University of Southern California Dissertations and Theses

Essays on access to prescription drugs

(USC Thesis Other)

Essays on access to prescription drugs

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Essays on Access to Prescription Drugs

by

Jianhui Xu

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA

In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
PUBLIC POLICY AND MANAGEMENT
August 2021

Copyright 2021 Jianhui Xu

ii

Acknowledgements

I am deeply indebted to my dissertation committee – John Romley, Darius Lakdawalla,
Geoff Joyce, and Erin Trish – for their help and support in this journey. John, my committee
chair, has been a wonderful mentor: always very accessible, and offers helpful advice on
conducting and communicating my research. I benefited greatly from taking courses with and
working as a TA for Darius, and I appreciate all the wisdom and thought-provoking questions. I
have been working with Erin and Geoff since before I started the PhD program. I am
tremendously grateful for their mentorship, support, and encouragement, which helped me get
through the challenges, especially during the pandemic.
During my time at USC, I have also been fortunate to work with and learn from a few
other faculty members. Rebecca Myerson served as an important member on my qualifying
exam committee. Neeraj Sood provided helpful input on several projects. T.J. McCarthy’s
methods courses and the TA experience with him helped me improve as a researcher and teacher.
My RA work with Don Lloyd at USC Roybal Institute led to the decision to pursue a PhD.
My research would have been much less smooth without the Schaeffer Center Data Core.
Patty St. Clair introduced me to Medicare data and answered numerous questions from me.
Laura Gascue has always been patient and helpful with my data and coding questions. Jillian
Wallis ably assisted me with IRB and data reuse applications. I am also grateful to members of
the administrative staff – Renelle Davis and Hanh Nguyen, who made scheduling a breeze.
I would like to thank my friends and colleagues for their support and the wonderful
times: Sean Angst, Yi Chen, Wendy Cheng, Yuchen Ding, Laura Henkhaus, Sushant Joshi, Kate
Jun, Zeewan Lee, Yougeng Lu, Xize Wang, Bo Wen, Richard Xie, Yifan Xu, Bo Zhou, Linna
iii

Zhu, and Yingying Zhu. I would also like to thank my parents and parents-in-law, who have
always been supportive of my education and career. Finally, I would like to thank my wife
Shanxi for her love, support, encouragement, and the haircuts in quarantine.
This dissertation was supported by the Oakley Endowed Fellowship from the USC
Graduate School.

iv

Table of Contents
Acknowledgements ......................................................................................................................... ii
Abstract .......................................................................................................................................... vi
Chapter 1. Introduction ................................................................................................................... 1
Chapter 2. Incorporating Prescription Drug Utilization Information into the Marketplace Risk
Adjustment Model Improves Payment Accuracy and Reduces Adverse Selection Incentives
,
..... 6
2.1 Introduction ........................................................................................................................... 7
2.2 Conceptual Framework ......................................................................................................... 9
2.3 New Contribution ................................................................................................................ 10
2.4 Methods ............................................................................................................................... 11
2.4.1 Data and sample ............................................................................................................ 11
2.4.2 Risk score ..................................................................................................................... 12
2.4.3 Predictive power and payment accuracy ...................................................................... 13
2.4.4 Statistical analysis......................................................................................................... 13
2.5 Study Results ....................................................................................................................... 16
2.5.1 Payment accuracy under the 2017 model ..................................................................... 16
2.5.2 Impact of 2018 changes incorporating prescription drug utilization on model predictive
power and payment accuracy ................................................................................................ 18
2.5.3 A further illustration of the mechanisms ...................................................................... 20
2.5.4 Partial-year enrollees .................................................................................................... 21
2.6 Discussion ........................................................................................................................... 21
2.6.1 Limitations .................................................................................................................... 26
2.7 Conclusion ........................................................................................................................... 27
2.8 References ........................................................................................................................... 28
2.9 Appendix ............................................................................................................................. 37
Chapter 3. Unsubsidized and Subsidized Part D Beneficiaries Face Different Financial
Incentives, and Respond Differently to Preferred Pharmacy Networks ....................................... 44
3.1 Introduction ......................................................................................................................... 45
3.2 Study Data and Methods ..................................................................................................... 47
3.2.1 Data and Sample ........................................................................................................... 47
3.2.2 Financial Incentives ...................................................................................................... 47
v

3.2.3 Analysis ........................................................................................................................ 48
3.2.4 Limitations .................................................................................................................... 50
3.3 Study Results ....................................................................................................................... 51
3.4 Discussion ........................................................................................................................... 53
3.5 Conclusion ........................................................................................................................... 55
3.6 References ........................................................................................................................... 56
3.7 Appendix ............................................................................................................................. 64
Chapter 4. Medicare Part D Beneficiaries’ Pharmacy Switching in Response to Preferred
Pharmacy Networks ...................................................................................................................... 69
4.1 Introduction ......................................................................................................................... 70
4.2 Data and Methods................................................................................................................ 72
4.2.1 Preferred Pharmacy Networks in PDPs ........................................................................ 72
4.2.2 Data and Sample ........................................................................................................... 73
4.2.3 The Design of Preferred Networks ............................................................................... 74
4.2.4 The Average Effects on the Use of Preferred Pharmacies ........................................... 75
4.2.5 Factors Associated With Beneficiaries’ Pharmacy Switching ..................................... 76
4.3 Results ................................................................................................................................. 77
4.3.1 Preferred Network Restrictiveness and Financial Incentives ....................................... 77
4.3.2 The Average Effects on the Use of Preferred Pharmacies ........................................... 79
4.3.3 Factors Associated With Beneficiaries’ Pharmacy Switching ..................................... 79
4.4 Discussion ........................................................................................................................... 80
4.5 References ........................................................................................................................... 83
4.6 Appendix ............................................................................................................................. 91
Bibliography ............................................................................................................................... 100

vi

Abstract

Prescription drugs are important in health care in the United States, and rising drug prices
are of great concern for the public. This dissertation consists of three essays, and together they
examine the impact of two understudied recent developments that aim at improving patient
access to prescription drugs and curbing price growth: One is more accurately compensating
health plans for enrollees who take high-cost drugs in health insurance exchanges created by the
Affordable Care Act (ACA), and the other is the implementation of preferred pharmacy networks
by Medicare Part D plans as an effort to negotiate lower prescription drug prices.
The first essay finds that in the exchanges, incorporating the information on the use of
high-cost, disease-modifying drugs into the risk adjustment model would improve the accuracy
of risk-adjusted payments, and thus mitigate plans’ incentives to skimp on the coverage of novel,
expensive drugs, thereby improving patient access. The second and third essays evaluate
preferred pharmacy networks in Medicare Part D. The second essay finds different financial
incentives for unsubsidized and subsidized patients under preferred networks' cost-sharing
structures. In addition, unsubsidized patients incurred higher out-of-pocket spending when they
used non-preferred pharmacies, while Medicare bore the additional financial burden for
subsidized patients who visited non-preferred pharmacies. The third essay estimates that the
implementation of preferred networks moderately increased preferred pharmacies’ claim share in
the first year among unsubsidized beneficiaries. Existing relationships with preferred
pharmacies, financial incentives, access to preferred pharmacies, and urban residence were
positively correlated with beneficiaries’ decisions to switch to these pharmacies.

1

Chapter 1. Introduction

Prescription drugs play an important role in health care in the United States. Spending on
prescription drugs reached $335 billion in 2018 (Centers for Medicare & Medicaid Services,
2019). A recent study attributed 35 percent of the life expectancy gains in the U.S. between 1990
and 2015 to prescription drugs (Buxbaum et al., 2020).
The prices of prescription drug have grown rapidly in recent years, and are a source of
great concern among Americans. Drug prices have consistently risen faster than general inflation
over the last decade, with the average list prices of brand-name and specialty drugs increasing
about 10 to 20 percent annually (Hernandez et al., 2019). These high and rising prices cause
affordability concerns for patients. One quarter of adults think it is “difficult” or “very difficult”
to afford their medications, and three in ten adults report skipping pills due to costs (Kirzinger et
al., 2019). Hence understandably, the public recently ranked “lowering prescription drug prices”
as their top health policy priority for Congress (Lopes et al., 2020).
My dissertation sheds light on the impact of two understudied recent developments that
aim at improving patient access to prescription drugs and curbing price growth amid high drug
prices. One development is more accurately compensating health plans for enrollees who take
high-cost drugs in health insurance exchanges created by the Affordable Care Act (ACA). The
other is the implementation of preferred pharmacy networks by Medicare Part D plans as an
effort to negotiate lower prescription drug prices.
The second chapter, titled “Incorporating Prescription Drug Utilization Information into
the Marketplace Risk Adjustment Model Improves Payment Accuracy and Reduces Adverse
Selection Incentives,” is coauthored with Erin Trish and Geoffrey Joyce. It evaluates the impact
2

of improved accuracy of payments to health plans in the ACA exchanges. Although expected
health spending varies widely across consumers, ACA exchange plans are prohibited from
charging sicker enrollees higher premiums. Risk adjustment – which transfers money from plans
with healthier enrollees to plans with sicker ones – is thus necessary to ensure that plans are
fairly compensated. Yet it is only as successful as the accuracy of the formula used to determine
transfer payments. For the first several years of this program, the formula did not incorporate
information on enrollees’ prescription drug utilization. However, given the growing importance
of novel, expensive drug therapies in the treatment of conditions such as HIV, hepatitis C, and
multiple sclerosis, leaving out drug use information generated payments to plans that were
inadequate for patients that used them. This payment inadequacy caused access concerns for
patients since plans might cover these drugs less generously, likely in an effort to avoid enrolling
consumers in need of them. For instance, one study found that HIV patients would pay an annual
average of $4,892 out-of-pocket per drug in some plans because of the high coinsurance imposed
on all HIV drugs, compared with $1,615 in other plans (Jacobs & Sommers, 2015).
To address these concerns, in 2018 the risk adjustment formula began including
indicators for the use of twelve classes of drugs, such as HIV, multiple sclerosis, hepatitis C, and
rheumatoid arthritis drugs. We study the policy change’s impact on payment accuracy.
Specifically, we calculate the ratio of predicted health spending to actual spending as a measure
of accuracy of the risk adjustment formula. Using the de-identified Clinformatics® DataMart
(OptumInsight, Eden Prairie, MN), we simulate and compare these ratios before and after the
policy change to incorporate prescription drug utilization. The results demonstrate that, for a
certain diagnosed health condition, before the change, risk adjustment considerably under-
compensates for individuals utilizing drug therapies, and over-compensates for those who are
3

not, since it combines the two subgroups in payment calculations, despite that the former tend to
have higher spending than the latter. By contrast, after the change, the risk adjustment formula
more accurately predicts spending, by both increasing payments for individuals taking
prescription drugs and decreasing payments for those who are not. As a result, incorporating
prescription drug utilization improves the accuracy of risk-adjusted payments, and thus mitigates
plans’ incentives to skimp on the coverage of novel, expensive drugs, thereby improving patient
access.
The third and fourth chapters focus on an understudied recent development in Medicare
Part D – the rise of preferred pharmacy networks. While manufacturers and pharmacy benefit
managers (PBMs) play an important role in determining drug prices, these prices also vary
widely across pharmacies (Arora et al., 2017). Although the impact of preferred provider
networks – or using differential cost sharing to incentivize patients to choose lower-cost, higher-
value physicians and hospitals – has been well-studied, very little is known about the impact of
preferred networks on incentivizing patients to use lower-cost pharmacies. But the topic is
important, particularly since preferred networks are predominant in Medicare Part D, having
grown from less than 9 percent of stand-alone plans in 2011 to over 98 percent in 2021 (authors’
calculation).
The third chapter, titled “Unsubsidized and Subsidized Part D Beneficiaries Face
Different Financial Incentives, and Respond Differently to Preferred Pharmacy Networks,” is
coauthored with Erin Trish and Geoffrey Joyce. Using the 20 percent Medicare claims data from
2010 through 2016, we examine the financial incentives, beneficiaries’ pharmacy switching, and
the financial consequences of pharmacy choice in preferred networks. Unsubsidized
beneficiaries faced substantial incentives – on average $147 annually in out-of-pocket spending –
4

and moderately switched towards preferred pharmacies, while subsidized beneficiaries were
insulated from the incentives and demonstrated little switching. Among those who continued to
use non-preferred pharmacies, on average, the unsubsidized paid more out-of-pocket ($94)
relative to using preferred pharmacies, while Medicare bore the extra spending ($170) for the
subsidized through cost-sharing subsidy. Given the importance of preferred pharmacy networks,
researchers and policymakers should study the impact on the quality of beneficiaries’ decision-
making and cost savings to fully evaluate these networks.
The fourth chapter, titled “Medicare Part D Beneficiaries’ Pharmacy Switching in
Response to Preferred Pharmacy Networks,” is coauthored with Erin Trish and Geoffrey Joyce.
Using the 20 percent Medicare claims data from 2010 through 2016, we examine the design of
preferred networks, their effects on the use of preferred pharmacies, and factors underlying
beneficiaries’ decisions of whether to switch to preferred pharmacies. We find that preferred
networks generally designated a restrictive set of preferred pharmacies, and the financial
incentives to use preferred pharmacies were substantial on average for unsubsidized
beneficiaries, while subsidized beneficiaries faced minimal incentive. Consequently, with a
difference-in-difference design, we estimate that the implementation of preferred networks on
average resulted in a 3.7-percentage point increase in preferred pharmacies’ claim share in the
first year among the non-LIS. Existing relationships with preferred pharmacies, financial
incentives, access to preferred pharmacies, and urban residence were positively correlated with
beneficiaries’ decisions to switch to these pharmacies.
Taken together, findings in this dissertation provide important insight on functioning of
prescription drug markets. The findings from my dissertation will enlighten policy interventions
aimed at reducing costs and improving access to prescription drugs.
5

References
Arora, S., Sood, N., Terp, S., & Joyce, G. (2017). The Price May Not Be Right: The Value of
Comparison Shopping for Prescription Drugs. American Journal of Managed Care, 23(7),
410–415.
Buxbaum, J. D., Chernew, M. E., Fendrick, A. M., & Cutler, D. M. (2020). Contributions Of
Public Health, Pharmaceuticals, And Other Medical Care To US Life Expectancy Changes,
1990-2015. Health Affairs, 39(9), 1546–1556. https://doi.org/10.1377/hlthaff.2020.00284
Centers for Medicare & Medicaid Services. (2019). National Health Expenditures 2018
Highlights. https://www.cms.gov/files/document/highlights.pdf
Hernandez, I., Good, C. B., Cutler, D. M., Gellad, W. F., Parekh, N., & Shrank, W. H. (2019).
The Contribution Of New Product Entry Versus Existing Product Inflation In The Rising
Costs Of Drugs. Health Affairs, 38(1), 76–83. https://doi.org/10.1377/hlthaff.2018.05147
Jacobs, D. B., & Sommers, B. D. (2015). Using Drugs to Discriminate—Adverse Selection in the
Insurance Marketplace. New England Journal of Medicine, 372(5), 399–402.
https://doi.org/10.1056/NEJMp1411376
Kirzinger, A., Lopes, L., Wu, B., & Brodie, M. (2019). KFF Health Tracking Poll – February
2019: Prescription Drugs. https://www.kff.org/health-costs/poll-finding/kff-health-tracking-
poll-february-2019-prescription-drugs/
Lopes, L., Hamel, L., Kearney, A., & Brodie, M. (2020, January 30). KFF Health Tracking Poll
– January 2020: Medicare-for-all, Public Option, Health Care Legislation And Court
Actions. KFF. https://www.kff.org/health-reform/poll-finding/kff-health-tracking-poll-
january-2020/

6

Chapter 2. Incorporating Prescription Drug Utilization Information into the
Marketplace Risk Adjustment Model Improves Payment Accuracy and
Reduces Adverse Selection Incentives
1, 2

Abstract
Beginning with the 2018 benefit year, the Centers for Medicare and Medicaid Services (CMS)
started incorporating select prescription drug utilization information into the Marketplace risk
adjustment model. There has been little evidence about the impact of this change on payment
accuracy and the mechanisms through which it may occur. Using commercial claims in 2017
from a large national health insurer, we find that the policy change improves payment accuracy
in our sample and would help to mitigate insurers’ selection incentives for some enrollees
through two channels: imputing missing diagnoses and varying risk scores to better capture the
heterogeneity in expenditures among patients with certain health conditions. However, while
improving payment accuracy overall, there are potential perverse incentives which could distort
treatment choice for marginal patients. Additionally, over- and under-compensation persists for
certain patient subgroups, suggesting an opportunity to further refine and improve the model.

1
This work was supported by the Leonard D. Schaeffer Center for Health Policy and Economics of the
University of Southern California.
2
This chapter is published in the journal Medical Care Research and Review: Xu, J., Trish, E., & Joyce,
G. (2019). Incorporating prescription drug utilization information into the Marketplace risk adjustment
model improves payment accuracy and reduces adverse selection incentives. Medical Care Research and
Review. https://doi.org/10.1177/1077558719870060
7

2.1 Introduction
The Affordable Care Act (ACA) imposed guaranteed issue and adjusted community
rating in the individual and small group health insurance markets, prohibiting insurers from
denying coverage and varying premiums based on health status or gender. In order to mitigate
insurers’ incentives to select only healthier enrollees as a result of these premium rating policies,
the ACA also created a permanent risk adjustment program, which transfers funds from plans
with low-risk enrollees to those with high-risk enrollees. The key component in the calculation
of transfer payments are enrollee risk scores, which are calculated from a concurrent risk
adjustment model developed for the Department of Health and Human Services (HHS) by the
Centers for Medicare and Medicaid Services (CMS), known as the HHS-HCC model. These risk
scores are based on enrollees’ age, gender, and the presence of hierarchical condition categories
(HCCs), defined by diagnoses recorded in their medical claims in that benefit year.
Although the risk adjustment program has generally been successful at adequately
compensating plans according to their enrollees’ risk in the first few years of existence (Jacobs,
Cohen, & Keenan, 2017), there have been some concerns about payment inaccuracies for certain
patient subgroups. Specifically, some patients’ diagnoses may be under-recorded in their medical
claims, especially because diagnoses must be recorded in their claims in that benefit year to
count toward their risk score (CMS, 2016a). Additionally, initial evidence suggests that the
model’s risk scores under-compensate plans for some conditions commonly treated with high
cost prescription drugs (Montz et al., 2016). This may be due in part to the fact that the model
was adapted from the CMS risk adjustment model used in Medicare Advantage, which does not
8

include prescription drug spending.
3
Prior research has shown that insurers in this market act on
these selection incentives by less generously covering drugs associated with unprofitable patient
groups, raising concerns about access (Jacobs & Sommers, 2015; Geruso, Layton, & Prinz,
2016).
In an effort to improve payment accuracy, CMS started incorporating additional
information from prescription drug utilization into risk scores for adult enrollees beginning with
the 2018 benefit year (CMS, 2016c). Specifically, CMS took a “hybrid” approach, where
prescription drug utilization could be used both to identify the presence of a condition not coded
in that individual’s medical claims (to be consistent with CMS’ language, we use the words
“impute” and “imputation” hereafter) and as a proxy for disease severity (or expected spending).
Thus, rather than simply using prescription drug utilization to assign an HCC to an enrollee,
CMS added twelve new prescription drug categories (RXCs), with associated risk scores, to the
model (Appendix Table 2.1). That is, an individual with an HCC only, with an RXC only, and
with both an HCC and an RXC for the same condition would each receive different risk scores.
An individual receives an RXC if they take one or more of a particular set of prescription drugs
associated with a given condition. CMS restricted these drugs to include only those that are
indicated and used specifically to treat the particular condition (HCC) associated with the RXC,
are relatively high cost, and are not used prophylactically (CMS, 2017b). In general, these
changes increased risk scores for enrollees with these conditions who also take prescription
drugs included in RXCs, and decreased risk scores for enrollees with the associated HCCs but
who do not take selected high cost, disease-specific prescription drugs (Appendix Table 2.2).

3 Most Medicare Advantage enrollees are in a plan that also providers Part D prescription drug coverage,
but plans are compensated for higher expected prescription drug costs through the Part D risk adjustment
formula rather than the Medicare Advantage risk adjustment formula.
9

However, the overall effect on payment accuracy under the new 2018 HHS-HCC model has not
been evaluated.

2.2 Conceptual Framework
Risk selection arises when the insurer is unable to charge a premium based on a
consumer’s expected health care costs, either because the consumer has private information
about their risk type or because regulations prohibit insurers from using certain information to
vary premiums, as is in the ACA Marketplace (Cutler and Zeckhauser, 2000; Breyer, Bundorf, &
Pauly, 2011). This creates an incentive for insurers to avoid sicker individuals. To act on this
incentive plans may, for example, offer benefit packages that are unattractive to them, such as
narrower provider networks or less generous coverage of certain medical services and
prescription drugs (Geruso & Layton, 2017). These contract distortions can lead to access
barriers and high out-of-pocket costs. Risk adjustment can help to mitigate these selection
incentives and instead promote competition on efficiency by more accurately aligning plan
compensation with the health risk of their enrollees, though the success is dependent on a strong
model and program design.
Despite being a widely adopted policy tool, diagnosis-based risk adjustment can face
challenges in eliminating risk selection in practice. The accuracy of the coding of health
conditions can affect payment accuracy (van de Ven & Ellis, 2000). In addition, since it does not
perfectly predict spending for every possible enrollee subset, plans may exploit the heterogeneity
in predictable spending not captured by the model (Brown, Duggan, Kuziemko, & Woolston,
2014). One example of such behavior is plans offering drug formularies which are unappealing
10

to patients who take prescription drugs indicative of high spending conditional on their predicted
risk level (Lavetti & Simon, 2018).
Incorporating prescription drug utilization has the potential to improve risk adjustment,
but is not without concerns (Gilmer, Kronick, Fishman, & Ganiats, 2001; CMS, 2016a). More
timely and complete, prescription drug information is a useful supplement to medical diagnoses
in determining the existence of health conditions. For example, diagnoses of chronic conditions
such as diabetes may not be recorded in a person’s medical claims every year. Additionally,
prescription drug information can serve as a proxy for disease severity and thus can better
account for the heterogeneity in spending related to medication use. However, the additional
payments derived from the utilization of certain drugs may distort treatment choice for patients
on the margin.

2.3 New Contribution
In this article, we evaluate the potential impact of CMS’ 2018 HHS-HCC risk adjustment
model incorporating prescription drug utilization on predictive power, payment accuracy, and
selection incentives for certain patient subgroups. We first motivate the interest in these changes
to the risk adjustment model by documenting the under- and over-compensation for certain
patient subgroups under the previous 2017 HHS-HCC model. We then evaluate the effect of
these changes to the 2018 model on predictive power and payment accuracy, highlighting the
relative contributions of the different mechanisms through which incorporating prescription drug
information into the HHS-HCC risk adjustment affects payment accuracy in practice – e.g.,
through imputation of missing diagnoses versus through establishing differential risk scores for
enrollees within an HCC according to their prescription drug utilization. Finally, we highlight
11

concerns about ongoing payment inaccuracies even after these changes to the model, particularly
undercompensating for patients without any chronic conditions, which may suggest a need for
further model refinements.

2.4 Methods
2.4.1 Data and sample
We evaluated the member information file, medical claims, and pharmacy claims from
the de-identified Clinformatics® DataMart (OptumInsight, Eden Prairie, MN), which contains all
consumers covered by a large national health insurance company. We primarily used the 2017
data, as well as a subset of the 2016 data for a longitudinal analysis. We restricted our main
sample to commercially insured enrollees ages 21 to 64 who were enrolled for the full year and
did not switch plans in 2017. In supplemental analyses, we examined those with partial-year
enrollment. Following CMS’ criteria for the calibration sample, we excluded individuals in
Health Maintenance Organizations (HMOs) to limit instances of incomplete claims information
on diagnoses and spending due to capitated contracts (CMS, 2016a).
4, 5
The sample included 5.16
million individuals, among whom 3,690 were individual market enrollees and the remainder
were in employer-sponsored plans.

4
Eleven percent of the records were capitated encounters for HMO enrollees. The results were
qualitatively unchanged if we kept them in the sample.
5
CMS also required that plans in the sample had prescription drugs and integrated mental
health/substance abuse coverage to ensure that they covered the essential health benefits mandated by the
ACA. No benefit design information was available in our data. However, Kaiser Family Foundation’s
Employer Health Benefit Survey shows that nearly all employer-sponsored plans cover prescription drugs
(Claxton et al., 2017). While this report does not directly address employer-sponsored coverage of mental
health benefits, we expect that comprehensive coverage of such benefits is common in our sample,
because during our study period individual and small group policies were subject to essential health
benefits requirements, including mental health and substance abuse, and large employers were subject to
federal mental health parity laws.
12

2.4.2 Risk score
We calculated the risk scores that each enrollee would have received under the 2017 and
2018 HHS-HCC models with the publicly-available model software for the two benefit years
(CMS, 2017a; CMS, 2018a).
6
The software maps diagnosis codes to HCCs and, for the 2018
model, National Drug Codes (NDCs) and Healthcare Common Procedure Coding System
(HCPCS) drug codes to RXCs, and calculates an enrollee-level risk score by summing the
coefficients for each HCC and/or RXC present for each enrollee. In addition to the presence of
HCCs, the models also include coefficients for age/gender categories and some severe illness
interactions. Some HCCs are combined into HCC Groups, which receive the same risk score
(e.g., HCCs 19, 20, and 21 are for diabetes with acute complications, with chronic complications,
and without complication, respectively, but these get combined into one “Group” (G01) and the
risk score is assigned at the Group rather than HCC level). The 2018 model adds RXCs and
HCC-RXC interactions, as well as adding a new unique HCC for hepatitis C. Factor coefficients
for each benefit year are estimated by CMS using a calibration sample. To focus on changes to
the algorithm, we set everyone’s plan metal level to gold, since on average plans in our sample
covered 85 percent of total spending, with minimal variation across plans.
7

6
We excluded coefficients for the two severity-only RXCs (RXC 11: ammonia detoxicants and RXC 12:
diuretics, loop and select potassium-sparing) as the associated risk scores were quite low and these have
since been removed from the HHS-HCC model starting with the 2019 benefit year (CMS, 2018b).
Severity-only RXCs lead to an increase in the risk score only in the presence of the associated HCCs.
7
Risk scores for a given HCC vary according to the metal level of the plan, because risk scores are
intended to reflect plan spending and plan spending varies with plan metal level. That is, the coefficient
for a given HCC will be higher if the patient is enrolled in a gold plan than it is if they are enrolled in a
silver plan. Other analyses (e.g., Kautter et al. 2014) have addressed variation in plan generosity of the
analytic sample by “re-processing” actual claims data under various sets of plan design features to
simulate plan spending for a constant sample under different simulated plan actuarial levels. We did not
take this approach in our analysis; this is a limitation to the extent that there is an association between
residual “unpredicted” risk (i.e., not accounted for by the risk score) and sorting into more or less
generous plans. However, as noted in the text, there is limited variation in generosity across the plans in
our sample.
13

2.4.3 Predictive power and payment accuracy
We used R
2
, derived from a linear regression of plan spending on risk score, as the
measure of model predictive power overall. A higher R
2
indicates better predictive power,
meaning that the model is able to explain a higher proportion of the variation in spending.
We assessed group-level payment accuracy for certain subsets of patients with predictive
ratios. We calculated the predictive ratio for a predefined patient subset of interest under a risk
adjustment model by taking the ratio of mean predicted plan spending to mean actual spending
covered by the plan for that patient subgroup. The numerator was equal to the sample’s mean
plan spending times the ratio of the group’s mean risk score to the sample’s mean risk score.
Each patient’s plan spending was the total costs of all medical and prescription drug claims less
patient cost-sharing, which included deductibles, copayments, and coinsurance.
8
A predictive
ratio less than one for a given patient subgroup indicates that plans are under-compensated for
them, on average. In contrast, a predictive ratio greater than one indicates that plans are
overcompensated for that patient subgroup, on average.
2.4.4 Statistical analysis
We first evaluated the distribution of patient enrollment and plan spending among
subgroups with zero, one, two, and three or more HCCs. To do so, we counted the number of
HCCs for each enrollee, counting an HCC Group as one single HCC. We then assessed the
predictive ratios under the 2017 model for enrollee subgroups with different numbers of HCCs.

8
The Optum data reports the standardized costs of a claim instead of the actual costs, in order to protect
confidential information. This should have little impact on our analyses, which were done at aggregate
levels and which focused on the variation of payment accuracy rather than the absolute level of spending.
Furthermore, the total costs by age groups in Table 2.1 were very close to the amounts reported by the
Health Care Cost Institute (2019).
14

To illustrate the magnitude of under- and over-prediction, we also computed the differences
between the mean predicted and the mean actual plan spending.
Next, to assess the potential impact of adding the prescription drug-based changes to the
model, we assessed the share of patients taking prescription drugs that would map to RXCs
based on the 2018 HHS-HCC model, and evaluated heterogeneity in payment accuracy among
those with and without RXCs. Specifically, we divided the same sample into: (1) individuals
with only HCCs under the 2018 model, i.e., those who did not take medications included in any
RXCs; (2) individuals with one or more HCCs and who would have one or more “associated”
RXCs under the 2018 model (i.e., those who only took medications in RXCs associated with the
HCCs they already had); and (3) individuals with only “unassociated” RXCs under the 2018
model, (i.e., those who only took medications in RXCs not associated with the HCCs they had,
including individuals with one or more RXCs but no HCC). Changes to risk scores after
incorporating RXCs for Groups 1 and 2 primarily represent the severity-adjustment component
whereas changes to risk scores for Group 3 represent both the imputation and severity adjustment
components, since the enrollees lacked medical diagnoses associated with their RXCs. We
excluded enrollees who had both associated and unassociated RXCs (less than 0.03 percent of
the sample).
Next, we analyzed the implications of incorporating information on prescription drug use
for model predictive power and aggregate payment accuracy. We compared the R
2
statistics of
our sample under the 2017 and the 2018 HHS-HCC models. We analyzed whether payment
accuracy improved under the latter by comparing risk scores and predictive ratios under the two
models.
15

Additionally, to examine the implications of including RXCs for payment accuracy and
plan incentives, we assessed plan spending, risk scores, and predictive ratios for subgroups of
patients with six prevalent and expensive health conditions associated with unique RXC(s) each:
HIV, multiple sclerosis, diabetes, hepatitis C, heart arrhythmia, and chronic kidney disease
(Appendix Table 2.1). We considered an individual as having a condition if she had either one or
more related HCC(s) and/or the associated RXC(s). Thus, there were three categories for each
condition: HCC only, RXC only, and those with both. Additionally, we extracted the subset of
individuals enrolled in our data in both 2016 and 2017 to examine persistence of the presence of
HCCs and RXCs over time.
Finally, while our main analyses were restricted to full-year enrollees, in supplemental
analyses, we assessed the impact of these changes on partial-year enrollees in our data. It is
possible that incorporating prescription drug utilization into the risk adjustment model may have
a stronger impact for this population, particularly for imputation, since individuals taking
medications for chronic conditions would likely fill prescriptions even if they were enrolled only
for a short period, though their diagnoses may not be captured in their medical claims. While we
believe that actual Marketplace data would be better suited for a thorough evaluation of the
impact on partial-year enrollees, since the enrollment duration and the health profile of partial-
year enrollees in the Marketplaces may be different than those in employment-based insurance
(Dorn, Garrett, & Epstein, 2018), the relative impact on partial-year enrollees as compared to
full-year enrollees in our dataset can still be informative on this point.

16

2.5 Study Results
Table 2.1 provides summary statistics for the sample. The average age was 42.5, and
women made up slightly less than half of all enrollees. Average total spending was $6,484 per
enrollee, while average plan spending was $5,521 (85 percent of the total spending). As
expected, spending increased with enrollee age, on average.
2.5.1 Payment accuracy under the 2017 model
Under the 2017 model, 80 percent of the sample had no HCCs and thus their risk scores
were based solely on age and gender; however, they accounted for less than one-third of plan
spending (Figure 2.1). Consistent with prior work (Kautter et al., 2014), we found that, on
average, the 2017 model under-predicted the spending of low-risk enrollees, while it over-
predicted that of high-risk enrollees (Figure 2.2).
9
For enrollees in our sample with 0 HCCs,
mean annual predicted plan spending based on the 2017 model was $453 less than mean actual
plan spending. In contrast, mean annual predicted plan spending was $328, $3,549, and $13,392
more than mean actual plan spending for enrollees with one, two, and three or more HCCs,
respectively.
However, there was substantial heterogeneity in risk-adjusted payment accuracy within
each of these groups depending on whether the enrollee utilized prescription drugs that would
have resulted in them having an RXC under the 2018 HHS-HCC model (4.7 percent of the
sample; Figure 2.3). This subset of individuals taking drugs included in RXCs, with or without
the associated HCCs, tended to have predicted plan spending that was substantially lower than

9
If healthier individuals in our sample tend to sort into less generous plans, then our approach of using
the gold plan factor coefficients across the entire sample may actually understate the magnitude of under-
prediction for these healthier individuals. That is because the demographic risk scores also vary with
metallic tier, so healthier enrollees in less generous plans would have even lower risk scores than they
would if they enrolled in a more generous plan.
17

actual plan spending under the 2017 model. This under-compensation was particularly large –
more than $10,000 per enrollee, on average – for individuals with no HCCs but with one or more
RXCs and for those with one HCC and one or more unassociated RXCs (i.e., those patients for
whom prescription drug information likely allows for imputation of a condition not coded in
their medical claims). Moreover, average plan spending among these individuals more closely
resembled that of enrollees with an additional HCC compared to average plan spending among
individuals with the same number of HCCs without any unassociated RXCs. For example, mean
actual plan spending among individuals with zero HCCs and one or more unassociated RXCs
was almost seven times higher than that among those with no HCCs or RXCs ($14,655 vs.
$2,143), but was more similar to average plan spending on enrollees with one HCC only
($10,656). This suggests that prescription drug utilization does indeed contain valuable
information to impute the presence of certain conditions and that making use of this information
in the risk adjustment model is likely to improve payment accuracy for patient subgroups taking
drugs included in RXCs for conditions not recorded in their medical claims.
In addition to the relative spending differences for individuals with unassociated RXCs,
we also found considerable differences in average plan spending among individuals with
associated RXCs compared to individuals with the same number of HCCs but without any
associated RXCs. The strongest contrast was among enrollees with three or more HCCs, where
mean actual plan spending among individuals with three or more HCCs and one or more
associated RXCs was $22,754 more than actual plan spending among individuals with three or
more HCCs and no associated RXCs, while predicted plan spending among these two groups
based on the 2017 model was relatively similar. This suggests that there is heterogeneity in
spending among individuals with the same number of HCCs and that prescription drug
18

utilization is one important factor associated with this variation. Assigning differential risk
scores for patients within a given condition who take prescription drugs versus those who do not
(via RXCs) may thus improve payment accuracy for these two subgroups.
2.5.2 Impact of 2018 changes incorporating prescription drug utilization on model predictive
power and payment accuracy
Basing payments on the 2018 HHS-HCC model (including RXCs) rather than the 2017
model improves predictive power, with the overall R
2
increasing from 0.377 to 0.399, consistent
with the increase from 0.356 to 0.409 when CMS applied the models to adults in gold plans in
the 2014 MarketScan data (CMS, 2016b; CMS, 2016c). More importantly, it mitigates under-
and over-prediction of spending for subgroups with different numbers of HCCs and RXCs.
Before evaluating subgroup level effects, it is important to highlight the size of groups
subject to imputation and severity adjustment. Enrollees affected by the policy change (i.e., those
with either one or more HCCs or RXCs) comprise 20 percent of the sample, but they account for
68 percent of total plan spending. Of these individuals, nearly all (97 percent) had no or only
associated RXCs, and thus the impact of incorporating prescription drug information under the
2018 HHS-HCC model is primarily through severity adjustment rather than imputation for those
enrolled throughout the year.
In addition to adding RXCs, there were other changes to the coefficients in the 2018
model, including reducing the coefficients for those HCCs associated with RXCs (i.e., reducing
risk scores for patients who had the HCC only; Appendix Table 2.2). Mean risk scores of
individuals in our sample with one HCC only would therefore decrease by 0.4, translating to
reductions in mean predicted plan spending of $1,037 (Table 2.2). Similarly, individuals with
two and three or more HCCs only would on average experience declines of 1.0 and 2.9 in risk
score, respectively, which represent decreases in predicted plan spending of $2,592 and $6,947,
19

respectively. As a result, while predictive ratios for these groups under the 2018 model are still
greater than one, the extent of their over-compensation would decline compared to that under the
2017 model. However, under-compensation would persist for enrollees with zero HCCs (80
percent of the sample); under the 2018 model, the predictive ratio for these individuals was 0.78.
Under the 2018 model, total risk scores increase for enrollees taking drugs included in
RXCs in our sample, except for those with more than three HCCs. Although smaller in size,
subgroups with only unassociated RXCs would experience larger increases in risk scores
compared to those with associated RXCs. Notably, the mean total risk score of individuals with
zero HCCs and one or more RXCs would increase from 0.5 to 3.6 (including the coefficients for
demographic adjustments). Consequently, predictive ratios would increase for all subgroups with
RXCs except those with three or more HCCs and only unassociated RXCs. Predictive ratios for
individuals without HCCs but with one or more unassociated RXCs – those whose plan spending
were most under-predicted by the 2017 model – increased considerably, from 0.14 to 1.08.
Although enrollees with one HCC and only unassociated RXCs would still have a predictive
ratio below one, all other subgroups with RXCs would yield predictive ratios greater than one.
The analyses of six prevalent and costly conditions further support our findings of
increased payment accuracy for enrollees with these conditions under the 2018 HHS-HCC model
(Table 2.3). Both risk scores and predictive ratios would decrease for HCC-only subgroups,
neutralizing overpayment for them compared to the 2017 model. Payment accuracy would also
improve for patients with an RXC associated with HIV, multiple sclerosis, diabetes, and hepatitis
20

C, however, under-compensation would persist for individuals with heart arrhythmia and chronic
kidney disease.
10
2.5.3 A further illustration of the mechanisms
Our analyses of patients with select health conditions also demonstrate more clearly the
mechanisms through which the inclusion of prescription drug information improves payment
accuracy. Risk scores increase under the 2018 model for subgroups with an RXC associated with
HIV, multiple sclerosis, diabetes, and hepatitis C, and predictive ratios would increase to above
one except for RXC-only diabetes patients (Table 2.3). These payment increases occur through a
combination of imputing missing diagnoses as well as using prescription drug utilization to
account for heterogeneity in severity, though many more beneficiaries are affected by the latter.
The majority of RXC-only patients either had the associated diagnoses or took medications in the
RXC in the previous year, suggesting that incorporating prescription drug utilization into risk
adjustment does help to capture persistent conditions (Appendix Figure 2.1). Additionally,
among individuals with each condition, the subgroup risk scores under the 2018 model would
follow the actual spending gradient. Thus, the 2018 hybrid model appears to improve payment
accuracy for varying levels of severity, although the spending patterns we report are not adjusted
for other factors which may vary across patients within each of these groups and affect spending
(such as the presence of other HCCs).

10
Consistent with the other conditions, the coefficients for these HCCs decrease overall, but the
additional RXC coefficients do not appear to be large enough to adequately compensate plans for
enrollees taking prescription drugs for these conditions.
21

2.5.4 Partial-year enrollees
Individuals with partial-year enrollment appeared to have fewer HCCs than those with
full-year enrollment. For example, 5, 0.7, and 0.3 percent of individuals enrolled for three
months or less had one, two, and three or more HCCs, respectively, compared to 15, 3, and 1
percent among those enrolled for the entire year (Appendix Table 2.3; Table 2.2). It is possible
that the lower prevalence of HCCs among partial-year enrollees could be due, at least in part, to
missing diagnoses, which could be mitigated to some extent by incorporating prescription drug
utilization information, particularly among individuals with very short duration. Among
enrollees with three months of enrollment or less, 12 percent of the individuals affected by the
policy change (i.e., those with either one or more HCCs or RXCs) had unassociated RXCs, and
thus would be subject to imputation, compared to 2 percent for full-year enrollees. However, the
effect of imputation would still be limited compared to full-year enrollees because the share of
total enrollees with 0 HCCs only (even after applying the 2018 model) was much higher among
partial-year enrollees compared to full-year enrollees.

2.6 Discussion
Overall, when applied to our sample of commercially-insured enrollees, the changes to
the 2018 HHS-HCC risk adjustment model, including incorporating the use of prescription drug
utilization, improve payment accuracy and help to mitigate the selection incentives compared to
the prior model. Interestingly, we find that using such information to impute missing diagnoses
affects a relatively small fraction of patients, particularly among full-year enrollees. Instead,
using prescription drug information to segment patients with a particular condition into groups
with different spending levels represents the much more significant effect from this policy
22

change. The introduction of these RXCs appears to statistically capture much of the spending
heterogeneity within HCCs and thus neutralizes payment incentives across patients who take
certain drugs and those who do not, conditional on having HCCs. This change to increase
payment accuracy through improvements to risk adjustment can promote competition in the
exchanges by encouraging plans to compete on quality and efficiency, rather than on selection of
profitable consumers.
However, we also find remaining considerable under-compensation for enrollees with 0
HCCs (the vast majority of enrollees) under the 2018 model, and evidence of over-compensation
for some subgroups of enrollees with conditions (e.g., subgroups with two and three or more
HCCs). Additional analyses suggest that, in our sample, among this group of enrollees with 0
HCCs, those with higher plan spending (over $2,000) are older and more likely to be women
(Appendix Table 2.4). Although the most common diagnoses among individuals with 0 HCCs
are similar for those with both higher and lower plan spending (including acute conditions – e.g.,
acute upper respiratory infection and acute pharyngitis; less severe conditions not included in
HCCs – e.g., lower back pain, essential hypertension, and hyperlipidemia; and anxiety), higher
spending individuals with 0 HCCs have a substantially higher prevalence of these conditions
than their lower spending counterparts (Appendix Table 2.5).
The fact that, despite improvements to the risk adjustment model, plans are still
undercompensated for enrollees with 0 HCCs, on average, creates a somewhat unique scenario
where sicker individuals may be less subject to selection concerns than healthy individuals in this
market. However, our findings suggest that further refinements to the model, such as increases to
the demographic risk scores that affect plan compensation for enrollees with no conditions and
23

addressing the issue of uncompensated mental health conditions (Montz et al., 2016), may be
needed.
Despite the overall improvements in payment accuracy under the 2018 model, the
inclusion of prescription drug information in the risk adjustment model also has the potential to
create perverse incentives that may distort treatment choice for marginal patients, and thus the
benefits should be considered in the context of these potential downsides. One concern is related
to the fact that drugs with multiple indications or off-label use and drugs often prescribed
prophylactically are excluded from generating an RXC. That is, patients only receive the RXC
coefficient if they take one of a subset of drugs used specifically to treat that condition. For
example, medications to treat multiple sclerosis are expensive, averaging around $5,698 per
month (Trish, Xu, & Joyce, 2018). While an RXC for multiple sclerosis was added to the model,
only a subset of prescription drugs used to treat it were included in the RXC. Natalizumab
(Tysabri) – a drug used by 6 percent (706 out of 11,430) of the multiple sclerosis patients in our
commercially-insured sample – is excluded from the RXC, likely because it is also used to treat
Crohn’s disease. Thus, these patients would not receive the RXC, although they would still be
equally costly to plans as multiple sclerosis patients taking a drug included in the RXC. This type
of scenario could encourage the utilization of an alternative therapy that is included in the RXC,
which may not necessarily be the best treatment choice for that patient, but has significant
financial implications for the plan. While we agree that it is important to exclude drugs with
multiple indications for the purposes of imputation, expanding the set of drugs included in RXCs
for severity adjustment among patients who also have the associated diagnosis recorded in their
medical claims to avoid these types of incentives may be warranted.
24

In addition to the potential concerns about differential incentives among alternative
therapies, plans may also have an incentive to encourage providers to over-prescribe medications
when the increase in risk adjustment revenues exceeds the costs of medications (Weiner, Trish,
Abrams, & Lemke, 2012). The potential increase in predicted plan spending when a patient
switches from having the HCC only to having both the HCC and the RXC is particularly large
for HIV, multiple sclerosis, hepatitis C, and chronic kidney disease, though to create a distorted
incentive the increased predicted plan spending would have to more than offset the plan's
increased cost associated with the patient taking the drug. Distorted incentives for treatment
choice can lead to wasteful spending and potential patient harm from side effects, while the
extent to which plans could or would actually encourage providers to engage in such behavior is
unclear.
Our findings also highlight some of the payment inaccuracies in the ACA’s risk
adjustment model prior to 2018, which may not be fully eliminated by incorporating the current
RXCs. The 2017 HHS-HCC model substantially under-predicted plan spending for enrollees
taking certain prescription drugs associated with a condition but lacking the diagnosis codes for
that condition in their medical claims. While we did not adjust for the conditions they had, their
spending on average resembled that of individuals with an additional HCC. Thus, although this is
a small group, the magnitude of the under-compensation for a given enrollee was considerable
and could be financially problematic for insurers, particularly to the extent that these individuals
and/or under-recording of diagnoses is not evenly distributed across plans. In addition,
conditional on having HCCs, the 2017 HHS-HCC model under-predicted plan spending for
patients taking drugs included in the 2018 RXCs, while over-predicting plan spending for those
who did not. Besides the financial concerns for insurers, these incentives may also have
25

problematic effects for patients, as insurers would face a financial incentive to engage in
selection conditional on HCCs ((Brown, Duggan, Kuziemko, & Woolston, 2014; Lavetti &
Simon, 2018). In fact, evidence suggests that insurers in this market had less generous coverage
of prescription drugs associated with unprofitable patients (Geruso, Layton, & Prinz, 2016).
Thus, poor payment accuracy inherent to the risk adjustment program could have contributed to
reduced access or higher out-of-pocket spending for certain patient populations in this market,
which may continue to exist for some of the conditions other than those associated with the
RXCs currently included in the model.
CMS has continued with the new model for the 2019 plan year with some relatively
minor adjustments, though CMS has also recognized the importance of monitoring the effects of
these changes over time. Our findings suggest that, incorporating prescription drug information
will improve payment accuracy overall and, in particular, reduce under-compensation for
enrollees who take prescription drugs included in RXCs, on average. However, over- and under-
compensation continue to exist for certain patient subgroups, with notable continued under-
compensation for enrollees with no conditions. Thus, further adjustments to the coefficients,
including the demographic adjusters, may be needed. Additionally, incorporating additional drug
categories through adding RXCs may help improve payment accuracy for other conditions. For
example, prior research has documented under-prediction for patients with mental health issues
(Montz et al., 2016). Including additional RXCs for mental health drug classes may help improve
payment accuracy for these enrollees. CMS had considered antipsychotics, anticonvulsants, and
antimanic agents (CMS, 2016a), but no mental health RXCs were included in the 2018 model.
How to properly incorporate them is an important area for future research.
26

2.6.1 Limitations
Our study had several limitations. First, our sample were predominantly in employer-
sponsored plans and thus may not necessarily reflect the health profile of exchange enrollees.
Additionally, coding may be more intensive in the actual Marketplace data than our employer-
based sample, since the presence of a risk adjustment program creates incentives for such
behavior (Vandagriff, Millen, Petroske, & Mattie, 2017), similar to what research has shown in
the case of Medicare Advantage (Geruso & Layton, 2015). Nevertheless, the MarketScan data
used for Marketplace model calibration also consists mainly of employer insurance plans, and
our sample’s distributions of age, gender, and number of HCCs closely resembled these of the
MarketScan sample’s, except that our sample had slightly lower percentages of older adults
(Appendix Table 2.6). Moreover, 89 percent of the enrollees with RXCs already had the
associated HCCs, and thus as we illustrated above, enrollees would be predominantly affected by
severity adjustment alone. Our findings on the small share of patients subject to imputation may
reflect an upper bound, if coding is more intensive among Marketplace enrollees than those in
our sample.
Second, our analyses of the impact of the policy change on partial-year enrollees were
limited in scope and may not reflect the effects in the exchanges. Actual Marketplace data would
be necessary in order to thoroughly assess the impact of incorporating prescription drug
utilization into risk adjustment on payment accuracy for partial-year enrollees, as the enrollment
duration and the health and utilization profile of partial-year enrollees in the exchanges are likely
different from partial-year enrollees in our data.
Third, our grouping of enrollees was based on the drug classes currently included in the
model. Our analysis thus did not comprehensively examine whether the risk adjustment models
27

adequately account for differences in health status across patients with and without prescription
drug use across other potential conditions.

2.7 Conclusion
CMS’s changes to the 2018 HHS-HCC model incorporating information on prescription
drug utilization are likely to improve payment accuracy in the exchanges and further reduce
plans’ incentives to engage in risk selection. Our findings suggest that the vast majority of such
improvements in payment accuracy are derived from RXCs generating risk scores that better
reflect the heterogeneity in expenditures across patients within an HCC according to whether
they use certain prescription drugs, rather than through the imputation of missing diagnoses.
Moreover, our findings suggest that considerable under-compensation persists for enrollees
without any conditions – the vast majority of enrollees in this market – and that further
refinements to neutralize incentives across enrollees with and without conditions in this market
may be needed.
28

2.8 References
Breyer, F., Bundorf, M. K., & Pauly, M. V. (2011). Health care spending risk, health insurance,
and payment to health plans. In Handbook of health economics (Vol. 2, pp. 691–762).
Elsevier. https://doi.org/10.1016/B978-0-444-53592-4.00011-6
Brown, J., Duggan, M., Kuziemko, I., & Woolston, W. (2014). How does risk selection respond
to risk adjustment? New evidence from the Medicare Advantage program. American
Economic Review, 104(10), 3335–3364. https://doi.org/10.1257/aer.104.10.3335
Centers for Medicare & Medicaid Services. (2016a). March 31, 2016, HHS-operated risk
adjustment methodology meeting discussion paper. Retrieved from
https://www.cms.gov/CCIIO/Resources/Forms-Reports-and-Other-
Resources/Downloads/RA-March-31-White-Paper-032416.pdf
Centers for Medicare & Medicaid Services. (2016b). Patient Protection and Affordable Care Act;
HHS notice of benefit and payment parameters for 2017. Retrieved from
https://www.govinfo.gov/content/pkg/FR-2016-03-08/pdf/2016-04439.pdf
Centers for Medicare & Medicaid Services. (2016c). Patient Protection and Affordable Care Act;
HHS notice of benefit and payment parameters for 2018; amendments to special
enrollment periods and the Consumer Operated and Oriented Plan Program. Retrieved
from https://www.govinfo.gov/content/pkg/FR-2016-12-22/pdf/2016-30433.pdf
Centers for Medicare & Medicaid Services. (2017a). 2017 Benefit Year Risk Adjustment SAS
Version of HHS-Developed Risk Adjustment Model Algorithm Software.
Centers for Medicare & Medicaid Services. (2017b). Creation of the 2018 benefit year HHS-
operated risk adjustment adult models draft prescription drug (RXCUIs) to HHS drug
classes (RXCs) crosswalk. Retrieved from
https://www.cms.gov/CCIIO/Resources/Regulations-and-Guidance/Downloads/Draft-
RxC-Crosswalk-Memo-9-18-17.pdf
Centers for Medicare & Medicaid Services. (2018a). 2018 Benefit Year Risk Adjustment SAS
Version of HHS-Developed Risk Adjustment Model Algorithm Software.
Centers for Medicare & Medicaid Services. (2018b). Patient Protection and Affordable Care Act;
HHS notice of benefit and payment parameters for 2019. Retrieved from
https://www.govinfo.gov/content/pkg/FR-2018-04-17/pdf/2018-07355.pdf
Claxton, G., Rae, M., Long, M., Damico, A., Foster, G., & Whitmore, H. (2017). 2017 Employer
Health Benefits Survey. Retrieved from http://files.kff.org/attachment/Report-Employer-
Health-Benefits-Annual-Survey-2017
29

Cutler, D. M., & Zeckhauser, R. J. (2000). The anatomy of health insurance. In Handbook of
health economics (Vol. 1, pp. 563–643). Elsevier. https://doi.org/10.1016/S1574-
0064(00)80170-5
Dorn, S., Garrett, B., & Epstein, M. (2018). New risk-adjustment policies reduce but do not
eliminate special enrollment period underpayment. Health Affairs, 37(2), 308–315.
https://doi.org/10.1377/hlthaff.2017.0970
Geruso, M., & Layton, T. (2015). Upcoding: Evidence from Medicare on squishy risk adjustment
(Working Paper No. 21222). National Bureau of Economic Research.
https://doi.org/10.3386/w21222
Geruso, M., & Layton, T. J. (2017). Selection in health insurance markets and its policy
remedies. Journal of Economic Perspectives, 31(4), 23–50.
https://doi.org/10.1257/jep.31.4.23
Geruso, M., Layton, T. J., & Prinz, D. (2016). Screening in Contract Design: Evidence from the
ACA Health Insurance Exchanges (Working Paper No. 22832). National Bureau of
Economic Research. https://doi.org/10.3386/w22832
Gilmer, T., Kronick, R., Fishman, P., & Ganiats, T. G. (2001). The Medicaid Rx model:
pharmacy-based risk adjustment for public programs. Medical Care, 39(11), 1188–1202.
Jacobs, D. B., & Sommers, B. D. (2015). Using drugs to discriminate — adverse selection in the
insurance Marketplace. New England Journal of Medicine, 372(5), 399–402.
https://doi.org/10.1056/NEJMp1411376
Jacobs, P. D., Cohen, M. L., & Keenan, P. (2017). Risk adjustment, reinsurance improved
financial outcomes for individual market insurers with the highest claims. Health Affairs,
36(4), 755–763. https://doi.org/10.1377/hlthaff.2016.1456
Kautter, J., Pope, G. C., Ingber, M., Freeman, S., Patterson, L., Cohen, M., & Keenan, P. (2014).
The HHS-HCC risk adjustment model for individual and small group markets under the
Affordable Care Act. Medicare & Medicaid Research Review, 4(3).
https://doi.org/10.5600/mmrr.004.03.a03
Lavetti, K., & Simon, K. (2018). Strategic formulary design in Medicare Part D plans. American
Economic Journal: Economic Policy, 10(3), 154–192.
https://doi.org/10.1257/pol.20160248
Montz, E., Layton, T., Busch, A. B., Ellis, R. P., Rose, S., & McGuire, T. G. (2016). Risk-
adjustment simulation: Plans may have incentives to distort mental health and substance
30

use coverage. Health Affairs, 35(6), 1022–1028.
https://doi.org/10.1377/hlthaff.2015.1668
The Health Care Cost Institute. (2019). 2017 Health Care Cost and Utilization Report. Retrieved
May 1, 2019, from https://www.healthcostinstitute.org/research/annual-
reports/entry/2017-health-care-cost-and-utilization-report
Trish, E., Xu, J., & Joyce, G. (2018). Growing number of unsubsidized Part D beneficiaries with
catastrophic spending suggests need for an out-of-pocket cap. Health Affairs, 37(7),
1048–1056. https://doi.org/10.1377/hlthaff.2018.0006
van de Ven, W. P. M. M., & Ellis, R. P. (2000). Risk adjustment in competitive health plan
markets. In Handbook of health economics (Vol. 1, pp. 755–845). Elsevier.
https://doi.org/10.1016/S1574-0064(00)80173-0
Vandagriff, A., Millen, B. N., Petroske, J., & Mattie, L. L. (2017). ACA risk adjustment
management: Cracking the code. Retrieved from
http://www.milliman.com/insight/2017/ACA-risk-adjustment-management-Cracking-the-
code/
Weiner, J. P., Trish, E., Abrams, C., & Lemke, K. (2012). Adjusting for risk selection in State
health insurance Exchanges will be critically important and feasible, but not easy. Health
Affairs, 31(2), 306–315. https://doi.org/10.1377/hlthaff.2011.0420

31

Table 2.1. Sample Characteristics
Full
Sample
Ages
21-25
Ages
26-44
Ages
45-54
Ages
55-64
Mean age 42.5 23.1 35.3 49.5 59.1
% Women 48.6% 48.3% 48.0% 49.0% 49.4%
Mean total costs ($) 6,484 3,697 4,920 7,276 10,178
Medical ($) 5,241 3,136 4,088 5,788 8,029
Prescription drug ($) 1,243 561 832 1,488 2,149
Mean total plan spending ($) 5,512 3,074 4,112 6,218 8,796
Medical ($) 4,457 2,598 3,401 4,953 6,982
Prescription drug ($) 1,056 475 711 1,264 1,814
N 5,162,983 535,279 2,271,238 1,261,720 1,094,746

Note: The sample includes commercially insured enrollees ages 21 to 64 that enrolled for the full
year and did not switch plans in 2017. The Optum data reports the standardized costs of a claim
instead of the actual costs. Plan spending was equal to total costs less patient cost-sharing, which
included deductibles, copayments, and coinsurance. All spending is in 2017 dollars.

32

Figure 2.1. Distribution of Enrollment and Plan Spending Among Groups With Different
Numbers of HCCs

Note: The sample includes commercially insured enrollees ages 21 to 64 that enrolled for the full
year and did not switch plans in 2017. A hierarchical condition category (HCC) Group was
counted as one single HCC.

33

Figure 2.2. Predictive ratios under the 2017 HHS-HCC model for groups with different
numbers of HCCs

Note: The sample includes commercially insured enrollees ages 21 to 64 that enrolled for the full
year and did not switch plans in 2017. A hierarchical condition category (HCC) Group was
counted as one single HCC.

34

Figure 2.3. Heterogeneity in payment accuracy under the 2017 HHS-HCC Model among
enrollees who would have received RXCs and who would not

Note: The sample includes commercially insured enrollees ages 21 to 64 that enrolled for the full
year and did not switch plans in 2017. A hierarchical condition category (HCC) Group was
counted as one single HCC. The two severity-only prescription drug categories (RXCs) were
excluded from the analysis. Individuals with “associated RXC(s)” were those taking medications
in RXC(s) associated with the HCC(s) they already had. Individuals with “unassociated RXC(s)”
were those taking medications in RXC(s) not associated with the HCC(s) they had. By definition,
individuals with zero HCCs did not receive associated RXCs. Enrollee subgroups with both
associated and unassociated RXCs were excluded due to small group size. All spending is in
2017 dollars.

35

Table 2.2. Payment accuracy under the 2017 and the 2018 HHS-HCC models for enrollees who would have received RXCs
and who would not

Mean Risk Score Predictive Ratio
Mean Plan
Spending ($) N
2017
Model
2018
Model
2017
Model
2018
Model
0 HCCs 0.4 0.4 0.81 0.78 2,143 4,110,875
0 HCC and only unassociated RXC(s) 0.5 3.6 0.14 1.08 14,655 17,705
1 HCC 3.0 2.6 1.16 1.06 10,656 639,454
1 HCC and only associated RXC(s) 2.8 4.0 0.69 1.05 16,816 145,939
1 HCC and only unassociated RXC(s) 3.6 5.3 0.58 0.91 25,926 6,086
2 HCCs 7.4 6.3 1.28 1.17 23,841 119,138
2 HCCs and only associated RXC(s) 6.5 7.6 0.86 1.08 31,303 45,160
2 HCCs and only unassociated RXC(s) 8.9 9.7 0.91 1.06 40,385 1,468
3+ HCCs 20.9 18.0 1.35 1.24 64,309 47,683
3+ HCCs and only associated RXC(s) 20.7 20.1 0.98 1.02 87,063 27,038
3+ HCCs and only unassociated RXC(s) 30.2 28.0 1.02 1.00 123,131 974

Note: The sample includes commercially insured enrollees ages 21 to 64 that enrolled for the full year and did not switch plans in
2017. A hierarchical condition category (HCC) Group was counted as one single HCC. The two severity-only prescription drug
categories (RXCs) were excluded from the analysis. Individuals with “associated RXC(s)” were those taking medications in
RXC(s) associated with the HCC(s) they already had. Individuals with “unassociated RXC(s)” were those taking medications in
RXC(s) not associated with the HCC(s) they had. By definition, individuals with zero HCCs did not receive associated RXCs.
Enrollee subgroups with both associated and unassociated RXCs were excluded due to small group size. All spending is in 2017
dollars.
36

Table 2.3. Payment accuracy for subgroups with six prevalent and expensive conditions associated with RXCs

Mean Risk Score Predictive Ratio
Mean Plan
Spending ($)

N 2017 Model 2018 Model 2017 Model 2018 Model
HIV - HCC only 13.2 4.6 2.58 0.95 21,087 994
HIV - RXC only 1.1 7.1 0.21 1.49 20,894 1,275
HIV - HCC and RXC 11.2 11.3 1.15 1.25 40,137 10,879
Multiple Sclerosis - HCC only 17.3 9.8 2.00 1.21 35,943 4,547
Multiple Sclerosis - RXC only 1.1 17.2 0.08 1.26 60,367 323
Multiple Sclerosis - HCC and RXC 15.4 21.1 0.82 1.20 77,762 6,560
Diabetes - HCC only 3.8 3.1 1.26 1.09 12,630 145,534
Diabetes - RXC only 3.3 3.8 0.69 0.85 19,604 15,695
Diabetes - HCC and RXC 4.4 4.5 0.92 0.99 20,089 160,915
Hepatitis C - HCC only 14.8 11.7 1.44 1.22 42,496 11,502
Hepatitis C - RXC only 3.3 29.5 0.18 1.70 76,916 291
Hepatitis C - HCC and RXC 7.9 33.8 0.36 1.63 91,258 1,656
Heart Arrhythmia - HCC only 8.8 7.9 1.26 1.21 28,984 43,303
Heart Arrhythmia - RXC only 6.9 6.3 0.83 0.81 34,298 1,620
Heart Arrhythmia - HCC and RXC 10.1 9.1 0.93 0.90 44,886 9,778
Chronic Kidney Disease - HCC only 24.6 21.4 1.36 1.26 75,202 7,121
Chronic Kidney Disease - RXC only 19.0 17.4 0.89 0.87 87,955 107
Chronic Kidney Disease - HCC and RXC 49.8 44.3 0.55 0.52 377,274 1,972

Note: The sample includes commercially insured enrollees ages 21 to 64 that enrolled for the full year and did not switch plans in
2017. An individual was considered as having a condition if she had either the hierarchical condition category (HCC) or the
associated prescription drug category (RXC). All spending is in 2017 dollars.

37

2.9 Appendix
Appendix Table 2.1. The twelve RXCs included in the 2018 HHS-HCC model

Prescription Drug Category (RXC) Use Condition
Anti-HIV agents Imputation and severity HIV
Anti-hepatitis c (HCV) agents Imputation and severity Hepatitis C
Antiarrhythmics Imputation and severity Heart arrhythmia
Phosphate binders Imputation and severity Chronic kidney disease
Inflammatory bowel disease agents Imputation and severity Inflammatory bowel disease
Insulin Imputation and severity Diabetes
Anti-diabetic agents, except insulin
and metformin
Imputation and severity Diabetes
Multiple sclerosis agents Imputation and severity Multiple Sclerosis
Immune suppressants and
immunomodulators
Imputation and severity Inflammatory bowel
disease, rheumatoid
arthritis, and lupus
Cystic fibrosis agents Imputation and severity Cystic fibrosis
Ammonia detoxicants Severity-only N/A
Diuretics, loop and select
potassium-sparing
Severity-only N/A

Source: HHS Notice of Benefit and Payment Parameters for 2018 (CMS, 2016c).
Note: The last two RXCs were excluded from our analyses.

38

Appendix Table 2.2. Comparison of risk score increases under the 2017 and the 2018 HHS-
HCC models for an enrollee in various scenarios

HIV Multiple Sclerosis
2017 Model 2018 Model 2017 Model 2018 Model
HCC only 8.450 0.409 13.194 6.087
RXC only N/A 6.050 N/A 16.192
Both HCC and RXC 8.450 8.862 13.194 19.105

Source: HHS Notices of benefit and payment parameters for benefit years 2017 and 2018 (CMS,
2016b; CMS, 2016c).

Note: This table uses HIV and multiple sclerosis as two examples to illustrate risk score
increases in various scenarios for a gold plan enrollee. Under the 2018 HHS-HCC model, an
individual with both the HCC and the RXC receives an additional interaction term, whose
coefficient can be either positive or negative. In the case of HIV, the score of the interaction term
is 2.403, and thus under the 2018 model, an individual with both the HCC and the RXC receives
a score of 8.862 (0.409+6.050+2.403). The case of multiple sclerosis is similar, except that the
interaction term has a negative coefficient (−3.174).
39

Appendix Figure 2.1. RXC-only patients' status in the previous year (2016)

Note: The sample includes enrollees with any of the six conditions who in 2017 only had the
associated prescription drug categories (RXCs) and who continuously enrolled in plans offered
by the insurer in both 2016 and 2017. HCC stands for hierarchical condition category.

40

Appendix Table 2.3. Partial-year enrollees who would have received RXCs and who would not under the 2018 HHS-HCC
model

Months Enrolled Full-Year
Enrollees
1-3 4-6 7-9 10-11
0 HCCs 93.15% 87.33% 81.96% 82.24% 79.62%
0 HCC and only unassociated RXC(s) 0.79% 0.58% 0.45% 0.36% 0.34%
1 HCC 4.12% 7.87% 10.81% 10.98% 12.39%
1 HCC and only associated RXC(s) 0.80% 1.75% 2.75% 2.27% 2.83%
1 HCC and only unassociated RXC(s) 0.04% 0.07% 0.10% 0.10% 0.12%
2 HCCs 0.57% 1.20% 1.86% 1.99% 2.31%
2 HCCs and only associated RXC(s) 0.17% 0.42% 0.78% 0.67% 0.87%
2 HCCs and only unassociated RXC(s) 0.01% 0.01% 0.02% 0.02% 0.03%
3+ HCCs 0.25% 0.51% 0.77% 0.86% 0.92%
3+ HCCs and only associated RXC(s) 0.08% 0.23% 0.46% 0.46% 0.52%
3+ HCCs and only unassociated RXC(s) 0.00% 0.01% 0.02% 0.02% 0.02%
N 1,147,274 1,018,063 1,117,051 454,150 5,162,983

Note: The sample includes commercially insured partial-year enrollees ages 21 to 64 that did not switch plans in 2017. A
hierarchical condition category (HCC) Group was counted as one single HCC. The two severity-only prescription drug categories
(RXCs) were excluded from the analysis. Individuals with “associated RXC(s)” were those taking medications in RXC(s)
associated with the HCC(s) they already had. Individuals with “unassociated RXC(s)” were those taking medications in RXC(s)
not associated with the HCC(s) they had. By definition, individuals with zero HCCs did not receive associated RXCs. Enrollee
subgroups with both associated and unassociated RXCs were excluded due to small group size.
41

Appendix Table 2.4. Mean age by level of plan spending of enrollees with 0 HCCs only
under the 2018 HHS-HCC model

Men Women N % Women
Plan spending=0 38.0 39.9 1,032,168 30%
Plan spending between 0 and $500 40.8 40.6 1,292,949 41%
Plan spending between $500 and $2,000 43.6 42.1 948,027 62%
Plan spending greater than $2,000 45.5 44.3 837,731 58%

Note: The sample includes enrollees with 0 HCCs only under the 2018 HHS-HCC Model. All
spending is in 2017 dollars.

42

Appendix Table 2.5. Top 15 diagnoses by share of patients for enrollees with 0 HCCs only under the 2018 HHS-HCC model

All Enrollees with 0 HCCs Only Enrollees with 0 HCCs Only and with Plan Spending>$2,000
Diagnosis ICD-10 % With
this Dx
Diagnosis ICD-10 % With
this Dx
No diagnosis (includes those with no
medical claims)
N/A 28%

Encounter for general adult medical
examination without abnormal findings
Z0000 33%
Encounter for general adult medical
examination without abnormal findings
Z0000 22%

Encounter for gynecological
examination (general) (routine) without
abnormal findings
Z01419 24%
Encounter for gynecological
examination (general) (routine) without
abnormal findings
Z01419 14%

Essential (primary) hypertension I10 22%
Essential (primary) hypertension I10 11% Encounter for immunization Z23 15%
Encounter for immunization Z23 9% Lower back pain M545 14%
Lower back pain M545 6%

Encounter for screening for malignant
neoplasm of colon
Z1211 13%
Acute upper respiratory infection,
unspecified
J069 6%

Hyperlipidemia, unspecified E785 11%
Hyperlipidemia, unspecified E785 5%

Gastro-esophageal reflux disease
without esophagitis
K219 11%
Acute pharyngitis, unspecified J029 5% Neck pain M542 10%
Cough R05 5% Anxiety disorder, unspecified F419 9%
Encounter for screening for malignant
neoplasm of colon
Z1211 4%

Acute upper respiratory infection,
unspecified
J069 9%
Other fatigue R5383 4% Cough R05 9%
Anxiety disorder, unspecified F419 4% Other fatigue R5383 9%
Gastro-esophageal reflux disease
without esophagitis
K219 4%

Vitamin D deficiency, unspecified E559 8%
Vitamin D deficiency, unspecified E559 4% Acute pharyngitis, unspecified J029 8%

Note: The sample includes enrollees with 0 HCCs only under the 2018 HHS-HCC Model.
43

Appendix Table 2.6. Comparison of sample characteristics of the study sample and the
adult calibration sample for the 2014 HHS-HCC model

Study Sample Calibration Sample
Men 21 to 24 4.2% 3.8%
Men 25 to 29 5.4% 4.3%
Men 30 to 34 6.2% 4.8%
Men 35 to 39 6.4% 5.2%
Men 40 to 44 6.0% 5.6%
Men 45 to 49 6.3% 6.0%
Men 50 to 54 6.1% 6.2%
Men 55 to 59 5.9% 5.8%
Men 60+ 4.8% 5.8%
Women 21 to 24 3.9% 4.0%
Women 25 to 29 4.9% 4.7%
Women 30 to 34 5.7% 5.3%
Women 35 to 39 6.0% 5.6%
Women 40 to 44 5.6% 6.1%
Women 45 to 49 6.0% 6.7%
Women 50 to 54 5.9% 7.0%
Women 55 to 59 5.8% 6.5%
Women 60+ 4.6% 6.5%
With 0 HCCs 80.0% 80.8%
With 1 HCC 15.3% 15.2%
With 2+ HCCs 4.7% 4.0%
N 5,162,983 14,220,503

Note: The sample includes commercially insured enrollees ages 21 to 64 that enrolled for the full
year and did not switch plans in 2017. A hierarchical condition category (HCC) Group was
counted as one single HCC.

44

Chapter 3. Unsubsidized and Subsidized Part D Beneficiaries Face Different
Financial Incentives, and Respond Differently to Preferred Pharmacy
Networks

Abstract

The percentage of Medicare stand-alone prescription drug plans with a preferred pharmacy
network has grown from less than 9 percent in 2011 to 98 percent in 2021. These plans use lower
cost-sharing to steer patients to preferred pharmacies. We simulated the potential savings for
beneficiaries of only using preferred versus non-preferred pharmacies, and empirically assessed
their use of pharmacies and its financial consequences. Unsubsidized beneficiaries faced
substantial incentives – on average $147 annually in out-of-pocket spending – and moderately
switched towards preferred pharmacies, while subsidized beneficiaries were insulated from the
incentives and demonstrated little switching. Among those who continued to mainly use non-
preferred pharmacies (half of the unsubsidized, and about two thirds of the subsidized), on
average, the unsubsidized paid more out-of-pocket ($94) relative to using preferred pharmacies,
while Medicare bore the extra spending ($170) for the subsidized through cost-sharing subsidy.
Further research is needed on the impact on the quality of beneficiaries’ decision-making and
cost savings to fully evaluate preferred networks.
45

3.1 Introduction
Prescription drug prices have received considerable policy attention recently. However,
despite the intense policy interest in lowering drug prices, there has been very little attention paid
to the variation in drug prices across different pharmacies, though evidence suggests that this
variation is large (Arora et al., 2017; Gellad et al., 2009; Luo et al., 2019), not unlike the price
differences observed across hospitals, physicians, and other types of providers.
There is a long history of health plans attempting to steer patients to lower-cost providers
through preferred provider networks (Cutler et al., 2000), including a robust literature that
identifies two primary mechanisms through which they contain costs: financially incentivizing
patients to choose lower-cost, higher-value physicians and hospitals (Frank et al., 2015; Gruber
& McKnight, 2016; Prager, 2020; Sinaiko & Rosenthal, 2014), and giving plans more leverage
in their negotiations with providers (Gowrisankaran et al., 2015; Ho & Lee, 2017; Town &
Vistnes, 2001). Additionally, recent efforts targeting value-based insurance design have
attempted to promote the use of certain high-value therapeutic classes or generic drugs through
the use of lower cost-sharing, favorable tier placement, or formulary exclusion of brand-name
drugs (Agarwal et al., 2018; Dusetzina et al., 2020; Hoadley et al., 2012; Socal et al., 2019).
However, less is known about the impact of efforts to encourage patients to use lower-cost
pharmacies.
In recent years, preferred pharmacy networks have emerged as a tool for plans to
encourage the use of lower-cost pharmacies. The percentage of Medicare Part D stand-alone
prescription drug plans (PDPs) with a preferred pharmacy network has grown rapidly from less
than 9 percent in 2011 to 98 percent in 2021 (Figure 3.1). The use of preferred networks has also
been expanding in other markets (Fein, 2017). In such networks, plans essentially divide retail
46

pharmacies into two tiers: non-preferred (or “standard”) pharmacies and preferred pharmacies,
which presumably offer greater price concessions. In exchange, plans presumably offer higher
expected patient volume by incentivizing patients to use preferred pharmacies through lower
cost-sharing than they would face at non-preferred pharmacies. Appendix Table 3.1 illustrates
copay changes across tiers in two of the major Part D plans that introduced a preferred network.
Preferred networks may have also contributed to the recent growth of direct and indirect
remuneration (DIR) in Part D, which includes manufacturer rebates and pharmacy fees received
by plans (Centers for Medicare and Medicaid Services, 2017). Coincidental with the emergence
of preferred networks, pharmacy fees increased from $229 million in 2013 to $9.1 billion in
2019. Accordingly, their share of total DIR rose from less than 2 percent to 18 percent during the
same period (Fein, 2020).
Despite their considerable growth in market share and potential implications for Part D
spending, little is known about the impact of preferred pharmacy networks in Part D. While one
prior study found a small shift towards preferred pharmacies and modest cost savings (Starc and
Swanson, 2019), it did not evaluate the financial incentives beneficiaries faced and the financial
consequences of their pharmacy choice. In this paper, using more recent Medicare claims data,
we simulated the potential savings for beneficiaries of only using preferred versus non-preferred
pharmacies, and empirically assessed their use of pharmacies and its financial consequences.

47

3.2 Study Data and Methods
3.2.1 Data and Sample
This study used beneficiary summary, Part D plan characteristics and tier cost-sharing
files, and prescription drug event claims from a nationally representative 20 percent sample of
Medicare beneficiaries from 2010 through 2016. The analyses focused on beneficiaries 65 years
and older and in PDPs with a preferred network during 2011-2016, the period of dramatic
increase in preferred network penetration among PDPs. We categorized beneficiaries as LIS if
they received Medicare low-income subsidy for premium and cost-sharing, and required that
beneficiaries’ LIS status remained unchanged during the year. Additional criteria included
enrolling in the same plan for the entire year and filling at least one prescription.
We augmented these data with the First Databank drug database and Part D pharmacy
network files obtained from Centers for Medicare and Medicaid Services (CMS). The drug
database provided information on whether a National Drug Code (NDC) is generic or brand-
name and its active ingredient. For each Part D plan in a given year, the pharmacy network files
contained the National Provider Identifier of every network pharmacy and indicators for whether
the pharmacy was a retail or mail-order pharmacy and for whether it was preferred. Additionally,
to illustrate the most recent trend of preferred network implementation, we collected Part D
landscape files and plan benefit design data from CMS for 2017-2021.
3.2.2 Financial Incentives
We evaluated the financial incentives for using preferred pharmacies by simulating
beneficiaries’ annual out-of-pocket (OOP) spending differentials between using non-preferred
and preferred pharmacies for all their prescriptions. Besides differential copays, predicted cost
48

differences for an individual depend on their other plan characteristics such as deductible and
gap coverage, their utilization of drugs, and their LIS status. To account for all these factors, we
computed beneficiaries’ total OOP spending over the course of a year if all their retail claims had
been at non-preferred pharmacies and if all had been at preferred pharmacies. We assumed that
substitution as a result of preferred networks occurred only among retail pharmacies, and thus
the use of mail-order pharmacies was unaffected.
11

We took as given point-of-sale prices and drug tier cost-sharing at preferred and non-
preferred pharmacies within a plan. Thus, we first constructed empirical formularies containing
each plan-NDC’s average prices and copays/coinsurance rates separately for preferred, non-
preferred, and mail-order pharmacies, information on whether the plan-NDC was generic or
branded, whether deductible applied to the plan-NDC, and whether it was covered in the
coverage gap. We then ran each beneficiary’s claims through the formulary of the specific plan
and plan characteristics including deductible, initial coverage limit, and catastrophic coverage
threshold, and calculated total claim costs and patient OOP payment for using a preferred and a
non-preferred pharmacy for each retail claim. For LIS beneficiaries, we applied cost-sharing
rules specific to each eligibility category and additionally computed low-income cost-sharing
subsidy (LICS) amounts for different pharmacies. See the appendix for more detail about the
simulation.
3.2.3 Analysis
We first assessed the financial incentives for using preferred instead of non-preferred
pharmacies. To do so, we first pooled all beneficiary-years in PDPs with a preferred network

11
Market share of mail-order pharmacy fills was low and only increased slightly during the study period.
49

from 2011 through 2016. We identified such plans by checking whether their networks were
divided into preferred and non-preferred pharmacies and whether patient cost-sharing differed
accordingly. We then examined separately for non-LIS and LIS beneficiaries the distribution of
patient OOP spending differentials between using non-preferred and preferred pharmacies only.
We excluded beneficiary-years with extreme values. That is, those with a differential in the top
or bottom 0.5 percent of the distribution. For the LIS, we also assessed the distribution the LICS
differentials. We computed the mean actual OOP spending and LICS amounts in order to
provide context for the distributions.
We then evaluated whether facing these financial incentives, beneficiaries shifted their
pharmacy use. Specifically, this analysis focused on comparing beneficiaries’ use of preferred
pharmacies in the year prior and plans’ first year of preferred networks, because we observed
substantial year-to-year changes in pharmacies’ preferred status within a plan. To do so, we
further restricted the sample to beneficiaries who remained in their plan when it implemented a
preferred network and whose LIS status and ZIP code remained unchanged during the two years.
There were 1,230,532 beneficiary-years, among which 28 percent were LIS. Then, for each two-
year pairs, we mapped pharmacies’ preferred status in a given plan in the post-period to the pre-
period, and calculated two measures of preferred pharmacy use each year for non-LIS and LIS
beneficiaries: preferred pharmacies’ claim share and the percentage of beneficiaries filling the
majority of their prescriptions at preferred pharmacies. Next, to obtain pre-period and post-
period means for these measures, we weighted the first measure by total number of claims, and
the second measure by total number of beneficiaries.
Because we found moderate switching towards preferred pharmacies among the non-LIS
and little switching among the LIS, we examined the amount of money beneficiaries left on the
50

table, based on their pharmacy use. We categorized non-LIS and LIS beneficiaries as “majority
preferred” if they filled the majority of their prescriptions in a given year at preferred
pharmacies. We categorized them as “majority non-preferred” and “majority mail-order”
similarly.
12
For each group, we then assessed the distribution of forgone savings for beneficiaries
and Medicare. We calculated forgone savings, or overspending, for a beneficiary as the
difference between their actual OOP spending and their simulated OOP spending if they had
used preferred pharmacies for all retail claims, holding drug utilization fixed. Similarly, for LIS
beneficiaries, we also computed forgone savings for Medicare as the difference between the
actual LICS and the simulated subsidy if the beneficiary had used preferred pharmacies for all
retail claims.
3.2.4 Limitations
By basing on current-year claims in simulating beneficiaries’ spending in different
pharmacy use situations, we implicitly assumed that pharmacy choice – and cost-sharing
resulting from that – had no impact on utilization. We did a sensitivity analysis by simulating
based on prior-year claims for a subset of the sample who were in the same plan during each
two-year period, and the results were similar. In addition, the analyses of pharmacy switching
focused on plans that transitioned to a preferred network, thus excluding new plans that entered
Part D with a preferred network. Nevertheless, due to the high level of inertia in Part D plan
enrollment,
13
the enrollment in the latter plans was only one fifth of that in the former. The

12
We excluded beneficiaries who did not use any type of pharmacies for the majority of their claims,
because they accounted for only 3 percent of the sample.
13
High inertia has been well-documented in Part D. Hoadley et al. (2013) reported that an annual average
of only 13 percent of Part D beneficiaries switched plans during 2006-2010. Abaluck and Gruber (2016)
and Heiss et al. (2016) found a rate of less than 10 percent during similar periods.
51

analyses also excluded beneficiaries who exited their plan at the implementation of a preferred
network. During the study period, for the non-LIS, the rate was slightly higher among majority
non-preferred beneficiaries than among majority preferred (8.9 percent versus 7.8 percent).
14
To
the extent that plan switching negatively correlated with the willingness to switch pharmacies,
the analyses might slightly overestimate the switching towards preferred pharmacies.

3.3 Study Results
The financial incentives for using preferred pharmacies differed for non-LIS and LIS
beneficiaries. As Table 3.1 shows, over the course of a year, the OOP spending differential
between filling all prescriptions at non-preferred and preferred pharmacies was sizable for about
half of non-LIS beneficiaries. On average, the non-LIS saw a differential of $147 (23 percent of
their mean OOP spending). The distribution was right skewed, with a median of $98 and a 75th
percentile of $214, which suggests that a quarter of non-LIS beneficiaries could have large
savings by using preferred pharmacies only. In contrast, because their cost-sharing was set by
law rather than by their plan, differential copays in preferred networks had minimal
consequences for LIS beneficiaries, with 48 percent of them having zero OOP spending
differential. Instead, Medicare shouldered those differentials, which equaled $190 on average for
the LICS program (10 percent of the mean LICS amount).
Facing these financial incentives, non-LIS beneficiaries moderately increased their use of
preferred pharmacies in the first year of preferred networks, while LIS beneficiaries

14
We chose not to report the rates among LIS beneficiaries because: (1) their plan switching was unlikely
to be related to their pharmacy preference, and (2) CMS automatically assign a large percentage of them
to basic plans with a premium below the regional benchmark, making their plan switching difficult to
interpret.
52

demonstrated minimal pharmacy switching. Among non-LIS beneficiaries in the year prior to
preferred network implementation, 28.5 percent of claims were filled at pharmacies that would
go on to become preferred pharmacies in the following year (we note that there was no particular
financial incentive to use these “to-be-preferred” pharmacies in the pre-implementation year).
After implementation of preferred pharmacy networks, this share increased by 3.2 percentage
points (Figure 3.2). In contrast, among LIS beneficiaries, only 22.5 percent of claims were filled
“to-be-preferred” pharmacies in the pre-implementation year, which was relatively unchanged
post-implementation. The patterns were similar for the percentage of beneficiaries using
preferred pharmacies for more than half of their prescription fills.
The use of non-preferred pharmacies was linked to significant foregone savings for non-
LIS beneficiaries and Medicare. Among the non-LIS, majority preferred and mail-order
beneficiaries had minimal forgone savings, with the means close to $0 and the 75th percentiles
around $20, while majority non-preferred beneficiaries (51 percent of the sample) overspent
considerably by continuing to visit non-preferred pharmacies (Table 3.2). Had they used
preferred pharmacies, they could have on average saved $94, which amounted to 12 percent of
their average OOP spending, and a quarter of them left $172 or more on the table. In comparison,
LIS beneficiaries using non-preferred pharmacies (64 percent of the sample) incurred
overspending to Medicare instead of themselves. Relative to the situation where they switched to
preferred pharmacies, Medicare on average paid an extra of $170 in cost-sharing subsidy (8
percent of the mean LICS amount) for them, and the amount reached $275 or more for a quarter
of them. As a result, LICS bore a higher share of the costs for them than for majority preferred
beneficiaries (42 percent versus 38 percent).

53

3.4 Discussion
We found that non-LIS and LIS beneficiaries faced different financial incentives in
preferred pharmacy networks and had different pharmacy switching responses. For the non-LIS,
the incentives created by the differential copays were substantial, with an average beneficiary
seeing a difference of $147 in their annual OOP spending between using non-preferred and
preferred pharmacies. In the first year of preferred networks, preferred pharmacies’ share of
claims increased moderately (from 28.5 to 31.7 percent) among those who stayed in their plan. In
contrast, because their cost-sharing was capped statutorily, the LIS faced minimal incentive and
did not seem to switch towards preferred pharmacies.
One reason why we observed moderate pharmacy switching among the non-LIS may be
that we focused on plans’ first year of preferred network implementation. The magnitude may
grow as beneficiaries’ exposure and experience increase over time (Prager, 2020; Sinaiko &
Mehrotra, 2020). Besides that, some beneficiaries might fail to fully understand the impact on
their spending, or lack convenient access to preferred pharmacies. Some might also have a higher
willingness to pay for convenience, services, and the relationship with their pharmacist.
Consequently, among non-LIS beneficiaries who remained in their plan when a preferred
network was implemented, about half were worse off relative to those using other types of
pharmacies, mainly because their pharmacy became non-preferred. They had forgone savings of
$94 on average, which was 12 percent of their mean OOP spending, and a quarter of them had
$172 or more.
Put in the broader context of Part D beneficiaries’ decision-making, the forgone savings
also raise concerns about preferred networks causing additional complexity for beneficiaries. The
mean forgone savings here is about a quarter of that due to less-than-optimal plan choice in Part
54

D (Abaluck & Gruber, 2016; Heiss et al., 2016; Ketcham et al., 2015; Zhou & Zhang, 2012).
Admittedly, not choosing a cost-minimizing pharmacy does not necessarily imply a poor choice
by a beneficiary, since they may have non-pecuniary considerations, as discussed above.
Nevertheless, the comparison underscores the financial significance of pharmacy choice in the
environment of preferred networks: a beneficiary may leave a nontrivial amount of money on the
table if they fail to account for their pharmacy’s preferred status. Therefore, such networks add
another layer of complexity to beneficiaries’ decision-making, when they are already having
difficulty navigating a complex menu of options (Freed et al., 2020; Koma et al., 2019), and the
year-to-year variation we observed of a pharmacy’s preferred status even within a plan would
exacerbate that. Indeed, preferred networks caused much confusion to beneficiaries in their early
days (Government Accountability Office, 2014). It is thus necessary for future research to assess
how these networks impact the quality of beneficiaries’ plan and pharmacy choices, since the
prior literature on Part D plan choice has only covered the periods before the emergence of
preferred networks, and it is important for policymakers to consider that when weighing the pros
and cons of preferred networks.
For roughly two thirds of the LIS beneficiaries who remained in their plan when a
preferred network was implemented, LICS bore the financial burden as a result of their use of
non-preferred pharmacies. Medicare overspent an average of $170 in LICS (8 percent of the
mean subsidy amount) for them. This likely had an effect on the growing expenditures of the
low-income subsidy program (Medicare Payment Advisory Commission, 2021), although not as
large as that of the rising drug prices, and the significance of this overspending may increase,
given recent proposals to expand the eligibility for the program (Cubanski et al., 2020). In
addition, our results of LICS taking up a higher share of the costs of non-preferred pharmacy
55

users suggest that preferred networks may allow plans to offload some risk of the spending of
LIS beneficiaries to Medicare, since they seldom switch to preferred pharmacies. This issue
relates to the broader discussion on the decreasing plan liability and proposals to realign the
incentives to manage spending in Part D (Medicare Payment Advisory Commission, 2020b;
Trish, 2019). Preferred networks’ net effect on government spending on Part D though, would
also depend on the magnitude of the cost savings.
But it is unclear how much cost savings preferred networks generated and how much was
passed through to beneficiaries. Despite moderate pharmacy switching, other evidence seems to
suggest that preferred networks have been lucrative for plans. Based on point-of-sale prices,
Starc and Swanson (2019) estimated cost savings of around 2 percent as a result of such
networks. The fast-growing pharmacy DIR fees may be the other important component of the
cost savings. However, we were unable to causally estimate the cost savings because detailed
pharmacy fee data were proprietary. Additionally, similar to the case of manufacturer rebates, it
is unclear to what extent higher fees benefited beneficiaries, although they did save some money
at the pharmacy counter when visiting preferred pharmacies. Therefore, it is important for
researchers and policymakers to uncover the magnitude of cost savings from preferred pharmacy
networks and the distribution of the savings among beneficiaries, Medicare, and plans.

3.5 Conclusion
The emergence of preferred pharmacy networks is an important new development in
Medicare Part D. The financial incentives to use preferred pharmacies, pharmacy switching, and
the financial consequences of continuing to use non-preferred pharmacies differed between non-
LIS and LIS beneficiaries. Further research is needed on their impact on the quality of
56

beneficiaries’ decision-making and cost savings in order to fully assess the costs and benefits of
preferred pharmacy networks.

3.6 References
Abaluck, J., & Gruber, J. (2016). Evolving Choice Inconsistencies in Choice of Prescription
Drug Insurance. American Economic Review, 106(8), 2145–2184.
https://doi.org/10.1257/aer.20130778
Agarwal, R., Gupta, A., & Fendrick, A. M. (2018). Value-Based Insurance Design Improves
Medication Adherence Without An Increase In Total Health Care Spending. Health
Affairs, 37(7), 1057–1064. https://doi.org/10.1377/hlthaff.2017.1633
Arora, S., Sood, N., Terp, S., & Joyce, G. (2017). The Price May Not Be Right: The Value of
Comparison Shopping for Prescription Drugs. American Journal of Managed Care,
23(7), 410–415.
Chernew, M. E., Rosen, A. B., & Fendrick, A. M. (2007). Value-Based Insurance Design. Health
Affairs, 26(Supplement 2), w195–w203. https://doi.org/10.1377/hlthaff.26.2.w195
Cubanski, J., Freed, M., Neuman, T., & Damico, A. (2020). Options to Make Medicare More
Affordable For Beneficiaries Amid the COVID-19 Pandemic and Beyond.
https://www.kff.org/report-section/options-to-make-medicare-more-affordable-for-
beneficiaries-amid-the-covid-19-pandemic-and-beyond-report/
Dusetzina, S. B., Cubanski, J., Nshuti, L., True, S., Hoadley, J., Roberts, D., & Neuman, T.
(2020). Medicare Part D Plans Rarely Cover Brand-Name Drugs When Generics Are
Available. Health Affairs, 39(8), 1326–1333. https://doi.org/10.1377/hlthaff.2019.01694
Fein, A. J. (2017). Yes, Commercial Payers Are Adopting Narrow Retail Pharmacy Networks.
https://www.drugchannels.net/2017/01/yes-commercial-payers-are-adopting.html
Freed, M., Koma, W., Cubanski, J., Biniek, J. F., & Neuman, T. (2020). More Than Half of All
People on Medicare Do Not Compare Their Coverage Options Annually.
https://www.kff.org/medicare/issue-brief/more-than-half-of-all-people-on-medicare-do-
not-compare-their-coverage-options-annually/
Gellad, W. F., Choudhry, N. K., Friedberg, M. W., Brookhart, M. A., Haas, J. S., & Shrank, W.
H. (2009). Variation in Drug Prices at Pharmacies: Are Prices Higher in Poorer Areas?
57

Health Services Research, 44(2p1), 606–617. https://doi.org/10.1111/j.1475-
6773.2008.00917.x
Government Accountability Office. (2014). Medicare Part D: CMS has implemented processes
to oversee Plan Finder pricing accuracy and improve website usability.
https://www.gao.gov/assets/670/660081.pdf
Heiss, F., McFadden, D., Winter, J., Wuppermann, A., & Zhou, B. (2016). Inattention and
Switching Costs as Sources of Inertia in Medicare Part D (Working Paper No. 22765).
National Bureau of Economic Research. https://doi.org/10.3386/w22765
Hoadley, J. F., Merrell, K., Hargrave, E., & Summer, L. (2012). In Medicare Part D Plans, Low
Or Zero Copays And Other Features To Encourage The Use Of Generic Statins Work,
Could Save Billions. Health Affairs, 31(10), 2266–2275.
https://doi.org/10.1377/hlthaff.2012.0019
Hoadley, J., Hargrave, E., & Summer, L. (2013). To Switch or Not to Switch: Are Medicare
Beneficiaries Switching Drug Plans To Save Money?
https://www.kff.org/medicare/issue-brief/to-switch-or-not-to-switch-are-medicare-
beneficiaries-switching-drug-plans-to-save-money/
Ketcham, J. D., Lucarelli, C., & Powers, C. A. (2015). Paying Attention or Paying Too Much in
Medicare Part D. American Economic Review, 105(1), 204–233.
https://doi.org/10.1257/aer.20120651
Koma, W., Cubanski, J., Jacobson, G., Damico, A., & Neuman, T. (2019). No Itch to Switch:
Few Medicare Beneficiaries Switch Plans During the Open Enrollment Period.
https://www.kff.org/medicare/issue-brief/no-itch-to-switch-few-medicare-beneficiaries-
switch-plans-during-the-open-enrollment-period/
Luo, J., Kulldorff, M., Sarpatwari, A., Pawar, A., & Kesselheim, A. S. (2019). Variation in
Prescription Drug Prices by Retail Pharmacy Type. Annals of Internal Medicine, 171(9),
605–611. https://doi.org/10.7326/M18-1138
Medicare Payment Advisory Commission. (2019). March 2019 Report to the Congress:
Medicare Payment Policy.
Medicare Payment Advisory Commission. (2021). March 2021 Report to the Congress:
Medicare Payment Policy. http://www.medpac.gov/docs/default-
source/reports/mar21_medpac_report_to_the_congress_sec.pdf?sfvrsn=0
58

Medicare Payment Advisory Commission. (2020b). June 2020 Report to the Congress:
Medicare and the Health Care Delivery System. http://medpac.gov/docs/default-
source/reports/jun20_reporttocongress_sec.pdf?sfvrsn=0
Medicare Payment Advisory Commission. (2020a). March 2020 Report to the Congress:
Medicare Payment Policy.
Prager, E. (2020). Healthcare Demand under Simple Prices: Evidence from Tiered Hospital
Networks. American Economic Journal: Applied Economics, 12(4), 196–223.
https://doi.org/10.1257/app.20180422
Sinaiko, A. D., & Mehrotra, A. (2020). Association of a national insurer’s reference-based
pricing program and choice of imaging facility, spending, and utilization. Health Services
Research, 55(3), 348–356. https://doi.org/10.1111/1475-6773.13279
Socal, M. P., Bai, G., & Anderson, G. F. (2019). Favorable Formulary Placement of Branded
Drugs in Medicare Prescription Drug Plans When Generics Are Available. JAMA
Internal Medicine, 179(6), 832–833. https://doi.org/10.1001/jamainternmed.2018.7824
Starc, A., & Swanson, A. (2018). Preferred Pharmacy Networks and Drug Costs (Working
Paper No. 24862). National Bureau of Economic Research.
https://doi.org/10.3386/w24862
Trish, E. E. (2019). Medicare Part D: Time for Re-Modernization? Health Services Research,
54(6), 1174–1183. https://doi.org/10.1111/1475-6773.13221
Zhou, C., & Zhang, Y. (2012). The Vast Majority Of Medicare Part D Beneficiaries Still Don’t
Choose The Cheapest Plans That Meet Their Medication Needs. Health Affairs, 31(10),
2259–2265. https://doi.org/10.1377/hlthaff.2012.0087

59

Figure 3.1. Percentage of Medicare Stand-Alone Prescription Drug Plans With a Preferred
Pharmacy Network

Source: Authors’ analysis of Part D prescription drug plan characteristics and tier cost-sharing
files from a 20 percent sample of Medicare beneficiaries (2011-2016), and Part D landscape files
and plan benefit design data (2017-2021).

Notes: The sample includes stand-alone plans offered in 50 states and DC.

0%
20%
40%
60%
80%
100%
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
60

Table 3.1. Simulated Spending Differentials Over the Course of a Year Between Filling All
Prescriptions at Non-Preferred Pharmacies and Preferred Pharmacies

Spending Category
Distribution of Differentials
Mean Actual
Amount N Mean P50 P75 P90
Non-LIS OOP spending ($) 147 98 214 364 641 6,400,330
LIS OOP spending ($) 5 0 4 20 75 1,545,436
LIS LICS ($) 190 103 275 516 1,837 1,545,436

Source: Authors’ analysis of beneficiary summary, Part D prescription drug plan characteristics,
tier cost-sharing, and prescription drug event files from a 20 percent sample of Medicare
beneficiaries (2010-2016), the First Databank drug database, and Part D pharmacy network files
(2011-2016).

Notes: Out-of-pocket (OOP), low-income cost-sharing subsidy (LICS). Low-income subsidy
(LIS) beneficiaries comprise those receiving both partial and full benefits. The sample includes
beneficiaries who were 65 years and older, who remained in the same stand-alone plan with a
preferred network for the entire year, whose LIS status stayed unchanged during the year, and
who filled at least one prescription. The table pools all beneficiary-years from 2011 through
2016. Spending differentials are based on current year’s claims and are reported in nominal
amounts. The table excludes those with a simulated OOP differential in the top or bottom 0.5
percent of the distribution. We also ran simulations based on prior year’s claims for a subset of
the sample who were in the same plan during each two-year period, and the results were similar.

61

Figure 3.2. Use of Preferred Pharmacies Pre- and Post-Preferred Pharmacy Networks, by
LIS Status

Source: Authors’ analysis of beneficiary summary, Part D prescription drug plan characteristics,
tier cost-sharing, and prescription drug event files from a 20 percent sample of Medicare
beneficiaries (2010-2016), the First Databank drug database, and Part D pharmacy network files
(2011-2016).

Notes: Low-income subsidy (LIS) beneficiaries comprise those receiving both partial and full
benefits. The sample includes beneficiaries who were 65 years and older, who remained in their
stand-alone plan when it implemented a preferred network, whose LIS status and ZIP code
remained unchanged during the two years, and who filled at least one prescription in both years.
Each bar represents the weighted average across all pre-periods or post-periods during 2010-
2016.

28.5
22.5
31.7
22.0
Non-LIS LIS
Preferred Claim Share (%)
The Year Prior
The First Year of Preferred Networks
27.5
26.2
31.0
26.4
Non-LIS LIS
Percent Using Preferred
Pharmacies for the Majority of
Claims (%)
62

Table 3.2. Forgone Savings in the First Year of Preferred Networks as a Result of Using
Non-Preferred Pharmacies

Non-LIS

Majority
Preferred
Majority
Non-Preferred
Majority
Mail-Order
Beneficiary Overspending ($)
Mean −14 94 5
Median 0 85 4
75th Percentile 19 172 22
Mean OOP Spending ($) 677 803 513
Mean Total Costs ($) 2,253 2,309 1,947
N 289,538 460,702 137,266
LIS
Beneficiary Overspending ($)
Mean 0 6 1
Median 0 0 0
75th Percentile 2 11 2
LICS Overspending ($)
Mean −19 170 16
Median 0 111 7
75th Percentile 40 275 36
Mean OOP Spending ($) 91 93 97
Mean LICS ($) 1,637 2,084 1,862
Mean Total Costs ($) 4,350 4,945 4,800
N 62,913 144,351 3,768

Source: Authors’ analysis of beneficiary summary, Part D prescription drug plan characteristics,
tier cost-sharing, and prescription drug event files from a 20 percent sample of Medicare
beneficiaries (2010-2016), the First Databank drug database, and Part D pharmacy network files
(2011-2016).

Notes: Out-of-pocket (OOP), low-income cost-sharing subsidy (LICS). Low-income subsidy
(LIS) beneficiaries comprise those receiving both partial and full benefits. The sample includes
beneficiaries who were 65 years and older, who remained in their stand-alone plan when it
implemented a preferred network, whose LIS status and ZIP code remained unchanged during
the two years, and who filled at least one prescription in both years. The table excludes those
with a simulated OOP differential in the top or bottom 0.5 percent of the distribution. The
“majority preferred” group includes beneficiaries who filled the majority of their prescriptions in
a given year at preferred pharmacies. The “majority non-preferred” and “majority mail-order”
groups are defined similarly. The table excludes beneficiaries who did not use any type of
pharmacies for the majority of their claims, because they accounted for only 3 percent of the
sample. Forgone savings, or overspending, for a beneficiary is the difference between their actual
OOP spending and their simulated OOP spending if they had used preferred pharmacies for all
retail claims, based on current year’s claims. For LIS beneficiaries, forgone savings for Medicare
63

is the difference between the actual LICS and the simulated subsidy if the beneficiary had used
preferred pharmacies. The table pools all beneficiary-years from 2011 through 2016. Forgone
savings and spending are reported in nominal amounts.

64

3.7 Appendix

Appendix Table 3.1. Cost-Sharing Changes in Two of the Major Plans That Transitioned
to a Preferred Network

AARP MedicareRx Preferred Cigna Medicare Rx Secure
2012 2013 2013 2014
Tier
All
Retail Preferred
Non-
preferred Tier
All
Retail Preferred
Non-
preferred
1 $4 $3 $6 1 $0 $0 $10
2 $8 $5 $10 2 $8 $5 $33
3 $41 $40 $45 3 $29.5 $32.5 $45
4 $95 $85 $95 4 $80 $82 $95
5 33% 33% 33% 5 25% 25% 25%

Source: Authors’ analysis of Part D prescription drug plan characteristics and tier cost-sharing
files from a 20 percent sample of Medicare beneficiaries, 2010-2016.

Notes: The AARP plan implemented a preferred network in 2013, while the Cigna plan
implemented in 2014. Copays are for 30-day prescriptions. The table presents the median if the
copay for a certain tier varies across regions. The drugs on a certain tier may vary between years.

Appendix Table 3.2. Regressions Assessing the Quality of Simulated Variables

Non-LIS LIS
Total
spending
OOP
spending
Total
spending
OOP
spending LICS
Simulated variable
0.987*** 0.918*** 0.987*** 0.929*** 0.985***

(0.000) (0.000) (0.000) (0.000) (0.000)
Constant term
−4.680*** 20.794*** 10.052*** 3.697*** 3.444***

(0.512) (0.134) (1.580) (0.036) (0.478)
R
2

0.959 0.923 0.959 0.911 0.958
N
4,053,937 4,053,937 1,205,117 1,205,117 1,205,117

Source: Authors’ analysis of beneficiary summary, Part D prescription drug plan characteristics,
tier cost-sharing, and prescription drug event files from a 20 percent sample of Medicare
beneficiaries (2010-2016), the First Databank drug database, and Part D pharmacy network files
(2011-2016).

Notes: * p<0.05, ** p<0.01, *** p<0.001. Out-of-pocket (OOP), low-income cost-sharing
subsidy (LICS). Low-income subsidy (LIS) beneficiaries comprise those receiving both partial
and full benefits. The sample includes beneficiaries who were 65 years and older, who remained
65

in the same stand-alone plan with a preferred network for the entire year, whose LIS status
stayed unchanged during the year, who filled at least one prescription, and who used only
preferred or non-preferred pharmacies for all claims. The regressions pool all beneficiary-years
from 2011 through 2016.

Simulating Out-of-Pocket Spending for Using Only Preferred Pharmacies and Using Only
Non-Preferred Pharmacies

Building Formularies
We built empirical NDC-level formularies for all plan-NDCs that appear in the prior year
or the first year of preferred networks, by taking point-of-sale prices and tier cost sharing as
exogenous to beneficiaries’ pharmacy choice. For each plan-NDC, we assembled the following
information: average 30-day prices at preferred, non-preferred, and mail-order pharmacies,
generic/brand indicator, and tier assignment and the corresponding cost sharing information in
pre-ICL and coverage gap at preferred, non-preferred, and mail-order pharmacies.
For prices, ideally we calculated insurer-benefit type-NDC-pharmacy type level prices.
But claims might be lacking for a certain plan-NDC at different types of pharmacies. We thus
computed the following measures in descending order of preference to obtain prices for missing
cells:
- Insurer-NDC-pharmacy type level prices
- Insurer-benefit type-clinical formulation ID-GNI-pharmacy type level prices by linking
to the First Databank database
15

- Insurer-clinical formulation ID-GNI-pharmacy type level prices

15
Clinical formulation ID is an identifier that is specific to a combination of active ingredient, dosage
form, strength, and route of administration. GNI is a brand/generic indicator.
66

- Average insurer-clinical formulation ID-GNI level prices across claims at any types of
pharmacies
For tier assignment, in an ideal case we collected insurer-benefit type-NDC level
information. However, there were plan-NDCs with missing tier assignment, which accounted for
a small fraction of all claims. For these plan-NDCs, we used the following measures in
descending order of preference:
- Insurer-benefit type-clinical formulation ID-GNI level
- Insurer-benefit type-American Hospital Formulary Service (AHFS) class-GNI level
With plan and tier identifiers, we linked plan-NDCs to the tier files to obtain tier cost
sharing information.
For generic/brand indicator, we used the plan-submitted indicator in the claim files. We
used First Databank’s GNI if it was missing.

Simulation
We ran each beneficiary’s claims through the formulary of the specific plan and plan
characteristics including deductible, initial coverage limit, and catastrophic coverage threshold,
and calculated total claim costs and patient out-of-pocket payment (OOP) in different scenarios
for each claim. First, for mail-order claims, we computed the total costs of a claim as (average
30-day prices at mail-order pharmacies × days supplied)/30, and for retail claims, we computed
the total costs at a preferred/non-preferred pharmacy as (average 30-day prices at preferred/non-
preferred pharmacies × days supplied)/30.

67

Then, we identified the benefit phase and compute patient OOP with an iterative process.
Below we describe the simulation using as an example the scenario where a beneficiary used
preferred pharmacies only for retail claims. Before we proceed, we define the acronyms for
benefit phases in Part D benefit design: D (deductible), P (pre-ICL), I (coverage gap), C
(catastrophic coverage). We characterize the benefit phase of a claim with the start phase and the
end phase. For example, a “DD” claim is entirely within the deductible. The vast majority of
claims fall into one phase, but the so-called “straddle claims” cross phases. For instance, a “PI”
claim starts in the pre-ICL phase and ends in the coverage gap.

Non-LIS
We used cumulative drug costs and true out-of-pocket (TrOOP) spending to determine
benefit phases. First, we computed patient OOP for claims whose benefit phase could be
completely determined with the former: DD, DP, and PP claims. We applied the plan’s cost
sharing rules for preferred pharmacies for the specific days supplied in each phase. When the
cost sharing type was copay, for non-standard days supplied (days other than 30, 60, and 90), we
used the 30-day copay if it was smaller than 30; we used (30-day copay × days supplied)/30 if it
was greater than 30. For straddle claims, we applied the cost sharing rules in the start phase to
the portion of the costs that fell into the start phase, and rules in the end phase to the portion that
fell into the end phase, subject to the constraint that the total OOP cannot exceed the total costs.
Second, we worked on DI/DC and PI/PC claims, whose phase was partially determined
with cumulative drug costs. We computed tentative patient OOP, assuming that their end phase
was the coverage gap (I). We then distinguished DI from DC and PI from PC based on whether
68

the cumulative TrOOP with the tentative OOP exceeded the catastrophic coverage threshold, and
adjusted the patient OOP accordingly.
Third, we took similar steps for II/IC claims, and processed CC claims in the end.

LIS
For LIS beneficiaries, we first applied cost sharing rules specific to each eligibility group
and computed their OOP payment. We then calculated what the patient OOP would have been if
they had been non-LIS. The LICS amount would then be equal to the non-LIS OOP minus the
LIS OOP.

Quality Check
To check the quality of the simulation, we ran beneficiary-year level regressions of actual
spending variables on simulated ones. We pooled beneficiaries who either used preferred or non-
preferred pharmacies only, and separately for LIS and non-LIS beneficiaries, regressed actual
total drug costs, out-of-pocket spending, and LICS (for LIS only) on their simulated
counterparts. For those who used preferred pharmacies only, the independent variable took the
simulated amount for preferred pharmacies, and for those who used non-preferred only, it took
that for non-preferred pharmacies. Hence ideally, the independent variable would have a
coefficient of one, the constant term would be zero, and the R
2
would equal one. Appendix Table
3.2 demonstrates that the results were very close to those in an ideal case.

69

Chapter 4. Medicare Part D Beneficiaries’ Pharmacy Switching in Response
to Preferred Pharmacy Networks

Abstract

Between 2011 and 2021, the percentage of Medicare stand-alone prescription drug plans
with a preferred pharmacy network has increased from less than nine percent to over 98 percent.
Such networks use lower cost-sharing to steer patients to the subset of preferred pharmacies.
Using the 20 percent Medicare claims data from 2010 through 2016, we examined the design of
preferred networks, their effects on the use of preferred pharmacies, and factors underlying
beneficiaries’ decisions of whether to switch to preferred pharmacies. We found that preferred
networks generally designated a restrictive set of preferred pharmacies. Unsubsidized
beneficiaries faced an average financial incentives of $129 to use these pharmacies, while
subsidized beneficiaries were insulated. Consequently, the implementation of preferred networks
on average resulted in a 3.7-percentage point increase in preferred pharmacies’ claim share in the
first year among the non-LIS. Existing relationships with preferred pharmacies, financial
incentives, access to preferred pharmacies, and urban residence were positively correlated with
beneficiaries’ decisions to switch to these pharmacies.

70

4.1 Introduction
Health plans have long been using preferred provider networks to steer patients to lower-
cost providers (Cutler et al., 2000). These networks contain costs by financially incentivizing
patients to choose lower-cost, higher-value physicians and hospitals (Frank et al., 2015; Gruber
& McKnight, 2016; Prager, 2020; Sinaiko & Rosenthal, 2014), and giving plans more leverage
in their negotiations with providers (Gowrisankaran et al., 2015; Ho & Lee, 2017; Town &
Vistnes, 2001). In addition, recent value-based insurance design efforts have attempted to
promote the use of high-value therapeutic classes and generic drugs with tools such as cost-
sharing, tier placement, and formulary coverage (Agarwal et al., 2018; Dusetzina et al., 2020;
Hoadley et al., 2012; Socal et al., 2019). However, at the same time, pharmacies offer different
value, and there is considerable price variation across pharmacies for the same drug (Arora et al.,
2017; Gellad et al., 2009; Luo et al., 2019). But since patients typically face the same copay for
that drug across all retail pharmacies in the network, they have little incentive to choose the one
with the lowest price.
In recent years, plans have been adopting preferred pharmacy networks to encourage the
use of low-cost pharmacies. Between 2011 and 2021, the percentage of Medicare stand-alone
prescription drug plans (PDPs) with a preferred pharmacy network has increased from less than
nine percent to over 98 percent (authors’ calculation). The use of preferred networks has also
been spreading in Medicare Advantage drug plans (authors’ calculation) and commercial
markets (Fein, 2017). In these networks, plans divide their retail pharmacies into two tiers: non-
preferred (or “standard”) pharmacies and preferred cost-sharing pharmacies (PCSPs). They
establish lower cost-sharing at preferred pharmacies as a price signal to steer patients to these
pharmacies.
71

Despite the importance of their potential impact, the prior literature is thin on patient
choice related to Part D preferred networks. One form of preferred physician and hospital
networks, tiered provider networks, bear particular resemblance to preferred pharmacy networks.
Research on tiered provider networks has generally shown patients’ preference for providers
with which they have existing relationships and a moderate response to financial incentives
(Frank et al., 2015; Sinaiko & Rosenthal, 2014). However, there are reasons to think that patients
may respond differently to preferred pharmacy networks. On one hand, the money at stake for
them is generally less in a pharmacy visit, compared with a physician or hospital visit. They also
visit pharmacies more frequently than physicians (Berenbrok et al., 2020), and thus
inconvenience as a result of switching pharmacies may add up over time. On the other hand,
pharmacies are generally viewed as more substitutable than physicians and hospitals. Yet to our
knowledge, Starc and Swanson (2019) conducted the only empirical study so far of Part D
preferred networks. They found that preferred networks were associated with lower point-of-sale
prices, and results from their pharmacy demand model revealed a small shift towards preferred
pharmacies and moderate cost savings as a result of these networks. However, the model was at
an aggregate level and estimated average effects.
In this study, we add to the literature on preferred pharmacy networks by evaluating
network restrictiveness and financial incentives, estimating the effects on beneficiaries’
pharmacy choices, and more importantly, focusing on patients most likely to be affected by the
policy – unsubsidized non-preferred pharmacy users – to assess factors underlying their
heterogeneous responses, such as existing relationships with preferred pharmacies, size of
financial incentives, access to preferred pharmacies, and urban/rural residence.

72

4.2 Data and Methods
4.2.1 Preferred Pharmacy Networks in PDPs
More than 47 million beneficiaries obtained prescription drug coverage through a
Medicare Part D plan in 2020 (Medicare Payment Advisory Commission, 2021). The majority of
Part D beneficiaries enroll in PDPs, although Medicare Advantage plans (including prescription
drug benefits) are becoming more popular over time.
In recent years, preferred pharmacy networks have emerged as a tool for plans to curb
rising prescription drug costs. Such networks use financial incentives created by differential cost-
sharing to steer patients to their preferred pharmacies. Appendix Table 4.1 illustrates cost-
sharing pre- and post-implementation for major plans – plans with an enrollment of 25,000 or
more in our 20 percent Medicare beneficiary sample – that implemented a preferred network
during our study period of 2011-2016. Differential cost-sharing (often copays) generally applied
to non-specialty tiers, which often were Tiers 1 through 4 in a five-tier formulary. Copay
strategies varied across plans: some plans – such as AARP MedicareRx Preferred, Humana
Enhanced, and First Health Part D Essentials – established moderate copay differentials, while
some – such as Blue MedicareRx Plus/Premier, WellCare Classic, and WellCare Extra –
instituted large differentials between preferred and non-preferred pharmacies. For instance, the
differential was $3 for a 30-day prescription of a Tier 1, preferred generic drug in the AARP plan
in 2013, while that could be as large as $8 in WellCare Classic in 2014. However, as discussed
below, the size of financial incentives for a particular patient would also depend on some other
factors.
Plans enter preferred pharmacy arrangements with pharmacy chains or individual or
groups of independent pharmacies (Fein, 2015). The breadth of the subset of preferred
pharmacies is not subject to Part D’s network adequacy standards, which apply to a pharmacy
73

network as a whole. Access to preferred pharmacies could thus be restrictive. This has prompted
the Centers for Medicare and Medicaid Services (CMS) to monitor and publish plans’ access
levels and to require those with extremely restrictive preferred networks to disclose that
information in their marketing materials (CMS, 2017). Additionally, the composition of a
preferred network may change over time. For instance, First Health’s two plans dropped CVS
and added Walgreens as their preferred pharmacies in 2016 (Fein, 2015; Fein, 2016).
In a preferred network, beneficiaries face a tradeoff between convenience and cost
savings: those using non-preferred pharmacies can switch pharmacies, or they can opt for
sticking with their pharmacies and paying more out-of-pocket. The financial incentives are
substantially weaker for beneficiaries who receive Medicare’s low-income cost-sharing subsidies
(LICS), since their cost-sharing is capped statutorily regardless of what pharmacy they use.
4.2.2 Data and Sample
In this study, we used beneficiary summary, Part D plan characteristics and tier cost-
sharing files, and prescription drug event claims from a nationally representative 20 percent
sample of Medicare beneficiaries from 2010 through 2016. To facilitate before-and-after
comparisons, we examined PDPs that transitioned to a preferred network during 2011-2016, a
period where preferred network penetration increased from less than nine percent to 85 percent.
We focused on plans’ first year of preferred networks, because we observed considerable year-
to-year changes in pharmacies’ preferred status within a plan. We referred to the year of
implementation for a given plan as Year t hereafter, and the year prior to that as Year t−1.
We restricted our main sample to include beneficiaries 65 years and older who enrolled in
the same plan in Year t−1 and Year t, and whose LIS status and ZIP code remained unchanged
during the two-year period. We defined LIS beneficiaries as those who received Medicare low-
74

income subsidy for premium and cost-sharing. The sample comprised 1,230,532 beneficiary-
years, among which 28 percent were LIS.
We complemented these data with the First Databank drug database and pharmacy
network files (2011-2016) obtained from CMS. The drug database provided information on
whether a National Drug Code (NDC) is generic or brand-name and its active ingredient. For
each Part D plan in a given year, the pharmacy network files contained the National Provider
Identifier of every pharmacy included in the network, its ZIP code, and indicators for whether
the pharmacy was a retail or mail-order pharmacy and whether it was preferred. In addition, we
obtained the 2010-2014 American Community Survey 5-year estimates data for beneficiaries’
ZIP code-level socioeconomic conditions, such as educational attainment, income level, and
vehicle ownership. Following Medicare Payment Advisory Commission (2012), we also used the
urban influence code developed by the U.S. Department of Agriculture to categorize each
beneficiary’s county of residence as a metropolitan, rural micropolitan, rural adjacent, or rural
nonadjacent county.
4.2.3 The Design of Preferred Networks
We first evaluated the tradeoff faced by beneficiaries by examining the restrictiveness of
preferred networks and the financial incentives to use preferred pharmacies as a result of the
differential copays. For network restrictiveness, the analysis focused on the nine major plans,
which accounted for three quarters of our sample population. We calculated the percentage of
network pharmacies designated as preferred in each region for each plan, both unweighted and
weighted by the number of claims submitted by each pharmacy in Year t−1.
We assessed financial incentives by simulating beneficiaries’ annual out-of-pocket
(OOP) spending differentials between using non-preferred and preferred pharmacies for all their
75

prescriptions. The differentials depend on differential copays, as well as other plan
characteristics such as deductible and gap coverage, their utilization of drugs, and their LIS
status. To account for all these factors, we computed beneficiaries’ total OOP spending over the
course of a year if all their retail claims had been at non-preferred pharmacies and if all had been
at preferred. Given that mail-order pharmacies’ overall market share was low and only increased
slightly during the study period, we assumed that substitution caused by preferred networks
occurred only among retail pharmacies, and thus the use of mail-order pharmacies was
unaffected.
Taking point-of-sale prices and tier cost-sharing as exogenous to beneficiaries’ pharmacy
choices, we first constructed empirical plan-NDC level formularies containing cost-sharing and
average prices at preferred, non-preferred, and mail-order pharmacies. We then ran each
beneficiary’s claims through the formulary of the specific plan and plan characteristics including
deductible, initial coverage limit, and catastrophic coverage threshold, and calculated total claim
costs and patient OOP payment for using a preferred and a non-preferred pharmacy for each
claim. For LIS beneficiaries, we applied cost-sharing rules specific to each eligibility category.
See the appendix for more detail about the simulation.
4.2.4 The Average Effects on the Use of Preferred Pharmacies
We then examined the average effects of preferred networks on beneficiaries’ switching
towards preferred pharmacies, with difference-in-differences models comparing the use of
preferred pharmacies pre- and post-implementation among non-LIS and LIS beneficiaries. The
latter served as the control group, since they were largely insulated from differential copays. For
each pre- and post-implementation year pairs, we mapped each pharmacy’s preferred status in a
given plan in Year t to Year t−1, and computed two outcome measures for each year: preferred
76

pharmacies’ claim share, and whether a beneficiary used preferred pharmacies for the majority of
claims. We estimated the following model:
𝑌 𝑖𝑦
= 𝛼 + 𝛽 𝑁𝑂𝑁𝐿𝐼𝑆 _ 𝑃𝐶 𝑆𝑃 𝑖𝑦
+ 𝛾 𝑖 + 𝜃 𝑦 + 𝜀 𝑖𝑦
,
where i indexes beneficiary-year of implementation, and y indexes year. The key independent
variable, NONLIS_PCSP, indicates treatment status, taking the value of one if the beneficiary
was non-LIS and their plan implemented a preferred network in Year y. The model also includes
beneficiary-year of implementation fixed effects 𝛾 𝑖 and year fixed effects 𝜃 𝑦 . Standard errors
were clustered at the plan-region level.
The identifying assumption was that in the absence of preferred networks, the use of
preferred pharmacies among non-LIS and LIS beneficiaries who stayed in their plans would have
followed a common trend. While we only observed one pre-period, and thus were unable to test
the pre-trends directly, we compared the percentages of beneficiaries changing their top
pharmacy from year to year for non-LIS and LIS in plans transitioning to preferred networks and
plans without a preferred network (Appendix Figure 1). The rate of switching remained flat for
LIS beneficiaries in both types of plans and non-LIS in plans without a preferred network, and it
fluctuated for non-LIS in plans implementing a preferred network, which lends support to the
common trend assumption.
4.2.5 Factors Associated With Beneficiaries’ Pharmacy Switching
Next, we evaluated factors underlying non-LIS beneficiaries’ pharmacy switching in
response to the implementation of preferred networks. This set of analyses focused on those who
did not fill the majority of their prescriptions in Year t−1 at preferred pharmacies. We estimated
a beneficiary-year of implementation-level logistic regression model where the outcome variable
77

was whether a beneficiary switched to using preferred pharmacies for the majority of claims in
Year t.
The model includes four key explanatory variables. First, a set of dummy variables
characterize the beneficiary’s preferred claim share in Year t−1 (0 percent, 0-25 percent, 25-50
percent), as existing relationships with pharmacies may affect their choice in Year t. Second,
simulated OOP spending differential between using non-preferred and using preferred
pharmacies for all of the beneficiary’s retail claims in Year t, which is in 100s of dollars,
captures the financial incentives to use preferred pharmacies. Third, we measured proximity to
preferred pharmacies with standardized preferred pharmacies’ claim share in the beneficiary’s 3-
digit ZIP area within the same plan in Year t−1, assuming that without differential copays,
beneficiaries choose the pharmacies most convenient for them. Fourth, a set of dummies indicate
whether the beneficiary’s county of residence was metropolitan, rural micropolitan, rural
adjacent, or rural nonadjacent. The model also controls for beneficiary demographics, including
age, gender, and whether the beneficiary was a racial/ethnic minority, total prescription drug
costs in Year t−1, the count of chronic conditions in the summary file as of the start of Year t,
and a vector of beneficiary’s ZIP code educational attainment, income, and car ownership levels.
In addition, it includes year of implementation and plan fixed effects, and standard errors were
clustered at the beneficiary level.

4.3 Results
4.3.1 Preferred Network Restrictiveness and Financial Incentives
Plans generally designated a very restrictive subset of pharmacies as preferred, and the
restrictiveness varied across regions. Overall during our study period, preferred pharmacies
78

accounted for just 26 percent of the claims in Year t−1, the year prior to the implementation of
preferred networks. Figure 1 illustrates the percentages of preferred pharmacies across PDP
regions in the year of implementation for the major plans, with each dot representing a region
and each box showing the 25th percentile, the median, and the 75th percentile. In the left panel,
the dots generally cluster between 10 percent and 30 percent, suggesting that depending on the
region, one in ten to three in ten network pharmacies obtained preferred status when plans
adopted a preferred network. The variation across regions within a plan was likely the result of
the geographic distribution of chain pharmacy locations. In addition, the weighted restrictiveness
in the right panel was similar to the unweighted except for EnvisionRxPlus Silver, whose
preferred pharmacies had lower-than-average market shares in the previous year. The overall
narrowness of preferred networks suggests that a large number of beneficiaries faced a tradeoff
between sticking with their old pharmacies and paying more out-of-pocket, and beneficiaries in
regions at the more restrictive end likely experienced limited access to preferred pharmacies.
Beneficiaries faced heterogeneous financial incentives to use preferred pharmacies. Table
4.1 shows the distribution of OOP spending differentials between filling all prescriptions at non-
preferred and preferred pharmacies for non-LIS and LIS beneficiaries. We excluded beneficiary-
years with a differential in the top or bottom 0.5 percent of the distribution. The non-LIS saw an
average differential of $129 (18 percent of their mean OOP spending). The median was $93,
which suggests that the differential was sizable for about half of them, and the 75th percentile
was $185. By contrast, because LIS beneficiaries’ cost-sharing was set statutorily, preferred
networks’ differential copays had minimal consequences for them, with 47 percent of them
having zero OOP spending differential.
79

The size of financial incentives did seem to affect beneficiaries’ pharmacy switching. For
the major plans, while the sample was dominated by the two largest plans with moderate
incentives and pharmacy switching – Humana Enhanced and AARP MedicareRx Preferred –
those with a larger mean OOP spending differential tended to experience more switching
towards preferred pharmacies among the non-LIS (Table 4.2). Notably, WellCare Extra
beneficiaries faced the largest incentives on average ($632) and switched to preferred pharmacies
at the highest rate. In comparison, LIS beneficiaries, facing minimal incentives, hardly increased
their use of preferred pharmacies (Table 4.3).
4.3.2 The Average Effects on the Use of Preferred Pharmacies
Table 4.3 presents unadjusted and adjusted difference-in-differences for the use of
preferred pharmacies. The top panel shows that preferred pharmacies’ claim share among the
non-LIS increased from 28.5 percent to 31.7 percent between Years t−1 and t, while that among
the LIS remained roughly unchanged. The difference between changes among the two groups
was thus 3.8 percentage points. Similarly, for the percentage of beneficiaries filling the majority
of prescriptions at preferred pharmacies, the difference between changes was 3.4 points.
The difference-in-differences estimates indicate that during the study period, the
implementation of preferred networks caused a 3.7-percentage point increase in preferred
pharmacies’ claim share in the first year among non-LIS beneficiaries. It also resulted in a 3.8-
point increase in the likelihood of using preferred pharmacies for the majority of one’s claims.
4.3.3 Factors Associated With Beneficiaries’ Pharmacy Switching
Appendix Table 4.2 compares the characteristics of non-LIS beneficiaries who switched
to filling the majority of prescriptions at preferred pharmacies and those who did not. In Table
4.4, we report logistic regression results as marginal effects when holding other covariates at
80

their means. Compared with those who never used preferred pharmacies in the year prior,
beneficiaries who did were 14.7 percentage points more likely to become “majority preferred” in
Year t. A $100 increase in OOP spending differential was associated with a 1.4-point increase in
the likelihood of switching to “majority preferred”. In addition, a one-standard deviation (19-
point) increase in the preferred claim share in a beneficiary’s 3-digit ZIP area in Year t−1 was
correlated with a 0.9-point increase in their likelihood of switching. The rurality of a
beneficiary’s residence also played a role. Compared with those in metropolitan (urban) counties,
beneficiaries in rural micropolitan, rural adjacent, and rural nonadjacent counties were 0.7, 1.9,
and 2.5 points less likely to switch, respectively.

4.4 Discussion
The dramatic rise of preferred pharmacy networks in recent years is an important yet
understudied development in Medicare Part D. Such networks use lower patient cost-sharing to
encourage the use of lower-cost pharmacies. This study is one of the first to examine
beneficiaries’ pharmacy choices in the context of preferred networks.
In examining the tradeoff presented by preferred networks, we found that the subset of
preferred pharmacies was generally very narrow. Non-LIS beneficiaries on average faced a
financial incentive of $129 to use preferred pharmacies, while LIS had minimal incentive to do
so. Consequently, we estimated that during 2011-2016, the implementation of preferred networks
on average caused a 3.7-percentage point increase in preferred pharmacies’ claim share in the
first year among the non-LIS. The effect was in line with the estimate of Starc and Swanson
(2019).
81

Moreover, we evaluated factors underlying beneficiaries’ decisions of whether to switch
to preferred pharmacies, since the moderate average effect might mask the impact of plans’
heterogeneous preferred network designs. Consistent with the research on tiered physician and
hospital networks (Frank et al., 2015; Prager, 2020; Sinaiko & Rosenthal, 2014), our results
suggest that patients with existing relationships with preferred pharmacies were more likely to
switch their prescriptions there, and beneficiaries responded to stronger financial incentives.
Given patients’ response to financial incentives and that plans with larger incentives
seemed to see more switching, it is unclear why several of the major plans in our sample
established moderate copay differentials between preferred and non-preferred pharmacies.
Notably, they were the predominant type when weighted by enrollment. One explanation may be
that plans need to balance cost savings and beneficiary satisfaction. Plans with more leverage
may be able to extract the same price concessions from pharmacies with smaller copay
differentials, which would cause less distaste among their beneficiaries. Future research may
examine plans’ copay strategies through the lens of plan-pharmacy bargaining.
Our analyses also revealed that beneficiaries with greater access to preferred pharmacies
and those in urban areas were more likely to switch to preferred pharmacies. Since using non-
preferred pharmacies would incur additional OOP spending for patients, to maintain equity,
CMS should continue to monitor and ensure access to preferred pharmacies, especially in rural
areas and disadvantaged communities, where access to pharmacies is already lacking (Qato et al.,
2014).
This study has several limitations. First, we only covered the first year of preferred
networks in our analyses, due to substantial year-to-year changes in a pharmacy’s preferred
status within a certain plan. We were thus unable to examine the role of inattention in
82

beneficiaries’ pharmacy switching. It is also possible that more patients would switch to
preferred pharmacies over time as their exposure and experience grow (Prager, 2020; Sinaiko &
Mehrotra, 2020), although preferred status changes may cause confusion and increase errors.
Second, our simulated OOP spending differentials were based on Year t claims, and did
not account for the impact of cost-sharing on utilization. While the demand for prescription
drugs is relatively inelastic (Einav et al., 2018), we simulated with prior-year claims as a
sensitivity analysis, and the results were similar.
Third, our analyses excluded beneficiaries who switched plans between Years t−1 and t.
The rate of plan switching was slightly higher among non-LIS beneficiaries who used non-
preferred pharmacies for the majority of their claims (8.9 percent) than among those who used
preferred (7.8 percent). This might produce a small upward bias in our estimates of the average
effect of preferred networks, to the extent that exiting a plan was negatively associated with the
willingness to switch pharmacies.
Fourth, in order to facilitate before-and-after comparisons, our sample of plans did not
include new plans that entered the PDP market with a preferred network. Because of the inertia
in Part D plan enrollment, their combined enrollment in the first year was only one fifth of the
size of our sample. However, further research may evaluate these plans, since beneficiaries may
pay more attention to the pharmacy network when they are new to a plan, or they may pick a
plan based on preferences for pharmacies.
Fifth, our measure of access to preferred pharmacies in a patient’s neighborhood – their
claim share in the year prior to preferred networks – did not capture individual patients’ travel
time to preferred pharmacies. Sixth, our data did not distinguish between pharmacy chains and
independent pharmacies. We thus did not control for pharmacy brand in the analyses.
83

In summary, our results show that PDP preferred pharmacy networks led to moderate
switching towards the narrow subset of preferred pharmacies among unsubsidized, non-LIS
beneficiaries, and that existing relationships with preferred pharmacies, the size of financial
incentives, proximity to a preferred pharmacy, and urban residence were positively associated
with their decisions to switch. Given the rise of preferred networks in both Part D and
commercial drug plans, researchers and policymakers should seek to better understand plans’
strategies and to assess whether communities have equitable access to preferred pharmacies.

4.5 References
Agarwal, R., Gupta, A., & Fendrick, A. M. (2018). Value-Based Insurance Design Improves
Medication Adherence Without An Increase In Total Health Care Spending. Health
Affairs, 37(7), 1057–1064. https://doi.org/10.1377/hlthaff.2017.1633
Arora, S., Sood, N., Terp, S., & Joyce, G. (2017). The Price May Not Be Right: The Value of
Comparison Shopping for Prescription Drugs. American Journal of Managed Care,
23(7), 410–415.
Berenbrok, L. A., Gabriel, N., Coley, K. C., & Hernandez, I. (2020). Evaluation of Frequency of
Encounters With Primary Care Physicians vs Visits to Community Pharmacies Among
Medicare Beneficiaries. JAMA Network Open, 3(7), e209132–e209132.
https://doi.org/10.1001/jamanetworkopen.2020.9132
Centers for Medicare & Medicaid Services. (2017). Announcement of Calendar Year (CY) 2018
Medicare Advantage Capitation Rates and Medicare Advantage and Part D Payment
Policies and Final Call Letter and Request for Information.
https://www.cms.gov/Medicare/Health-
Plans/MedicareAdvtgSpecRateStats/Downloads/Announcement2018.pdf
Chernew, M. E., Rosen, A. B., & Fendrick, A. M. (2007). Value-Based Insurance Design. Health
Affairs, 26(Supplement 2), w195–w203. https://doi.org/10.1377/hlthaff.26.2.w195
Dusetzina, S. B., Cubanski, J., Nshuti, L., True, S., Hoadley, J., Roberts, D., & Neuman, T.
(2020). Medicare Part D Plans Rarely Cover Brand-Name Drugs When Generics Are
Available. Health Affairs, 39(8), 1326–1333. https://doi.org/10.1377/hlthaff.2019.01694
84

Einav, L., Finkelstein, A., & Polyakova, M. (2018). Private Provision of Social Insurance: Drug-
Specific Price Elasticities and Cost Sharing in Medicare Part D. American Economic
Journal: Economic Policy, 10(3), 122–153. https://doi.org/10.1257/pol.20160355
Fein, A. J. (2015). EXCLUSIVE: How the Eight Biggest Retail Chains (and Independents) Are
Participating in 2016’s Part D Preferred Networks.
https://www.drugchannels.net/2015/11/exclusive-how-eight-biggest-retail.html
Fein, A. J. (2016). Behind Diplomat Pharmacy’s Plunge: A Primer on DIR Fees in Medicare
Part D. https://www.drugchannels.net/2016/11/behind-diplomat-pharmacys-plunge-
primer.html
Fein, A. J. (2017). Yes, Commercial Payers Are Adopting Narrow Retail Pharmacy Networks.
https://www.drugchannels.net/2017/01/yes-commercial-payers-are-adopting.html
Frank, M. B., Hsu, J., Landrum, M. B., & Chernew, M. E. (2015). The Impact of a Tiered
Network on Hospital Choice. Health Services Research, 50(5), 1628–1648.
https://doi.org/10.1111/1475-6773.12291
Gellad, W. F., Choudhry, N. K., Friedberg, M. W., Brookhart, M. A., Haas, J. S., & Shrank, W.
H. (2009). Variation in Drug Prices at Pharmacies: Are Prices Higher in Poorer Areas?
Health Services Research, 44(2p1), 606–617. https://doi.org/10.1111/j.1475-
6773.2008.00917.x
Hoadley, J. F., Merrell, K., Hargrave, E., & Summer, L. (2012). In Medicare Part D Plans, Low
Or Zero Copays And Other Features To Encourage The Use Of Generic Statins Work,
Could Save Billions. Health Affairs, 31(10), 2266–2275.
https://doi.org/10.1377/hlthaff.2012.0019
Luo, J., Kulldorff, M., Sarpatwari, A., Pawar, A., & Kesselheim, A. S. (2019). Variation in
Prescription Drug Prices by Retail Pharmacy Type. Annals of Internal Medicine, 171(9),
605–611. https://doi.org/10.7326/M18-1138
Medicare Payment Advisory Commission. (2012). Report to the Congress: Medicare and the
Health Care Delivery System (June 2012) (p. 259).
Medicare Payment Advisory Commission. (2021). March 2021 Report to the Congress:
Medicare Payment Policy. http://www.medpac.gov/docs/default-
source/reports/mar21_medpac_report_to_the_congress_sec.pdf?sfvrsn=0
Prager, E. (n.d.). Health Care Demand under Simple Prices: Evidence from Tiered Hospital
Networks. American Economic Journal: Applied Economics.
https://doi.org/10.1257/app.20180422
85

Qato, D. M., Daviglus, M. L., Wilder, J., Lee, T., Qato, D., & Lambert, B. (2014). ‘Pharmacy
Deserts’ Are Prevalent In Chicago’s Predominantly Minority Communities, Raising
Medication Access Concerns. Health Affairs, 33(11), 1958–1965.
https://doi.org/10.1377/hlthaff.2013.1397
Sinaiko, A. D., & Mehrotra, A. (2020). Association of a national insurer’s reference-based
pricing program and choice of imaging facility, spending, and utilization. Health Services
Research, 55(3), 348–356. https://doi.org/10.1111/1475-6773.13279
Sinaiko, A. D., & Rosenthal, M. B. (2014). The Impact of Tiered Physician Networks on Patient
Choices. Health Services Research, 49(4), 1348–1363. https://doi.org/10.1111/1475-
6773.12165
Socal, M. P., Bai, G., & Anderson, G. F. (2019). Favorable Formulary Placement of Branded
Drugs in Medicare Prescription Drug Plans When Generics Are Available. JAMA
Internal Medicine, 179(6), 832–833. https://doi.org/10.1001/jamainternmed.2018.7824
Starc, A., & Swanson, A. (2018). Preferred Pharmacy Networks and Drug Costs (Working
Paper No. 24862). National Bureau of Economic Research.
https://doi.org/10.3386/w24862

86

Figure 1. The Percentages of Preferred Pharmacies Across Regions for Major Plans That
Transitioned to a Preferred Network

Notes: This figure includes plans with 2,5000 or more beneficiaries. Each label on the vertical
axis includes the name of the plan and the year in which it implemented a preferred network.
Year t−1 is the year prior to the implementation. For a certain plan, each dot represents a region,
and each box shows 25th, 50th, and 75th percentiles.

87

Table 4.1. Simulated OOP Spending Differentials Over the Course of a Year Between
Filling All Prescriptions at Non-Preferred Pharmacies and Preferred Pharmacies

Distribution of Differentials ($)
Mean Actual
OOP ($) N Mean P25 P50 P75 P90
Non-LIS 129 31 93 185 299 716 950,100
LIS 8 0 0 8 29 89 238,749

Notes: OOP, out-of-pocket. Low-income subsidy (LIS) beneficiaries comprise those receiving
both partial and full benefits. The sample includes beneficiaries who were 65 years and older,
who remained in their plan when it implemented a preferred network, whose LIS status and ZIP
code remained unchanged during the year prior and the implementation year, and who filled at
least one prescription in both years. The table pools all beneficiary-years from 2011 to 2016.
OOP spending differentials are based on current year’s claims and are reported in nominal
amounts. The table excludes those with a simulated OOP differential in the top or bottom 0.5
percent of the distribution

88

Table 4.2. Changes of Preferred Claim Share and Simulated OOP Spending Differentials
for Non-LIS Beneficiaries in Major Plans

Plan
Change of preferred
claim share between
t−1 and t (p.p.)
Mean simulated
OOP spending
differential ($) N
Humana Enhanced, 2013 −0.2 68 192,722
First Health Part D Essentials, 2014 0.2 101 48,343
Express Scripts Medicare - Value, 2015 9.4 102 12,517
AARP MedicareRx Preferred, 2013 4.0 117 470,921
Cigna Medicare Rx Secure, 2014 4.0 178 16,614
Blue MedicareRx Plus/Premier, 2014 4.6 209 35,549
WellCare Classic, 2014 8.9 215 70,043
EnvisionRxPlus Silver, 2015 1.2 279 3,067
WellCare Extra, 2014 33.8 632 20,659

Notes: OOP, out-of-pocket; LIS, low-income subsidy. This table includes plans with 2,5000 or
more beneficiaries. The sample includes non-LIS beneficiaries who were 65 years and older,
who remained in their plan when it implemented a preferred network, whose LIS status and ZIP
code remained unchanged during the year prior and the implementation year, and who filled at
least one prescription in both years. OOP spending differentials are based on current year’s
claims and are reported in nominal amounts. The table excludes those with a simulated OOP
differential in the top or bottom 0.5 percent of the distribution.

89

Table 4.3. The Average Effects of Preferred Network Implementation on Preferred
Pharmacies' Claim Share and Beneficiaries' Likelihood of Using Preferred Pharmacies for
the Majority of Their Claims

Preferred claim share Being majority preferred
Simple means
Non-LIS LIS Non-LIS LIS
Pre (%) 28.5 22.5 27.5 26.2
Post (%) 31.7 22.0 31.0 26.4
Change (p.p.) 3.2 -0.6 3.6 0.2
DID estimates
0.037*** 0.038***
(0.002) (0.002)
N 1,181,143 1,181,143

Notes: * p<0.05, ** p<0.01, *** p<0.001. DID, difference-in-differences. Low-income subsidy
(LIS) beneficiaries comprise those receiving both partial and full benefits. The sample includes
beneficiaries who were 65 years and older, who remained in their plan when it implemented a
preferred network, whose LIS status and ZIP code remained unchanged during the year prior and
the implementation year, and who filled at least one prescription in both years. Standard errors
are clustered at the plan-region level and are in parentheses.

90

Table 4.4. Factors Associated With Pharmacy Switching

Marginal
effect 95% CI
Ever used preferred pharmacies in the year prior
0.147*** [0.144, 0.150]
Simulated OOP spending differential between preferred and
non-preferred in t (in 100s) 0.014*** [0.013, 0.014]
Standardized preferred claim share in 3-digit ZIP area in t-1
0.009*** [0.009, 0.010]
Rurality of county of residence (ref.=Metropolitan)

Rural micropolitan
−0.007*** [−0.008, −0.005]
Rural adjacent
−0.019*** [−0.021, −0.017]
Rural nonadjacent
−0.025*** [−0.027, −0.023]
N
648,334

Notes: * p<0.05, ** p<0.01, *** p<0.001. The table reports marginal effects when holding other
variables at their means. OOP, out-of-pocket; LIS, low-income subsidy; CI, confidence interval.
The sample includes non-LIS beneficiaries who were 65 years and older, who remained in their
plan when it implemented a preferred network, whose LIS status and ZIP code remained
unchanged during the year prior and the implementation year, and who filled at least one
prescription in both years but did not use preferred pharmacies for the majority of their claims in
Year t−1. We exclude those with a simulated OOP differential in the top or bottom 0.5 percent
of the distribution.
91

4.6 Appendix

Appendix Table 4.1. Plans' Cost-Sharing Pre- and Post-Preferred Pharmacy Networks

AARP MedicareRx Preferred Humana Enhanced
2012 2013 2012 2013
Tier
All
retail Preferred
Non-
preferred Tier
All
retail Preferred
Non-
preferred
1 $4 $3 $6 1 $7 $2 $5
2 $8 $5 $10 2 $38 $5 $7
3 $41 $40 $45 3 $73 $41 $41
4 $95 $85 $95 4 33% $90 $90
5 33% 33% 33% 5 33% 33%

Blue MedicareRx Plus/Premier Cigna Medicare Rx Secure
2013 2014 2013 2014
Tier
All
retail Preferred
Non-
preferred Tier
All
retail Preferred
Non-
preferred
1 $2 $1 $9 1 $0 $0 $10
2 $7 $7 $15 2 $8 $5 $33
3 $45 $40 $45 3 $29.5 $32.5 $45
4 $90 $90 $95 4 $80 $82 $95
5 33% 33% 33% 5 25% 25% 25%
6 33% 33% 33%

First Health Part D Essentials WellCare Classic
2013 2014 2013 2014
Tier
All
retail Preferred
Non-
preferred Tier
All
retail Preferred
Non-
preferred
1 $1 $1 $3 1 $6 $0 $8
2 25% 15% 16% 2 $42 $14 $29
3 45% 44% 45% 3 $94 $40 $45
4 33% $90 $95
5 33% 33%

92

WellCare Extra Express Scripts Medicare - Value
2013 2014 2014 2015
Tier
All
retail Preferred
Non-
preferred Tier
All
retail Preferred
Non-
preferred
1 $0 $0 $8 1 $2 $0 $3
2 25% $0 $29 2 $7 $4 $10
3 25% $38 $45 3 25% 23% 25%
4 50% $68 $95 4 50% 48% 50%
5 33% 33% 33% 5 25% 25% 25%

EnvisionRxPlus Silver
2014 2015
Tier
All
retail Preferred
Non-
preferred
1 $9 $2 $10
2 25% 15% 25%
3 $45 37% 42%
4 45% 25% 25%
5 25%
6 $10

Notes: Copays are for 30-day prescriptions. The table presents the median if the copay for a
certain tier varies across regions. The drugs on a certain tier may vary between years.

93

Appendix Figure 1. The Rate at Which Beneficiaries Changed Their Top Pharmacies
Between Years

Notes: PCSP, preferred cost-sharing pharmacies. Low-income subsidy (LIS) beneficiaries
comprise those receiving both partial and full benefits. The sample includes beneficiaries who
were 65 years and older, who enrolled in the same plan during the two years, whose LIS status
and ZIP code stayed the same, and who filled the majority of their prescriptions at one pharmacy
in each of the two years.

94

Appendix Table 4.2. Characteristics of Non-LIS Beneficiaries Who Switched to Majority
Preferred and Those Who Did Not When Their Plans Implemented Preferred Networks

Whether switched to majority
preferred in Year t

Yes No Overall
Switched to majority preferred in t−1 1 0 0.075
Age in t−1 74.8 76.3 76.2
Female 0.608 0.609 0.609
Nonwhite 0.076 0.069 0.070
Visited a preferred pharmacy in t−1 0.503 0.150 0.176
Simulated OOP spending differential between preferred
and non-preferred in t ($)
171 115 119
Preferred claim share in 3-digit ZIP area in t−1 0.315 0.257 0.261
Total drug costs in t−1 ($) 2,175 2,165 2,166
Number of chronic conditions as of the beginning of t 6.812 7.077 7.057
ZIP code characteristics
Fraction of high school graduates, 65+ 0.826 0.819 0.820
Fraction owning vehicles, households 65+ 0.891 0.888 0.889
Fraction with income below FPL, 65+ 0.081 0.084 0.083
Fraction with income 100-400% FPL, 65+ 0.540 0.545 0.544
Fraction with income above 400% FPL, 65+ 0.380 0.372 0.372
Rurality of county of residence
Metropolitan 0.820 0.747 0.752
Rural micropolitan 0.115 0.134 0.133
Rural adjacent 0.043 0.071 0.069
Rural nonadjacent 0.023 0.048 0.047
Year of preferred network implementation
2011 0.011 0.005 0.006
2012 0.011 0.014 0.013
2013 0.695 0.761 0.756
2014 0.220 0.176 0.179
2015 0.044 0.026 0.027
2016 0.020 0.018 0.018
N 48,444 601,090 649,534

Notes: OOP, out-of-pocket; FPL, federal poverty level. The sample includes non-LIS
beneficiaries who were 65 years and older, who remained in their plan when it implemented a
preferred network, whose LIS status and ZIP code remained unchanged during the year prior and
the implementation year, and who filled at least one prescription in both years but did not use
preferred pharmacies for the majority of their claims in Year t−1. We exclude those with a
simulated OOP differential in the top or bottom 0.5 percent of the distribution.
95

Appendix Table 4.3. Regressions Assessing the Quality of Simulated Variables

Non-LIS LIS
Total
spending
OOP
spending
Total
spending
OOP
spending LICS
Simulated variable
0.970*** 0.913*** 0.986*** 0.927*** 0.989***

(0.000) (0.000) (0.000) (0.001) (0.000)
Constant term
22.371*** 28.533*** 10.309** 4.475*** −2.764*

(1.422) (0.400) (3.939) (0.096) (1.272)
R
2

0.951 0.915 0.957 0.922 0.956
N
600,128 600,128 189,247 189,247 189,247

Notes: * p<0.05, ** p<0.01, *** p<0.001. OOP, out-of-pocket. LICS, low-income cost-sharing
subsidy. Low-income subsidy (LIS) beneficiaries comprise those receiving both partial and full
benefits. The sample includes non-LIS beneficiaries who were 65 years and older, who remained
in their plan when it implemented a preferred network, whose LIS status and ZIP code remained
unchanged during the year prior and the implementation year, who filled at least one prescription
in both years, and who used only preferred or non-preferred pharmacies for all claims.

Simulating Out-of-Pocket Spending for Using Only Preferred Pharmacies and Using Only
Non-Preferred Pharmacies

Building Formularies
We built empirical NDC-level formularies for all plan-NDCs that appear in the prior year
or the first year of preferred networks, by taking point-of-sale prices and tier cost sharing as
exogenous to beneficiaries’ pharmacy choice. For each plan-NDC, we assembled the following
information: average 30-day prices at preferred, non-preferred, and mail-order pharmacies,
generic/brand indicator, and tier assignment and the corresponding cost sharing information in
pre-ICL and coverage gap at preferred, non-preferred, and mail-order pharmacies.
For prices, ideally we calculated insurer-benefit type-NDC-pharmacy type level prices.
But claims might be lacking for a certain plan-NDC at different types of pharmacies. We thus
96

computed the following measures in descending order of preference to obtain prices for missing
cells:
- Insurer-NDC-pharmacy type level prices
- Insurer-benefit type-clinical formulation ID-GNI-pharmacy type level prices by linking
to the First Databank database
16

- Insurer-clinical formulation ID-GNI-pharmacy type level prices
- Average insurer-clinical formulation ID-GNI level prices across claims at any types of
pharmacies
For tier assignment, in an ideal case we collected insurer-benefit type-NDC level
information. However, there were plan-NDCs with missing tier assignment, which accounted for
a small fraction of all claims. For these plan-NDCs, we used the following measures in
descending order of preference:
- Insurer-benefit type-clinical formulation ID-GNI level
- Insurer-benefit type-American Hospital Formulary Service (AHFS) class-GNI level
With plan and tier identifiers, we linked plan-NDCs to the tier files to obtain tier cost
sharing information.
For generic/brand indicator, we used the plan-submitted indicator in the claim files. We
used First Databank’s GNI if it was missing.

Simulation
We ran each beneficiary’s claims through the formulary of the specific plan and plan
characteristics including deductible, initial coverage limit, and catastrophic coverage threshold,

16
Clinical formulation ID is an identifier that is specific to a combination of active ingredient, dosage
form, strength, and route of administration. GNI is a brand/generic indicator.
97

and calculated total claim costs and patient out-of-pocket payment (OOP) in different scenarios
for each claim. First, for mail-order claims, we computed the total costs of a claim as (average
30-day prices at mail-order pharmacies × days supplied)/30, and for retail claims, we computed
the total costs at a preferred/non-preferred pharmacy as (average 30-day prices at preferred/non-
preferred pharmacies × days supplied)/30.
Then, we identified the benefit phase and compute patient OOP with an iterative process.
Below we describe the simulation using as an example the scenario where a beneficiary used
preferred pharmacies only for retail claims. Before we proceed, we define the acronyms for
benefit phases in Part D benefit design: D (deductible), P (pre-ICL), I (coverage gap), C
(catastrophic coverage). We characterize the benefit phase of a claim with the start phase and the
end phase. For example, a “DD” claim is entirely within the deductible. The vast majority of
claims fall into one phase, but the so-called “straddle claims” cross phases. For instance, a “PI”
claim starts in the pre-ICL phase and ends in the coverage gap.

Non-LIS
We used cumulative drug costs and true out-of-pocket (TrOOP) spending to determine
benefit phases. First, we computed patient OOP for claims whose benefit phase could be
completely determined with the former: DD, DP, and PP claims. We applied the plan’s cost
sharing rules for preferred pharmacies for the specific days supplied in each phase. When the
cost sharing type was copay, for non-standard days supplied (days other than 30, 60, and 90), we
used the 30-day copay if it was smaller than 30; we used (30-day copay × days supplied)/30 if it
was greater than 30. For straddle claims, we applied the cost sharing rules in the start phase to
98

the portion of the costs that fell into the start phase, and rules in the end phase to the portion that
fell into the end phase, subject to the constraint that the total OOP cannot exceed the total costs.
Second, we worked on DI/DC and PI/PC claims, whose phase was partially determined
with cumulative drug costs. We computed tentative patient OOP, assuming that their end phase
was the coverage gap (I). We then distinguished DI from DC and PI from PC based on whether
the cumulative TrOOP with the tentative OOP exceeded the catastrophic coverage threshold, and
adjusted the patient OOP accordingly.
Third, we took similar steps for II/IC claims, and processed CC claims in the end.

LIS
For LIS beneficiaries, we first applied cost sharing rules specific to each eligibility group
and computed their OOP payment. We then calculated what the patient OOP would have been if
they had been non-LIS. The LICS amount would then be equal to the non-LIS OOP minus the
LIS OOP.

Quality Check
To check the quality of the simulation, we ran beneficiary-year level regressions of actual
spending variables on simulated ones. We pooled beneficiaries who either used preferred or non-
preferred pharmacies only, and separately for LIS and non-LIS beneficiaries, regressed actual
total drug costs, out-of-pocket spending, and LICS (for LIS only) on their simulated
counterparts. For those who used preferred pharmacies only, the independent variable took the
simulated amount for preferred pharmacies, and for those who used non-preferred only, it took
that for non-preferred pharmacies. Hence ideally, the independent variable would have a
99

coefficient of one, the constant term would be zero, and the R
2
would equal one. Appendix Table
4.3 demonstrates that the results were very close to those in an ideal case.

100

Bibliography
Abaluck, J., & Gruber, J. (2016). Evolving Choice Inconsistencies in Choice of Prescription
Drug Insurance. American Economic Review, 106(8), 2145–2184.
https://doi.org/10.1257/aer.20130778
Agarwal, R., Gupta, A., & Fendrick, A. M. (2018). Value-Based Insurance Design Improves
Medication Adherence Without An Increase In Total Health Care Spending. Health
Affairs, 37(7), 1057–1064. https://doi.org/10.1377/hlthaff.2017.1633
Arora, S., Sood, N., Terp, S., & Joyce, G. (2017). The Price May Not Be Right: The Value of
Comparison Shopping for Prescription Drugs. American Journal of Managed Care,
23(7), 410–415.
Berenbrok, L. A., Gabriel, N., Coley, K. C., & Hernandez, I. (2020). Evaluation of Frequency of
Encounters With Primary Care Physicians vs Visits to Community Pharmacies Among
Medicare Beneficiaries. JAMA Network Open, 3(7), e209132–e209132.
https://doi.org/10.1001/jamanetworkopen.2020.9132
Breyer, F., Bundorf, M. K., & Pauly, M. V. (2011). Health Care Spending Risk, Health
Insurance, and Payment to Health Plans. In Handbook of Health Economics (Vol. 2, pp.
691–762). Elsevier. https://doi.org/10.1016/B978-0-444-53592-4.00011-6
Brown, J., Duggan, M., Kuziemko, I., & Woolston, W. (2014). How Does Risk Selection
Respond to Risk Adjustment? New Evidence from the Medicare Advantage Program.
American Economic Review, 104(10), 3335–3364.
https://doi.org/10.1257/aer.104.10.3335
Centers for Medicare & Medicaid Services. (2016a). March 31, 2016, HHS-Operated Risk
Adjustment Methodology Meeting Discussion Paper.
https://www.cms.gov/CCIIO/Resources/Forms-Reports-and-Other-
Resources/Downloads/RA-March-31-White-Paper-032416.pdf
Centers for Medicare & Medicaid Services. (2016b). Patient Protection and Affordable Care
Act; HHS Notice of Benefit and Payment Parameters for 2017.
https://www.govinfo.gov/content/pkg/FR-2016-03-08/pdf/2016-04439.pdf
Centers for Medicare & Medicaid Services. (2016c). Patient Protection and Affordable Care
Act; HHS Notice of Benefit and Payment Parameters for 2018; Amendments to Special
Enrollment Periods and the Consumer Operated and Oriented Plan Program.
https://www.govinfo.gov/content/pkg/FR-2016-12-22/pdf/2016-30433.pdf
101

Centers for Medicare & Medicaid Services. (2017a). 2017 Benefit Year Risk Adjustment SAS
Version of HHS-Developed Risk Adjustment Model Algorithm Software.
Centers for Medicare & Medicaid Services. (2017b). Announcement of Calendar Year (CY) 2018
Medicare Advantage Capitation Rates and Medicare Advantage and Part D Payment
Policies and Final Call Letter and Request for Information.
https://www.cms.gov/Medicare/Health-
Plans/MedicareAdvtgSpecRateStats/Downloads/Announcement2018.pdf
Centers for Medicare & Medicaid Services. (2017c). Creation of the 2018 Benefit Year HHS-
Operated Risk Adjustment Adult Models Draft Prescription Drug (RXCUIs) to HHS Drug
Classes (RXCs) Crosswalk. https://www.cms.gov/CCIIO/Resources/Regulations-and-
Guidance/Downloads/Draft-RxC-Crosswalk-Memo-9-18-17.pdf
Centers for Medicare & Medicaid Services. (2018a). 2018 Benefit Year Risk Adjustment SAS
Version of HHS-Developed Risk Adjustment Model Algorithm Software.
Centers for Medicare & Medicaid Services. (2018b). Patient Protection and Affordable Care
Act; HHS Notice of Benefit and Payment Parameters for 2019.
https://www.govinfo.gov/content/pkg/FR-2018-04-17/pdf/2018-07355.pdf
Centers for Medicare and Medicaid Services. (2017). Medicare Part D – Direct and Indirect
Remuneration (DIR) | CMS. https://www.cms.gov/newsroom/fact-sheets/medicare-part-
d-direct-and-indirect-remuneration-dir
Claxton, G., Rae, M., Long, M., Damico, A., Foster, G., & Whitmore, H. (2017). 2017 Employer
Health Benefits Survey. http://files.kff.org/attachment/Report-Employer-Health-Benefits-
Annual-Survey-2017
Cubanski, J., Freed, M., Neuman, T., & Damico, A. (2020). Options to Make Medicare More
Affordable For Beneficiaries Amid the COVID-19 Pandemic and Beyond.
https://www.kff.org/report-section/options-to-make-medicare-more-affordable-for-
beneficiaries-amid-the-covid-19-pandemic-and-beyond-report/
Cutler, D. M., McClellan, M., & Newhouse, J. P. (2000). How Does Managed Care Do It? The
RAND Journal of Economics, 31(3), 526. https://doi.org/10.2307/2600999
Cutler, D. M., & Zeckhauser, R. J. (2000). Chapter 11 The anatomy of health insurance. In
Handbook of Health Economics (Vol. 1, pp. 563–643). Elsevier.
https://doi.org/10.1016/S1574-0064(00)80170-5
102

Dorn, S., Garrett, B., & Epstein, M. (2018). New Risk-Adjustment Policies Reduce But Do Not
Eliminate Special Enrollment Period Underpayment. Health Affairs, 37(2), 308–315.
https://doi.org/10.1377/hlthaff.2017.0970
Dusetzina, S. B., Cubanski, J., Nshuti, L., True, S., Hoadley, J., Roberts, D., & Neuman, T.
(2020). Medicare Part D Plans Rarely Cover Brand-Name Drugs When Generics Are
Available. Health Affairs, 39(8), 1326–1333. https://doi.org/10.1377/hlthaff.2019.01694
Einav, L., Finkelstein, A., & Polyakova, M. (2018). Private Provision of Social Insurance: Drug-
Specific Price Elasticities and Cost Sharing in Medicare Part D. American Economic
Journal: Economic Policy, 10(3), 122–153. https://doi.org/10.1257/pol.20160355
Fein, A. J. (2015). EXCLUSIVE: How the Eight Biggest Retail Chains (and Independents) Are
Participating in 2016’s Part D Preferred Networks.
https://www.drugchannels.net/2015/11/exclusive-how-eight-biggest-retail.html
Fein, A. J. (2016). Behind Diplomat Pharmacy’s Plunge: A Primer on DIR Fees in Medicare
Part D. https://www.drugchannels.net/2016/11/behind-diplomat-pharmacys-plunge-
primer.html
Fein, A. J. (2017). Yes, Commercial Payers Are Adopting Narrow Retail Pharmacy Networks.
https://www.drugchannels.net/2017/01/yes-commercial-payers-are-adopting.html
Fein, A. J. (2020). Pharmacy DIR Fees Hit a Record $9 Billion in 2019—That’s 18% of Total
Medicare Part D Rebates. https://www.drugchannels.net/2020/02/pharmacy-dir-fees-hit-
record-9-billion.html
Frank, M. B., Hsu, J., Landrum, M. B., & Chernew, M. E. (2015). The Impact of a Tiered
Network on Hospital Choice. Health Services Research, 50(5), 1628–1648.
https://doi.org/10.1111/1475-6773.12291
Freed, M., Koma, W., Cubanski, J., Biniek, J. F., & Neuman, T. (2020). More Than Half of All
People on Medicare Do Not Compare Their Coverage Options Annually.
https://www.kff.org/medicare/issue-brief/more-than-half-of-all-people-on-medicare-do-
not-compare-their-coverage-options-annually/
Gellad, W. F., Choudhry, N. K., Friedberg, M. W., Brookhart, M. A., Haas, J. S., & Shrank, W.
H. (2009). Variation in Drug Prices at Pharmacies: Are Prices Higher in Poorer Areas?
Health Services Research, 44(2p1), 606–617. https://doi.org/10.1111/j.1475-
6773.2008.00917.x
103

Geruso, M., & Layton, T. (2015). Upcoding: Evidence from Medicare on Squishy Risk
Adjustment (Working Paper No. 21222). National Bureau of Economic Research.
https://doi.org/10.3386/w21222
Geruso, M., & Layton, T. J. (2017). Selection in Health Insurance Markets and Its Policy
Remedies. Journal of Economic Perspectives, 31(4), 23–50.
https://doi.org/10.1257/jep.31.4.23
Geruso, M., Layton, T. J., & Prinz, D. (2016). Screening in Contract Design: Evidence from the
ACA Health Insurance Exchanges (Working Paper No. 22832). National Bureau of
Economic Research. https://doi.org/10.3386/w22832
Gilmer, T., Kronick, R., Fishman, P., & Ganiats, T. G. (2001). The Medicaid R x Model:
Pharmacy-Based Risk Adjustment for Public Programs. Medical Care, 39(11), 1188–
1202. JSTOR.
Government Accountability Office. (2014). Medicare Part D: CMS has implemented processes
to oversee Plan Finder pricing accuracy and improve website usability.
https://www.gao.gov/assets/670/660081.pdf
Gowrisankaran, G., Nevo, A., & Town, R. (2015). Mergers When Prices Are Negotiated:
Evidence from the Hospital Industry. American Economic Review, 105(1), 172–203.
https://doi.org/10.1257/aer.20130223
Gruber, J., & McKnight, R. (2016). Controlling Health Care Costs through Limited Network
Insurance Plans: Evidence from Massachusetts State Employees. American Economic
Journal: Economic Policy, 8(2), 219–250. https://doi.org/10.1257/pol.20140335
Heiss, F., McFadden, D., Winter, J., Wuppermann, A., & Zhou, B. (2016). Inattention and
Switching Costs as Sources of Inertia in Medicare Part D (Working Paper No. 22765).
National Bureau of Economic Research. https://doi.org/10.3386/w22765
Ho, K., & Lee, R. S. (2017). Insurer Competition in Health Care Markets. Econometrica, 85(2),
379–417. https://doi.org/10.3982/ECTA13570
Hoadley, J. F., Merrell, K., Hargrave, E., & Summer, L. (2012). In Medicare Part D Plans, Low
Or Zero Copays And Other Features To Encourage The Use Of Generic Statins Work,
Could Save Billions. Health Affairs, 31(10), 2266–2275.
https://doi.org/10.1377/hlthaff.2012.0019
Hoadley, J., Hargrave, E., & Summer, L. (2013). To Switch or Not to Switch: Are Medicare
Beneficiaries Switching Drug Plans To Save Money?
104

https://www.kff.org/medicare/issue-brief/to-switch-or-not-to-switch-are-medicare-
beneficiaries-switching-drug-plans-to-save-money/
Jacobs, D. B., & Sommers, B. D. (2015). Using Drugs to Discriminate—Adverse Selection in the
Insurance Marketplace. New England Journal of Medicine, 372(5), 399–402.
https://doi.org/10.1056/NEJMp1411376
Jacobs, P. D., Cohen, M. L., & Keenan, P. (2017). Risk Adjustment, Reinsurance Improved
Financial Outcomes For Individual Market Insurers With The Highest Claims. Health
Affairs, 36(4), 755–763. https://doi.org/10.1377/hlthaff.2016.1456
Kautter, J., Pope, G. C., Ingber, M., Freeman, S., Patterson, L., Cohen, M., & Keenan, P. (2014).
The HHS-HCC Risk Adjustment Model for Individual and Small Group Markets under
the Affordable Care Act. Medicare & Medicaid Research Review, 4(3).
https://doi.org/10.5600/mmrr.004.03.a03
Ketcham, J. D., Lucarelli, C., & Powers, C. A. (2015). Paying Attention or Paying Too Much in
Medicare Part D. American Economic Review, 105(1), 204–233.
https://doi.org/10.1257/aer.20120651
Koma, W., Cubanski, J., Jacobson, G., Damico, A., & Neuman, T. (2019). No Itch to Switch:
Few Medicare Beneficiaries Switch Plans During the Open Enrollment Period.
https://www.kff.org/medicare/issue-brief/no-itch-to-switch-few-medicare-beneficiaries-
switch-plans-during-the-open-enrollment-period/
Lavetti, K., & Simon, K. (2018). Strategic Formulary Design in Medicare Part D Plans.
American Economic Journal: Economic Policy, 10(3), 154–192.
https://doi.org/10.1257/pol.20160248
Luo, J., Kulldorff, M., Sarpatwari, A., Pawar, A., & Kesselheim, A. S. (2019). Variation in
Prescription Drug Prices by Retail Pharmacy Type. Annals of Internal Medicine, 171(9),
605–611. https://doi.org/10.7326/M18-1138
Medicare Payment Advisory Commission. (2012). June 2012 Report to the Congress: Medicare
and the Health Care Delivery System (p. 259). http://medpac.gov/docs/default-
source/reports/jun12_entirereport.pdf?sfvrsn=0
Medicare Payment Advisory Commission. (2021). March 2021 Report to the Congress:
Medicare Payment Policy. http://www.medpac.gov/docs/default-
source/reports/mar21_medpac_report_to_the_congress_sec.pdf?sfvrsn=0
105

Medicare Payment Advisory Commission. (2020b). June 2020 Report to the Congress:
Medicare and the Health Care Delivery System. http://medpac.gov/docs/default-
source/reports/jun20_reporttocongress_sec.pdf?sfvrsn=0
Montz, E., Layton, T., Busch, A. B., Ellis, R. P., Rose, S., & McGuire, T. G. (2016). Risk-
Adjustment Simulation: Plans May Have Incentives To Distort Mental Health And
Substance Use Coverage. Health Affairs, 35(6), 1022–1028.
https://doi.org/10.1377/hlthaff.2015.1668
Prager, E. (n.d.). Health Care Demand under Simple Prices: Evidence from Tiered Hospital
Networks. American Economic Journal: Applied Economics.
https://doi.org/10.1257/app.20180422
Prager, E. (2020). Healthcare Demand under Simple Prices: Evidence from Tiered Hospital
Networks. American Economic Journal: Applied Economics, 12(4), 196–223.
https://doi.org/10.1257/app.20180422
Qato, D. M., Daviglus, M. L., Wilder, J., Lee, T., Qato, D., & Lambert, B. (2014). ‘Pharmacy
Deserts’ Are Prevalent In Chicago’s Predominantly Minority Communities, Raising
Medication Access Concerns. Health Affairs, 33(11), 1958–1965.
https://doi.org/10.1377/hlthaff.2013.1397
Sinaiko, A. D., & Mehrotra, A. (2020). Association of a national insurer’s reference-based
pricing program and choice of imaging facility, spending, and utilization. Health Services
Research, 55(3), 348–356. https://doi.org/10.1111/1475-6773.13279
Sinaiko, A. D., & Rosenthal, M. B. (2014). The Impact of Tiered Physician Networks on Patient
Choices. Health Services Research, 49(4), 1348–1363. https://doi.org/10.1111/1475-
6773.12165
Socal, M. P., Bai, G., & Anderson, G. F. (2019). Favorable Formulary Placement of Branded
Drugs in Medicare Prescription Drug Plans When Generics Are Available. JAMA
Internal Medicine, 179(6), 832–833. https://doi.org/10.1001/jamainternmed.2018.7824
Starc, A., & Swanson, A. (2019). Preferred Pharmacy Networks and Drug Costs (SSRN
Scholarly Paper ID 3240967). Social Science Research Network.
https://doi.org/10.2139/ssrn.3240967
The Health Care Cost Institute. (2019). 2017 Health Care Cost and Utilization Report.
https://www.healthcostinstitute.org/research/annual-reports/entry/2017-health-care-cost-
and-utilization-report
106

Town, R., & Vistnes, G. (2001). Hospital competition in HMO networks. Journal of Health
Economics, 20(5), 733–753. https://doi.org/10.1016/S0167-6296(01)00096-0
Trish, E. E. (2019). Medicare Part D: Time for Re-Modernization? Health Services Research,
54(6), 1174–1183. https://doi.org/10.1111/1475-6773.13221
Trish, E., Xu, J., & Joyce, G. (2018). Growing Number Of Unsubsidized Part D Beneficiaries
With Catastrophic Spending Suggests Need For An Out-Of-Pocket Cap. Health Affairs,
37(7), 1048–1056. https://doi.org/10.1377/hlthaff.2018.0006
van de Ven, W. P. M. M., & Ellis, R. P. (2000). Chapter 14 Risk adjustment in competitive
health plan markets. In Handbook of Health Economics (Vol. 1, pp. 755–845). Elsevier.
https://doi.org/10.1016/S1574-0064(00)80173-0
Vandagriff, A., Millen, B. N., Petroske, J., & Mattie, L. L. (2017). ACA risk adjustment
management: Cracking the code. Milliman.
http://www.milliman.com/insight/2017/ACA-risk-adjustment-management-Cracking-the-
code/
Weiner, J. P., Trish, E., Abrams, C., & Lemke, K. (2012). Adjusting For Risk Selection In State
Health Insurance Exchanges Will Be Critically Important And Feasible, But Not Easy.
Health Affairs, 31(2), 306–315. https://doi.org/10.1377/hlthaff.2011.0420
Zhou, C., & Zhang, Y. (2012). The Vast Majority Of Medicare Part D Beneficiaries Still Don’t
Choose The Cheapest Plans That Meet Their Medication Needs. Health Affairs, 31(10),
2259–2265. https://doi.org/10.1377/hlthaff.2012.0087

Abstract (if available)

Abstract Prescription drugs are important in health care in the United States, and rising drug prices are of great concern for the public. This dissertation consists of three essays, and together they examine the impact of two understudied recent developments that aim at improving patient access to prescription drugs and curbing price growth: One is more accurately compensating health plans for enrollees who take high-cost drugs in health insurance exchanges created by the Affordable Care Act (ACA), and the other is the implementation of preferred pharmacy networks by Medicare Part D plans as an effort to negotiate lower prescription drug prices. ? The first essay finds that in the exchanges, incorporating the information on the use of high-cost, disease-modifying drugs into the risk adjustment model would improve the accuracy of risk-adjusted payments, and thus mitigate plans? incentives to skimp on the coverage of novel, expensive drugs, thereby improving patient access. The second and third essays evaluate preferred pharmacy networks in Medicare Part D. The second essay finds different financial incentives for unsubsidized and subsidized patients under preferred networks' cost-sharing structures. In addition, unsubsidized patients incurred higher out-of-pocket spending when they used non-preferred pharmacies, while Medicare bore the additional financial burden for subsidized patients who visited non-preferred pharmacies. The third essay estimates that the implementation of preferred networks moderately increased preferred pharmacies? claim share in the first year among unsubsidized beneficiaries. Existing relationships with preferred pharmacies, financial incentives, access to preferred pharmacies, and urban residence were positively correlated with beneficiaries? decisions to switch to these pharmacies.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Thesis on Medicare Part D

PDF

Essays in opioid use and abuse

PDF

Economic, clinical, and behavioral outcomes from medical and pharmaceutical treatments

PDF

US biosimilar market entry and uptake: US commercial payer preferences in access decision making

PDF

Health care utilization and spending of the U.S. aging population

PDF

Evaluating treatment options for metastatic, castration-resistant prostate cancer: a comprehensive value assessment

PDF

Essays in pharmaceutical and health economics

PDF

Essays on the economics of infectious diseases

Pdf

Alzheimer's disease: risk factors, value, and alternative payment models

PDF

Value in oncology care and opportunities for improvement

PDF

Impact of pharmacy-based transitional care on healthcare utilization and costs

PDF

Value in health in the era of vertical integration

PDF

Essays in health economics and provider behavior

PDF

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

PDF

Three essays on estimating the effects of government programs and policies on health care among disadvantaged population

PDF

Three essays in health economics

PDF

Three essays on comparative economic and clinical outcomes of antipsychotics using retrospective claims data

PDF

The value of novel antihyperlipidemic treatments in the U.S. healthcare system: Reducing the burden of cardiovascular diseases and filling the gap of low adherence in statins

PDF

Designing health care provider payment systems to reduce potentially preventable medical needs and patient harm: a simulation study

PDF

The impact of Patient-Centered Medical Home on a managed Medicaid plan

Asset Metadata

Creator Xu, Jianhui (author)

Core Title Essays on access to prescription drugs

School School of Policy, Planning and Development

Degree Doctor of Philosophy

Degree Program Public Policy and Management

Degree Conferral Date 2021-08

Publication Date 01/16/2023

Defense Date 05/13/2021

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

Tag ACA exchanges,financial incentive,Medicare Part D,OAI-PMH Harvest,pharmacy choice,preferred pharmacy networks,prescription drugs,risk adjustment,risk selection

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Romley, John (committee chair), Joyce, Geoffrey (committee member), Lakdawalla, Darius (committee member), Trish, Erin (committee member)

Creator Email jianhui.frank.xu@gmail.com,jianhuix@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC15610960

Unique identifier UC15610960

Legacy Identifier etd-XuJianhui-9789

Document Type Dissertation

Format application/pdf (imt)

Rights Xu, Jianhui

Type texts

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

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

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

Repository Email cisadmin@lib.usc.edu