Lichtenberg & Sun, 2007; Schneeweiss, et al., 2009; Yin, et al., 2008).
However, these findings do not take into consideration the implications of the cost-containment
provisions – or cost-sharing requirements, plan premiums, or coverage gap – of the MMA, and
the impact of those provisions on cumulatively disadvantaged beneficiaries. In other words, there
was no differentiation of groups reported in these findings; beneficiaries were regarded as a
homogeneous group that gained access to a prescription drug benefit and lowered their
medication expenditures. It is important to recognize the potential implications of Medicare Part

66

D cost-sharing provisions on older females and Black Americans. For example, one implication
of the cost-sharing requirements is that access to medications is not in fact the same for all
beneficiaries. Those with lower income and increased morbidity, such as is often experienced by
older women and Black Americans, may be negatively and disproportionately influenced by
cost-sharing requirements both because of their higher medication expenses and their limited
income. This, in turn, can impact their ability to access medication and maintain adherence.
Thus, existing research on the impact of MMA does not recognize the differential consequences
of these provisions
Research examining changes in CRN before and after Medicare Part D is extremely
limited. One of the main reasons for this limitation is the restrictions placed on the availability of
data. The Centers for Medicare and Medicaid Services has historically been a primary source of
data for researchers. However, a provision contained in the MMA stipulated that private health
care plans were not required to release Medicare Part D claims data for nonpayment-related
purposes, which included research, public health purposes, and health care analysis (Crystal,
2008, November; Stuart, 2008). This stipulation largely accounts for the limitation with existing
research. Fortunately, this was modified in 2009, and data are currently in the process of being
released.
When considering studies conducted after the implementation of Medicare Part D, Safran
et al. (2009) reported that for older individuals with no or a meager drug benefit, enrollment
resulted in lower out-of-pocket prescription drug expenditures, higher medication utilization, and
overall lower CRN. However, low-income, chronically ill beneficiaries were significantly more
likely to report CRN and difficulty affording their medications. Similarly, Madden et al. (2008)
found that while there was a modest but significant decrease in overall CRN, no decrease was

67

observed among individuals in fair to poor health, with at least four morbidities, or having an
annual income less than $25,000. Although differences associated with race or gender were not
analyzed in either study, the findings shed some light on the experiences of these groups.
Because older women and Black Americans are more likely to be low income, rate their health
as poor or fair, and experience higher rates of chronic illness and health problems, these studies
suggest that enrolling in the Medicare Part D benefit may have little or no impact on CRN for
these groups.
While previous studies have not specifically examined the effect of the Medicare drug
benefit on racial or gender disparities in CRN, research suggests ethnic disparities continue to
exist. One study conducted after the implementation of Medicare Part D, for example, found that
Hispanics were more likely to report CRN than non-Hispanics. Further, CRN was significantly
greater for non-Hispanic Blacks than non-Hispanic Whites. When examining gender, both
Hispanic and non-Hispanic females were more likely to report nonadherence due to cost than
males (Frankenfield, et al., 2010).
CRN and the Low-Income Subsidy. It remains unknown to what extent the LIS has
affected CRN among the non-dual eligible population, or individuals required to apply for this
assistance. Further, research has not examined how the provision of this assistance has impacted
racial and gender differences in nonadherence. Studies examining the LIS have focused solely on
dually eligible beneficiaries transitioning from Medicaid prescription drug plans to Medicare
Part D, did not differentiate by age, race, or gender, and only examined out-of-pocket costs,
utilization, and access – or factors that can suggest changes in adherence may have occurred –
rather than examining adherence itself. Prescription drug claim-based analyses concluded that for
dually eligible beneficiaries, mean out-of-pocket expenditures on certain drugs did not change

68

significantly (Basu, Yin, & Alexander, 2010; Millett, et al., 2010) or were reduced (Chen, et al.,
2008; Shrank et al., 2008), and prescription drug utilization remained stable (Shrank, et al.,
2008). Other studies utilizing a survey-based approach found that dual eligibles with a mental
illness had problems accessing medications because of prior authorization requirements or other
restrictions (Hall, Kurth, & Moore, 2007; West, et al., 2007), or increased copayment
requirements (Huskamp et al., 2009). Further, while one study reported that prescription drug
utilization did not change (Safran, et al., 2009), another found utilization decreased among
medications commonly used to treat certain chronic conditions (Farley & Dusetzina, 2010).
It appears that in general, the transition from Medicaid prescription drug plans to
Medicare Part D may have had little impact on dually eligible beneficiaries and medication
adherence may not have changed. In contrast, certain vulnerable subpopulations such as
individuals with a mental illness may have experienced difficulties accessing medications and
maintaining adherence because of their health status. Moreover, several studies utilized a claimbased approach, and the results only reflected successfully filled prescriptions and not attempted
fills where access problems and subsequent CRN may have occurred (Polinski, et al., 2009).
Because of these issues, the lack of differentiation by age, race, or gender within the samples,
and the absence of research pertaining to non-dual eligibles, it is difficult to conclude how racial
and gender disparities in CRN have been impacted by the LIS. Further, studies have examined
factors that can impact adherence, such as out-of-pocket cost, utilization, or access. They did not
specifically explore CRN for dual eligibles after the transition from Medicaid to Medicare Part D
and the LIS.
Overall, studies suggest that gender and racial disparities in CRN prior to Medicare Part
D existed even after accounting for Medicaid drug coverage. However, it remains unknown

69

whether or to what extent Medicare Part D enrollment or the LIS program has affected
disparities in nonadherence due to cost. Studies have primarily investigated the effect of
Medicare Part D and the LIS provision utilizing pharmaceutical claims and health care plan data
and do not differentiate beneficiaries by demographics. While a few studies have used nationally
representative datasets to examine CRN, racial and gender differences in change in medication
adherence before and after Medicare Part D have not been examined and the impact of the LIS
has not been considered. The one study examining disparities in CRN after the implementation
of the benefit was cross sectional as opposed to longitudinal and did not account for LIS status.
Thus, research has not investigated whether racial and gender disparities in CRN have changed
for individuals enrolled in the Medicare Part D benefit or the impact of the LIS; therefore, the
MMA’s effectiveness in eliminating racial and gender differences in nonadherence due to cost is
currently unknown.
This study builds on existing knowledge and provides important insights about CRN
under Medicare Part D. First, it uses data from a nationally representative longitudinal dataset to
analyze CRN for beneficiaries that enrolled in Medicare Part D. This allows for the examination
of gender and racial differences in CRN both before and after the implementation of this benefit.
Second, this study investigates whether the LIS directly and indirectly affects the relationship
between race, gender, health, and socioeconomic status, and CRN over time. Examining whether
disparities persist despite applying for or automatically receiving the subsidy can help us
understand the impact of the LIS, a key provision of the MMA. It also furthers our understanding
of how factors associated with cumulative disadvantage can directly and indirectly impact the
likelihood of nonadherence due to cost for older females and Black Americans enrolled in this
benefit while accounting for LIS status.

70

Based on existing literature, there are six hypotheses in this study:
5)

As compared to older Whites, older Black Americans enrolled in Medicare Part
D are more likely to report CRN over time

6)

As compared to older males, older females enrolled in Medicare Part D are
more likely to report CRN over time

7)

LIS status will mediate the effect of racial and gender differences in CRN over
time

8)

Controlling for race and gender, LIS status will decrease the likelihood of CRN
over time

9)

Controlling for LIS status, health and socioeconomic status will mediate the
effect of racial and gender differences in CRN over time

10)

Controlling for race, gender, and LIS status, poorer health status and lower
socioeconomic status will increase the likelihood of CRN over time
Methodology

Sample
This study utilizes secondary data from the 2004 and 2006 waves of The University of
Michigan (U-M) Health and Retirement Study (HRS) as well as data from the 2005 and 2007
waves of the HRS Prescription Drug Study (PDS) to conduct a longitudinal analysis. Launched
in 1992, the HRS is a nationally representative population study of more than 20,000 Americans
over the age of 50 designed to assess the relationship between health and retirement. The core
HRS survey collects information pertaining to physical and mental health, insurance coverage,
financial status, family support systems, retirement situation, and work status (Juster & Suzman,
1995). The primary mode of data collection is by telephone; however, in situations where health

71

limitations would make the hour-plus telephone session difficult or where there is no telephone
in the household, the survey is conducted in the respondent’s home (HRS, 2011). Primarily
funded through the National Institute on Aging (NIA U01AG009740), the HRS is designed,
administered, and conducted by the Institute for Social Research (ISR) at the U-M. The overall
interview response rate for the core 2004 HRS was 86.2% and for the core 2006 HRS, the overall
response rate was 88.5% (Ofstedal, 2009, June).
The PDS is a subsample of the HRS, and is a two-wave mail survey conducted in 2005 and
2007. It was specifically designed to examine prescription drug utilization as Medicare Part D
was implemented. The baseline wave was administered pre-Medicare Part D in 2005 by the ISR,
and captured information pertaining to prescription drug use, coverage, and knowledge of the
Medicare prescription drug benefit. The second wave was conducted post-Medicare Part D in
2007, and captured similar information as well as data pertaining to the impact of the MMA costcontainment provisions such as the coverage gap and restrictions. The sample was drawn from
respondents who participated in the HRS in 2004. To be eligible for inclusion in the PDS,
respondents needed to be age 65 or older in 2007 (born in 1942 or later), or have Medicare or
Medicaid coverage at some time between 2002 and 2004. Because the Consumption and
Activities Mail Survey (CAMS) was fielded at the same time as the PDS, approximately 40% of
eligible participants were excluded because of their involvement in the CAMS. However, the
exclusion did not affect the representation of the PDS sample because CAMS participants were
randomly selected (J. Faul, Personal Communication, January 4, 2012). Of those who were
eligible for inclusion in the PDS, 4,684 individuals or 88.1% completed the PDS in 2005, and
3,536 persons or 74% responded to the survey in 2007 (HRS, 2011). Given that this is a
relatively new topic area, there are no reliability or validity measures associated with the PDS (J.

72

Faul, Personal Communication, February 3, 2012).
To be included in the analytic sample for this study, respondents were required to 1) be
either non-Hispanic White/Caucasian or non-Hispanic Black/African American; 2) be age 65 or
older in 2005 and therefore eligible for Medicare when the PDS was administered; 3) have
reported taking prescription medications in 2005 or 2007; 4) have responded to questions about
prescription drug adherence in the 2005 or 2007 PDS; and 5) have reported being enrolled in
Medicare Part D in 2007 PDS. Of the 4,808 responding to the 2005 or 2007 PDS, 4,248 were
either non-Hispanic White/Caucasian or non-Hispanic Black/African American; 3,606 were age
65 and over; 3,323 reported taking prescription medications; 3,181 reported data on prescription
drug adherence or nonadherence; and 1,406 reported enrolling in the Part D benefit in the 2007
PDS. Thus, a total 1,466 respondents to the 2005 (n=1,273) or 2007 (n=1,262) PDS were
included in the final analytic sample.
The combination of HRS and PDS provides an excellent source of data for testing the
proposed hypotheses because indicators of cumulative advantage/disadvantage that could
decrease or increase older adults’ risk of CRN are available through these datasets. Variables
from the 2005 and 2007 PDS included CRN, LIS status, and Medicare Part D enrollment status.
All other variables were obtained from the 2004 and 2006 RAND HRS data files. These data
files contain a broad range of measures across HRS waves, and are developed and maintained by
the RAND Center for the Study of Aging, with support provided by the NIA and Social Security
Administration (RAND Center for the Study of Aging, 2010).
In regards to Institutional Review Board (IRB) approval, this study relied exclusively on
secondary data to complete all analysis. The HRS and PDS files contain de-identified individual

73

data that are available to the public. Therefore, the IRB at Michigan State University (MSU)
does not require human subjects review.
Measures
Dependent variable. A summary indicator of any CRN was constructed as the
dependent variable. In the 2005 and 2007 PDS, CRN was evaluated using questions about the
following three behaviors: 1) not filling a new prescription because of cost; 2) stopping taking a
medication because of cost; or 3) skipping doses of a medication to save money. If the
respondent indicated that they engaged in any of these behaviors, they were considered
nonadherent in this study. Earlier research typically used a single measure to assess CRN, and
focused solely on whether respondents failed to fill a prescription in the past year because of
cost. However, this measure did not assess for multiple ways CRN could occur. Thus, the
additional questions were developed to help monitor and identify changes in CRN, and have
demonstrated test-retest reliability (Pierre-Jacques, et al., 2008) and construct validity (Soumerai,
et al., 2006). This composite variable approach was originally developed by Safran et al. (2003)
based on a series of validated survey questions concerning CRN to prescription regimens,
confirmed using factor analysis techniques, and used as a main outcome in subsequent research
examining drug nonadherence practices of older adults (Bambauer, et al., 2007; Gellad, et al.,
2007; Madden, et al., 2008; Safran, et al., 2005; Soumerai, et al., 2006; I. B. Wilson, et al., 2004;
Zivin, et al., 2009). Because all three measures of CRN, specifically filling, stopping, or skipping
doses of a prescription because of cost, were highly correlated with comorbidity, self-reported
health status, and socioeconomic status (Bambauer, et al., 2007; Safran, et al., 2005; Safran, et
al., 2003; Soumerai, et al., 2006), it is important that they be included in this study. Older
females and Black Americans are more likely to have a lower income and multiple health

74

problems due to disadvantages throughout the life course and thus, accounting for all behaviors
can further understanding of CRN before and after the implementation of Medicare Part D.
In this study, a dichotomous variable indicating any nonadherence or no nonadherence to
the three CRN behaviors in the PDS – not filling a new prescription because of cost, stopping
taking a medication because of cost, or skipping doses of a medication to save money – was
created for each time point, or for both the 2005 and 2007 measures. The CRN variable was
initially incorporated as a continuous variable indicating a sum total of the three behaviors.
However, results were significantly skewed because the majority of the sample were adherent to
their medication regimen and similar to previous research, the variable was dichotomized
(Bambauer, et al., 2007; Gellad, et al., 2007; Madden, et al., 2008; Safran, et al., 2005; Soumerai,
et al., 2006; I. B. Wilson, et al., 2004; Zivin, et al., 2009). Using this global, dichotomous
measure is common in existing literature, and will help provide a meaningful comparison to
previous research.
Independent variables. The main predictors of interest were race (1 = “Black
American” and 0 = “White”), gender (1 = “female” and 0 = “male”), and LIS status (also
referred to as “Extra Help”) status. A three-category variable was used to indicate LIS status (0 =
“no LIS,” 1 = “applied for LIS,” “2 = “dual eligible/receiving LIS”) at each time point.
Individuals whose income and assets exceed Medicaid eligibility criteria are required to apply
for the LIS through the Social Security Administration whereas dually eligible beneficiaries
automatically receive the LIS because they qualify for both Medicare and Medicaid. Further,
level of LIS assistance is dependent on eligibility status, and the subsidy provided to dual
eligibles is greater than that provided to non-dual eligibles who applied for assistance. Thus, the
including the measure as an ordinal variable is warranted. Respondents were asked three

75

questions about LIS status. Questions related to whether or not respondents who were not duallyeligible applied for LIS differed slightly in 2005 and 2007. First, in the 2005 PDS, respondents
were asked “Do you intend to apply to Social Security for Extra Help paying for prescription
drugs?” Responses included 1 = “yes,” 5 = “no,” and 8 = “I have already applied.” Respondents
who indicated that they had applied for or intended to apply for the LIS were coded as 1 =
“applied for LIS” in the 2005 measure. Separate analyses were conducted using the threecategory variable prior to combining those that actually applied and planned to apply into one
category. The results are almost identical; in other words, significance was the same and odds
ratios changed slightly. Further, the 2005 PDS was administered during the initial LIS
application period when this subsidy first became available to respondents. Considerable
outreach efforts by the Social Security Administration were conducted during this timeframe to
help beneficiaries apply for and receive this subsidy; for this reason, I believe that intent to apply
is a good measure of having applied for LIS. Next, in the 2007 PDS, individuals were asked
“Have you ever applied for Extra Help?” Those who said “yes” were coded as 1 = “applied for
LIS” for the 2007 measure. Finally, in both the 2005 and 2007 PDS, respondents were asked
whether they were receiving Medicaid. Those who said “yes” were coded as 2 = “dual
eligible/receiving LIS” at each time point.
A time variable was also created to evaluate CRN over time (2005 and 2007) using the
“reshape” command in Stata version 11 and was used in all regression analyses as a main
predictor. The “reshape” command converted the data to a long format and provided a single
variable that was used to measure CRN at both time points. Using literature pertaining to
cumulative advantage/disadvantage as a guide, other independent variables included

76

socioeconomic characteristics and health status. Additionally, age and marital status were used as
control variables.
Socioeconomic characteristic indicators included attained education and annual
household income. Disadvantages can accumulate throughout the life course as the result of
decreased educational attainment and lower income, and result in decreased ability to afford
prescription medications and maintain adherence. The level of attained education was a
continuous variable ranging from 0 to 17 or more years of education. Cumulatively
disadvantaged individuals tend to have lower levels of education. Subsequently, they are less
likely to have prescription drug coverage and the financial resources needed to purchase
prescribed medications. Utilizing this measure will help clarify how educational attainment can
impact CRN. Additionally, it can further our understanding of how education can indirectly
impact CRN among older women and Black Americans. Annual household income was a
continuous variable ranging from $0 to $1,242,996. The variable was recoded into quartiles:
$12,720 or less; $12,721 to $23,363; $23,364 to $43,479; and $43,480 or above. The lower two
quartiles roughly correspond to the 100% and 150% federal poverty thresholds, and help
understand how CRN has been impacted for the poor and near-poor while controlling for LIS
status. Further, the measure was used as a continuous variable in Chapter Two and provided
similar results. That is, as income increased, the likelihood of CRN significantly decreased.
Health status variables included number of chronic medical conditions and self-reported
health status. Increased morbidity is associated with cumulative disadvantage and increased
reliance on pharmaceuticals, particularly among older women and Black Americans. Thus, it is
important to understand how health status impacts CRN. First, a continuous variable indicating
the number of chronic medical conditions was used. Respondents were asked, “Has a doctor ever

77

told you that you have [name of chronic medical condition]” or “our records from your last
interview show that you have [name of chronic medical condition].” This question was asked for
a series of eight chronic medical conditions and included: high blood pressure, diabetes, cancer,
lung disease, heart disease, stroke, psychiatric problems, and arthritis. The second health-related
variable included self-reported health status. Respondents were asked, “Would you say your
health is excellent, very good, good, fair, or poor?” Responses ranged from 0 = “excellent” to 4 =
“poor” with a higher score indicating that the respondent perceived their health status as being
more poor.
Finally, other control variables included age and marital status. Age was a continuous
variable ranging from age 65 to 101. Marital status was a seven-category variable: married,
married with spouse absent, partnered, separated, divorced, widowed, and never married.
Analytic strategy
Univariate descriptive statistics summarized all measures at baseline. Cross-tabulations
were conducted to determine the bivariate association between respondent characteristics and
CRN at baseline and in 2007. The Rao-Scott chi-square, a design-adjusted version of the Pearson
chi-square, was used to test the significance of the association (Rao & Scott, 1981). Mixedeffects logistic regression models were used for multivariate analysis to estimate racial and
gender differences in CRN over time for those enrolled in Medicare Part D, and to examine the
direct and indirect effect of the LIS, health and socioeconomic status, and demographic
characteristics on the relationship between race and gender and CRN.
Mixed regression models are useful for this type of analysis for several reasons
(Blackwell, et al., 2006; Rabe-Hesketh & Skrondal, 2008). First, they do not require that
respondents be measured on the same number of time points. In other traditional methods such as

78

repeated-measures ANOVA, cases are dropped if some measurements are missing (Blackwell, et
al., 2006). This is important because some participants did not respond to questions pertaining to
CRN in both surveys. In particular, 72.9% of respondents indicated adherence or nonadherence
in both the 2005 and 2007 PDS. Of the 27.1% not responding at both waves, 51.4% completed
the CRN questions in the 2005 PDS and 48.6% indicated adherence or nonadherence in the 2007
PDS. In general, by allowing the inclusion of cases with missing data, the sample size is larger
and thus more closely approximates the general population (Ott & Lngnecker, 2001).
Second, mixed regression models do not require that respondents complete the survey at
precisely the same time interval. This, again, is an essential point when considering the
administration of the PDS. The survey was mailed to respondents in October, and the field
period continued through March. It was completed at varying times over a six-month period.
Finally, mixed regression models are advantageous when analyzing repeated-measures data
because they provide an estimate of individual-specific change across time and also model
patterns of change between individuals. By modeling both fixed and random effects, there is the
ability to account for individual differences over time as well as deviations from the group trend
(Van Dongen, et al., 2004).
A series of six mixed regression models were run with the direct and indirect effects of
each set of variables on the relationship between race, gender, and CRN over time examined in
each model. Model 1 included only the time variable. Race and gender were added in Model 2
and in Model 3, LIS status variables were included. Demographic variables were added in Model
4. Model 5 added the health status variables and in Model 6, socioeconomic status variables were
included. A likelihood-ratio test was performed for each model to determine whether including
time as a fixed and random effect provided a better fit than the random-intercept only model. A

79

null model was first fit with the dependent variable only to determine the level of betweenperson variance. Next, the fixed effect of time and a random effect associated with the intercept
of each respondent were added to the model. A third model then added a second random effect
that allowed each person to have a unique outcome at each time point. With the exception of the
null model, this process was repeated for Model 1 through Model 6. An alpha of .05 was used as
the cutoff for determining statistical significance. In every analysis, the likelihood-ratio test was
not significant. Thus, time was included only as a fixed effect in the regression models.
All statistical analyses were performed in the Stata version 11.0 statistical package
(StataCorp). Stata’s survey commands were used in univariate statistics and bivariate tests of
association to account for the complex multistage clustered sampling design of the HRS when
computing standard errors. The PDS questionnaire weights (a product of the HRS sampling
weights) were applied to adjust for sample selection probability or oversampling of select
minority groups and for interviewee nonresponse in univariate and bivariate analysis only (HRS,
2011). Stata does not allow the incorporation of the survey command or application of
probability weights in mixed-effect logistic regression analyses. Thus, mixed regression models
did not control for the design effect and were not weighted. A consequence of not utilizing the
questionnaire weights or controlling for the design effect is the potential for inaccurate point
estimates and/or inaccurate standard errors, which in turn can impact the ability to infer results to
the general population. Despite this limitation, mixed-effect regression analyses can provide a
strong indication of racial and gender difference in CRN over time because unlike other methods
such as time-series ANOVA, cases are not dropped if there are missing values. Further, there
was no substantive difference in outcomes in bivariate models that were weighted and controlled
for the design effect as compared to multivariate models. Thus, the potential that the results will

80

inaccurately reflect the general population is minimal. Significance in multivariate analysis was
tested at the p<.05 level using a one-tailed test.
Results
Description of Sample at Baseline
A detailed description of the sample at baseline or in 2005 is provided in Table 3.1. Of
the 1,273 individuals responding to the question about medication adherence at baseline and
enrolling in Medicare Part D in 2007, 10.3% self-identified as non-Hispanic Black or African
American and the majority (61.3%) were female. In regards to LIS status, 17.2% applied for the
LIS and 15.9% dual eligibles receiving the LIS. Respondents ranged in age from 65 to 101 with
a mean age of 73.8 years, and slightly less than half (49.7) were married. Respondent self-reports
indicated that the majority felt their health was good (33.2%), very good (27.8%), or fair
(21.2%), followed by excellent (9.2%), and poor (8.6%). The average number of chronic health
conditions was 2.43 (SD = 1.38). Over a quarter of respondents had annual incomes between
$23,364 and 43,479 (26.7%) or $43,480 and greater (26.7%), while 22.3% had annual incomes
between $12,721 and $23,363, and 21.4%% had annual incomes $12,720 and less. the mean
years of education was 12.29 (SD = 2.92). Finally, 18.8% of respondents reported cost-related
medication nonadherence in 2005.
Bivariate Results
Bivariate analyses of medication adherence and nonadherence in 2005 and 2007 are
presented in Table 3.2. In regards to the main predictor variables, a higher proportion of Black
Americans compared to Whites were nonadherent both before and after Medicare Part D, while a
greater proportion of females were nonadherent compared to males before Medicare Part D only.

81

TABLE 3.1
Description of Sample of Individuals Age 65 and Over Reporting Medication
Adherence/Nonadherence in 2005 and Enrolling in Medicare Part D (n=1,273)
Variable
Race
Black American
White
Gender
Female
Male
2
Low-Income Subsidy Status
Dual eligible/receiving Low-Income Subsidy
Applied for Low-Income Subsidy
No Low-Income Subsidy
Age (range 65 to 101)
2
Marital status
Married
Married, spouse absent
Partnered
Separated
Divorced
Widowed
Never married
Number of chronic medical conditions (range: 0 to 7)
2
Self-reported health
Excellent
Very good
Good
Fair
Poor
2
Education (range: 0 to 17)
Annual household income
$12,720 and less
$12,721 to $23,363
$23,364 to $43,479
$43,480 and above
Cost-related adherence or nonadherence
Adherent
Nonadherent
1
2

1

% / SD

224
1,049

10.3
89.7

805
468

61.3
38.7

180
212
615
73.8

15.9
17.2
66.9
7.12

653
13
27
17
111
415
37
2.43

49.7
1.0
1.8
1.3
8.9
33.9
3.4
1.38

113
333
421
288
117

9.2
27.8
33.2
21.2
8.6

12.13

3.08

314
309
325
325

21.4
22.3
26.7
29.6

1,007
266

81.2
18.8

Weighted %, unweighted N
Sample size varies slightly due to missing data within PDS or HRS

82

1

N /M

In regards to LIS status, a significantly larger proportion of respondents that applied for the LIS
reported nonadherence in both 2005 and 2007.
In terms of control variables, age was significantly related to CRN. Nonadherent older
individuals reported a significantly lower age than adherent older adults before and after
Medicare Part D. Marital status was not significantly related to CRN.
In regards to health status variables, respondents who reported CRN had a significantly
greater number of chronic conditions compared to those that maintained adherence in 2005 and
2007. A smaller proportion of respondents in excellent health reported nonadherence than those
in other categories with those in fair or poor health reporting the most nonadherence. This
relationship was only significant in 2007.
In terms of socioeconomic status variables, annual household income and educational
attainment were not significantly related to nonadherence before or after Medicare Part D.
Bivariate analyses suggest that without controlling for other factors, older Black
Americans were more likely to report CRN than older Whites before and after the
implementation of Medicare Part D as expected. Gender appeared to impact CRN before
Medicare Part D only. Respondents that applied for the LIS were more likely to report CRN over
time as compared to those who did not apply. Older individuals who were younger or had a
greater number of chronic medical conditions appeared more likely to report nonadherence
before and after Medicare Part D. Respondents that self-reported their health as being very good,
good, fair, or poor (as compared to excellent) appeared more likely to be nonadherent due to cost
in 2007 only.

83

TABLE 3.2
Bivariate Relationships Between Independent Variables and Medication Adherence/Nonadherence, 2005 and 2007
2005 (n=1,273)
2007 (n=1,262)
Adherence
Nonadherence
Adherence
Nonadherence
1

1

1

1

N
Race
Black American
White
2

! , df=1
Gender
Female
Male

%

N

%

N

%

N

%

163
844

73.5
82.1

61
205

26.5
17.9

139
887

71.6
84.4

56
180

28.4
15.6

82.3
84.2

150
86

17.7
15.8

81.6
75.7
84.2

20
54
160

18.4
24.3
15.8

3

4.03*

611
396

2

! , df=1

3

77.5
87.1

194
72

22.5
12.9

20.62***

2

88.5
66.8
87.2

79
27
82

11.5
33.2
12.8

18.99***

2

Age (M±SD)

df=1, 51

3

3

Marital status
Married
Married, spouse absent
Partnered
Separated
Divorced

71.74 ± 6.41

46.42***
523
13
20
15
86

3

82
127
801
3

3.18*

74.31 ± 7.55
2

634
392
0.70*

2

Low-Income Subsidy (LIS) Status
Dual eligible/full LIS
153
Applied for LIS
133
No LIS
533
! , df=1

3

15.49***

76.18 ± 70.3
3

74.22 ± 5.87

21.75***
82.3
1.0
77.3
89.9
81.9

129
0
7
2
25

17.7
0.0
22.7
10.1
18.1

84

537
16
13
11
83

84.3
93.6
66.8
87.0
83.5

116
1
9
2
23

15.7
6.4
33.2
13.0
17.5

TABLE 3.2
Bivariate Relationships Between Independent Variables and Medication Adherence/Nonadherence, 2005 and 2007
2005 (n=1,273)
2007 (n=1,262)
Adherence
Nonadherence
Adherence
Nonadherence
Marital status (con’t)
Widowed
Never married

1

1

N

%

N

%

N

%

324
25

2

79.9
69.0

91
12

20.1
31.0

329
26

81.5
83.9

76
5

18.5
16.1

3

1.24

2.38 ± 1.37

2.67 ± 1.43

8.58**

99
277
330
219
81
3
2.39

2

3

2

Annual household income
$12,720 and less
$12,721 to $23,363

86.1
84.4
80.5
79.2
73.5

12.18 ± 3.25

14
56
91
69
36

13.9
15.6
19.5
20.8
26.5

2.54 ± 1.37
3
13.82***
75
282
354
207
95
3
4.36**

11.92 ± 3.25

0.84

242
237

3

0.96

3

df=1, 51
2
Self-reported health
Excellent
Very good
Good
Fair
Poor
2
! , df=4
df=1, 51

1

%

2

! , df=6
Number of chronic
2
medical conditions

Education (M±SD)

1

N

93.0
84.1
84.8
76.9
82.3

12.17 ± 2.96

3

2.95 ± 1.40

9
52
75
69
26

7.0
15.9
15.2
23.1
17.7

11.93 ± 3.02

1.38
80.2
79.2

72
72

19.8
20.8

85

200
237

81.7
81.5

51
68

18.3
18.5

TABLE 3.2
Bivariate Relationships Between Independent Variables and Medication Adherence/Nonadherence, 2005 and 2007
2005 (n=1,273)
2007 (n=1,262)
Adherence
Nonadherence
Adherence
Nonadherence
N

1

2

! , df=3
1
2
3

3

1

1

N

%

N

%

N

%

79.2
85.3

71
51

20.8
14.7

279
299

81.6
87.0

67
46

18.4
13.0

2

Annual household income (con’t)
$23,364 to $43,479
254
$43,480 and above
274

1

%

1.44

3

1.59

Weighted %, unweighted N
Sample size varies slightly due to missing data within Prescription Drug Study (PDS) or Health and Retirement Study (HRS)
Test statistic is Rao-Scott chi-square for categorical variables and F means test for continuous variables

*Significant at p<.05; **Significant at p<.01; ***Significant at p<.001

86

Mixed-Effect Logistic Regression Results
Table 3.3 shows mixed-effects logistic regression models for CRN outcomes. Model 1
projected CRN over time only. Respondents enrolling in Medicare Part D were 22% (OR=.78)
less likely to report CRN in 2007 as compared to 2005. Next, Model 2 estimated the effects of
race and gender on CRN over time, controlling for each other and time. Compared to older
Whites, Black Americans were over twice as likely to report CRN both before and after the
implementation of Medicare Part D (OR=2.22). Females were 1.50 times more likely than males
to experience CRN in both 2005 and 2007.
The effect of race on CRN over time remained significant with the addition of LIS status
variable in Model 3. Applying for or receiving the LIS appeared to have little indirect effect on
the relationship between race and CRN. However, the odds ratio for the relationship between
females and males decreased 26% (OR=1.24) and was no longer significant with the inclusion of
the LIS status variable. Two separate dichotomous variables indicating whether the respondent
applied for the LIS (versus no LIS) and whether the respondent was a dual eligible receiving the
LIS (versus no LIS) were created to determine which LIS status category indirectly affected the
relationship between gender and CRN over time. The effect of gender on CRN did not change
when adding the dual eligible/receiving LIS variable to the model. However, the relationship was
no longer significant with the addition of the applied for LIS variable (additional analysis, not
shown). This suggests that gender differences in nonadherence over time are driven substantially
by whether an individual applied for the LIS. A test of mediation confirmed this finding (Baron
& Kenny, 1986). That is, gender was related to applying for the LIS, gender was associated with
CRN, and applying for the LIS was related to CRN. Further, time was no longer a significant
predictor of CRN with the addition of the applied for LIS variable, indicating that applying for

87

TABLE 3.3
Mixed-Effects Logistic Regression Results of Medication Nonadherence
Model 1
Model 2
OR
OR
Year
2007
.78*
.79*
2005 (reference group)
--Race
Black American
2.22***
White (reference group)
-Gender
Female
1.50**
Male (reference group)
-Low Income Subsidy (LIS) status
Dual eligible/full LIS
Applied for LIS
No LIS
Age (continuous)
Marital status
Currently married (reference category)
Married, spouse absent
Partnered
Separated
Divorced
Widowed
Never married
Number of chronic medical conditions (continuous)
Self-reported health
Excellent (reference group)
Very good

88

Model 3
OR

Model 4
OR

Model 5
OR

Model 6
OR

1.06
--

1.18
--

1.10
--

1.10
--

1.97***
--

1.75**
--

1.53*
--

1.51*
--

1.24
--

1.26
--

1.26
--

1.24
--

.81
3.13***
--

.95
3.16***
-.94***

.79
2.63***
-.94***

.77
2.57***
-.94***

-.14*
2.81*
.39
.80
1.14
1.17

-.14*
2.94*
.38
.76
1.08
1.18
1.22***

-.13*
2.74*
.41
.79
1.10
1.22
1.21**

-1.36

-1.33

TABLE 3.3
Mixed-Effects Logistic Regression Results of Medication Nonadherence
Model 1
Model 2
OR
OR
Self-reported health (con’t)
Good
Fair
Poor
Education (continuous)
Annual household income
$12,720 and less
$12,721 to $23,363
$23,364 to $43,479
$43,480 and above

Model 3
OR

Model 4
OR

Model 5
OR

Model 6
OR

1.47
1.96*
1.86*

1.39
1.82*
1.74*
.98
1.08
1.45
1.54*
--

*Significant at p<.05; **Significant at p<.01; ***Significant at p<.001
Model 1:
Model 2:
Model 3:
Model 4:
Model 5:
Model 6:

Year (N=1,466)
Year, race, and gender (N=1,466)
Year, race, gender, and Low-Income Subsidy status (N=1,425)
Year, race, gender, and Low-Income Subsidy status, age, and marital status (N=1,421)
Year, race, gender, and Low-Income Subsidy status, age, marital status, and health status (N=1,420)
Year, race, gender, and Low-Income Subsidy status, age, marital status, health, and socioeconomic status (N=1,420)

89

the LIS indirectly affected nonadherence before and after Medicare Part D. It also appears that
applying for the LIS had a direct effect on CRN in 2005 and 2007 when controlling for race,
gender, and time. As compared to those with no LIS, individuals who applied for the LIS were
over three times more likely (OR=3.13) to report CRN before and after Medicare Part D.
The effect of race on CRN over time remained significant with introduction of age and
marital status variables in Model 4, although the odds of older Black Americans reporting CRN
decreased 22% (OR=1.75). Thus, the addition of these variables appeared to have some indirect
effect on the relationship between race and CRN over time. The impact of age and marital status
on the relationship between LIS status and CRN over time was minimal. Further, as compared to
older individuals who were currently married, odds of CRN were greater for those who were
partnered (OR=2.81) and less for those who were married but their spouse was absent (OR=.14).
Finally, respondents who were older had a decreased likelihood of CRN (OR=.94).
The inclusion of health status variables in Model 5 appeared to mediate some of the
effect of race on CRN over time, although the significant relationship was maintained. The odds
of older Black Americans experiencing CRN in 2005 and 2007 decreased 22% (OR=1.53). The
effect of applying for the LIS on CRN over time remained significant, though the odds of
nonadherence were reduced 53% (OR=2.63) indicating that the addition of these variables
indirectly affected CRN before and after Medicare Part D. Further, both number of chronic
medical conditions and self-reported health had a direct effect on CRN over time when
controlling for race, gender, time, LIS status, age, and marital status. Respondents with a greater
number of chronic medical conditions had an increased likelihood of CRN (OR=1.22). Further,
as compared to older individuals who reported their health as being excellent, odds of CRN were
greater for those who rated their health as fair (OR=1.96) or poor (OR=1.86).

90

With the addition of socioeconomic status variables in Model 6, the effect of race and
LIS status on CRN over time remained significant. Educational attainment and annual household
income appeared to have very little indirect effect on the relationship between race and CRN or
applying for the LIS and CRN as only a slight change in odds ratios was observed. Further,
annual household income had a direct effect on CRN before and after Medicare Part D. As
compared to respondents in the highest income group, individuals with an annual income
between $23,364 and $43,479 were 1.54 times more likely to experience CRN over time.
Discussion
The implementation of the MMA represented an essential policy initiative. The Medicare
Part D benefit and LIS program were clearly needed to help older persons gain access to
medications and reduce out-of-pocket prescription drug expenditures, yet the extent to which the
policy reduced or eliminated racial and gender disparities was unknown. This is the first
longitudinal study to examine whether and to what extent nonadherence due to cost over time
persists for older females and Black Americans enrolled in Medicare Part D, and how the LIS
directly and indirectly affects the relationship between race, gender, and CRN over time. Further,
it investigates how factors associated with cumulative disadvantage directly and indirectly affect
the likelihood of CRN over time for older individuals enrolled in the benefit while controlling for
LIS status.
In terms of hypotheses organizing the present analyses, there are several findings to
discuss. The first hypothesis that older Black Americans enrolled in Medicare Part D would be
more likely than Whites to report CRN over time was supported. This finding expands previous
research illustrating racial disparities in CRN (Gellad, et al., 2007; Klein, et al., 2004; Soumerai,
et al., 2006; Steinman, et al., 2001) by demonstrating that despite enrolling in Medicare Part D,

91

older Black Americans were significantly more likely to report CRN both before and after the
implementation of this benefit. It is therefore suggested that cost continues to be a barrier to
maintaining adherence for older Black Americans.
The second hypothesis predicted that compared to older males, older females enrolled in
Medicare Part D would be more likely to report CRN over time. This hypothesis was supported.
This study replicates previous findings showing that women were more likely to experience
CRN than men (Heisler, et al., 2005; Piette, et al., 2004c; Zivin, et al., 2010), and expands this
research by illustrating that gender disparities existed over time despite Medicare Part D
enrollment. This suggests that although a drug benefit became available under Medicare, women
continue to have more difficulty than men affording medications and maintaining adherence. A
potential reason why disparities persist is that Medicare Part D’s cost-sharing provisions
continue to present barriers to maintaining adherence.
Next, it was predicted that LIS status would mediate the effect of racial and gender
differences in CRN over time. Results for this hypothesis are mixed depending on whether
respondents were dually eligible and automatically received the LIS or applied for LIS on their
own. Specifically, dual eligibility did not mediate the effect of either race or gender. This
finding is similar to research conducted before the MMA that found continuing racial and gender
disparities in CRN among Medicaid recipients (Bambauer, et al., 2007; Soumerai, et al., 2006).
Though it was anticipated that some out-of-pocket costs could be lower under the LIS than state
Medicaid plans (Guyer & Schneider, 2004), it appears that differences in CRN persisted over
time despite the transition from Medicaid drug plans to Medicare Part D.
Applying for the LIS mediated the effect of gender on CRN over time, but not the effect
of race. This is a bit perplexing given that lower income individuals were more likely to apply

92

for the LIS (Davidoff, et al., 2010). Additional analyses (not shown) demonstrated that in 2005
and 2007, a significantly higher proportion of Black Americans had an annual household income
in the lowest quartile than Whites and thus, had a greater likelihood of qualifying for the LIS.
One potential explanation for this is that compared to Whites, Black Americans at or below
150% of poverty are significantly less likely to be aware of the LIS benefit (Neuman, et al.,
2007). This is the case among this sample as well (additional analysis, not shown). Only 36% of
older Black Americans reported being aware of the LIS in 2007, as compared to 59% of Whites.
While research has not specifically examined whether they are more likely to encounter barriers
with the LIS applications process, it is suggested that lack of awareness is related barriers within
the medical setting which are more common from Black Americans (Smedley, Stith, & Nelson,
2003). Thus, the importance of outreach to increase awareness of the benefit is certainly
demonstrated, particularly in light of how applying for the LIS affected gender differences in
CRN over time. In this study, 23% of older females applied for the LIS in 2005 as compared to
11% of older males, and in 2007, 18% of older women applied as compared to 8% of older
males.
The fourth hypothesis that while controlling for race and gender, LIS status would
decrease the likelihood of CRN over time was not supported. While CRN was less for dual
eligibles than non-dual eligibles before Medicare Part D (Gellad, et al., 2007; Soumerai, et al.,
2006), there was no difference in CRN over time between dually eligible individuals and those
with no LIS. A potential reason could involve increased copayment requirements (Huskamp, et
al., 2009). It was expected that for some dual eligibles, cost-sharing obligations were expected to
be higher under the LIS and Medicare Part D than Medicaid for some beneficiaries (Guyer &
Schneider, 2004). This, in turn, could impact CRN differences that previously existed.

93

Additionally, those non-dual eligibles that applied for the subsidy were significantly more likely
than those with no LIS to report CRN over time. It seems somewhat counterintuitive that
applying for the LIS would result in an increased likelihood of CRN given that cost-sharing
assistance can substantially reduce out-of-pocket expenditure requirements. However, there are
two potential explanations. First, this study included all respondents that enrolled in Medicare
Part D, regardless of income, assets, and whether or not they qualified for the LIS. In other
words, if an individual did not apply for the LIS, it may be because they exceeded the income or
asset guidelines. Those with the greatest need in terms of socioeconomic disadvantage and
poorer health status – and therefore, at greater risk of CRN – were most likely to apply for the
LIS (Davidoff, et al., 2010). Second, even though an individual applied for the LIS, cost-sharing
requirements still exist and can vary depending on income. Though research has not examined
how cost-sharing under the LIS affects CRN, it is suggested that copayments could pose
difficulties for some beneficiaries and thus impact adherence given that income and asset
requirements for LIS coverage are relatively low.
The fifth hypothesis that controlling for LIS status, health and socioeconomic status will
mediate the effect of racial and gender differences in CRN over time was not supported.
Although the odds of CRN for older Black Americans decreased with the addition of these
variables compared to older Whites, the effects were small and racial differences in CRN over
time remained significant. In other words, racial disparities continued even after accounting for
factors associated with cumulative disadvantage and LIS status. Regarding gender differences,
the inclusion of the LIS status variable in Model 3 mediated the relationship between gender and
CRN over time. In additional analysis (not shown), the LIS variable was removed in an attempt
to understand whether health and socioeconomic status indirectly affected the relationship

94

between gender and CRN. While the odds of CRN decreased over time, older females were
significantly more nonadherent than males. Thus, this suggests that gender differences in CRN
were largely influenced by LIS status and factors related to cumulative disadvantage had little
impact on this relationship.
With regard to racial disparities, it is difficult to assess the reasons why differences
persist after controlling for factors associated with cumulative disadvantage – specifically LIS
status, health, and socioeconomic status – because of limitations with existing research. There
are, however, two potential explanations. First, this study could not control for individual
differences within Part D plans. There are over 40 Medicare Part D plans (Kaiser Family
Foundation, 2011b). Cost-sharing can vary extensively within plans (Hoadley, et al., 2006) and
exist, regardless of LIS status. Each plan’s out-of-pocket expenditure requirements can impact
CRN. For example, cost-sharing requirements are typically greater for individuals with diabetes
than most other chronic illnesses under Medicare Part D plans (Ettner, et al., 2010). Because
older Black Americans are more likely to be diabetic than Whites (Schoenbaum & Waidmann,
1997; Schoenborn & Heyman, 2009) and have comorbid health problems requiring multiple
medications (Skarupski, et al., 2009), their chances of incurring cost-sharing under Part D plans
is typically greater. Further, Black Americans had a significantly greater number of medications
in 2005 and 2007 as compared to Whites (additional analysis, not shown). Thus, the variation in
cost-sharing may have a greater impact on older Black Americans than Whites because their
health status is typically poorer and as a result, they often have higher prescription drug needs.
This, in turn, can impact their ability to purchase prescribed medications and maintain adherence.
A second explanation involves factors involving the physician-patient relationship and
trust. The physician-patient relationship has been found to influence CRN (I. B. Wilson, et al.,

95

2004). Higher out-of-pocket prescription drug expenditures have been associated with a greater
likelihood to forgo medications because of cost if trust in the physician is low (Piette, et al.,
2005). Further, communication regarding costs can influence CRN. Individuals that have
experienced problems paying for their medications are often not asked if they can afford the cost
(Piette, et al., 2004b). It is possible that for some older Black Americans in this study, physician
trust and/or communication were low and contributed to CRN. For example and as discussed
previously, a lower proportion of Black Americans knew about the LIS program than Whites.
This could be an indication that physicians are communicating with older Black Americans
about the benefit less frequently than they are with Whites. More research is needed to
understand the role of physician trust/communication as it relates to CRN under Medicare Part
D.
Finally, it was predicted that controlling for race, gender, and LIS status, poorer health
status and lower socioeconomic status would increase the likelihood of CRN over time.
Regarding health status, poorer health directly affected CRN as predicted. In this study, older
persons with two or more chronic medical conditions were more likely than those with no or one
medical condition(s) to report CRN before and after Medicare Part D. Additionally,
nonadherence due to cost was more likely for individuals reporting fair or poor health than those
reporting excellent, very good, or good health. This is consistent with previous findings that
individuals in poor health with multiple comorbidities have high and persistent CRN over time
(Briesacher, et al., 2007; Madden, et al., 2008; Safran, et al., 2005). CRN can be caused by
higher out-of-pocket expenditures due to multiple medications (Frankenfield, et al., 2010; Zivin,
et al., 2010). While enrolling in Medicare Part D may have alleviated some of the financial
burden for sicker beneficiaries, it is suggested that cost-sharing requirements under the benefit

96

can still be problematic regardless of whether or not the beneficiary is receiving or has applied
for the LIS and thus, have little impact on CRN.
When considering socioeconomic status, this prediction was partially supported. Similar
to previous cross-sectional studies conducted before the implementation of Medicare Part D,
lower annual household income increased the likelihood of nonadherence due to cost (Gellad, et
al., 2007; Mojtabai & Olfson, 2003; Soumerai, et al., 2006; Steinman, et al., 2001; Zivin, et al.,
2010). In this study, older adults with an annual household income between $23,364 and $43,479
had a greater likelihood of nonadherence compared to those with an income of $43,480 or more.
Interestingly, CRN was not significantly different when comparing individuals with an annual
income of $23,363 or less to those with an income of $43,480 and above. There are two potential
explanations. First, after removing the LIS status variable from the model (additional analysis,
not shown), older persons in the second quartile or in the $12,721 to $23,363 income range had a
significantly greater likelihood of CRN over time than those with an income in the fourth
quartile or $43,480 and above. Differences appeared to be mediated by whether an individual
applied for the LIS suggesting that the subsidy may have alleviated some of the medication cost
burden for lower income beneficiaries. The LIS can also help us understand why significant
differences persist when comparing individuals earning between $23,364 and $43,479 annually
to those earning $43,480 or more. Individuals with near-poor incomes of 150% to 200% of the
Federal Poverty Level (FPL) were ineligible for the LIS, but typically have greater out-of-pocket
health care costs compared to higher income individuals due to poorer health status (Briesacher
et al., 2010). Thus, maintaining adherence may be difficult for older adults with annual income
between $23,364 and $43,479 because while they experience a heavier financial burden for
health-related costs, they are minimally ineligible for the LIS. In other words, a portion of

97

respondents within the income category would have a near-poor income of 150% to 200% FPL
which disqualifies them from receiving assistance.
While removing the LIS status variables from the model helps clarify findings regarding
the lower income quartiles, it fails to explain the lack of significant difference between the
lowest income group and the highest income group or levels of educational attainment. This can
be a bit confounding given the significant relationships in Chapter Two. A potential reason
involves the representation of the sample. The sample in Chapter Two contained far more
variation in source of prescription drug coverage because it included all adults aged 65 and older.
The current study only included Medicare Part D enrollees. Those enrolling in the Part D benefit
had a greater likelihood of being less educated, lower income, and having multiple morbidities as
compared to those who did not enroll (Skarupski, et al., 2009). Thus, respondents may have been
more homogenous in terms of income and this could account for the lack of differences
observed.
An additional finding worthy of note involves the age variable. As indicated in Chapter
Two, it was somewhat surprising that persons the likelihood of CRN over time decreased as age
increased, particularly because research shows lower prescription drug coverage rates among
older age groups (Kanavos & Gemmill-Toyama, 2010). This finding, however, is consistent with
cross-sectional studies using HRS to examine CRN before the implementation of Medicare Part
D (Klein, et al., 2004; Zivin, et al., 2010). One potential reason involved the sampling of the
HRS and exclusion of institutionalized persons. Because risk of institutionalization or nursing
home placement increases with age, it was suggested that the older age groups participating in
the HRS are disproportionately healthy (Klein, et al., 2004). This explanation, however, could

98

not be confirmed through additional analysis. Therefore, future research is needed to understand
why younger persons were more likely to experience CRN than older age groups.
Strengths of this study include the use of a large nationally representative sample of older
adults and its ability to examine multiple predictors simultaneously and over time. However,
there are a number of limitations that must be considered. In the future, a longer prepolicy and
postpolicy series would provide more clarity and permit a stronger assessment of Part D
enrollment. Other unmeasured factors may have influenced the results. For example, the
transition of dual eligibles from Medicaid to Medicare drug plans in 2006 was confusing for
many older adults (Kaiser Family Foundation, 2006) and resulted in difficulties obtaining
medications (West, et al., 2007). A significantly greater proportion of older Black Americans in
this study were dually eligible as compared to Whites, and thus, had a greater likelihood of being
impacted by the transition. As a result, CRN may have been overestimated because data
collection occurred shortly after the launch of Medicare Part D and at a time of initial confusion.
However, the PDS only contained two waves of data at the time this study was conducted. While
this research provides important evidence on the impact of the policy on racial and gender
disparities in CRN, the results should be considered early evidence until longer-term data
become available.
Second, socioeconomic and health status were measured in the 2004 and 2006 HRS, and
CRN was measured in the 2005 and 2007 PDS. Thus, there is the potential that income or certain
health conditions may have changed because these measures were not taken simultaneously or
when CRN was reported. While this seems less likely with income because older adults are
largely dependent on Social Security, there is greater potential for fluctuation in health status.
Depression, for example, can be transitory (Zivin, et al., 2010). Older females and Black

99

Americans who enrolled in Medicare Part D have a greater prevalence of depression as
compared to older males and Whites (Skarupski, et al., 2009). Thus, the indirect effect of health
status on the relationship between race, gender, and CRN may have been inaccurately
represented in the results.
Third, this study investigated nonadherence due to cost only and did not account for other
factors that can impact adherence. For example, an individuals’ concerns or beliefs regarding
treatment through the use of medications can certainly impact adherence (Aikens & Piette,
2009). Some medications may be viewed as more essential than others (Lau, et al., 2008; Piette,
et al., 2006) and there could be racial and gender differences in whether a prescription is
perceived as necessary, particularly when faced with cost concerns. As compared to older males
and Whites, older females’ and Black Americans’ ability to afford out-of-pocket expenditure
requirements is typically less due to lower socioeconomic status (Wei, et al., 2006). Thus, they
may perceive the need for certain medications differently because of economic concerns.
Additionally, physician trust and communication can be a factor in medication adherence (Piette,
et al., 2005). It is possible that for some older females and Black Americans in this study,
physician trust and/or communication were low and contributed to CRN. While this could
account for some variation in CRN, these factors were not the focus of this research but will be
examined in future studies.
Fourth, older Black Americans had a significantly greater proportion of missing data
when asked whether they applied for LIS in the 2005 and 2007 PDS than older Whites.
Individuals can be confused by or unaware of Medicare Part D and the LIS provision because of
its complexity (Davidoff, et al., 2010; Heiss, et al., 2006; Hsu, et al., 2008). As previously stated,
older Black Americans were significantly less likely to be aware of the LIS, and lack of

100

awareness about the subsidy could have contributed to a the greater proportion of missing data.
This, in turn, could bias the results and underestimate the racial effect of the LIS on CRN over
time. Further, this study relied on self-reported survey measures. Self-reporting can be subject to
social desirability bias (Nederhof, 2006). Respondents may be reluctant to admit failure to
adhere to a prescribed medication regimen – particularly if it is due to cost factors. Because older
women and Black Americans are more likely to be economically disadvantaged, they may incur
greater difficulty affording the cost of prescribed medications – and, admitting failure to adhere
because of economic reasons. This bias could contribute to underestimating the prevalence of
CRN.
Fifth, there are limitations with measures contained in the HRS and PDS. First, there is
the need for better representation of non-married partners in the HRS and other samples used in
health benefits research. Results suggest that partnered individuals enrolled in Medicare Part D
have a greater likelihood of CRN than married individuals. However, less than 2% (n=27) of the
sample indicated they were in a partnered relationship when asked about marital status, and it is
therefore difficult to generalize these findings. Failure to address this limitation can potentially
result in underestimating CRN among partnered individuals. It is important to understand
whether differences in CRN exist between partnered and married individuals in order to
effectively design social work interventions. Second, the PDS does not ask respondents whether
they took less than the prescribed dosage of a medication or if they cut pills in half to make a
medication last longer. The survey question asked if a dosage of a medication was skipped to
save money, and this measure could be a potential indicator of these behaviors. However, failure
to ask if these behaviors specifically occurred could underestimate CRN. Third, the PDS only
asks respondents if they applied for or are receiving the LIS. It does not ask individuals if they

101

use this assistance when purchasing their medications. Respondents who applied for or are
receiving the LIS may not necessarily utilize the subsidy; however, an exact measure of LIS
utilization was not available within the 2005 or 2007 PDS or the HRS core survey. In my
experience as an agency administrator, Medicare beneficiaries who apply for or receive the LIS
typically utilize this assistance. For this reason, I believe that these measures of LIS status are an
adequate indicator of utilization.
Finally and as stated previously, this study utilized a composite measure to estimate CRN
as opposed to including all three behaviors as separate indicators: 1) not filling a new
prescription because of cost; 2) stopping taking a medication because of cost; or 3) skipping
doses of a medication to save money. While this is consistent with previous research and
therefore useful for cross-study comparison (Bambauer, et al., 2007; Gellad, et al., 2007;
Madden, et al., 2008; Safran, et al., 2005; Soumerai, et al., 2006; I. B. Wilson, et al., 2004; Zivin,
et al., 2009), measuring each behavior separately may provide important insights for social work
practice on racial and gender disparities in CRN. For example, one behavior could be more
prevalent than others among older women or Black Americans. By understanding which types or
types of nonadherence are more common in older women and Black Americans, interventions
that specifically target certain behaviors can be designed. Thus, future research will examine
each CRN item as an outcome.
Within the constraints of these limitations, the study makes a strong and significant
contribution to our understanding of whether and to what extent racial and gender disparities
exist in CRN before and after enrolling in Medicare Part D. It also extends our understanding of
how the LIS directly and indirectly affects the relationship between race, gender, socioeconomic
and health status, and CRN over time. Evidence has shown that the MMA provided much-

102

needed assistance to those previously lacking prescription drug coverage, and has been effective
in reducing out-of-pocket prescription costs. However, it was unclear whether the policy
eliminated racial and gender differences in CRN for those enrolled in the Part D benefit. Overall,
the study suggests that racial and gender disparities in CRN persisted after accounting for health
and socioeconomic status. The LIS mediated gender differences. However, the subsidy did not
affect racial differentials. Applying for the LIS, poorer health status, and having a lower annual
income increased the likelihood of CRN, however differences between those earning $12,721 to
$23,363 and the highest income group were mediated by applying for the LIS.
Findings are consistent with policy expectations that the LIS would lower prescription drug costs
for beneficiaries with limited income and increase access to medications. As demonstrated,
gender differences in CRN were substantially driven by whether an individual applied for the
LIS as were differentials between lower and higher income groups. However, older Black
Americans, the near-poor, and individuals in poor health continue to be disproportionately
nonadherent. These findings indicate the need to target outreach efforts towards these groups to
raise awareness of the LIS and increase participation. Additionally, policymakers should
consider raising the LIS income requirements to 200% FPL.

103

CHAPTER FOUR
MANUSCRIPT THREE
ABSTRACT
The objectives of the fourth chapter of this multiple manuscript dissertation were to
determine: 1) to what extent race and gender impact the likelihood of cost-related medication
nonadherence (CRN) under Medicare Part D plans; and 2) to what extent the Medicare Part D
coverage gap and coverage restrictions directly and indirectly affect the relationship between
race, gender, socioeconomic and health status and CRN. To evaluate CRN, secondary data from
the 2006 wave of the Health and Retirement Study (HRS) and data from the 2007 wave of the
Prescription Drug Study (PDS) were utilized. The analytic sample for this cross-sectional
analysis consisted of 1,353 respondents age 65 and older who were enrolled in Medicare Part D,
taking at least one prescribed medication, and who responded to questions about adherence or
nonadherence in 2007. The outcome variable of interest was CRN, which included not filling,
stopping, or skipping doses of a prescription because of cost. The main predictors were race
(Black American and White), gender (male and female), Medicare Part D coverage gap (reached
gap, did not reach gap, and no coverage gap), and coverage restrictions (experienced restrictions
and did not experience restrictions). Covariates included demographics, socioeconomic, and
health status. The analysis was conducted using logistic regression models. Results indicated that
racial disparities in CRN existed under Medicare Part D after accounting for demographics,
health status, socioeconomic status, and the coverage restrictions. However, the inclusion of the
coverage gap variable mediated differences in CRN between older Black Americans and Whites.
Having a coverage gap or coverage restrictions had a direct effect on CRN. Poorer health status,
lower income, and having applied for the Low-Income Subsidy (LIS) increased the likelihood of

104

nonadherence, even after accounting for the coverage gap and coverage restrictions. In
conclusion, racial differences in CRN were substantially driven by the Medicare Part D coverage
gap. The coverage gap and restrictions, having a lower annual income, poorer health, and
applying for the LIS were associated with CRN under Medicare Part D.

105

CHAPTER FOUR
MEDICARE PART D COVERAGE GAP AND RESTRICTIONS: RACE, GENDER, AND
COST-RELATED MEDICATION NONADHERENCE
Background
The Medicare Prescription Drug, Improvement, and Modernization Act of 2003 (MMA)
is the largest benefit expansion in the history of the program, and represents the most meaningful
health care policy in the past three decades because it reduces out-of-pocket health care costs. To
respond to the need to provide older adults with drug coverage, the MMA added the Part D
benefit to Medicare (Madden, et al., 2008; Mayes, 2005). In 2008, it was estimated that 90
percent of Medicare’s 45 million beneficiaries had prescription drug coverage (Joyce, et al.,
2009) as compared to 76 percent before the implementation of the benefit (Levy & Weir, 2009).
While overall the MMA helped facilitate the purchase of needed health services, core
stipulations of the Act can potentially prevent access to medications because out-of-pocket
expenditure requirements can be substantial.
As a means of controlling the cost of the benefit, provisions contained in the MMA allow
Medicare Part D plans to utilize coverage restrictions. The vast majority of Medicare Part D
plans do not offer coverage within the gap (Hoadley, et al., 2008). Entering the gap means that
prescriptions that were previously covered become the sole responsibility of the Medicare
beneficiary. Under the 2010 standard benefit, beneficiaries pay a $310 deductible and 25 percent
copay per prescription until they incur $2,830 in total drug cost at which point they enter the
coverage gap. Once beneficiaries incur $6,440 in total drug costs, they then become eligible for
“catastrophic coverage,” and Medicare and the benefit provider pay for 95 percent of their drug

106

costs for the remainder of the year (Kaiser Family Foundation, 2009a). The cycle renews
annually.
There is concern that these cost-containment provisions can inhibit the use of medications
particularly for older women and Black Americans. Compared to older men and Whites, older
women and Black Americans tend to have a greater number of chronic health problems (Clark,
1997; Kelley-Moore & Ferraro, 2004a) and are, therefore, more reliant on prescription
medications (Rogowski, et al., 1997; Sambamoorthi, et al., 2003). This means that older women
and Black Americans tend to have higher copayments and are more likely to reach the coverage
gap each year. Further, they have a greater likelihood of encountering coverage restrictions, such
as limitations on the number and/or type of medications paid for by the plan and increased costsharing for certain drugs, because their utilization is higher. Combined with their overall lower
socioeconomic status, therefore, older women and Blacks are less able to afford out-of-pocket
expenditure requirements associated with the MMA’s cost-containment provisions (Wei, et al.,
2006) and more likely to have difficulties adhering to their medication regimen.
Previous research has shown the connection between cost-related medication
nonadherence (CRN) and cost-containment provisions similar to those present in Medicare Part
D plans (Adams, et al., 2001; T. B. Gibson, et al., 2005; Tseng, et al., 2004; Wang, et al., 2010).
CRN can include not filling, stopping, or skipping doses of a prescription because of cost. Prior
to the MMA, racial and gender disparities in CRN clearly existed (Heisler, et al., 2005; Klein, et
al., 2004). Medicare Part D may continue to impede access to medications because the benefit’s
cost-containment provisions can increase out-of-pocket expenditures (Hsu, et al., 2008), and may
have a disproportionately greater impact on older females and Black Americans because of their

107

economic status, health status, and greater reliance on prescription medications (Wei, et al.,
2006).
To date, studies have not examined whether racial and gender disparities in CRN persist
under Medicare Part D plans or how the coverage gap and restrictions affect differences in
nonadherence. This cross-sectional analysis builds upon prior research by using a cumulative
advantage/disadvantage framework to examine to what extent racial and gender disparities in
CRN exist for older women and Black Americans enrolled in the benefit. Further, this study
seeks to understand whether the coverage gap and restrictions directly and indirectly affect the
relationship between race, gender, socioeconomic and health status and CRN. The results of this
research can help inform Medicare policy by identifying whether older women and Black
Americans are disproportionately nonadherent under the Part D benefit, and how the coverage
gap and restrictions directly and indirectly impact CRN.
Literature Review
Cumulative Advantage/Disadvantage and Medicare Part D
Cumulative advantage/disadvantage theory is useful as a foundation for understanding
and evaluating the effectiveness of Medicare Part D. The theory has increasingly been used as a
paradigm in the field of aging to describe stratification and growing trajectories over time
(Dannefer, 1987; Farkas, 2003; Kail, et al., 2009; A. E. Wilson, et al., 2007). Systematic
disadvantages have a considerable impact on later life, and it is suggested that they not only
accrue throughout the life course but are magnified and can have a feedback effect (Crystal &
Shea, 1990a; Ferraro & Kelley-Moore, 2003). In other words, “the more disadvantages
individuals experience, the more likely they are to accrue subsequent and greater disadvantages”
(Kail, et al., 2009, p. 557). Because of cumulative advantages and disadvantages throughout the

108

life course, heterogeneity increases with age, and inequalities are greatest in later life (Crystal &
Shea, 1990a; Dannefer, 1987; Ross & Wu, 1996). When deliberating social policy initiatives
intended to target older individuals, the aged should not be stereotyped (Butler, 1985) or treated
as a homogeneous group (Neugarten, 1982). Inequities resulting from disadvantages throughout
the life course need to be considered in order to ensure equitable access to benefits.
This study will focus on racial and gender disparities in CRN that result from one
particular social policy, Medicare Part D. As the result of cumulative advantages and
disadvantages throughout the life course, it is anticipated that higher out-of-pocket expenditures
will be related to personal characteristics associated with early disadvantage as measured by
income, education, health insurance coverage, and health status. These factors, in turn, can
potentially impact CRN under the benefit.
Income, education, and prescription insurance coverage. The ability to afford
pharmaceuticals and maintain adherence to a prescribed medication regimen are largely
dependent on retirement earnings. Income in later life is essentially based on pre-retirement
experiences, and racial and gender inequities continue to increase with advancing age (Crystal &
Shea, 1990a). Wages for females and Black Americans have consistently lagged behind those
received by White males, and women and Black Americans are more likely to have been
segregated into positions without retirement or pension benefits (Gonyea & Hooyman, 2005).
Since Social Security is based on a 35-year work history, older women and Black Americans are
typically subject to lower benefits in later life than White males due to lower wages and fewer
years in the labor force (Olson, 1994) and thus have less income available to purchase prescribed
medications.

109

Source of prescription insurance coverage in later life is often determined by past
employment experiences and those who are disadvantaged – or have lower income and education
– typically have a greater dependence on public sources for drug insurance (Hopkins & Kidd,
1996; Kanavos & Gemmill-Toyama, 2010; Lillard, Rogowski, & Kington, 1997; Pourat, Rice,
Kominski, & Snyder, 2000). Older males (Gonyea & Hooyman, 2005) and Whites (Zuvekas &
Taliaferro, 2003) are more likely to have employer-provided prescription drug coverage, largely
due to higher-paying positions that provide subsidized drug benefits in retirement.
Comparatively, older females and Black Americans are more likely to have held positions
without health benefits (Marquis & Long, 1995; Pol, Mueller, & Adidam, 2002) and as a
consequence, were more likely to enroll in Medicare Part D (Bakk, 2012b; Davidoff, et al., 2010;
Millett, et al., 2010; Neuman, et al., 2007; Rudolph & Montgomery, 2010; Skarupski, et al.,
2009). Source of coverage can directly impact cost sharing and ability to maintain adherence.
Compared to older individuals enrolled in Medicare Part D, out-of-pocket expenditures and
nonadherence due to cost were substantially lower for those with employer-sponsored coverage
(Bakk, 2012b; Neuman, et al., 2007; Safran, et al., 2009). Thus, older women and Black
Americans’ disproportionately greater reliance on the Medicare Part D can place them at an
overall increased risk of CRN because cost sharing is greater under Part D than employer-based
plans. Further, older females and Black Americans are less able to afford out-of-pocket
expenditure requirements due to their lower socioeconomic status (Wei, et al., 2006), which can
also impact ability to afford medications and maintain adherence within Medicare Part D.
Health status. Lower income is typically accompanied by a higher prevalence of chronic
health conditions among older women and Black Americans. Racial and gender inequities in
morbidity exist throughout the life course and health trajectories continue to diverge with

110

increasing age (Clark, 1997; Kelley-Moore & Ferraro, 2004a). Women are more likely to report
functional limitations, a higher number of chronic conditions, and rate their health as poor or fair
than their male counterparts (Murtagh & Hubert, 2004). As the result of increased morbidity,
females and Black Americans are more reliant on pharmaceuticals (Goulding, 2005). Yet, their
ability to afford prescribed medications is often compromised because of lower socioeconomic
status. Compared to White males, older women and Black Americans spend a higher proportion
of their income on pharmaceuticals (Rogowski, et al., 1997; Sambamoorthi, et al., 2003). For
these reasons, it is expected that they will continue to experience a disproportionately heavier
financial burden under Medicare Part D (Wei, et al., 2006) and have a greater risk of incurring
the benefit’s cost-containment provisions. This, in turn, can subsequently affect their ability to
purchase prescribed medications and maintain adherence.
Medication Access and Adherence Under Medicare Part D
When analyzing the MMA from certain economic perspectives, the policy can appear
highly successful. Before Medicare Part D, older adults lacking prescription drug insurance spent
more on their medications than those with coverage, despite filling fewer prescriptions (Safran,
et al., 2005; Soumerai, et al., 2006). Further, it is estimated that out-of-pocket spending on
prescription drugs decreased between 13 and 18.4 percent even as the number of medications
used by older adults increased from 5.9 to 12.8 percent since the implementation of the MMA
(Chen, et al., 2008; Joyce, et al., 2009; Lichtenberg & Sun, 2007; Schneeweiss, et al., 2009; Yin,
et al., 2008). However, these findings do not take into consideration the implications of race or
gender. Further, cost-containment provisions – or the coverage gap or coverage restrictions – of
the MMA are not accounted for in these studies, and the potential impact of those provisions on
cumulatively disadvantaged beneficiaries is not recognized. In other words, there was no

111

differentiation of groups reported in these findings; beneficiaries were regarded as a
homogeneous group that gained access to a prescription drug benefit and lowered their
medication expenditures. It is important to recognize the potential implications of Medicare Part
D cost-sharing provisions on older females and Black Americans. For example, one implication
of the cost-sharing requirements is that access to medications is not in fact the same for all
beneficiaries. Those with lower income and increased morbidity, such as is often experienced by
older women and Black Americans, may be negatively and disproportionately influenced by
cost-sharing requirements both because of their higher medication expenses and their limited
income. This, in turn, can impact their ability to access medication and maintain adherence.
Thus, existing research on the impact of MMA does not recognize the differential consequences
of these provisions.
While studies demonstrated that racial (Bambauer, et al., 2007; Gellad, et al., 2007;
Heisler, et al., 2005; Klein, et al., 2004; Piette, et al., 2004c; Roth, et al., 2009; Roth & Ivey,
2005; Soumerai, et al., 2006; Steinman, et al., 2001; Zivin, et al., 2010) and gender (Heisler, et
al., 2005; Piette, et al., 2004c; Zivin, et al., 2010) disparities in CRN existed before Medicare
Part D, research examining CRN after the implementation of the benefit is limited. One of the
main reasons for this limitation is the restrictions placed on the availability of data. The Centers
for Medicare and Medicaid Services has historically been a primary source of data for
researchers. However, a provision contained in the MMA stipulated that private health care plans
were not required to release Medicare Part D claims data for nonpayment-related purposes,
which included research, public health purposes, and health care analysis (Crystal, 2008,
November; Stuart, 2008). While this was modified in 2009 and data are currently in the process
of being released, this stipulation largely accounts for the limitation with existing research.

112

When considering studies conducted after the implementation of Medicare Part D, Safran
et al. (2009) reported that for older individuals with no or a meager drug benefit, enrollment
resulted in lower out-of-pocket prescription drug expenditures, higher medication utilization, and
overall lower CRN. However, low-income, chronically ill beneficiaries were significantly more
likely to report CRN and difficulty affording their medications. Similarly, Madden et al. (2008)
found that individuals in fair to poor health, with at least four morbidities, or having an annual
income less than $25,000 continued to report greater CRN under Medicare Part D. Although
differences associated with race or gender were not analyzed in either study, the findings shed
some light on the experiences of these groups. Because older women and Black Americans are
more likely to be low income, rate their health as poor or fair, and experience higher rates of
chronic illness and health problems, these studies suggest that enrolling in the Medicare Part D
benefit may have little or no impact on CRN for these groups.
While previous studies have not specifically examined the effect of the Medicare drug
benefit on racial or gender disparities in CRN, research suggests ethnic disparities continue to
exist. One study conducted after the implementation of Medicare Part D, for example, found that
Hispanics were more likely to report CRN than non-Hispanics. Further, CRN was significantly
greater for non-Hispanic Blacks than non-Hispanic Whites. When examining gender, both
Hispanic and non-Hispanic females were more likely to report nonadherence due to cost than
males (Frankenfield, et al., 2010).
Medicare Part D Cost-Containment Provisions and Adherence
Due to declining state budgets and cost increases in the area of prescription medications
particularly over the last decade, most state Medicaid programs utilize cost-containment
mechanisms or coverage restrictions and coverage caps similar to those present in Medicare Part

113

D plans in an effort to contain spending (Cunningham, 2005). Coverage restrictions are intended
to reduce non-essential drug use and plan expenditures, and include restricting the number and/or
type of medications paid for by the plan and requiring increased cost-sharing or a copayment for
certain drugs (Nelson, Reeder, & Dickson, 1984; Reeder & Nelson, 1985; Soumerai, et al.,
1987). Similarly, coverage caps limit the number of prescriptions paid for by a plan on a yearly
basis and are used to control the annual cost of a benefit (Cox, Jerigan, Coons, & Draugalis,
2001). While coverage caps and restrictions have provided considerable savings for state
Medicaid programs (Cunningham, 2005; Soumerai, et al., 1987), they also produced unintended
consequences. For example, requiring a small copayment (fifty cents per prescription) in
Medicaid drug plans reduced prescription drug use and spending across several essential
medications (Reeder & Nelson, 1985; Soumerai, et al., 1987). Yearly coverage gaps – similar to
the Medicare Part D coverage gap – have resulted in decreased utilization and nonadherence
(Cox, et al., 2001; Joyce, Goldman, Karaca-Mandic, & Zheng, 2007; Tseng, et al., 2004).
The impact of coverage caps and restrictions under state Medicaid programs can help
clarify beneficiary behavior when faced with out-of-pocket expenditure requirements due to
cost-containment policies. Despite the potential consequences of cost-sharing provisions, studies
examining whether experiencing the Medicare Part D coverage gap and restrictions are limited,
and even less is known about how these provisions indirectly affect racial and gender differences
in adherence. This is important because those with a lower income and increased morbidity –
which are more common for older females and Black Americans – have a greater risk of
encountering the benefit’s cost-containment provisions and subsequent difficulty affording their
medications.

114

CRN and the Coverage Gap. Affording medications is a primary concern of Medicare
Part D beneficiaries experiencing the coverage gap (Neuman & Cubanski, 2009). Once a
beneficiary reaches this benefit threshold, they are responsible for 100% of their prescription
drug costs until spending approximately $4,550 out-of-pocket for prescription medications.
Overall, reaching the coverage gap resulted in lower drug consumption (Pedan, et al., 2009; Sun
& Lee, 2007; Zhang, et al., 2009) and decreased adherence (Cronk, Humphries, Delate, Clark, &
Morris, 2008; Hsu, et al., 2008; Raebel, Delate, Ellis, & Bayliss, 2008) because of increased outof-pocket expenditure requirements. Furthermore, Hsu et al. (2008) found that 36% percent of
respondents encountering the coverage gap stopped taking their medication, reduced the
recommended dosage, or switched to an alternate drug.
While research has not specifically examined the impact of the Medicare Part D coverage
gap on older women and Black Americans, beneficiaries with certain chronic illnesses – such as
depression and diabetes – have a greater risk of this cost-sharing provision because they spend
considerably more on medications and are, therefore, at greater risk of CRN (Duru et al., 2010;
Ettner, et al., 2010; Hoadley, et al., 2008; Schmittdiel et al., 2009; Stuart, et al., 2005). For
example, 51% of older diabetics who reached the coverage gap or benefit threshold had a decline
in out-of-pocket spending on medications, suggesting that they may have become noncompliant
(Hoadley, et al., 2008). In comparing older adults with and without depressive symptoms,
individuals with depressive symptoms were more likely to report CRN and forego basic needs to
pay for medications after reaching this benefit threshold (Zivin, et al., 2009). Because older
women and Black Americans experience higher rates of diabetes (Gellad, et al., 2006;
Schoenborn & Heyman, 2009), they may face a greater risk of CRN under Medicare Part D.
While depression is more prevalent in females than males (Skarupski, et al., 2009), the severity

115

and persistence of depressive disorders is higher in Black Americans as compared to Whites
(Williams et al., 2007). Thus, greater prevalence, severity, and persistence of depressive
symptoms can increase the need for pharmaceuticals and subsequent risk of CRN due to the
coverage gap.
Coverage restrictions. Only one study examined Medicare Part D coverage restrictions
and found that cost-sharing requirements did not impact the number of prescriptions used by
Medicare Part D enrollees. Further, having a copayment requirement did not affect gender
differences in drug use (Goedken, et al., 2010). While this study can provide insights on how
cost sharing impacts medication use under Medicare Part D plans, there are several limitations.
First, the sample consisted of Medicare beneficiaries with higher than average income and
education and therefore, generalization of results to older women and Black Americans is
difficult due to their increased likelihood of having a lower socioeconomic status (Crystal, 1986;
Crystal & Shea, 1990a; Olson, 1994). Second, the study only assessed drug use and did not
specifically examine adherence. Finally, racial differences in medication use were not
investigated. Thus, the ability to understand how the Medicare Part D coverage restrictions
directly or indirectly affect CRN is largely unknown.
In sum, research has not investigated whether or to what extent racial and gender
disparities in CRN exist under Medicare Part D, or how the benefit’s cost-containment impacts
racial and gender differences in nonadherence due to cost. Studies primarily investigated the
effect of Medicare Part D utilizing pharmaceutical claims and health care plan data and did not
differentiate beneficiaries by demographics. While a few studies have used nationally
representative datasets to investigate CRN, racial and gender differences have not been examined
and the indirect effect of the cost-containment provisions has not been considered. The one study

116

examining disparities in CRN after the implementation of the benefit focused on ethnic
differences and did not account for the coverage gap or restrictions (Frankenfield, et al., 2010).
Further, the impact of the Medicare Part D coverage restrictions on CRN is essentially unknown.
This study builds on existing knowledge and provides important insights about CRN
under Medicare Part D. First, it uses data from a nationally representative dataset to analyze
CRN for beneficiaries who enrolled in Medicare Part D. In particular, this study examines
whether and to what extent race and gender increase the likelihood of CRN under the Medicare
Part D benefit. Second, this study investigates whether and to what extent the Medicare Part D
cost-containment provisions directly and indirectly affect the relationship between race, gender,
socioeconomic and health status and CRN. This allows us to consider how the coverage gap and
restrictions influence racial and gender differences in CRN. It also expands our knowledge of
how the Medicare Part D cost-containment provisions impact the likelihood of nonadherence due
to cost, and whether these provisions affect the relationship between factors associated with
cumulative disadvantage and CRN.
Based on existing literature, there are four hypotheses in this study:
11)

As compared to older Whites, older Black Americans enrolled in Medicare Part
D are more likely to report CRN

12)

As compared to older males, older females enrolled in Medicare Part D are
more likely to report CRN

13)

Experiencing the Medicare Part D coverage gap and restrictions will increase
the likelihood of CRN

14)

Experiencing the Medicare Part D coverage gap and restrictions will mediate
the effect of racial, gender, socioeconomic, and health differences in CRN

117

Methodology
Sample
This study utilizes secondary data from the 2006 wave of The University of Michigan (UM) Health and Retirement Study (HRS) and 2007 wave of the HRS Prescription Drug Study
(PDS) to conduct a cross-sectional analysis. Launched in 1992, the HRS is a nationally
representative population study of more than 20,000 Americans over the age of 50 designed to
assess the relationship between health and retirement. The core HRS survey collects information
pertaining to physical and mental health, insurance coverage, financial status, family support
systems, retirement situation, and work status (Juster & Suzman, 1995). The primary mode of
data collection is by telephone; however, in situations where health limitations would make the
hour-plus telephone session difficult or where there is no telephone in the household, the survey
is conducted in the respondent’s home (HRS, 2011). Primarily funded through the National
Institute on Aging (NIA U01AG009740), the HRS is designed, administered, and conducted by
the Institute for Social Research (ISR) at the U-M. The overall interview response rate for the
core 2006 HRS was 88.5% (Ofstedal, 2009, June).
The PDS is a subsample of the HRS, and is a two-wave mail survey conducted in 2005 and
2007. It was specifically designed to examine prescription drug utilization as Medicare Part D
was implemented. The baseline wave was administered pre-Medicare Part D in 2005 by the ISR,
and captured information pertaining to prescription drug use, coverage, and knowledge of the
Medicare prescription drug benefit. The second wave was conducted post-Medicare Part D in
2007, and captured similar information as well as data pertaining to the impact of the MMA costcontainment provisions or coverage gap and restrictions. The sample was drawn from
respondents who participated in the HRS in 2004. To be eligible for inclusion in the PDS,

118

respondents needed to be age 65 or older in 2007 (born in 1942 or later), or have Medicare or
Medicaid coverage at some time between 2002 and 2004. Because the Consumption and
Activities Mail Survey (CAMS) was fielded at the same time as the PDS, approximately 40% of
eligible participants were excluded because of their involvement in the CAMS. However, the
exclusion did not affect the representation of the PDS sample because CAMS participants were
randomly selected (J. Faul, Personal Communication, January 4, 2012). Of those who were
eligible, 3,536 individuals or 74% completed the PDS in 2007 (HRS, 2011). Given that this is a
relatively new topic area, there are no reliability or validity measures associated with the PDS (J.
Faul, Personal Communication, February 3, 2012).
To be included in the analytic sample for this study, respondents were required to 1) be
either non-Hispanic White/Caucasian or non-Hispanic Black/African American; 2) be age 65 or
older in 2007 and therefore eligible for Medicare when the PDS was administered; 3) have
reported taking prescription medications in 2007; 4) have responded to questions about
prescription drug adherence in the 2007 PDS; and 5) have reported being enrolled in Medicare
Part D in 2007 PDS. Of the 3,536 responding to the 2007 PDS, 3,139 were either non-Hispanic
White/Caucasian or non-Hispanic Black/African American; 2,996 were age 65 and over; 2,797
reported taking prescription medications; 1,419 reported Medicare Part D enrollment; and 1,352
reported data on prescription drug adherence or nonadherence. Thus, the final analytic sample
consisted of 1,352 respondents.
The combination of HRS and PDS provides an excellent source for testing the proposed
hypotheses because indicators of cumulative advantage/disadvantage that could decrease or
increase older adults’ risk of CRN are available through these datasets. Variables from the PDS
included CRN, Low-Income Subsidy status, plan restrictions, coverage gap status, and Medicare

119

Part D enrollment status. All other variables were obtained from the 2006 RAND HRS data files.
These data files contain a broad range of measures across HRS waves, and are developed and
maintained by the RAND Center for the Study of Aging, with support provided by the NIA and
Social Security Administration (RAND Center for the Study of Aging, 2010).
In regards to Institutional Review Board (IRB) approval, this study relied exclusively on
secondary data to complete all analyses. The HRS and PDS files contain de-identified individual
data that are available to the public. Therefore, the IRB at Michigan State University (MSU)
does not require human subjects review.
Measures
Dependent variable. A summary indicator of any CRN was constructed as the
dependent variable. In the 2007 PDS, CRN was evaluated using questions about the following
three behaviors: 1) not filling a new prescription because of cost; 2) stopping taking a medication
because of cost; or 3) skipping doses of a medication to save money. If the respondent indicated
that they engaged in any of these behaviors, they were considered nonadherent in this study.
Earlier research typically used a single measure to assess CRN, and focused solely on whether
respondents failed to fill a prescription in the past year because of cost. However, this measure
did not assess for multiple ways CRN could occur. Thus, the additional questions were
developed to help monitor and identify changes in CRN, and have demonstrated test-retest
reliability (Pierre-Jacques, et al., 2008) and construct validity (Soumerai, et al., 2006). This
composite variable approach was originally developed by Safran et al. (2003) based on a series
of validated survey questions concerning CRN to prescription regimens, confirmed using factor
analysis techniques, and used as a main outcome in subsequent research examining drug
nonadherence practices of older adults (Bambauer, et al., 2007; Gellad, et al., 2007; Madden, et

120

al., 2008; Safran, et al., 2005; Soumerai, et al., 2006; I. B. Wilson, et al., 2004; Zivin, et al.,
2009). Because all three measures of CRN, specifically filling, stopping, or skipping doses of a
prescription because of cost, were highly correlated with comorbidity, self-reported health status,
and socioeconomic status (Bambauer, et al., 2007; Safran, et al., 2005; Safran, et al., 2003;
Soumerai, et al., 2006), it is important that they be included in this study. Older females and
Black Americans are more likely to have a lower income and multiple health problems due to
disadvantages throughout the life course and thus, accounting for all behaviors can further
understanding of CRN before and after the implementation of Medicare Part D.
In this study, a dichotomous variable indicating any nonadherence or no nonadherence to
the three CRN behaviors in the PDS – not filling a new prescription because of cost, stopping
taking a medication because of cost, or skipping doses of a medication to save money – was
created. The CRN variable was initially incorporated as a continuous variable indicating a sum
total of the three behaviors. However, results were significantly skewed and similar to previous
research, the variable was dichotomized (Bambauer, et al., 2007; Gellad, et al., 2007; Madden, et
al., 2008; Safran, et al., 2005; Soumerai, et al., 2006; I. B. Wilson, et al., 2004; Zivin, et al.,
2009). Using this global, dichotomous measure is common in existing literature, and will help
provide a meaningful comparison to previous research.
Independent variables. The main predictors of interest were race (Black and White),
gender (female and male), coverage restrictions, and coverage gap status. The Medicare Part D
coverage gap and restrictions can impact CRN because they require that the beneficiary pay an
additional cost to receive their medication(s). Since older women and Black Americans often
have a greater number of prescribed medications, their risk of incurring these cost-sharing
requirements is greater. Thus, it is important to understand how these provisions affect CRN.

121

First, three dichotomous variables indicating the types of coverage restrictions present in the
respondents’ Medicare Part D plan were used. Respondents to the PDS were asked, “Some
prescription drug insurance plans restrict the number, type, or dollar amount of prescriptions they
will pay for. Check any of the following types of restrictions that your plan has [Mark (X) all
that apply].” This question included the following response choices: “My plan won’t pay at all
for some types of drugs, “My plan makes me pay more for some types of drugs,” and “My plan
only pays up to a certain amount of money each year.” Responses were coded 0 =
“no/beneficiary does not have restriction” and 1 = “yes/beneficiary has restriction.”
Although having a plan restriction does not mean the respondent will incur that
restriction when attempting to purchase a prescribed medication, a more precise measure
indicating that the respondent actually incurred the restriction was not available in either the
2007 PDS or the HRS core survey. However, in my experience as an agency administrator,
Medicare beneficiaries are often unaware that their plan has a restriction until it is incurred, or
until they are required to pay an additional cost for their medication. For this reason, I believe
these measures are an adequate indicator of experience with plan restrictions.
Respondents to the 2007 PDS were also asked, “Some prescription drug insurance plans
have what’s called a coverage gap, or doughnut hole, where you have to pay all or nearly all of
the costs of your drugs for part of the year after the insurance has paid up to a certain amount.
Have you reached the coverage gap in your prescription drug plan this year? [Mark (X) ONE].”
Responses included “yes,” “no,” and “my plan does not have a coverage gap.”
Using literature pertaining to cumulative advantage/disadvantage as a guide, other
independent variables included socioeconomic characteristics and health status. Additionally,
age and marital status were used as control variables.

122

Socioeconomic characteristic indicators included education, annual household income,
and Low-Income Subsidy (LIS) status. Cumulatively disadvantaged individuals tend to have
lower levels of education. Subsequently, they are less likely to have prescription drug coverage
and the financial resources needed to purchase prescribed medications. Utilizing these measures
will help clarify how income and educational attainment can directly and indirectly impact CRN
among older women and Black Americans. The level of attained education was a continuous
variable ranging from 0 to 17 or more years of education. Annual household income was a
continuous variable ranging from $228 to $1,242,996. Finally, a three-category variable was
used to indicate LIS status (0 = “no LIS, “1 = “applied for LIS,” 2 = “dual eligible/receiving
LIS”). Individuals whose income and assets exceed Medicaid eligibility criteria are required to
apply for the LIS whereas dually eligible beneficiaries automatically receive the LIS because
they qualify for both Medicare and Medicaid. Further, level of LIS assistance is dependent on
eligibility status, and the subsidy provided to dual eligibles is greater than that provided to nondual eligibles who applied for assistance. The LIS was intended to offset prescription drug costs
for beneficiaries with limited income and resources by providing financial assistance. Thus, it is
expected that older women and Black Americans in particular are impacted by the LIS because
of their lower-income status.
Health status variables included number of chronic medical conditions and self-reported
health status. Increased morbidity is associated with cumulative disadvantage and increased
reliance on pharmaceuticals, particularly among older women and Black Americans. Thus, it is
important to understand how health status impacts CRN. First, a continuous variable indicating
the number of chronic medical conditions was used. Respondents were asked, “Has a doctor ever
told you that you have [name of chronic medical condition]” or “our records from your last

123

interview show that you have [name of chronic medical condition].” This question was asked for
a series of eight chronic medical conditions and included: high blood pressure, diabetes, cancer,
lung disease, heart disease, stroke, psychiatric problems, and arthritis. The second health-related
variable included self-reported health status. Respondents were asked, “Would you say your
health is excellent, very good, good, fair, or poor?” Responses ranged from 0 = “excellent” to 4 =
“poor” with a higher score indicating that the respondent perceived their health status as being
more poor.
Finally, other control variables included age and marital status. Age was a continuous
variable ranging from age 65 to 99. The variable was recoded into a three-category measure to
analyze CRN in relation to the respondent’s age cohort (65 to 74, 75 to 84, and 85 or above).
Research conducted before the implementation of Medicare Part D found that older-aged
individuals were less likely to have prescription drug coverage (Kanavos & Gemmill-Toyama,
2010). For this study, three age groups were utilized to analyze differences in adherence between
age cohorts after the Part D benefit became available, and to determine how the benefit may have
impacted adherence. Further, the measure was used as a continuous variable in Chapter Two and
provided similar results. That is, as income increased, the likelihood of CRN significantly
decreased.
It was expected lack of coverage could impact CRN. Marital status was a seven-category
variable: married, married with spouse absent, partnered, separated, divorced, widowed, and
never married.
Analytic strategy
Univariate descriptive statistics summarized all measures. To determine the bivariate
association between respondent characteristics and CRN in 2007, the Rao-Scott chi-square, a

124

design-adjusted version of the Pearson chi-square, was used for categorical variables (Rao &
Scott, 1981) and an F means test for continuous variables. Logistic regression models were used
for multivariate analysis to estimate racial and gender differences in CRN under Medicare Part D
plans, and to examine whether and to what extent the Medicare Part D coverage gap and
coverage restrictions directly and indirectly affect the relationship between race, gender,
socioeconomic and health status and CRN. A series of eight regression models were run. Model
1 included only race and gender. Model 2 introduced the age and marital status variables.
Socioeconomic status variables were added in Model 3 and in Model 4, health status variables
were included. Models 5, 6, 7 and 8 added the coverage restriction and coverage gap measures
separately to assess how each mediates the effect of race, gender, socioeconomic status, and
health status on CRN, as well as determine how the individual measures directly impact
adherence.
All statistical analyses were performed in the Stata version 11.0 statistical package
(StataCorp). Stata’s survey commands were used in all statistical analyses to account for the
complex sampling design (stratification and clustering) of the HRS when computing standard
errors for calculated estimates. The PDS questionnaire weights (a product of the HRS sampling
weights) were applied to adjust for sample selection probability or oversampling of select
minority groups and for interviewee nonresponse (HRS, 2011), and to obtain estimates that
reflect the general population age 65 and older. Significance in all analyses was tested at the
p<.05 level using a one-tailed test.

125

Results
Description of Sample at Baseline
A detailed description of the sample is provided in Table 4.1. Of the 1,352 individuals
reporting medication adherence (81.7%) or nonadherence (18.3%) due to cost, 9.6% selfidentified as non-Hispanic Black or African American and the majority (60.2%) were female.
Respondents ranged in age from 65 to 96 with a mean age of 75.2 years (SD = 7.35), and slightly
more than half (53.6%) were married. The average household income was $46,752 (SD =
$64,747) and the mean years of education was 12.29 (SD = 2.92). In regards to LIS status, 13%
applied for the LIS and 4.8% were dual eligibles receiving the LIS. Respondent self-reports
indicated that the majority felt their health was good (33.3%), very good (26.9%), or fair
(22.7%), followed by poor (10.1%), and excellent (7.0%). The average number of chronic health
conditions was 2.55 (SD = 1.38). In regards to coverage restrictions under Medicare Part D
plans, over a third (35.8%) of respondents were required to pay more for some drugs, 26.3% had
a yearly cost limitation, and 21.8% had drugs that were not covered. Finally, 15% reported
reaching the coverage gap, 67% did not reach the gap, while the remainder (18%) did not have a
gap in coverage.
Bivariate Results
Bivariate analyses of medication adherence and nonadherence are presented in Table 4.2.
In regards to the main predictor variables, a higher proportion of Black Americans compared to
Whites were nonadherent. Gender was not significantly related to CRN. In terms of the coverage
restriction variables, a significantly greater proportion of respondents whose plans did not pay
for some medications, were required to pay more for some drugs, or had yearly cost limitations
reported nonadherence as compared to those without these restrictions. A much larger proportion

126

TABLE 4.1
Description of Sample of Individuals Age 65 and Over Reporting Medication
Adherence/Nonadherence Under Medicare Part D or MA Plan in 2007 (N=1,352)
1

1

N /M

2

Low-Income Subsidy (LIS) status
Dual eligible receiving LIS
Applied for LIS
No LIS
2
Number of chronic medical conditions (range: 0 to 7)
Self-reported health2
Excellent
Very good
Good
Fair
Poor

127

47.9
36.7
15.4

749
16
23
9
111
395
34

Annual household income (range: 228 to 1,242,996)

60.2
39.8

704
469
179

2

9.6
90.1

831
521

Race
Black American
White
Gender
Female
Male
Age
65 to 74
75 to 84
85 and above
2
Marital status
Married
Married, spouse absent
Partnered
Separated
Divorced
Widowed
Never married
2
Education (range: 0 to 17)

% / SD

200
1,152

Variable

52.6
1.0
1.4
1.0
8.0
33.0
3.0

12.29

2.92

46,752

64,747

69
196
1,071

4.8
13.0
82.2

2.55

1.38

94
364
468
293
116

7.0
26.9
33.3
22.7
10.1

TABLE 4.1
Description of Sample of Individuals Age 65 and Over Reporting Medication
Adherence/Nonadherence Under Medicare Part D or MA Plan in 2007 (N=1,352)
1

Coverage Restriction
Plan won't pay for some drugs
Yes
No
Plan requires paying more for some drugs
Yes
No
Plan has cost limit per year
Yes
No
2
Coverage Gap status
Reached gap
Did not reach gap
Do not have coverage gap
1
2

1

N /M

Variable

% / SD

300
1,052

21.8
78.2

461
891

35.8
64.2

362
990

26.3
73.7

176
774
190

15.0
67.0
18.0

Weighted %, unweighted N

Sample size varies slightly due to missing data within Prescription Drug Study (PDS) or
Health and Retirement Study (HRS)

128

of respondents that reached the coverage gap reported CRN than those not reaching or without
this benefit threshold.
In terms of control variables, age was significantly related to CRN. The highest
proportion of respondents reporting nonadherence occurred among the young-old, or those age
65 to 74, and CRN decreased as age increased. Marital status was not significantly related to
CRN.
In terms of socioeconomic status variables, LIS status was significantly related to CRN.
A greater proportion of those that applied for the LIS or were dual eligibles receiving the LIS
reported nonadherence. Nonadherent older individuals reported a significantly lower annual
income (M = $35,608, SD = $34,310) than adherent older adults (M = $49,244, SD = $69,973).
In regards to health status variables, respondents who reported CRN had a significantly
greater number of chronic conditions (M = 2.96, SD = 1.40) compared to those that maintained
adherence (M = 2.46, SD = 1.35). A smaller proportion of respondents in excellent health
reported nonadherence than those in other categories with those in fair or poor health reporting
the most nonadherence.
Bivariate analyses suggest that without controlling for other factors, older Black
Americans were more likely to report CRN under Medicare Part D than older Whites as
expected. Respondents whose plan did not pay for certain medications, required paying more for
some medications, or had yearly cost limitations, were more likely to report CRN as compared to
those without these restrictions. Reaching the coverage gap increased the likelihood of CRN.
Older individuals that were younger (age 65 to 74), had a lower annual income, greater number
of chronic medications, or self-reported their health as being very good, good, fair, or poor (as
compared to excellent) appeared more likely to be nonadherent due to cost.

129

TABLE 4.2
Bivariate Relationships Between Independent Variables and Medication Adherence/Nonadherence Under Medicare Part D or
MA Plan in 2007 (N=1,352)
Adherence
Nonadherence
Test
1
1
2
%
%
Statistic
N
N
df
p
Race
Black American
138
69.6
62
30.4
18.02
1
<.001
White
947
83.0
205
17.0
Gender
Female
660
79.4
171
20.6
1.67
1
0.202
Male
425
81.6
96
18.4
Age
65 to 74
549
79.3
155
20.7
5.49
2
<.01
75 to 84
375
80.9
94
19.1
85 and above
161
91.1
18
8.9
3
Marital status
Currently married
609
82.8
140
17.2
0.50
6
0.788
Married, spouse absent
15
93.1
1
6.9
Partnered
16
78.6
7
21.4
Separated
7
79.2
2
20.8
Divorced
84
80.6
27
19.4
Widowed
315
79.9
80
20.1
Never married
28
83.5
6
16.5
3
Education (M±SD)
12.35±2.82
12.04±2.90
1.73
1, 51
0.195
3

Annual household income (M±SD)

49,244±69,973

35,608±34,310

13.88

1, 51

<.001

Low Income Subsidy (LIS) status
Dual eligible receiving LIS
Applied for LIS
No LIS

52
130
890

17
66
181

7.77

2

<.01

3

73.5
70.8
83.8

130

26.5
29.2
16.2

TABLE 4.2
Bivariate Relationships Between Independent Variables and Medication Adherence/Nonadherence Under Medicare Part D or
MA Plan in 2007 (N=1,352)
Adherence
Nonadherence
Test
1
1
2
%
%
Statistic
N
N
df
p
3

Number/chronic conditions (M±SD)
3

2.46±1.35

Self-reported health
Excellent
84
Very good
305
Good
382
Fair
217
Poor
75
Coverage Restrictions
Plan won’t pay for some drugs
Yes
201
No
884
Plan requires paying more for some drugs
Yes
340
No
745
Plan has yearly cost limit
Yes
269
No
816
3
Coverage Gap status
Reached gap
118
Did not reach gap
631
Do not have coverage gap
165
1
3

2.96±1.40

17.04

1, 51

<.001

93.0
83.8
83.4
75.5
77.1

10
59
86
76
31

7.0
16.2
16.6
24.5
22.9

4.64

4

<.01

69.9
85.0

99
168

30.1
15.0

28.45
--

1
--

<.001
--

75.2
85.3

121
146

24.8
14.7

17.14

1

<.001

75.6
83.8

93
174

24.4
16.2

11.05

1

<.01

68.1
83.0
88.3

58
143
25

31.9
17.0
11.7

10.39

2

<.001

2

Weighted %, unweighted N; Test statistic is Rao-Scott chi-square for categorical variable/F means test for continuous variables
Sample size varies slightly due to missing data within Prescription Drug Study (PDS) or Health and Retirement Study (HRS)

131

Logistic Regression Results
Table 4.3 shows logistic regression models for CRN outcomes. Model 1 estimated the
effects of race and gender on CRN, controlling for each other. Compared to older Whites, older
Black Americans were over twice as likely to report CRN under Medicare Part D (OR=2.09).
Gender was not related to nonadherence. The relationship between race and CRN remained the
same with the inclusion of demographic, socioeconomic, and health status variables in Models 2,
3 and 4 (not shown).
TABLE 4.3
Logistic Regression Results of Cost-Related Medication Nonadherence Under
Medicare Part D or MA Plan in 2007: Race and Gender (N=1,353)
Model 1
OR
Race
Black American
2.09***
White (reference group)
-Gender
Female
1.18
Male (reference group)
-***Significant at p<.001
Table 4.4 presents the logistic regression results after the inclusion of the Medicare Part
D cost-containment provision measures. The effect of race on CRN was maintained with the
addition of the coverage restriction measures in Model 5 (OR=1.57), Model 6 (OR=1.68), and
Model 7 (OR=1.58). However, when the coverage gap variable was added in Model 8, the
relationship between race and CRN was no longer statistically significant. A test of mediation
(Baron & Kenny, 1986) was performed. While reaching the coverage gap was not associated
with race, the test confirmed that not reaching the gap mediated the relationship between race
and CRN. That is, race was related to not reaching the coverage gap, race was associated with

132

TABLE 4.4
Logistic Regression Results of Cost-Related Medication Nonadherence Under Medicare Part D or MA Plan in 2007
Model 5
Model 6
Model 7
OR
OR
OR
Race
Black American
1.57*
1.68*
1.58*
White (reference group)
---Gender
Female
1.12
1.03
1.04
Male (reference group)
---Age
65 to 74 (reference group)
---75 to 84
.80
.80
.77
85 and above
.31***
.31***
.30***
Marital status
Currently married (reference category)
---Married, spouse absent
.39
.28
.28
Partnered
1.50
1.41
1.56
Separated
.93
1.07
1.19
Divorced
.85
.90
.79
Widowed
1.09
1.15
1.10
Never married
.80
.74
.76
Education (continuous)
1.03
1.03
1.04
Annual household income (continuous)
.99*
.99**
.99*
Low Income Subsidy (LIS) status
Dual eligible receiving LIS
1.09
1.46
1.38
Applied for LIS
1.49*
1.60**
1.57**
No LIS
---Number of Chronic Medical Conditions (continuous)
1.20**
1.22**
1.22**

133

Model 8
OR
1.25
-1.06
--.76
.27***
-.37
1.33
1.67
.83
.96
.69
1.03
.99*
2.11
1.58*
-1.23**

TABLE 4.4
Logistic Regression Results of Cost-Related Medication Nonadherence Under Medicare Part D or MA Plan in 2007
Model 5
Model 6
Model 7
OR
OR
OR
Self-reported health
Excellent (reference group)
---Very good
2.57*
2.63*
2.33
Good
2.33*
2.24*
2.08*
Fair
3.04*
3.15*
2.91*
Poor
2.57*
2.16*
2.42*
Coverage Restrictions
Plan won’t pay for some drugs
Yes
2.20***
--No (reference group)
---Plan requires paying more for some drugs
Yes
-2.33***
-No (reference group)
---Plan has yearly cost limit
Yes
--1.82***
No (reference group)
---Coverage Gap status
Reached gap
---Did not reach gap
---Do not have coverage gap (reference group)
----

Model 8
OR
-2.20*
1.85*
2.36*
2.43*

------4.34***
1.92*
--

*Significant at p<.05; **Significant at p<.01; ***Significant at p<.001
Model 5:
Model 6:
Model 7:
Model 8:

Race, gender, age, marital, socioeconomic, LIS, and health status, and restriction (some drugs not covered) (N=1,317)
Race, gender, age, marital, socioeconomic, LIS, and health status, and restriction (addn’l pymt./some drugs) (N=1,317)
Race, gender, age, marital, socioeconomic, LIS, and health status, and restriction (yearly cost limit) (N=1,317)
Race, gender, age, marital, socioeconomic, LIS, and health status, and coverage gap (N=1,118)

134

CRN, and not reaching the coverage gap was related to CRN. Thus, this suggests that racial
disparities in nonadherence are driven substantially by not reaching this benefit threshold. In
other words, simply having a gap in coverage can be associated with CRN – even if this benefit
threshold is not reached. Individuals often engage in cost-coping behaviors to postpone or delay
reaching this benefit threshold, which include becoming nonadherent to a prescribed medication
regimen, particularly if medication costs are problematic (Bakk, McGuire, & Woodward, 2012).
The coverage gap and restrictions seemed to have very little indirect effect on the relationship
between socioeconomic status and CRN or health status and CRN. No significant changes were
found. Further, both coverage restrictions and the coverage gap had a direct effect on CRN.
Respondents whose plan did not pay for certain medications (OR 2.20), required paying more for
some medications (OR 2.33), and had a yearly cost limitation (1.82), had a greater likelihood of
reporting CRN than those without these restrictions. The odds of CRN were over four times
greater (OR 4.34) for respondents that reached the coverage gap and nearly twice as likely
(OR=1.92) for those who did not reach the coverage gap as compared to respondents without this
benefit threshold.
In terms of socioeconomic and health status variables, annual income, LIS status, number
of chronic medical conditions, and self-reported health had a direct effect on CRN. Having a
greater annual income was associated with a lower likelihood of nonadherence due to cost
(OR=.99). As compared to those with no LIS, individuals who applied for the LIS were over one
and a half times as likely (OR=1.58) to report CRN. Respondents with a greater number of
chronic medical conditions had an increased likelihood of CRN (OR=1.23). Further, as
compared to older individuals who reported their health as being excellent, odds of CRN were

135

greater for those who rated their health as very good (OR=2.20), good (1.95), fair (OR=2.36) or
poor (OR=2.43).
Finally, those aged 85 and older were 73% (OR=.27) as likely to report CRN as
compared to the youngest respondents.
Discussion
Medicare Part D represents the most significant change to the program since its inception.
While the benefit provided much-needed coverage to older adults struggling with prescription
drug expenses, it includes cost-containment provisions that can potentially inhibit access to
medications for more vulnerable populations, most notably older women and Black Americans.
This is the first study to examine whether and to what extent racial and gender disparities in
nonadherence due to cost exist under Medicare Part D, and how the coverage gap and restrictions
affect the relationship between race, gender, socioeconomic and health status and CRN.
In this study, there was a significant disparity in CRN between older Black Americans
and Whites enrolled in Medicare Part D. Racial differences remained significant after controlling
for age, marital status, socioeconomic status, health status, and the coverage restrictions, but
were mediated by not reaching Medicare Part D coverage gap. Experiencing the coverage gap
and restrictions had a direct effect on CRN. Further, applying for the Low-Income Subsidy (LIS)
or having a lower annual income or poorer health increased the likelihood of nonadherence, even
after accounting for the coverage gap and coverage restrictions.
In terms of hypotheses organizing the present analyses, there are several findings to
discuss. The first hypothesis that compared to older Whites, older Black Americans enrolled in
Medicare Part D are more likely to report CRN was supported. This finding expands previous
research conducted before the implementation of the MMA illustrating racial disparities in CRN

136

(Gellad, et al., 2007; Klein, et al., 2004; Soumerai, et al., 2006; Steinman, et al., 2001) and
demonstrates that despite enrolling in Medicare Part D, older Black Americans reported
significantly greater nonadherence due to cost. This finding is important, particularly given that
older Black Americans have disproportionately greater reliance on Medicare Part D as compared
to Whites (Bakk, 2012b; Davidoff, et al., 2010; Millett, et al., 2010; Neuman, et al., 2007;
Rudolph & Montgomery, 2010; Skarupski, et al., 2009). While the benefit was intended to
increase access to medications, it appears that cost continues to pose barriers to adherence for
older Black Americans.
The lack of gender differences in CRN contrasts with previous research conducted before
Medicare Part D that found that older women had a greater likelihood of CRN than men (Heisler,
et al., 2005; Piette, et al., 2004c; Zivin, et al., 2010). One potential explanation involves the
availability of the Low-Income Subsidy (LIS). The LIS was established with the enactment of
the MMA and provides assistance with Part D costs (monthly premiums, annual deductible,
medication copayments) for beneficiaries with limited income and resources. As demonstrated in
Manuscript Two (Bakk, 2012a), gender differences in CRN over time were substantially driven
by whether an individual applied for the LIS. Further, subanalysis (not shown) confirmed that in
this study, 18% of older women applied for the LIS as compared to 8% of older males. Thus, it is
suggested that the availability of the LIS with the enactment of the MMA largely accounts for
nonsignificant relationship between gender and CRN under Medicare Part D in this study. Older
women were more likely than old men to apply for the subsidy, and this in turn appears to have
had an impact on gender differences in nonadherence.
Next, the prediction that experiencing the Medicare Part D coverage gap and restrictions
will increase the likelihood of CRN was supported. This study replicates previous findings

137

showing that compared to older adults without a gap in coverage, those reaching this benefit
threshold are more likely to report CRN (Cronk, et al., 2008; Hsu, et al., 2008; Raebel, et al.,
2008), and expands this research by illustrating that beneficiaries not reaching the coverage gap
also have a greater likelihood of nonadherence due to cost. It may seem counterintuitive that not
reaching the coverage gap increases the risk of CRN because Medicare Part D plans pay
approximately 75% of the cost of medications before reaching the gap. However, beneficiaries
who are aware that their plan has a gap in coverage often engage in cost-coping behaviors –
which includes decreasing their adherence to certain medications – to postpone or delay reaching
this benefit threshold, particularly if they experienced the gap during the prior year (Bakk, et al.,
2012). Thus, simply having a gap in coverage appears to have an impact on maintaining
adherence, even if a beneficiary has not incurred additional spending. When considering the
coverage restrictions, older adults whose plan did not pay for certain medications, required
paying more for some medications, or had yearly cost limitations were more likely to report
CRN as compared to those without these restrictions as expected. This finding is consistent with
research examining implications of cost-sharing requirements under state Medicaid plans
(Reeder & Nelson, 1985; Soumerai, et al., 1987), and provides important insights on how each of
these Medicare Part D restrictions impact adherence.
Finally, it was predicted that experiencing the Medicare Part D coverage gap and
restrictions would mediate the effect of racial, gender, socioeconomic, and health differences in
CRN. Support for this hypothesis was limited. Racial differences in CRN were primarily driven
not reaching the coverage gap. Findings suggest that having a gap in coverage can be associated
with CRN – even if this benefit threshold is not reached. Individuals often engage in cost-coping
behaviors to postpone or delay reaching this benefit threshold, which include becoming

138

nonadherent to a prescribed medication regimen, particularly if medication costs are problematic
(Bakk, McGuire, & Woodward, 2012). Interestingly, reaching the coverage gap was not
associated with race. It is difficult to determine why this relationship was not significant because
of the lack of research examining racial difference in CRN under Medicare Part D. A potential
explanation could be missing data. Older Black Americans in this study had a significantly
greater proportion of missing data when asked whether they reached the coverage gap than older
Whites. This could underestimate the findings. Future research is needed to better understand the
implications of both reaching and not reaching the coverage gap for older Black Americans.
In regards to the coverage restrictions, having this cost-sharing provision did not impact
the relationship between race and CRN. This is a bit confounding – particularly considering that
the gap had an indirect effect on the relationship between race and CRN but the coverage
restrictions did not. In other words, both the coverage gap and coverage restrictions can
potentially increase out-of-pocket expenditures, yet only the gap had an impact on racial
differences in CRN. A potential explanation for this finding involves the amount of cost sharing
required when entering the coverage gap. Under Medicare Part D, beneficiaries pay a 25%
copayment until reaching this benefit threshold; they then become responsible for 100% of their
medication costs. The coverage restrictions examined in this study – not paying for or requiring
an additional copayment for certain drugs and only paying for a certain number of medications
each year – may have presented less of a financial burden. For example, it is possible that the
coverage restrictions required a minimal copayment amount or only excluded certain
medications. An additional explanation involves potential confusion regarding whether coverage
restrictions exist under a beneficiaries’ Medicare Part D plan. Subanalyses (not shown) found
that a significantly higher proportion of older Black Americans had missing data because they

139

did not know if they had a gap in coverage. The PDS did not allow beneficiaries to respond that
they were unsure if they had coverage restrictions; they were only asked to check the specific
restrictions within their plan. Individuals are often uncertain whether cost-sharing requirements
exist within their Part D plan because of its complexity (Davidoff, et al., 2010; Heiss, et al.,
2006; Hsu, et al., 2008). Given that a greater proportion of older Black Americans were
uncertain whether they had a gap in coverage, it is speculated that this same uncertainty could
exist for coverage restrictions, thus underestimating the impact of this cost-containment
provision.
In regards to socioeconomic and health status, the coverage gap and restrictions had little
indirect effect on these relationships. Slight decreases in odds ratios were observed with the
addition of the cost-containment provision variables but the significance did not change. A
potential explanation involves higher medication costs due to comorbid chronic health problems.
Individuals in poor health with multiple comorbidities have high and persistent CRN over time
(Briesacher, et al., 2007; Madden, et al., 2008; Safran, et al., 2005). CRN can be caused by
higher out-of-pocket expenditures due to multiple medications (Frankenfield, et al., 2010; Zivin,
et al., 2010), and having a lower annual household income can further diminish ability to pay for
medications and increase the likelihood of nonadherence (Gellad, et al., 2007; Mojtabai &
Olfson, 2003; Soumerai, et al., 2006; Steinman, et al., 2001; Zivin, et al., 2010). It is suggested
that any cost-sharing under the Medicare Part D can be problematic – regardless of whether or
not it increases due to reaching the gap or incurring a coverage restriction. For lower-income,
chronically ill older adults, affording medications can be difficult even when copayment
requirements are minimal (Klein, et al., 2004; Soumerai, et al., 1987).

140

An additional finding worthy of note involves the impact of factors associated with
cumulative disadvantage on the relationship between race and CRN. Though the odds of CRN
for older Black Americans decreased with the addition of the socioeconomic and health status
variables, racial differences in CRN remained significant. A potential explanation involves
factors involving the physician-patient relationship and trust. The physician-patient relationship
has been found to influence CRN (I. B. Wilson, et al., 2004). Higher out-of-pocket prescription
drug expenditures have been associated with a greater likelihood to forgo medications because of
cost if trust in the physician is low (Piette, et al., 2005). Further, prior research has found that
Black Americans view their physician less favorably (Doescher, Saver, Franks, & Fiscella, 2000)
and have had worse experiences with primary care providers than Whites (Shi, 1999). Thus, for
older Black Americans in this study, physician trust may have played a contributing role in CRN.
They may have been less willing to discuss cost concerns or perceived benefit of a prescribed
medication because the physician-patient relationship was not satisfactory.
Strengths of this study include the use of a large nationally representative sample of older
adults and its ability to examine multiple predictors simultaneously. However, there are a
number of limitations that must be considered. First, the PDS was administered one year after the
implementation of the MMA. In the future, a longer post-policy period would provide more
clarity and permit a stronger assessment of Part D enrollment. Other unmeasured factors may
have influenced the results. For example, the implementation of Medicare Part D was a
confusing time for many older adults (Davidoff, et al., 2010; Heiss, et al., 2006; Hsu, et al., 2008;
Kaiser Family Foundation, 2006). As a result, CRN may have been overestimated because data
collection occurred shortly after the launch of Medicare Part D and at a time of initial confusion.
Additionally, this study was cross-sectional and thus, only provides a snapshot of CRN at one

141

point in time. Other factors such as the development of an acute health condition or sudden
change in economic status could have impacted ability to afford prescribed medications. An
examination of CRN over time would provide a stronger assessment of whether differences in
nonadherence persist under Medicare Part D. While this research provides important evidence on
the impact of the policy on racial and gender disparities in CRN and the impact of the MMA’s
coverage gap and restrictions, the results should be considered early evidence until longer-term
data become available.
Second, socioeconomic and health status were measured in 2006 HRS, and CRN was
measured in the 2007 PDS. There is the potential that income or certain health conditions may
have changed because these measures were not taken simultaneously or when CRN was
reported. While this seems less likely with income because older adults are largely dependent on
Social Security, there is greater potential for fluctuation in health status. Depression, for
example, can be transitory (Zivin, et al., 2010). Older Black Americans who enrolled in
Medicare Part D have a greater prevalence of depression as compared to older Whites
(Skarupski, et al., 2009). Thus, the indirect effect of health status on the relationship between
race and CRN may have been inaccurately represented in the results.
Third, this study investigated nonadherence due to cost only and did not account for other
factors that can impact adherence. For example, an individuals’ concerns or beliefs regarding
treatment through the use of medications can certainly impact adherence (Aikens & Piette,
2009). Some medications may be viewed as more essential than others (Lau, et al., 2008; Piette,
et al., 2006) and there could be racial differences in whether a prescription is perceived as
necessary, particularly when faced with cost concerns. As compared to older Whites, older Black
Americans’ ability to afford out-of-pocket expenditure requirements is typically less due to lower

142

socioeconomic status (Wei, et al., 2006). Thus, they may perceive the need for certain
medications differently because of economic concerns. Additionally, physician trust and
communication can be a factor in medication adherence (Piette, et al., 2005). It is possible that
for some older Black Americans in this study, physician trust and/or communication were low
and contributed to CRN. While this could account for some variation in CRN, these factors were
not the focus of this research but will be examined in future studies.
Fourth and as described above, older Black Americans had a significantly greater
proportion of missing data when asked whether they reached the coverage gap than older Whites.
Individuals can be confused by or unaware of the Medicare Part D coverage gap because of its
complexity (Davidoff, et al., 2010; Heiss, et al., 2006; Hsu, et al., 2008). This, in turn, could bias
the results and overestimate or underestimate the racial effect of the coverage gap. Further, this
study relied on self-reported survey measures. Self-reporting can be subject to social desirability
bias (Nederhof, 2006). Respondents may be reluctant to admit failure to adhere to a prescribed
medication regimen – particularly if it is due to cost factors. Because older Black Americans are
more likely to be economically disadvantaged, they may incur greater difficulty affording the
cost of prescribed medications – and, admitting failure to adhere because of economic reasons.
This bias could contribute to underestimating the prevalence of CRN.
Fifth, there are limitations with adherence measures. The PDS did not ask respondents
whether they took less than the prescribed dosage of a medication or if they cut pills in half to
make a medication last longer. The survey questioned if a dosage of a medication was skipped to
save money, and this measure could be a potential indicator of these behaviors. However, failure
to ask if these behaviors specifically occurred could have underestimated CRN.

143

Finally and as stated previously, this study utilized a composite measure to estimate CRN
as opposed to including all three behaviors as separate indicators: 1) not filling a new
prescription because of cost; 2) stopping taking a medication because of cost; or 3) skipping
doses of a medication to save money. While this is consistent with previous research and
therefore useful for cross-study comparison (Bambauer, et al., 2007; Gellad, et al., 2007;
Madden, et al., 2008; Safran, et al., 2005; Soumerai, et al., 2006; I. B. Wilson, et al., 2004; Zivin,
et al., 2009), measuring each behavior separately may provide important insights for social work
practice on racial and gender disparities in CRN. For example, one behavior could be more
prevalent than others among older women or Black Americans. Future research will examine
each CRN item as an outcome.
In conclusion, the study makes a strong and significant contribution to our understanding
of whether and to what extent racial and gender disparities in CRN exist under Medicare Part D
despite these limitations. It extends our knowledge of how the benefit’s cost-containment
provisions – or coverage gap and restrictions – directly and indirectly affect the relationship
between race, gender, socioeconomic and health status, and CRN. While the MMA provided
much-needed assistance to those previously lacking prescription drug coverage, the Part D
benefit contains high levels of cost sharing that can impact adherence among vulnerable
subpopulations. Overall, the study suggests that while gender differences did not exist in CRN
for those enrolled in the Medicare Part D benefit, racial disparities in CRN existed after
accounting for demographics, health status, socioeconomic status, and the coverage restrictions.
Reaching and not reaching the coverage gap substantially accounted for racial differences in
CRN. The coverage gap and restrictions, having a lower annual income, poorer health, and
applying for the LIS was associated with CRN under the Medicare Part D benefit.

144

CHAPTER FIVE
CONCLUSIONS
Overview
The purpose of this dissertation was to better understand how racial and gender
disparities in cost-related medication nonadherence (CRN) have been impacted by Medicare Part
D. Prior to the implementation of the Part D benefit, CRN was greater among older women than
men (Heisler, et al., 2005; Piette, et al., 2004c; Zivin, et al., 2010), and numerous studies found
that Black Americans were more likely than Whites to forgo or delay taking prescribed
medications because of cost (Bambauer, et al., 2007; Klein, et al., 2004; Roth, et al., 2009; Roth
& Ivey, 2005; Soumerai, et al., 2006; Steinman, et al., 2001). While gender and racial disparities
in CRN prior to Medicare Part D clearly existed, studies had not examined whether these
differences persist since the implementation of the Part D benefit.
The research in this dissertation addresses this current gap in literature. Hypotheses were
developed based on previous research. Cumulative advantage/disadvantage theory was used as a
foundation for understanding the effectiveness of Medicare Part D, and both longitudinal and
cross-sectional analyses were conducted using a nationally representative secondary dataset. The
findings in Chapters Two, Three, and Four shed light on how racial and gender differences in
CRN have been impacted by the Part D benefit, and deepen our understanding of how factors
associated with the benefit impact adherence. In this chapter, key findings are summarized and
the linkages between studies are considered. Conclusions are drawn in relation to racial and
gender differences in CRN and Medicare Part D, and implications for social work practice,
policy and education are discussed. The chapter concludes by summarizing the limitations of this
study and presenting recommendations for future research.

145

Summary of Findings
The first paper (Chapter Two) examined whether racial and gender disparities in CRN
have changed since the implementation of Medicare Part D. Specifically, it was expected that
compared to older males and Whites, older females and Black Americans would be more likely
to report CRN before and after the implementation of Medicare Part D and less likely to report a
decline in CRN over time. Additionally, it was anticipated that socioeconomic status, health
status, and prescription insurance coverage would affect CRN over time and explain racial and
gender variations.
As expected, there was a significant disparity in CRN between older Black Americans
and Whites both before and after the implementation of the benefit; however, these differences
did not change over time. Similarly, older females were more likely than males to report CRN in
both 2005 and 2007. Interestingly, contrary to expectations, older women experienced a
significantly greater decline in nonadherence over time compared to men. The availability of the
Low-Income Subsidy (LIS) may be one reason one reason for this finding because older women
tend to have fewer financial resources and, therefore, are more likely to qualify for the subsidy.
Also contrary to expectations, disparities remained significant after controlling for factors
associated with cumulative advantage/disadvantage – specifically, socioeconomic status and
health status. This was true across all three studies and potential reasons for this finding will be
explored later in this section. Finally, as predicted, lower socioeconomic status, poorer health
status, and less generous or no prescription insurance coverage had a direct effect on CRN over
time.
While Chapter Two included respondents who met study criteria regardless of source of
prescription drug coverage, Chapter Three included only a subset respondents that enrolled in

146

Medicare Part D. Racial and gender disparities in CRN existed over time among enrollees were
examined as well as the direct and indirect effects of the Low-Income Subsidy (LIS). Hypotheses
were similar to Chapter Two, with the additional hypothesis that LIS status would mediate racial
and gender differences in CRN and decrease the likelihood of nonadherence over time.
As in Chapter Two, racial and gender disparities in CRN persisted even after accounting
for health and socioeconomic status. Also, lower income and poorer health status increased the
likelihood of nonadherence. There was limited support for the predictions regarding LIS status.
As was speculated in Chapter Two, gender differences in CRN were largely driven by whether
an individual applied for the LIS. However, LIS status did not affect racial differences in CRN.
Previous studies found that Black Americans at or below 150% of poverty were significantly less
likely to be aware of the LIS benefit (Neuman, et al., 2007; Skarupski, et al., 2009). Bivariate
analysis established that this may be the case in the present sample as well. Further, applying for
the LIS resulted in an increased likelihood of CRN. This may be because all respondents
enrolling in Medicare Part D were included in the sample, regardless of income and assets.
Unlike the other chapters, Chapter Four used cross-sectional analysis to focus specifically
on the impact of the Medicare Part D coverage gap and restrictions on racial and gender
differences in CRN. In addition to racial and gender differences in CRN, it was expected that
experiencing the Medicare Part D coverage gap and restrictions would mediate the effects of
race, gender, socioeconomic and health status differences in CRN and increase the likelihood of
nonadherence.
As with previous chapters, racial disparities in CRN existed after controlling for
socioeconomic and health status. While the coverage restrictions did not affect differences in
CRN between older Black Americans and Whites, the coverage gap mediated racial differences.

147

As predicted, both the coverage gap and restrictions directly affected CRN. Contrary to
expectations, poorer health and lower income resulted in an increased likelihood of CRN after
controlling for the coverage gap and restrictions. Higher medication costs due to comorbidities
could be a reason for this. CRN is often caused by higher out-of-pocket spending resulting from
the use of several medications (Frankenfield, et al., 2010; Zivin, et al., 2010). Low income
individuals often have more difficulty affording medications and have a greater risk of CRN
(Gellad, et al., 2007; Mojtabai & Olfson, 2003; Soumerai, et al., 2006; Steinman, et al., 2001;
Zivin, et al., 2010), even when copayment requirements are small (Klein, et al., 2004; Soumerai,
et al., 1987). Finally, gender differences in CRN did not exist in this study. The availability of
the LIS could be a reason for this finding.
Cumulative Advantage/Disadvantage Theory and CRN
This study used cumulative advantage/disadvantage theory as a foundation for
understanding how Medicare Part D impacted CRN. There were interesting findings with regards
covariates related to cumulative advantage/disadvantage theory across all three studies. It was
hypothesized that CRN would be associated with early disadvantage – measured by income,
education, and health status – and these factors, in turn, would have an indirect effect on the
relationship between race, gender, and CRN. In all studies, racial differences persisted after
controlling for these variables. Further, in Chapters Two and Three, gender disparities existed
both before and after Medicare Part D after accounting for these covariates.
There are several reasons why factors related to cumulative disadvantage had little impact
on these relationships. First, the physician-patient relationship has been found to influence CRN
(I. B. Wilson, et al., 2004). Higher out-of-pocket prescription drug expenditures have been
associated with a greater likelihood to forgo medications because of cost if trust in the physician

148

is low (Piette, et al., 2005). It is possible that for some older Black Americans and females in this
study, trust was low and contributed to CRN. Physician trust may hold particular relevance when
considering Black Americans given their experiences with the health care system. Older Black
Americans first-contact experiences with primary care providers have been worse than Whites,
and they have experienced greater barriers obtaining appointments and longer wait times during
an appointment (Shi, 1999). Further, Black Americans are more likely to perceive race-based
discrimination in their interactions with providers (Bird & Bogart, 2001). These experiences, in
turn, could impact physician trust and subsequent desire to maintain adherence, particularly
when faced with cost difficulties.
Second, physician-patient communication regarding costs can influence CRN.
Individuals experiencing problems paying for their medications are typically not asked if they
can afford the cost by their physician (Alexander, Casalino, & Meltzer, 2003; Piette, et al.,
2004b), and almost half of persons reporting CRN did not discuss cost concerns their physicians
(Piette, et al., 2004b; I. B. Wilson et al., 2007). For older Black Americans, physician-patient
communication could be even lower because the relationship may be less patient-centered
(Johnson, Roter, Powe, & Cooper, 2004) and participatory (Cooper-Patrick et al., 1999). In other
words, the physicians may be more directive and therefore less likely to engage older Black
Americans in a conversation about ability to afford prescribed medications, particularly in racediscordant dyads (Cooper-Patrick, et al., 1999). Physicians can also hold unconscious racial
biases or negative beliefs that can influence their interactions with patients (Doescher, et al.,
2000; van Ryn & Burke, 2000). For example, physicians may view Blacks as being less willing
to comply with treatment (Gregory, Wells, & Leake, 1987) and thus, be less likely to discuss cost
concerns when prescribing a medication.

149

Third, cultural factors can play a role in adherence with a medication regimen. A
patient’s personal beliefs or values regarding the use of pharmaceuticals to treat a certain
condition can subsequently impact CRN, and they may respond differentially when facing costs
for essential versus nonessential treatments (Piette, et al., 2006). For example, if a prescribed
medication is perceived as unnecessary (Bakk, et al., 2012) or harmful (Aikens & Piette, 2009),
nonadherence can increase – especially if cost is an issue.
Finally, other unexplored cost-related factors may have accounted for the lack of effect.
This study could not control for individual differences within Part D plans. There are over 40
Medicare Part D plans (Kaiser Family Foundation, 2011b). Cost-sharing can vary extensively
within plans (Hoadley, et al., 2006). Each plan’s out-of-pocket expenditure requirements can
impact CRN. For example, cost-sharing requirements are typically greater for individuals with
diabetes than most other chronic illnesses under Medicare Part D plans (Ettner, et al., 2010).
Because older Black Americans are more likely to be diabetic than Whites (Schoenbaum &
Waidmann, 1997; Schoenborn & Heyman, 2009) and have comorbid health problems requiring
multiple medications (Skarupski, et al., 2009), their chances of incurring cost-sharing under Part
D plans are typically greater. Further, in this study, older females and Black Americans had a
significantly greater number of medications in both 2005 and 2007 as compared to older males
and Whites. Thus, the variation in cost sharing between plans may have a greater impact on older
women and Black Americans because their health status is typically poorer and as a result, they
often have higher prescription drug needs. This, in turn, can impact their ability to purchase
prescribed medications and maintain adherence.

150

Implications for Social Work Practice, Policy, and Education
This dissertation provides important insights into Medicare Part D’s effectiveness in
eliminating racial and gender differences in CRN and factors associated with the benefit that
impact compliance. The collective and individual findings suggest implications for practice,
policy, and education.
Social Work Practice
Communication about difficulties affording medications is essential in order to lower or
prevent instances of CRN, particularly given the association between CRN, worse health
outcomes (Heisler, et al., 2004), and increased medical costs (Soumerai, et al., 1994; Soumerai,
et al., 1991; West, et al., 2007). Social work literature has not examined communication patterns
regarding adherence between practitioners and clients, and studies examining physician-patient
dialogue about CRN are quite limited. Only 35 percent of physicians asked patients about their
ability to afford a prescribed drug (Alexander, et al., 2003), and nearly 40 percent of older adults
reporting CRN did not discuss problems affording medications with their physicians (Piette, et
al., 2004b; I. B. Wilson, et al., 2007). As suggested previously, physician-patient communication
could be even worse older Black Americans. Thus, lack of dialogue about ability to afford
prescribed medications has specific implications for direct practice as more attention to and
discussion about adherence appears needed. Physicians should not be the sole source of
communication about cost difficulties. As described below, there are multiple avenues where
social workers have the ability to intervene and provide this much-needed assistance.
First, findings suggest that older Black Americans and females were more likely to report
CRN before and after Medicare Part D than older Whites and males. Further, having poorer
health and lower annual income – which is more common for older Black Americans and

151

females – increased the likelihood of nonadherence due to cost. Despite the establishment of a
drug benefit under Medicare, social workers should be aware that CRN can still be problematic,
and racial and gender differences in the ability to maintain adherence continue to exist. Because
social workers often assume positions where they actively participate in performing
biopsychosocial assessments and linking persons to needed services and resources, they are
among the first to come in contact with older adults struggling with prescription drug costs. For
example, social workers may work in Area Agencies on Aging or county-specific aging
organizations providing information and assistance which includes locating resources to meet the
identified needs of callers. As part of the intake and referral process, questions pertaining to
prescription drug use and adherence could be integrated into the assessment instrument.
Resources such as pharmaceutical company prescription assistance programs and county
prescription drug discount programs can then be provided to older individuals unable to afford
medications, thus increasing their ability to maintain adherence. An additional personal example
involving practice implications pertains to a Home-Delivered Meals (HDM) program in
southeast Michigan. This research directly informed organizational practices and how older
adults are assessed for the program. It was discovered that despite its mandate to serve older
individuals most in need – primarily, low-income and minority persons (Gelfand, 2006) – the
HDM program was not asking participants about medication adherence during intake. The
screening instrument has since been modified, and individuals having difficulty affording
medications are now referred to their local Area Agency on Aging Information and Referral
program for pharmaceutical resources that can assist during periods of non-coverage or limited
coverage.

152

Second, emphasis on the use of a collaborative, interdisciplinary model for service
delivery is becoming more prevalent in social work (Bronstein, 2003; Mizrahi & Abramson,
2000). For example, section 3026 of the Patient Protection and Affordable Care Act of 2009
(PPACA) provided substantial funding to community-based aging organizations to reduce
hospitalizations for high-risk Medicare beneficiaries through collaboration with hospitals,
physicians, nurses, health care organizations, and community organizations. This five-year
program is slated to begin in mid-2012 and stresses an interdisciplinary approach to improve
quality of care for comorbid older adults. Findings from this dissertation may have particular
relevance not only in settings such as these where collaboration is encouraged, but also
demonstrate the need to form interdisciplinary relationships. In Chapter Four, racial differences
in CRN were substantially driven by having a gap in coverage. Further, having a coverage gap or
restrictions, poorer health status, or lower income increased the likelihood of nonadherence. By
collaborating with other disciplines, social workers can help older clients that have a greater risk
of CRN – those having a coverage gap, restrictions, poorer health, or lower income – avoid
CRN. For example, by discussing at-risk clients with physicians, less costly generic alternatives,
therapeutic substitutions (Sharkey, Ory, & Browne, 2005; I. B. Wilson, et al., 2007) or
medication samples (Bakk, et al., 2012) may be provided. Further, hospitals and community
mental health often have programs or resources that can assist persons struggling with
medication costs. Social workers can work with these organizations to help clients that are more
likely to have problems affording their medications. By forming collaborative relationships with
other disciplines, practitioners can help clients identified by this research as having more
difficulty with prescription drug costs to maintain adherence.

153

Third, findings illustrate the importance of outreach to increase awareness of the LIS
benefit. In Chapters Two and Three provided evidence that the LIS influenced gender, but not
racial, differences. These findings indicate that the LIS can potentially impact adherence. A
reason why racial disparities persisted after accounting for whether or not they applied for the
subsidy is awareness. As compared to older Whites, Black Americans were less likely to know
about the LIS benefit (Neuman, et al., 2007). In this study, only 36% of older Black Americans
reported being aware of the LIS in 2007 as compared to 59% of Whites. Further, 57% of older
females were aware of the LIS in 2007 compared to 43% of males. Thus, the importance of
outreach to increase awareness of the benefit is certainly demonstrated, particularly in light of
how applying for the LIS affected gender differences in CRN over time. Despite efforts to
promote the LIS, nonparticipation of eligible beneficiaries was estimated at 3 million or slightly
less than 50% (Federman et al., 2009). Social work can assume a key role by conducting
aggressive, targeted outreach efforts to increase LIS awareness and enrollment. This can include
providing information and education about the benefit in senior centers, physicians’ offices,
churches, hospitals, public health departments, community-based organizations and places that
may have high concentrations of lower income individuals, particularly Black Americans.
Additionally, since health insurance presentations were associated with greater awareness of
pharmaceutical cost-assistance programs among low-income older adults (Federman, et al.,
2009), this could be a particularly important strategy to promote knowledge of the program.
Finally, questions pertaining to LIS eligibility and assistance should also be incorporated into
screening tools used by programs such as care management and Home Delivered Meals as an
attempt to increase awareness of this important program.

154

Another area where social work should consider providing outreach and assistance is
Medicare Part D plan enrollment. Each year, Part D plans have the opportunity to change benefit
packages. Some of these changes are substantial. For example, the monthly premium for the
Humana Standard plan – the largest Medicare Part D plan – increased 329 percent from 2006 to
2009. Additionally, plan restrictions and amount of cost-sharing required when reaching the gap
change from year to year (Cahill, 2010). These changes, in turn, can significantly increase
beneficiaries’ out-of-pocket costs and lead to difficulties affording medications, particularly
among older women and Black Americans because of their lower income and higher utilization.
Despite increased costs, most beneficiaries are not changing their coverage plans (Jackson &
Axelsen, 2008), in part because of plan complexity (Bakk, et al., 2012; Heiss, et al., 2006). The
Part D benefit is very confusing, and older adults often have difficulty understanding how the
benefit is structured, plan restrictions, and the coverage gap (Heiss, et al., 2006; Hsu, et al., 2008;
Rosenbaum & Teitelbaum, 2005; Wilk et al., 2008). As a result, less than 10 percent are in the
most cost-effective plan for their medication regimen (Hanoch, Rice, Cummings, & Wood, 2009;
Neuman & Cubanski, 2009). Because plans can vary significantly, enrolling in the most
appropriate plan is critical. For example, in 2009, the cost of Aricept (a commonly prescribed
drug for Alzheimer’s disease) ranged from as little as $20 in one plan to $88 in another (Neuman
& Cubanski, 2009). Finally, the best way to choose the most appropriate Part D plan is through
the Centers for Medicare and Medicaid Services (CMS) website. This form of delivery can pose
barriers because computer and Internet use can vary by age, income, education, and comorbidity
status (Cutler, Hendricks, & Guyer, 2003; Morris, 1996). For example, individuals with more
income and education are more likely to own a computer (Cheeseman Day, Janus, & Davis,
2005). Further, for many older adults, Internet access is either unaffordable or unavailable

155

(Woodward et al., 2011). It cannot be assumed that all groups are able to access the benefit
equally or have the ability or desire to use computer technology.
Thus, social workers should not only educate older adults about the need to evaluate their
coverage plan on a yearly basis, but also need to provide assistance with the enrollment process.
For example, Medicare Part D enrollment days could be conducted in the community at venues
such as Area Agencies on Aging, hospitals, or senior centers where social workers provide oneon-one assistance in evaluating coverage options and plan enrollment. Additionally, communitybased case managers should evaluate older clients’ Medicare Part D plans on an annual basis to
ensure they are in the most appropriate plan based on their prescription drug regimen. Finally,
while all older persons may not have a desire to use the computer and Internet (Cheeseman Day,
et al., 2005), those that want utilize Medicare’s online resource to choose their Part D plan
should receive the training needed to annually evaluate their options. Social work can assume a
key role in providing computer education and training to older adults so they are empowered to
make this choice.
Social Work Policy
The results of Chapter Four hold particular relevance as we consider social work’s role in
public policy. As demonstrated, racial differences in CRN were substantially driven by having a
gap in coverage. The need to reduce or eliminate the Part D coverage gap has been recognized in
this dissertation as well as previous research (Kaiser Family Foundation, 2010a, 2011a; Pedan, et
al., 2009; Sun & Lee, 2007; Zhang, et al., 2009). The PPACA called for a gradual “phasing out”
of the Medicare Part D coverage gap by reducing cost sharing for brand-name medications and
providing partial coverage for generic medications (Cahill, 2010; Kaiser Family Foundation,
2011a; Traynor, 2010) and is an important step toward reconciling racial, gender, economic, and

156

health disparities CRN. In 2011, beneficiaries received a 50 percent discount for some brand
name prescription medications in the coverage gap. By 2020, the discount will increase to 75
percent. This is significant, but perhaps a bit misleading because “phasing out” the gap could be
construed as eliminating copayment requirements altogether. Phasing out the gap simply brings
the benefit in line with the coverage before the gap began; beneficiaries will be responsible for
25 percent of their prescription drug costs. Further, the PPACA does not address those problems
faced by near-poor older adults who experience difficulty affording copayments because they
minimally exceed LIS income or asset limitations (Bakk, et al., 2012). As demonstrated in
Chapter Three, beneficiaries whose income minimally exceeded LIS income requirements
continued to have a greater likelihood of CRN as compared to the highest income group.
While the PPACA is certainly a positive step in reducing disparities, there are several
reasons to be cautious. First, insurers could react to these provisions by increasing Part D
premiums to manage the change (Davidoff, et al., 2010). Because the cost of prescription drugs
continues to increase faster than inflation (Kaiser Family Foundation, 2009a), medication costs
may still pose significant difficulties for economically vulnerable populations, especially those
who exceed the LIS income/asset thresholds because cost-sharing is still required. Second, while
the law in its final form calls for a significant reduction in cost sharing (to 25 percent), the
remaining costs will still be out of reach for many Part D beneficiaries (Ettner, et al., 2010).
Failing to completely close the coverage gap will still present economic burdens that lead to
older adults rationing their own medications (Bakk, et al., 2012). Finally, given mounting
pressure to reduce the nation’s debt and federal deficit, there are currently threats to repeal the
law (Herszenhorn & Pear, 2011, January 19). While a full repeal is extremely unlikely because
the Senate has not passed the bill and President Obama has promised to veto it (Kaiser Health

157

News, 2011, January 6), some parts of the bill may be in danger. If reforms to Part D provisions
that assist the worst off beneficiaries, such as narrowing the coverage gap, are not maintained,
the most vulnerable adults will face even greater economic hardship and more serious health
outcomes (Bakk & Sosulski, in press) because of the increasing cost of medications coupled with
their greater susceptibility to CRN due to poorer health status and lower income.
Social workers need to be vigilant and continually monitor the progress of Medicare
policy and Part D benefits, in particular. The profession holds a critical role in advocating for
greater social justice, and fair and equitable access to benefits and services. The Medicare Part D
coverage gap can cause the most vulnerable groups to become nonadherent to their medication
regimen and potentially result in a deterioration in health status (Heisler, et al., 2004) as well as
increased health care expenditures (Soumerai, et al., 1994; Soumerai, et al., 1991; West, et al.,
2007). Social service providers and other policymakers should advocate not only that the
PPACA coverage gap provisions remain intact, but also that greater consideration be placed on
how the gap provision impacts more vulnerable groups in future policy decisions. Failure to
address these differentials can unintentionally perpetuate racial and gender disparities. Further,
given that the LIS mediated gender differences and differences in between those earning $12,721
to $23,363 and the highest income group, it is important to advocate for the expansion of this
program. Finally, advocacy efforts are needed to simplify the structure of the benefit. As
illustrated in the previous section, there are multiple problems faced by older adults because of
the complexity of the benefit, and these issues are particularly relevant for more vulnerable
populations. Policymakers should considering decreasing the number of Medicare Part D plans
and delivering the benefit in a less complicated manner. In this age of reform, gender and race

158

should be fully considered in order to create changes that enhance older adults’ health and wellbeing and reduce rather than increase overall medical spending.
Social Work Education
Social workers will assume a key role in providing assistance to older adults with health
care benefits as the population continues to grow. By 2050, it is projected that the number of
Americans age 65 and older will be 88.5 million, or over double the 2010 population of 40.2
(Vincent & Velkoff, 2010). It is also anticipated that the older adult population will be much
more racially diverse. In 2006, Black Americans represented 9 percent of all those age 65 and
over; by 2050, Black Americans will comprise 12 percent of the older adult population (The
Federal Interagency Forum on Aging-Related Statistics, 2008). One area that will increasingly
demand the attention of the profession is access to pharmaceuticals under Medicare Part D
because prescription drugs are an increasingly important part of a treatment regimen (Medicare
Payment Advisory Commission, 2006).
The results of this dissertation have distinct implications for social work education.
Despite the implementation of Medicare Part D, maintaining adherence continues to be
problematic for older females and Black Americans. Thus, it is important to include this topic in
policy, practice, and continuing education curricula in order to facilitate medication access and
avoid nonadherence, particularly among vulnerable older adults, as well as exercise sensitivity to
the differences within the population. Social workers should be educated on the intricacies of the
benefit so they have the ability to assist older persons in choosing the most feasible drug plan for
their medication regimen, and recognize that benefit choice – which determines the amount of
cost sharing required when encountering the coverage gap or restrictions – can subsequently
influence medication compliance. Additionally, findings demonstrate the need to include this

159

topic in social work curriculum in order to raise awareness of resources available to older adults,
such as the LIS program, prescription assistance programs and medication samples, which can
benefit persons during periods of non-coverage. Finally, social workers should understand that
older adults cannot be treated a homogeneous group when considering social policy initiatives
(Neugarten, 1982), in particular Medicare Part D. Disparities need to be recognized in order to
ensure equitable access to benefits.
Limitations and Future Research
As discussed in each chapter, this dissertation has several limitations. This section
discusses limitations that are common in Chapters Two, Three, and Four, and provides
suggestions for future research.
First, a longer post-policy period would provide more clarity and permit a stronger
assessment of Part D enrollment. Other unmeasured factors may have influenced the results. For
example, the transition of dual eligibles from Medicaid to Medicare drug plans in 2006 was
confusing for many older adults (Kaiser Family Foundation, 2006) and resulted in difficulties
obtaining medications (West, et al., 2007). A significantly greater proportion of older Black
Americans in this study were dually eligible as compared to Whites, and thus, had a greater
likelihood of being impacted by the transition. As a result, CRN may have been overestimated
because data collection occurred shortly after the launch of Medicare Part D and at a time of
initial confusion. While this research provides important evidence on the impact of the policy on
racial and gender disparities in CRN, the results should be considered early evidence until
longer-term data become available. Thus, future studies should provide a longer post-policy
series to more adequately assess how Medicare Part D has impacted disparities in CRN.

160

Second, variables were measured at differing time points. For example, socioeconomic
and health status were measured in the 2004 and 2006 HRS, and CRN was measured in the 2005
and 2007 PDS. Thus, there is the potential that income or certain health conditions may have
changed because these measures were not taken simultaneously or when CRN was reported.
While this seems less likely with income because older adults are largely dependent on Social
Security, there is greater potential for fluctuation in health status. Depression, for example, can
be transitory (Zivin, et al., 2010). Older females and Black Americans who enrolled in Medicare
Part D have a greater prevalence of depression as compared to older males and Whites
(Skarupski, et al., 2009). Thus, the indirect effect of health status on the relationship between
race, gender, and CRN may have been inaccurately represented in the results. Future research
should contain measures taken at the same time point.
Third, this study investigated nonadherence due to cost only and did not account for other
factors that can impact adherence. As stated previously, an individuals’ concerns or beliefs
regarding treatment through the use of medications can certainly impact adherence (Aikens &
Piette, 2009). Some medications may be viewed as more essential than others (Lau, et al., 2008;
Piette, et al., 2006) and there could be racial and gender differences in whether a prescription is
perceived as necessary, particularly when faced with cost concerns. As compared to older males
and Whites, older females’ and Black Americans’ ability to afford out-of-pocket expenditure
requirements is typically less due to lower socioeconomic status (Wei, et al., 2006). Thus, they
may perceive the need for certain medications differently because of economic concerns.
Additionally, physician trust and communication can be a factor in medication adherence (Piette,
et al., 2005). It is possible that for some older females and Black Americans in this study,
physician trust and/or communication were low and contributed to CRN. While this could

161

account for some variation in CRN, these factors were not the focus of this research but should
be examined in future studies.
Fourth, this study relied on self-reported survey measures. Self-reporting can be subject
to social desirability bias (Nederhof, 2006). Respondents may be reluctant to admit failure to
adhere to a prescribed medication regimen – particularly if it is due to cost factors. Because older
women and Black Americans are more likely to be economically disadvantaged, they may incur
greater difficulty affording the cost of prescribed medications – and, admitting failure to adhere
because of economic reasons. This bias could contribute to underestimating the prevalence of
CRN. Other methods of measurement, such as the use of Medicare claims data, should be
considered in future studies.
Fifth, there are limitations with adherence measures. The PDS did not ask respondents
whether they took less than the prescribed dosage of a medication or if they cut pills in half to
make a medication last longer. The survey questioned if a dosage of a medication was skipped to
save money, and this measure could be a potential indicator of these behaviors. However, failure
to ask if these behaviors specifically occurred could have underestimated CRN. Future studies
should account for these factors as they can impact adherence.
Sixth, mixed-effect logistic regression models were used in Chapters Two and Three to
examine racial and gender differences in CRN over time. The analyses were performed in the
Stata version 11.0 statistical package (StataCorp). Stata’s survey commands do not allow the
incorporation of the survey command or application of HRS probability weights in mixed-effect
logistic regression analyses. Thus, the models did not control for the design effect and were not
weighted. A consequence of not utilizing the questionnaire weights or controlling for the design
effect is the potential for inaccurate point estimates and/or inaccurate standard errors, which in

162

turn can impact the ability to infer results to the general population. Future research is needed to
determine whether there are any substantive differences between these findings and comparative
analyses using a methodology that accounts for the design effect and probability weights.
Finally, this study utilized a composite measure to estimate CRN as opposed to including
all three behaviors as separate indicators: 1) not filling a new prescription because of cost; 2)
stopping taking a medication because of cost; or 3) skipping doses of a medication to save
money. While this is consistent with previous research and therefore useful for cross-study
comparison (Bambauer, et al., 2007; Gellad, et al., 2007; Madden, et al., 2008; Safran, et al.,
2005; Soumerai, et al., 2006; I. B. Wilson, et al., 2004; Zivin, et al., 2009), measuring each
behavior separately may provide important insights for social work practice on racial and gender
disparities in CRN. For example, one behavior could be more prevalent than others among older
women or Black Americans. By understanding which types or types of nonadherence are more
common in older women and Black Americans, interventions that specifically target certain
behaviors can be designed. Future research should examine each CRN item as an outcome.
In addition to the above suggestions for future studies, there are other areas where more
research is needed. First, it was somewhat surprising that persons over the age of 75 were less
likely to experience CRN than those age 65 to 74, particularly because research shows lower
prescription drug coverage rates among older age groups (Kanavos & Gemmill-Toyama, 2010).
This finding, however, is consistent with cross-sectional studies using HRS to examine CRN
before the implementation of Medicare Part D (Klein, et al., 2004; Zivin, et al., 2010). One
potential reason involved the sampling of the HRS and exclusion of institutionalized persons
within the sample. Because risk of institutionalization or nursing home placement increases with
age, older age groups residing within the community and participating in the HRS may be

163

disproportionately healthy (Klein, et al., 2004). This explanation, however, could not be
confirmed through additional analysis. Therefore, future research is needed to understand why
younger persons were more likely to experience CRN than older age groups.
Second, this study examined racial disparities in CRN between older Whites and Black
Americans only. The need to reduce health disparities between Whites and minority racial and
ethnic groups, which includes Blacks, Hispanics, Asian/Pacific Islanders, and American
Indians/Alaska Natives, has increasingly garnered the attention of the federal government (U.S.
Department of Health and Human Services, 2011, April 8; Walker, Mays, & Warren, 2004).
Overall, racial and ethnic minority groups are less likely to receive needed health services, even
after accounting for factors associated with cumulative disadvantage (Institute of Medicine of the
National Academies, 2003). Thus, in order to more fully understand how Medicare Part D has
impacted disparate access to medications, future research should broaden this comparison.
Finally, the ramifications of CRN with a prescribed medication regimen prior to
Medicare Part D have been established. Individuals who restricted medication use because of
cost were more likely to exhibit a deterioration in overall health status over time (Heisler, et al.,
2004). Further, persons with coverage restrictions were twice as likely to enter a nursing home,
and more likely to use hospital and outpatient services than those without restrictions (Soumerai,
et al., 1994Soumerai, et al., 1994; Soumerai, et al., 1991). Given that the addition of the Part D
benefit represented the largest expansion in the history of the Medicare program, future research
should consider the long-term health implications associated with enrollment in the benefit. This
is particularly relevant when considering how the LIS mediated gender differences. In other
words, it is important to understand whether improved health outcomes are associated with this
reduction in CRN and whether the policy has lowered health care costs.

164

Conclusions
The findings presented in this dissertation raise questions of whether the Part D benefit is
sufficient for older adults with increased risk of CRN. It is important to look at not only the
overall effect of Medicare Part D but also the relative effect for vulnerable populations. Results
suggest that older Black Americans and females were more likely to report CRN before and after
Medicare Part D than older Whites and males. While applying for the LIS mediated gender
differences in CRN, racial disparities appeared to be driven by having a Medicare Part D plan
with a gap in coverage. Poorer health, lower annual income, less generous or no prescription
drug coverage, applying for the LIS, and experiencing coverage gap and restrictions increased
the likelihood of CRN.
Understanding how Medicare Part D has affected adherence can help social work
practitioners recognize that clients may need additional resources and assistance in order to avoid
CRN. Additionally, findings can help substantiate the need to account for race and gender when
evaluating policy alternatives in order to promote more equitable access to medications and
avoid disparate health outcomes. Equity and fairness should receive consideration as policy
debates continue.

165

BIBLIOGRAPHY

166

BIBLIOGRAPHY
Adams, A. S., Soumerai, S. B., & Ross-Degnan, D. (2001). The case for a Medicare drug
coverage benefit: A critical review of the empirical evidence. Annual Review of Public
Health, 22, 49-61.
Aikens, J. E., & Piette, J. D. (2009). Diabetic patients' medication underuse, illness outcomes,
and beliefs about antihyperglycemic and anthypertensive treatments. Diabetes Care,
32(1), 19-24.
Alexander, G. C., Casalino, L. P., & Meltzer, D. O. (2003). Patient-physican communication
about out-of-pocket costs. JAMA, 290(7), 953-958.
Anderson, P. D., Long, J. S., & Krauze, T. K. (1982). Cumulative advantage and inequality in
science. American Sociological Review, 47, 615-625.
Bakk, L. (2012a). Cost-related medication nonadherence under Medicare Part D and the LowIncome Subsidy: A racial and gender comparison, Manuscript in preparation.
Bakk, L. (2012b). Racial and gender disparities in source of drug coverage and impact on
nonadherence due to cost, Manuscript in preparation.
Bakk, L., McGuire, J., & Woodward, A. T. (2012). The Medicare Part D coverage gap:
Medication access, adherence, and factors impacting compliance, Manuscript in
preparation.
Bakk, L., & Sosulski, M. R. (in press). A social justice perspective on Medicare Part D in an age
of reform: Critical implications of trends in health care policy and advocacy. Journal of
Sociology & Social Welfare.
Bambauer, K. Z., Safran, D. G., Ross-Degnan, D., Zhang, F., Adams, A. S., Gurwitz, J., . . .
Soumerai, S. B. (2007). Depression and cost-related medication nonadherence in
Medicare beneficiaries. Archives of General Psychiatry, 64, 602-608.
Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social
psychological research: Conceptual, strategic, and statistical considerations. Journal of
Personality and Social Psychology, 51(6), 1173-1182.
Bass, S. A. (2005). Medicare reform - A wolf in sheep's clothing. Journal of Aging & Social
Policy, 17(3), 1-17.
Basu, A., Yin, W., & Alexander, G. C. (2010). The impact of Medicare Part D on MedicareMedicaid dual-eligible beneficiaries' prescription utilization and expenditures. Health
Services Research, 45(1), 133-151.
Bird, S. T., & Bogart, L. M. (2001). Perceived race-based and socioeconomic status (SES)-based
discrimination in interactions with health care providers. Ethnicity & Disease, 11(3), 554563.

167

Blackwell, E., Mendes de Leon, C. F., & Miller, G. (2006). Applying mixed regression models
to the analysis of repeated-measures data in psychosomatic medicine. Psychosomatic
Medicine, 68, 870-878.
Blustein, J. (2000). Drug coverage and drug purchases by Medicare beneficiaries with
hypertension. Health Affairs, 19(2), 219-230.
Briesacher, B. A., Gurwitz, J. H., & Soumerai, S. B. (2007). Patients at-risk for cost-related
medication nonadherence: a review of literature. Journal of General Internal Medicine,
22(6), 864-871.
Briesacher, B. A., Ross-Degnan, D., Wagner, A. K., Fouayzi, H., Zhang, F., Gurwitz, J. H., &
Soumerai, S. B. (2010). Out-of-pocket burden of health care spending and the adequacy
of the Medicare Part D Low-Income Subsidy. Medical Care, 48(6), 503-509.
Bronstein, L. R. (2003). A model for interdisciplinary collaboration. Social Work, 3(1), 2003.
Butler, R. N. (1985). Why survive? Being old in America. New York: Harper and Row.
Cahill, J. A. (2010). Medicare Part D coverage gap discount program beginning in 2011. Journal
of Managed Care Pharmacy, 16(5), 367-368.
Calasanti, T. M., & Slevin, K. F. (2001). Gender, social inequities, and aging. Walnut Creek,
CA: AltaMira Press.
Cheeseman Day, J., Janus, A., & Davis, J. (2005). Computer and internet use in the United
States: 2003. Washington, D.C.: U. S. Census Bureau.
Chen, H., Nwangwa, A., Aparasu, R., Essien, E., Sun, S., & Lee, K. (2008). The impact of
Medicare part D on psychotropic utilization and financial burden of community-based
seniors. Psychiatric Services, 59(10), 1191-1197.
Clark, D. O. (1997). U.S. trends in disability and institutionalization among older Blacks and
Whites. American Journal of Public Health, 87, 438-440.
Clark, D. O., & Maddox, G. L. (1992). Racial and social correlates of age-related changes in
functioning. Journal of Gerontology: Social Sciences, 47, S222-S232.
Cole, J. R., & Cole, S. (1973). Social stratification in science. Chicago, IL: University of
Chicago Press.
Collins, C. A., Estes, C. L., & Bradsher, J. E. (2001). Inequality and aging: The creation of
dependency. In C. L. Estes and Associates (Ed.), Social policy & aging: A critical
perspective (pp. 137-163). Thousand Oaks, CA: Sage Publications.
Cooper-Patrick, L., Gallo, J. J., Gonzales, J. J., Vu, H. T., Powe, N. R., Nelson, C., & Ford, D. E.
(1999). Race, gender, and partnership in the patient-physician relationship. Journal of the
Amerian Medical Association, 282(6), 583-589.

168

Cox, E. R., Jerigan, C., Coons, S. J., & Draugalis, J. R. (2001). Medicare beneficiaries'
management of capped prescription drug benefits. Medical Care, 39(3), 296-301.
Cronk, A., Humphries, T. L., Delate, T., Clark, D., & Morris, B. (2008). Medication strategies
used by Medicare beneficiaries who reach the Part D standard drug benefit threshold.
American Journal of Health System Pharmacy, 65(11), 1062-1070.
Crystal, S. (1986). Measuring income and inequality among the elderly. The Gerontologist, 26,
56-59.
Crystal, S. (2008, November). Medicare policy challenges for a new administration:
Prescription drugs, privatization, costs and Medicare's future. Paper presented at the 61st
Annual Scientific Meeting of The Gerontological Society of America, Washington, D.C.
Crystal, S., & Shea, D. (1990a). Cumulative advantage, cumulative disadvantage, and inequality
among elderly people. The Gerontologist, 30, 437-443.
Crystal, S., & Shea, D. (1990b). The economic well-being of the elderly. Review of Income and
Wealth, 36(3), 227-247.
Cunningham, P. J. (2005). Medicaid cost containment and access to prescription drugs. Health
Affairs, 24(3), 780-789.
Cutler, S., Hendricks, J., & Guyer, A. (2003). Age differences in home computer availability and
use. The Journals of Gerontology, 58B, S271-S280.
Dannefer, D. (1987). Aging as interacohort differentiation: Accentuation, the Matthew Effect,
and the life course. Sociological Forum, 2(2), 211-229.
Dannefer, D. (2003). Cumulative advantage/disadvantage and the life course: Cross-fertilizing
age and social science theory. Journal of Gerontology: Social Sciences, 58, S327-S337.
Davidoff, A. J., Stuart, B., Shaffer, T., Shoemaker, J. S., Kim, M., & Zacker, C. (2010). Lessons
learned: Who didn't enroll in Medicare drug coverage in 2006, and why? Health Affairs,
29(6), 1255-1263.
Doescher, M. P., Saver, B. G., Franks, P., & Fiscella, K. (2000). Racial and ethnic disparities in
perceptions of physician style and trust. Archives of Family Medicine, 9, 1156-1163.
Donohue, J. M., Huskamp, H. A., & Zuvekas, S. H. (2009). Dual eligibles with mental disorders
and Medicare Part D: How are they faring? Health Affairs, 28(3), 746-759.
Duru, O. K., Mangione, C. M., Hsu, J., Steers, W. N., Quiter, E., Turk, N., . . . Tseng, C. (2010).
Generic-only drug coerage in the Medicare Part D gap and effect on medication costcutting behaviors for patients with diabetes mellitus: The translating research into action
for diabetes study. Journal of the American Geriatrics Society, 58(5), 822-828.

169

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. Journal of General Internal Medicine, 25(6), 568-574.
Families USA. (2000). Cost overdose: Growth in drug spending for the elderly, 1992-2010.
Retrieved from http://www.familiesusa.org/assets/pdfs/drugod852b.pdf
Farkas, G. (2003). Human capital and the long-term effects of education on late-life inequality.
In S. Crystal, D. Shea & K. W. Schaie (Eds.), Annual review of gerontology and
geriatrics (Vol. 22, pp. 138-154). New York, NY: Springer Publishing Company.
Farley, J. F., & Dusetzina, S. B. (2010). Medicaid prescriptoini drug utilization and expenditures
following Part D. Journal of Health Care for the Poor and Underserved, 21, 715-728.
Federman, A. D., Safran, D. G., Keyhani, S., Cole, H., Halm, E. A., & Siu, A. L. (2009).
Awareness of pharmaceutical cost-assistance programs among inner-city seniors. The
American Journal of Geriatric Pharmacotherapy, 7(2), 117-129.
Ferraro, K. F., & Kelley-Moore, J. A. (2003). Cumulative disadvantage and health: Long-term
consequences of obesity. American Sociological Review, 68(5), 707-729.
Fillenbaum, G. G., Hanlon, J. T., Corder, E. H., Brock, D., & et al. (1993). Prescription and
nonprescription drug use among black and white community-residing elderly. American
Journal of Public Health, 83(11), 1577.
Finkle, T., Hartmann, H., & Lee, S. (2007). The economic security of older women and men in
the United States. Institute for Women's Policy Research: Briefing Paper. Retrieved
from http://www.iwpr.org
Frankenfield, D. L., Wei, I. I., & Anderson, K. K. (2010). Prescription medication cost-related
non-adherence among Medicare CAHPS respondents: Disparity by Hispanic ethnicity.
Journal of Health Care for the Poor and Underserved, 21(2), 518-543.
Gelfand, D. E. (2006). The aging network: Programs and services. New York, NY: Springer
Publishing Company.
Gellad, W. F., Haas, J. S., & Safran, D. G. (2007). Race/ethnicity and nonadherence to
prescription medications among seniors: Results of a national study. Journal of General
Internal Medicine, 22(11), 1572-1578.
Gellad, W. F., Huskamp, H. A., Phillips, K. A., & Haas, J. S. (2006). How the new Medicare
drug benefit could affect vulnerable populations. Health Affairs, 25(1), 248-255.
Gibson, M. J., & Brangan, N. (1998). Out-of-pocket spending on health care by women age 65
and over in fee-for-service Medicare: 1998 projections. Out-of-Pocket Costs and
Affordability of Medicare. Retrieved from
http://www.aarp.org/research/medicare/outofpocket/aresearch-import-677-FS72.html

170

Gibson, T. B., Ozimkowski, R. J., & Goetzel, R. Z. (2005). The effects of prescription drug cost
sharing: A review of the evidence. American Journal of Managed Care, 11(11), 730740.
Goedken, A. M., Urmie, J. M., Farris, K. B., & Doucette, W. R. (2010). Impact of cost sharing
on prescription drugs used by Medicare beneficiaries. Research in Social &
Administrative Pharmacy, 6, 100-109.
Gonyea, J. G., & Hooyman, N. R. (2005). Reducing poverty among older women: Social
Security reform and gender equity. Families in Society, 86(3), 338-346.
Goulding, M. R. (2005). Trends in prescribed medicine use and spending by older Americans,
1992-2001. Trends in Health and Aging. Retrieved from
http://www.cdc.gov/nchs/data/ahcd/agingtrends/05medicine.pdf
Gregory, K., Wells, K. B., & Leake, B. (1987). Medical students' expectations for encounters
with minority and nonminority patients. Journal of the National Medical Association,
79(4), 137-144.
Guyer, J., & Schneider, S. (2004). Implications of the new Medicare law for dual eligibles: 10
key questions and answers. The Kaiser Commission on Medicaid and the Uninsured.
Retrieved from http://www.kff.org/medicaid/4160.cfm
Hall, J. P., Kurth, N. K., & Moore, J. M. (2007). Transition to Medicare Part D: An early
snapshot of barriers experienced by younger dual eligibles iwth disabilities. American
Journal of Managed Care, 13, 14-18.
Hanoch, Y., Rice, T., Cummings, J., & Wood, S. (2009). How much choice is too much? The
case of the Medicare prescription drug benefit. Health Services Research, 44(4), 11571168.
Hatch, S. L. (2005). Conceptualizing and identifying cumulative adversity and protective
resources: Implications for understanding health inequalities. The Journals of
Gerontology, 60B, 130-135.
Heisler, M., Langa, K. M., Eby, E. L., Fendrick, A. M., Kabeto, M. U., & Piette, J. D. (2004).
The health effects of restricting prescription medication use because of cost. Medical
Care, 42(7), 626-634.
Heisler, M., Wagner, T. H., & Piette, J. D. (2005). Patient strategies to cope with high
prescription medication costs: Who is cutting back on necessities, increasing debt, or
underusing medications? Journal of Behavioral Medicine, 28(1), 43-51.
Heiss, F., McFadden, D., & Winter, J. (2006). Who failed to enroll in Medicare Part D and why?
Early results. Health Affairs 25, w344-w354.

171

Herszenhorn, D. M., & Pear, R. (2011, January 19). House votes for repeal of health law in
symbolic act. The New York Times. Retrieved from
http://www.nytimes.com/2011/01/20/health/policy/20cong.html
Hoadley, J., Hargrave, E., Cubanski, J., & Neuman, T. (2006). The Medicare drug benefit: An
in-depth examination of formularies and other features of Medicare drug plans. Retrieved
from http://www.kff.org/medicare/upload/7489.pdf
Hoadley, J., Hargrave, E., Cubanski, J., & Neuman, T. (2008). The Medicare Part D coverage
gap: Costs and consequences in 2007. The Henry J. Kaiser Family Foundation Retrieved
from http://www.kff.org/medicare/upload/7811.pdf
Hoadley, J., Summer, L., Hargrave, E., Cubanski, J., & Neuman, T. (2010). Medicare Part D
2010 data spotlight. Medicare prescription drug plans in 2010 and key changes over five
years: Summary of findings. Retrieved from http://www.kff.org/medicare/8096.cfm
Hopkins, S., & Kidd, M. P. (1996). The determinants of demand for private health insurance
under Medicare. Applied Economics, 28, 1623-1632.
HRS. (2011). Health and retirement study: A longitudinal study of health, retirment, and aging.
Retrieved from http://hrsonline.isr.umich.edu/
Hsu, J., Fung, V., Price, M., Huang, J., Brand, R., Hui, R., . . . Newhouse, J. P. (2008). Medicare
beneficiaries' knowledge of Part D prescription drug program benefits and responses to
drug costs. Journal of the American Medical Association, 299(16), 1929-1936.
Huskamp, H. A., Stevenson, D. G., Donohue, J. M., Newhouse, J. P., & Keating, N. L. (2007).
Coverage and prior authorization of psychotropic drugs under Medicare Part D.
Psychiatric Services, 58(3), 308-310.
Huskamp, H. A., West, J. C., Rae, D. S., Rubio-Stipec, M., Regier, D. A., & Frank, R. D. (2009).
Part D and dually eligible patients with mental illness: Medication access problems and
use of intensive services. Psychiatric Services, 60(9), 1169-1174.
Institute of Medicine of the National Academies. (2003). Unequal treatment: Confronting racial
and ethnic disparities in health care. Washington, D.C.: National Academies Press.
Jackson, E. A., & Axelsen, K. J. (2008). Medicare Part D formulary coverage since program
inception: Are beneficiaries choosing wisely? The American Journal of Managed Care,
14(11), SP29-SP35.
Johnson, R. L., Roter, D., Powe, N. R., & Cooper, L. A. (2004). Patient race/ethnicity and
quality of patient-physician communication during medical visits. American Journal of
Public Health, 94(12), 2084-2090.
Joyce, G. F., Goldman, D. P., Karaca-Mandic, P., & Zheng, Y. (2007). Pharmacy benefit caps
and the chronically ill. Health Affairs, 26(5), 1333-1343.

172

Joyce, G. F., Goldman, D. P., Vogt, W. B., Sun, E., & Jena, A. B. (2009). Medicare Part D after
2 years. The American Journal of Managed Care, 15(8), 536-544.
Juster, T., & Suzman, R. (1995). Overview of the Health and Retirement Study. Journal of
Human Resources, 30, S7-56.
Kail, B. L., Quadagno, J., & Keene, J. R. (2009). The political economy perspective of aging. In
V. L. Bengtson, D. Gans, N. M. Putney & M. Silverstein (Eds.), Handbook of theories of
aging (pp. 555-571). New York, NY: Springer Publishing Company, Inc.
Kaiser Family Foundation. (2006). Voices of beneficiaries: Early experiences with the
Medicare drug benefit. Retrieved from http://www.kff.org/medicare/upload/7504.pdf
Kaiser Family Foundation. (2008). Fact sheet: Prescription drug trends. Retrieved from
http://www.kff.org/rxdrugs/upload/3057_07.pdf
Kaiser Family Foundation. (2009a). The Medicare prescription drug benefit. Retrieved from
http://www.kff.org/medicare/upload/7044-10.pdf
Kaiser Family Foundation. (2009b). Low income assistance under the Medicare drug benefit.
Retrieved from http://www.kff.org/medicare/upload/7327-05.pdf
Kaiser Family Foundation. (2009c). Affordable Health Care for America Act (H.R. 3962).
Retrieved from http://www.kff.org/healthreform/upload/housebill_final.pdf
Kaiser Family Foundation. (2010a). Focus on health reform: Summary of new health reform
law. Retrieved from http://kff.org/healthreform/upload/8061.pdf
Kaiser Family Foundation. (2010b). U.S. Health Care Costs. Retrieved from
http://www.kaiseredu.org/Issue-Modules/US-Health-Care-Costs/Background-Brief.aspx
Kaiser Family Foundation. (2011a). Understanding the effects of the Medicare Part D coverage
gap in 2008 and 2009: Costs and consequences prior to improvements in coverage
established by the 2010 Health Reform Law. Retrieved from
http://www.kff.org/medicare/8221.cfm
Kaiser Family Foundation. (2011b). Medicare prescription drug plan information, by state.
Retrieved from http://www.kff.org/medicare/7426.cfm
Kaiser Health News. (2011, January 6). President Obama promises health law repeal veto.
Retrieved from http://www.kaiserhealthnews.org/Stories/2011/January/07/White-HousePromises-health-law-repeal-veto.aspx
Kanavos, P., & Gemmill-Toyama, M. (2010). Prescription drug coverage among elderly and
disabled Americans: can Medicare-D reduce inequities in access? International Journal
of Health Care Finance and Economics, 10, 203-218.

173

Kelley-Moore, J. A., & Ferraro, K. F. (2004a). The Black/White disability gap: Persistent
inequality in later life? The Journals of Gerontology, 59B(1), S34-S43.
Kelley-Moore, J. A., & Ferraro, K. F. (2004b). The black/white disability gap: Persistent
inequality in later life? The Journals of Gerontology, 59B(1), S34.
Klein, D., Turvey, C., & Wallace, R. (2004). Elders who delay medication because of cost:
Health insurance, demographic, health, and financial correlates. The Gerontologist, 44(6),
779-787.
Lau, D. T., Briesacher, B., Mercaldo, N. D., Halpern, L., Osterberg, E. C., Jarzebowski, M., . . .
Mazor, K. (2008). Older patients' perceptions of medication importance and worth: An
exploratory study. Drugs Aging, 25(12), 1061-1075.
Levy, H., & Weir, D. (2009). Take-up of Medicare Part D: Results from the Health and
Retirement Study. NBR Working Paper. Retrieved from
http://www.nber.org/papers/w14692
Lichtenberg, F. R., & Sun, S. X. (2007). The impact of Medicare Part D on prescription drug use
by the elderly. Health Affairs, 26(6), 1735-1744.
Lillard, L., Rogowski, J., & Kington, R. (1997). Long-term determinants of patterns of health
insurance coverage in the Medicare population. The Gerontologist, 37(3), 314-323.
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 Medicare Part D. Journal of the American Medical
Association, 299(16), 1922-1928.
Manton, K. G., & Stallard, E. (1997). Health and disability differences among racial and ethnic
groups. In L. G. Martin & B. J. Soldo (Eds.), Racial and ethnic differences in the health
of older Americans (pp. 43-105). Washington, D.C.: National Academy Press.
Marquis, M. S., & Long, S. H. (1995). Worker demand for helath insurance in the non-group
market. Journal of Health Economics, 14, 47-63.
Mayes, R. (2005). Medicare and America's healthcare system in transition: From the death of
managed care to the Medicare Modernization Act of 2003 and beyond. Journal of Health
Law, 38(3), 391-415.
Medicare Payment Advisory Commission. (2000). Report to the Congress: Selected Medicare
issues. Retrieved from http://permanent.access.gpo.gov/lps4816/Jun00+Entire+report.pdf
Medicare Payment Advisory Commission. (2006). Report to the Congress: increasing the value
of Medicare. (June, 2006). Retrieved from
http://www.medpac.gov/documents/jun06_entirereport.pdf
Merton, R. K. (1968a). The Matthew Effect in science. Science, 159(3810), 56-63.

174

Merton, R. K. (1968b). The sociology of science: Theoretical and empirical investigations.
Chicago: The University of Chicago Press.
Millett, C., Everett, C. J., Matheson, E. M., Bindman, A. B., & Mainous, A. G. (2010). Impact of
Medicare Part D on seniors' out-of-pocket expenditures on medications. Arch Intern Med,
170(15), 1325-1330.
Minkler, M., & Stone, R. (1985). The feminization of poverty and older women. The
Gerontologist, 25(4), 351-357.
Mizrahi, T., & Abramson, J. S. (2000). Collaboration between social workers and physicians:
Perspectives on a shared case. Social Work in Health Care, 31(3), 1-24.
Mojtabai, R., & Olfson, M. (2003). Medication costs, adherence, and health outcomes among
Medicare beneficiaries. Health Affairs, 22(4), 220-229.
Moody, H. R. (2009). Aging: Concepts and controversies. Los Angeles, CA: Pine Forge Press.
Morris, D. C. (1996). Ownership, use, and non-use of personal computers by older adults.
Journal of Educational Technology Systems, 25, 5-12.
Murtagh, K. N., & Hubert, H. B. (2004). Gender differences in physical disability among an
elderly cohort. American Journal of Public Health, 94(8), 1406-1411.
Nederhof, A. J. (2006). Methods of coping with social desirability bias: A review. European
Journal of Social Psychology, 15(3), 263-280.
Nelson, A. A., Jr., Reeder, C. E., & Dickson, W. M. (1984). The Effect of a Medicaid Drug
Copayment Program on the Utilization and Cost of Prescription Services. Medical Care,
22(8), 724-736.
Neugarten, B. (1982). Age or need? Public policies for older people. Beverly Hills: Sage
Publications.
Neuman, P., & Cubanski, J. (2009). Medicare Part D update - Lessons learned and unfinished
business. New England Journal of Medicine, 361(4), 406-414.
Neuman, P., Strollo, M. K., Guterman, S., Rogers, W. H., Li, A., Rodday, A. M., & Safran, D. G.
(2007). Medicare prescription drug benefit progress report: Findings from a 2006
national survey of seniors. Health Affairs - Web Exclusive, W630-W643.
O'Rand, A. M. (2003). Cumulative advantage and gerontological theory. In S. Crystal, D. Shea
& K. W. Schaie (Eds.), Annual review of gerontology and geriatrics (Vol. 22, pp. 14-30).
New York, NY: Springer Publishing Company, Inc.
O'Rand, A. M., & Hamil-Luker, J. (2005). Processes of cumulative adversity: Childhood
disadvantage and increased risk of heart attach across the life course. Journal of
Gerontology: Social Sciences, 60, 117-124.

175

Ofstedal, M. F. (2009, June). The Health and Retirement Study: An overview. Paper presented at
the 2009 Summer Institute in Survey Research, Health and Retirement Study Workshop,
Ann Arbor, MI.
Olson, L. K. (1994). Women and Social Security: A progressive approach. Journal of Aging &
Social Policy, 6(1), 43-56.
Ott, R. L., & Lngnecker, M. (2001). An introduction to statistical methods and data analysis.
Pacific Grove, CA: Duxbury Thompson Learning.
Pedan, A., Lu, J., & Varasten, L. T. (2009). Assessment of drug consumption patterns for
Medicare Part D patients. The American Journal of Managed Care, 15(5), 323-327.
Pierre-Jacques, M., Safran, S. G., Zhang, F., Ross-Degnan, D., Adams, A. S., Gurwitz, J., . . .
Soumerai, S. B. (2008). Reliability of new measures of cost-related medication
nonadherence. Medical Care, 46(4), 444-448.
Piette, J. D., Heisler, M., Krein, S., & Kerr, E. A. (2005). The role of patient-physician trust in
moderating medication nonadherence due to cost pressures. Archives of Internal
Medicine, 165, 1749-1755.
Piette, J. D., Heisler, M., & Wagner, T. H. (2004a). Cost-related medication underuse among
chronically ill adults: The treatments people forgo, how often, and who is at risk.
American Journal of Public Health, 94(10), 1782-1787.
Piette, J. D., Heisler, M., & Wagner, T. H. (2004b). Cost-related medication underuse: Do
patients with chronic illnesses tell their doctors? Archives of Internal Medicine, 164,
1749-1755.
Piette, J. D., Heisler, M., & Wagner, T. H. (2004c). Problems paying out-of-pocket medication
costs among older adults with diabetes. Diabetes Care, 27(2), 384-391.
Piette, J. D., Heisler, M., & Wagner, T. H. (2006). Medication characteristics beyond cost alone
influence decisions to underuse pharmacotherapy in response to financial pressures.
Journal of Clinical Epidemeiology, 56(7), 739-746.
Pol, L. G., Mueller, K. J., & Adidam, P. T. (2002). Health insurance in the near elderly
population. Population Research and Policy Review, 19(2), 97-112.
Polinski, J. M., Mohr, P. E., & Johnson, L. (2009). Impact of Medicare Part D on access to and
cost sharing for specialty biologic medications for beneficiaries with rheumatoid arthritis.
Arthritis & Rheumatism, 61(6), 745-754.
Pourat, N., Rice, T., Kominski, G., & Snyder, R. E. (2000). Socioeconomic differences in
Medicare supplemental coverage. Health Affairs, 19(5), 186-196.
Quinn, P. (1996). Identifying gendered outcomes of gender-neutral policies. AFFILIA, 11(2),
195-206.

176

Rabe-Hesketh, S., & Skrondal, A. (2008). Multilevel and longitudinal modeling using Stata.
College Station, TX: Stata Press.
Raebel, M. A., Delate, T., Ellis, E. L., & Bayliss, E. A. (2008). Effects of reaching drug benefit
threshold on Medicare members' healthcare utilization during the first year of Medicare
Part D. Medical Care, 46(10), 1116-1122.
RAND Center for the Study of Aging. (2010). RAND HRS Data Documentation, Version J.
Retrieved from http://hrsonline.isr.umich.edu/index.php?p=showcbk
Rao, J. N. K., & Scott, A. J. (1981). The analysis of categorical data from complex sample
surveys: Chi-squared tests for goodness of fit and independence in two-way tables.
Journal of the American Statistical Association, 76(374), 221-230.
Reeder, C. E., & Nelson, A. A. (1985). The differential impact of copayment on drug use in a
Medicaid population. Inquiry, 22(4), 396-403.
Rogowski, J., Lillard, L. A., & Kington, R. (1997). The financial burden of prescription drug use
among elderly persons. The Gerontologist, 37(4), 475-482.
Rosenbaum, S., & Teitelbaum, J. (2005). Implementing the Medicare Part D prescription drug
benefit: Implications for public health policy and practice. Public Health Reports,
120(4), 467-469.
Ross, C. E., & Wu, C. L. (1996). Education, age, and the cumulative advantage in health.
Journal of Health and Social Behavior, 37, 104-120.
Roth, M. T., Esserman, D. A., Ivey, J. L., & Weinberger, M. (2009). Racial disparities in the
quality of medication use in older adults: Baseline finding from a longitudinal study.
Journal of General Internal Medicine, 25(3), 228-234.
Roth, M. T., & Ivey, J. L. (2005). Self-reported medication use in community-residing older
adults: A pilot study. The American Journal of Geriatric Pharmacotherapy, 3(3), 196204.
Rudolph, N. V., & Montgomery, M. A. (2010). Low-income Medicare beneficiaries and their
experiences with the Part D prescription drug benefit. Inquiry, 41(2), 162-172.
Safran, D. G., Neuman, P., Schoen, C., Kitchman, M. S., Cooper, B., Wilson, I. B., . . . Rogers,
W. (2005). Prescription drug coverage and seniors: Findings from a 2003 national
survey. Health Affairs, 24, 152-167.
Safran, D. G., Neuman, P., Schoen, C., Montgomery, J. E., & et al. (2003). Prescription drug
coverage and seniors: How well are states closing the gap? Health Affairs, W253-W268.
Safran, D. G., Strollo, M. K., Guterman, S., Li, G., Rogers, W. H., & Neuman, P. (2009).
Prescription coverage, use and spending before and after Part D implementation: A
national longitudinal panel study. Journal of General Internal Medicine, 25(1), 10-17.

177

Sambamoorthi, U., Shea, D., & Crystal, S. (2003). Total and out-of-pocket expenditures for
prescription drugs among older persons. The Gerontologist, 43(3), 345-359.
Saver, B. G., Doescher, M. P., Jackson, J. E., & Fishman, P. (2004). Seniors with chronic health
conditions and prescription drugs: Benefits, wealth, and health. Value in Health, 7(2),
133-143.
Schmittdiel, J. A., Ettner, S. L., Fung, V., Huang, J., Turk, N., Quiter, E., . . . Mangione, C. M.
(2009). Medicare Part D coverage gap and diabetes beneficiaries. The American Journal
of Managed Care, 15(3), 189-193.
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 Affairs, 28(2), w305-w316.
Schoenbaum, M., & Waidmann, T. (1997). Race, socioeconomic status, and health: Accounting
for race differences in health. The Journals of Gerontology, 52B(Special Issue), 61-73.
doi: 10.1093/geronb/52B.Special_Issue.61
Schoenborn, C. A., & Heyman, K. M. (2009). Health characteristics of adults aged 55 years and
over: United States, 2004-2007. National Health Statistics Report. Retrieved from
http://www.cdc.gov/nchs/data/nhsr/nhsr016.pdf
Sharkey, J. R., Ory, M. G., & Browne, B. A. (2005). Determinants of self-management strategies
to reduce out-of-pocket prescription medication expense in homebound older people.
Journal of the American Geriatrics Society, 53, 666-674.
Shi, L. (1999). Experience of primary care by racial and ethnic groups in the United States.
Medical Care, 37(10), 1068-1077.
Shrank, W. H., Patrick, A. R., Pedan, A., Polinski, J. M., Varasteh, L., Levin, R., . . .
Schneeweiss, S. (2008). The effect of transitioning to Medicare Part D drug coverage in
seniros dually eligible for Medicare and Medicaid. Journal of the American Geriatrics
Society, 56, 2304 - 2310.
Skarupski, K. A., Mendes de Leon, C. F., Barnes, L. L., & Evans, D. A. (2009). Medicare Part D
enrollment in a biracial community-based population of older adults. The Gerontologist,
49(6), 828-838.
Smedley, B. D., Stith, A. Y., & Nelson, A. R. (Eds.). (2003). Unequal treatment: Confronting
racial and ethnic disparities in health care. Washington, D.C.: The National Academies
Press.
Soumerai, S. B., Avorn, J., Ross-Degnan, D., & Gortmaker, S. (1987). Payment restrictions for
prescription drugs under Medicaid: Effects on therapy, cost and equity. New England
Journal of Medicine, 317(9), 550-556.

178

Soumerai, S. B., McLaughlin, T. J., Ross-Degnan, D., Casteris, C. S., & Bollini, P. (1994).
Effects of limiting Medicaid drug-reimbursement benefits on the use of psychotropic
agents and acute mental health services by patients with schizophrenia. The New England
Journal of Medicine, 331(10), 650-655.
Soumerai, S. B., Pierre-Jacques, M., Zhang, F., D. Ross-Degnan, Adams, A. S., Gurwitz, J., . . .
Safran, D. G. (2006). Cost-related medication nonadherence among elderly and disabled
Medicare beneficiaries: A national survey 1 year before the Medicare drug benefit.
Archives of Internal Medicine, 166(17), 1829-1835. doi: 10.1001/archinte.166.17.1829
Soumerai, S. B., & Ross-Degnan, D. (1999). Inadequate prescription-drug coverage for Medicare
enrollees--a call to action. The New England Journal of Medicine, 340(9), 722-729.
Soumerai, S. B., Ross-Degnan, D., Avorn, J., McLaughlin, T. J., & Choodnovskiv, I. (1991).
Effects of Medicaid drug-payment limits on admission to hospitals and nursing homes.
The New England Journal of Medicine, 325(15), 1072-1077.
Steinman, M. A., Sands, L. P., & Covinsky, K. E. (2001). Self-restriction of medications due to
cost in seniors without prescriptions drug coverage: A national survey. Journal of
General Internal Medicine, 16, 793-799.
Stuart, B. (2008). Where are the Medicare Part D claims data? Annals of Internal Medicine,
148(3), 239-240.
Stuart, B., Simoni-Wastila, L., & Chauncey, D. (2005). Assessing the impact of coverage gaps in
the Medicare Part D drug benefit. Health Affairs, 24, 167-180.
Sun, S. X., & Lee, K. Y. (2007). The Medicare Part D doughnut hole: Effect on pharmacy
utilization. Managed Care Interface, 20(9), 51-59.
The Federal Interagency Forum on Aging-Related Statistics. (2008). Older Americans 2008:
Key indicators of well-being Retrieved from
http://www.agingstats.gov/agingstatsdotnet/main_site/data/2008_Documents/OA_2008.p
df
Traynor, K. (2010). Health reform law offers relief in Medicare Part D coverage gap. American
Journal of Health System Pharmacy, 67(13), 1049.
Tseng, C. W., Brook, R. H., Keeler, E., Steers, W. N., & Mangione, C. M. (2004). Cost-lowering
strategies used by Medicare beneficiaries who exceed benefit caps and have a gap in drug
coverage. Journal of the American Medical Association, 292(8), 952-960.
U.S. Department of Health and Human Services. (2011, April 8). HHS announces plan to reduce
health disparities. Retrieved from
http://www.hhs.gov/news/press/2011pres/04/20110408a.html

179

Van Dongen, H. P. A., Olofsen, E., Dinges, D. F., & Maislin, G. (2004). Mixed-model regression
analysis and ealing with interindividual differences. Methods of Enzymology, 384, 139171.
van Ryn, M., & Burke, J. (2000). The effect of patient race and socio-economic status on
physicians' perceptions of patients. Social Science & Medicine, 50, 813-828.
Vincent, G. K., & Velkoff, V. A. (2010). The next four decades: The older population in the
United States 2010 to 2050. Current Population Reports. Retrieved from
http://www.aoa.gov/AoARoot/Aging_Statistics/future_growth/DOCS/p25-1138.pdf
Walker, B., Mays, V. M., & Warren, R. (2004). The changing landscape for the elimination of
racial/ethnic health status disparities. Journal of Health Care for the Poor and
Underserved, 15(4), 506-521.
Wang, P. S., Patrick, A. R., Dormuth, C., Maclure, M., Avorn, J., Canning, C. F., &
Schneeweiss, S. (2010). Impact of drug cost sharing on service use and adverse clinical
outcomes in elderly receiving antidepressants. The Journal of Mental Health Policy and
Economics, 13, 37-44.
Wei, W., Akincigil, A., Crystal, S., & Sambamoorthi, U. (2006). Gender differences in out-ofpocket prescription drug expenditures among the elderly. Research on Aging, 28(4), 427453.
West, J. C., Wilk, J. E., Muszynski, I. L., Rae, D. S., Rubio-Stipec, M., Alter, C. L., . . . Regier,
D. A. (2007). Medication access and continuity: The experiences of dual-eligible
psychiatric patients during the first 4 months of the Medicare prescription drug benefit.
The American Journal of Psychiatry, 164(5), 789-796.
Wilk, J. E., West, J. C., Rae, D. S., Rubio-Stipec, M., Chen, J. J., & Regier, D. A. (2008).
Medicare Part D prescription drug benefits and administrative burden in care of dually
eligible psychiatric patients. Psychiatric Services, 59(1), 34-39.
Williams, D. R. (2002). Racial/ethnic variations in women's health: The social embeddedness of
health. American Journal of Public Health, 92(4), 588-597.
Williams, D. R., Gonzalez, H. M., Neighbors, H., Nesse, R., Abelson, J. M., Sweetman, J., &
Jackson, J. S. (2007). Prevalence and distribution of major depressive disorder in African
American, Caribbean Blacks, and Non-Hispanic Whites. Archives of General Psychiatry,
64, 305-315.
Wilson, A. E., Shuey, K. M., & Elder, G. H. (2007). Cumulative advantage processes as
mechanisms of inequality in life course health. American Journal of Sociology, 112,
1886-1924.
Wilson, I. B., Rogers, W. H., Chang, H., & Safran, D. G. (2004). Cost-related skipping of
medications and other treatments among Medicre benficiaries between 1998 and 2000:
Results of a national study. Journal of General Internal Medicine, 20, 715-720.
180

Wilson, I. B., Schoen, C., Neuman, P., Strollo, M. K., Rogers, W. H., Chang, H., & Safran, D. G.
(2007). Physician-patient communication about prescription medication nonadherence:
A 50-state study of America's seniors. Journal of General Internal Medicine, 22, 6-12.
Woodward, A. T., Freddolino, P. P., Blaschke-Thompson, C. M., Wishart, D. J., Bakk, L.,
Kobayashi, R., & Tupper, C. (2011). Technology and Aging Project: Training outcomes
and efficacy from a randomized field trial. Ageing International, 36, 46-65.
Wright, J. D., & Devine, J. A. (1994). Poverty among the elderly. Journal of Long-Term Home
Health Care, 13(1), 5-16.
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.
Annals of Internal Medicine, 148(3), 169-177.
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 Affairs, 28(2), w-317w325.
Zivin, K., Madden, J. M., Graves, A. J., Zhang, F., & Soumerai, S. B. (2009). Cost-related
medication nonadherence among beneficiaires with depression following Medicare Part
D. American Journal of Geriatric Psychiatry, 17(12), 1068-1076.
Zivin, K., Ratliff, S., Heisler, M. M., Langa, K. M., & Piette, J. D. (2010). Factors influencing
cost-related nonadhherence to medication in older adults: A conceptually-based
approach. Value in Health, 13(4), 338-345.
Zuvekas, S. H., & Taliaferro, G. S. (2003). Pathways to access: Health insurance, the health care
delivery system, and racial/ethnic disparities, 1996-1999. Health Affairs, 22, 139-153.

181