The Impact of Medicare Part D on
Medication Treatment of Hypertension
Yuting Zhang, Julie M. Donohue, Judith R. Lave, and
Walid F. Gellad
Objective. To evaluate Medicare Part D’s impact on use of antihypertensive med-
ications among seniors with hypertension.
Data Sources. Medicare-Advantage plan pharmacy data from January 1, 2004 to
December 12, 2007 from three groups who before enrolling in Part D had noor limited
drug benefits, and a comparison group with stable employer-based coverage.
Study Design. Pre–post intervention with a comparison group design was used to
study likelihood of use, daily counts, and substitutions between angiotensin-converting
enzyme inhibitors and angiotensin-II receptor blockers (ARBs).
Principal Findings. Antihypertensive use increased most among those without prior
drug coverage: likelihood of use increased (odds ratio51.40, 95 percent confidence
interval [CI] 1.25–1.56), and daily counts increased 0.29 (95 percent CI 0.24–0.33).
Proportion using ARBs increased from 40 to 46 percent.
Conclusions. Part D was associated with increased antihypertensive use and use of
ARBs over less expensive alternatives.
Key Words. Medicare, pharmacy, access/demand/utilization, hypertension
Over half of those aged 60–69 years and three-quarters of those aged 70 and
above have hypertension (Agency for Health Care Research and Quality
2007). While many antihypertensive medications effectively treat hyperten-
sion, o80 percent of Americans aged 60 years or older with hypertension
receive treatment and only 40 percent of those have their blood pressure
appropriately controlled (Egan, Zhao, and Axon 2010). Previous studies have
shown that restrictions on drug coverage reduce the use of antihypertensive
medications and increase the likelihood of emergency room and inpatient
visits (Hsu et al. 2006). Furthermore, results from clinical trials have shown
antihypertensive therapy reduced major cardiovascular events and mortality
(SHEP Cooperative Research Group 1991).
rHealth Research and Educational Trust
Health Services Research
The Medicare drug benefit (Part D), which provides outpatient pre-
scription drug coverage forover26 million Medicare beneficiaries, took effect
of medications, thereby increasing appropriate drug use and improving ben-
among patients with hypertension.
Medicare Part D could affect not only the overall use of antihyperten-
sives but also the types of medications used. Previous studies have shown that
when drug coverage improved, individuals increased overall medication use
(Zhang et al. 2009). Increases in out-of-pocket cost were also associated with
substitution of cheaper generic medications for more expensive brand names
(Huskamp et al. 2003). Hence, as drug coverage improves, patients might be
more likely to initiate or switch to more expensive drug subclasses. Two
groups of hypertension treatments that may be substitutes are angiotensin-
converting enzyme (ACEs) inhibitors and angiotensin-II receptor blockers
(ARBs). ACEs and ARBs have no significantly different blood pressure low-
ering or renal-protective effects (Kunz et al. 2008; Matchar et al. 2008), yet
they differ substantially in cost as ARBs are all brand name without any
generic substitutes, whereas ACEs exist widely in generic form at one-eighth
of the price (drugstore.com prices). While there are differences in the side
effect profilesofthetwosubclasses, examining the substitution betweenACEs
and ARBs may be the best opportunity among hypertension medications to
studywhetherinsuranceaffectsthe useofbranded versusgeneric medications
that are equally effective but very different in price.
In this paper we address the following questions about the effect of Part
these beneficiaries increase use of antihypertensive medications overall? and
(2) Did they increase use of some subclasses of antihypertensive medications
more than others; in particular, did beneficiaries shift to newer, more expen-
further explore whether the shifting was due to switching from ACEs to ARBs
or higher rates of initiating ARBs over ACEs.
University of Pittsburgh, 130 De Soto Street, Crabtree Hall A664, Pittsburgh, PA 15261; e-mail:
firstname.lastname@example.org. Judith R. Lave, Ph.D., Julie M. Donohue, Ph.D., are with the Department of
Health Policy and Management, University of Pittsburgh, Pittsburgh, PA. Walid Gellad, M.D.,
M.P.H., is with theVA Pittsburgh, RAND Health, Pittsburgh, PA. WalidGellad, M.D., M.P.H., is
also with the Division of General Medicine, University of Pittsburgh, Pittsburgh, PA.
186 HSR: Health Services Research 46:1, Part I (February 2011)
We used a pre–post intervention with a comparison group design to evaluate
the changes in medication use 2 years before and 2 years after the imple-
mentation of Part D in four groups of Medicare beneficiaries enrolled in
Medicare-Advantage plans sold by a large insurance company in Pennsylva-
nia. The three intervention groups who were automatically enrolled in the
plan’s Part D products in 2006 included those who had (1) no previous drug
coverage (no coverage), (2) poor previous drug coverage (U.S.$150 quarterly
to that offered under Part D (U.S.$350 quarterly cap of plan payment
[U.S.$350 cap]). Pre-Part D, the level of drug coverage in the latter two groups
depended on members’ county of residence (i.e., the insurer only offered
either U.S.$150 cap or U.S.$350 cap in one county). After Part D, the three
intervention groups faced monthly copayments of U.S.$8 (generic) and
gap between U.S.$2,250 (U.S.$2,400 in 2007) and U.S.$5,100 (U.S.$5,451 in
2007) of annual drug spending. Approximately 70 percent of members in the
intervention groups had generic coverage in the donut hole. All beneficiaries
had 95 percent coverage above the catastrophic limit (U.S.$5,451 in 2007).
The comparison group had generous employer-sponsored drug cover-
age that depended solely on whether members’ former employers offered it.
to beneficiaries’ selecting into an intervention or comparison group is small.
The comparison group faced copayments of U.S.$10–20 per monthly pre-
scription and had no coverage gap or catastrophic limits during the entire
study period. Its coverage did not change after Part D was implemented.
Data Source and Population
We obtained enrollment, benefits, and pharmacy claims on a 40 percent
random sample of individuals enrolled in the plan between January 2003 and
least two claims in 2003 with a diagnosis coded for hypertension (ICD-9 401,
402, 403, 404) and were continuously enrolled in the plan between 2004 and
2007, 2 years before and after Part D’s implementation. (We also defined a
subpopulation based on diagnosis and baseline use of antihypertensives. Be-
cause these results were quantitatively similar we only present results on the
Impact of Part D on Antihypertensive Use 187
We examined the proportion of members in each group who ever filled any
antihypertensive medications as well as drugs in each subclass, including b-
blockers, diuretics, ACEs, ARBs, and calcium channel blockers each year
between 2004 and 2007. We also examined average daily counts of any an-
tihypertensive filled each year between 2004 and 2007.
either in each year between 2004 and 2007 in each study group. In addition,
we examined whether changes in use of ARBs relative to ACEs was due to
an ACE but not on ARB in 2004 and 2005 who switched to an ARB in 2006
and 2007; and (2) the proportion of patients not on an ACE or ARB in 2003
who initiated an ACE versus ARB in the pre- and post-Part D periods.
We created an indicator variable for each intervention group relative to the
comparison group. We created a post-Part D indicator variable, which took
the value 1 after January 1, 2006. The ‘‘Part D Effect’’ is measured by the
interaction terms between the post-Part D and the three Prior Coverage in-
dicatorvariables; thiscapturesthe changesinoutcomebeforeand afterPart D
in each intervention group relative to those in the comparison group.
Our pre–post intervention with a comparison group design guards
againstselection bias,which islikely smallasdiscussedpreviously.We further
used propensity score weighting to enhance the comparability between each
intervention group and the comparison group in two steps (Hirano and Im-
bens 2001). First, we estimated the probability of being in each intervention
group relative to the comparison group using three logistic regressions, con-
trolling for zip-code level income and race, residence in an urban area, and
scores were calculated using the Risk Grouper software from DxCG to adjust
prior-year medical diagnoses and spending. The risk scores are similar to the
CMS-HCC weights used to adjust Medicare-Advantage plan payments, with
higher scores indicating worse health status and greater expected future med-
ical spending (Pope et al. 2004).
in general estimating equations (GEEs). This essentially assigned a higher
weight to those individuals in the comparison group with more similarity to
individuals in the intervention group. GEE adjusted for correlations across 4
188 HSR: Health Services Research 46:1, Part I (February 2011)
also used traditional multivariable regression models with adjustments for the
same covariates used in the logistic regressions described above.
Background Characteristics of Study Population
Table 1 shows the baseline characteristics of each group. The comparison
group was younger, although prospective risk scores were similar across the
groups. Members in the U.S.$150-cap group were more likely to live in the
suburbs and in zip-code areas with higher proportions of whites. Members in
the no-coverage group were more likely to have emergency department visits
but had fewer number of outpatient visits per year. Medical spending was
similar across groups.
Likelihood of Use of Antihypertensive Medications
subclass before and after Part D. Before Part D, members in the no-coverage
group were less likely to use any antihypertensive than those in the other three
groups (p-value o.05). The likelihood of use of any antihypertensive in the
comparison group did not change (88.1 percent pre to 89.1 percent post).
Relative to the comparison group, the likelihood of any antihypertensive use
did not change in the U.S.$150-cap and U.S.$350-cap groups, but the propor-
tion of individuals with hypertension in the no-coverage group who used at
least one antihypertensive medication increased from 59.8 to 69.7 percent
(odds ratio [OR]51.40, 95 percent confidence interval [CI] 1.25–1.56).
Among members in the no-coverage group, the increase in the likeli-
hood of antihypertensive use is largest for ARBs (OR51.53, 95 percent CI
1.35–1.75), followed by b-blockers (OR51.44, 95 percent CI 1.30–1.59), and
then by ACE (OR51.34, 95 percent CI 1.20–1.49) and diuretics (OR51.34,
95 percent CI 1.21–1.47). The likelihood of b-blocker use increased slightly in
the U.S.$150-cap and U.S.$350-cap groups as well. Results from multivariate
regressions are quantitatively similar.
Average Daily Counts of Antihypertensive Medications
Table 2 panel B shows the average daily counts of antihypertensive medica-
tions before and after Part D. After adjusting for propensity score weights and
secular trends in the comparison group, the no-coverage group increased the
Impact of Part D on Antihypertensive Use189