xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 22 Ka 4500
controlfield tag 007 cr-bnu---uuuuu
008 s2009 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0003200
Medicare part d program :
b prescription drug plan copayment structure and premium sensitivity
h [electronic resource] /
by Rui Dai.
[Tampa, Fla] :
University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains X pages.
Dissertation (Ph.D.)--University of South Florida, 2009.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
ABSTRACT: Since January 2006 Medicare beneficiaries have the option to purchase prescription drug benefits from Medicare under the Part D program. The addition of outpatient drugs to the Medicare programs reflects Congress' recognition of the fundamental change in recent years in how medical care is delivered in the U.S. It recognizes the vital role of prescription drugs in the health care delivery system and the need to modernize Medicare to assure their availability to Medicare beneficiaries. The Medicare Prescription Drug Improvement and Modernization Act of 2003 (MMA) created the Medicare drug benefit and specified a standard plan. The law also enables plans to offer alternative benefit packages that are either actuarially equivalent or provide enhanced benefits above the basic benefits. A majority of these alternative plans offer multitiered formulary where different medications have different patient copayments. Different from traditional Medicare, Part D benefits are provided by private sector plans through a competitive bidding process. Firms submit a bid to the Center for Medicare and Medicaid Services (CMS) which represents the expected cost to the firm for providing basic benefits to an individual of average health. The competition between plans was expected to drive premiums down toward marginal cost, ensuring that the beneficiaries receive maximum benefits for a given public expenditure (Biles et al. 2004). This dissertation examines the stand-alone Medicare Prescription Drug Plans (PDPs) bid and premium from the following perspectives using the 2006-2008 PDP data. First, we examine the use of multiple-tier copayment structures. In particular, we tend to discover the relationship between enrollee cost sharing at each tier and prescription drug plan (PDP) bids. Bids are equivalent to the total premiums charged by an insurer. This includes the premium paid by the consumer and the portion paid by the federal government. Further, we decompose plan bid and premium changes between 2006 and 2008 into two components, the proportion due to changes in plan characteristics and the proportion due to changes in marginal price. By doing so, we estimate whether the actuarial methods used to price those characteristics play a role in explaining the plan bid and premium difference across years. Finally, we measure the Medicare beneficiaries' sensitivity to price in the PDP market, specifically the elasticity and semi-elasticity of enrollment with respect to PDP premium.
Advisor: Gabriel A. Picone, Ph.D.
Medicare Part D
t USF Electronic Theses and Dissertations.
Medicare Part D Program: Prescription Drug Plan Copayment Structure and Premium Sensitivity by Rui Dai A Dissertation Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Economics College of Business Administration University of South Florida Major Professor: Gabr iel A. Picone, Ph.D. John M. Robst, Ph.D. Yi Deng, Ph.D. Murat K. Munkin, Ph.D. Christopher R. Thomas, Ph.D. Date of Approval: October 16, 2009 Keywords: Medicare Part D, tiered copa yments, elasticity, risk score, formulary Copyright 2009, Rui Dai
i Acknowledgements I am grateful to many individuals who s upported my work on this dissertation. First, I would like to thank my dissertation committee chair, Dr. Gabriel A. Picone, for his continuous guidance to my academic studies and research. I also would like to express thanks to Dr. John M. Robst, whose guidance was invaluable. This dissertation is not possible without their support and advice. I must also thank my other committee members, Dr. Yi Deng, Dr. Murat K. Munkin and Dr. Christopher R. Thomas for their helpful comments and advice. I also want to thank the rest of the Economics Department faculty for their academic instructions and support. Finally, I would like to take this opportuni ty to thank the financial support from the Gaiennie Foundation at th e University of South Florida, College of Business Administration. This grant provided me with the necessary funding to purchase the data used in this dissertation.
ii Table of Contents List of Tables .......................................................................................................................v List of Figures ................................................................................................................... vii Abstract ............................................................................................................................ viii Chapter One Introduction ....................................................................................................1 1.1 Medicare ............................................................................................................1 1.1.1 Eligibility ............................................................................................1 1.1.2 Administration and Financing.............................................................1 1.1.3 Medicare Benefits: ..............................................................................2 1.1.4 Medicare Supplemental Coverage ......................................................5 1.1.5 Reimbursement Method and Risk Scores ...........................................5 1.1.6 Medicare Advantage Bidding Process ................................................7 1.1.7 Current Status and Challenges ............................................................8 1.2 Medicare Part D ...............................................................................................10 1.2.1 Eligibility and Enrollment Process ...................................................10 1.2.2 MAPD vs. PDP .................................................................................11 1.2.3 Part D Standard Benefits ...................................................................12 1.2.4 Part D Alternative Benefits ...............................................................13 1.2.5 Plan Formularies ...............................................................................15 1.2.6 Part D Bidding Process and Beneficiary Premium ...........................16 1.2.7 Part D Reimbursement Method and Risk Scores ..............................18 1.2.8 Government Subsidy .........................................................................20 1.2.9 Current Status ....................................................................................21 1.2.10 Challenges .......................................................................................23 Chapter Two Literature Review .........................................................................................26 2.1 Medicare Part D Program ................................................................................26 2.2 Effects of Insurance a nd Plan Characteristics ..................................................29 2.2.1 Cost Sharing ......................................................................................29 2.2.2 Utilization Control Tools ..................................................................32 2.2.3 Premium and Premium Elasticity .....................................................33 2.3 Summary ..........................................................................................................35 Chapter Three Research Design .........................................................................................37 3.1 Objectives and Hypotheses ..............................................................................37 3.2 Description of Data ..........................................................................................38 3.2.1 Prescription Drug Plan and Pharmacy Network Files ......................39 220.127.116.11 Plan Information File .........................................................39 18.104.22.168 Formulary File ...................................................................40 22.214.171.124 Beneficiary Cost File .........................................................40
iii 126.96.36.199 Pharmacy Network File .....................................................40 188.8.131.52 Geographic Locator File ....................................................41 184.108.40.206 Supporting Files .................................................................41 3.2.2 Other CMS Data ...............................................................................41 220.127.116.11 Part D Risk Score File........................................................42 18.104.22.168 PDP Penetration File ..........................................................42 22.214.171.124 CMS PDP Monthly Enrollment File ..................................43 126.96.36.199 CMS Landscape File ..........................................................43 3.2.3 Other Data .........................................................................................43 3.2.4 Data Compilation ..............................................................................44 3.3 Description of Variables ..................................................................................45 3.3.1 Dependent Variables .........................................................................46 3.3.2 Explanatory Variables .......................................................................49 3.4 Methodology ....................................................................................................51 3.4.1 Hedonic Pricing Model .....................................................................52 188.8.131.52 Missing Variable Problem .................................................54 184.108.40.206 Firm Fixed Effects Model Specification ............................55 3.4.2 Decomposition Model .......................................................................57 3.4.3 Premium Elasticities .........................................................................58 220.127.116.11 Mean Utility Function ........................................................58 18.104.22.168 Instrumental Variables .......................................................61 22.214.171.124 Premium Elasticity Definition ...........................................61 Chapter Four Research Results ..........................................................................................63 4.1 Hedonic Pricing Model Results .......................................................................63 4.1.1 Descriptive Statistics .........................................................................63 4.1.2 Statistical Analysis ............................................................................68 4.1.3 Firm Fixed Effects Model Results ....................................................73 4.1.4 Low and High Risk Region Analysis ................................................81 4.1.5 Other Model Forms, Function Forms and Variables ........................83 4.2 Decomposition Model Results .........................................................................85 4.2.1 Descriptive Statistics .........................................................................85 4.2.2 Firm Fixed Effects Model Results ....................................................88 4.3 Premium Elasticities ........................................................................................94 4.3.1 Descriptive Statistics .........................................................................95 4.3.2 OLS and 2SLS Model Results ..........................................................97 4.4. Summary .......................................................................................................103 Chapter Five Discussion ..................................................................................................105 5.1 Conclusions ....................................................................................................105 5.2 Limitations .....................................................................................................109 5.3 Future Research .............................................................................................113
iv References ........................................................................................................................115 Websites ...........................................................................................................................120 Appendix A: Tables .........................................................................................................121 Appendix B: Figures ........................................................................................................137 About the Author ................................................................................................... End Page
v List of Tables Table 1 Medicare Part D Defined Standard Benefits ................................................... 13 Table 2 National Average Part D Numbers ................................................................. 46 Table 3 Descriptive Statistics of Variables for the Firm Fixed Effects Model ............ 65 Table 4 Average Bids by Cost Sharing at Each Tier ................................................... 69 Table 5 Statistics by Tier Member Cost Sharing ......................................................... 71 Table 6 Regression Resu lts: Firm Fixed Effects Model Dependent Variable: Ln (Bid) ......................................................................... 74 Table 7 Differentiating between Low and High Risk Regions Dependent Variable: Ln (Bid) ........................................................................ 82 Table 8 Descriptiv e Statistics, (2006 and 2008 Data) ................................................. 86 Table 9 Firm fixed effects Model Estimates Dependent Variable: Ln(Bid), (2006 and 2008 Data) .................................... 90 Table 10 Firm fixed effects Model Estimates Dependent Variable: Ln(Premium), (2006 and 2008 Data) ........................... 92 Table 11 Descriptive Sta tistics, 2008 Non-benchmark Plans ....................................... 96 Table 12 Regression Results Assuming Composite Outside Goods ............................ 98 Table 13 Regression Results Assuming MAPDs as Outside Goods .......................... 100 Table A 1 Median Negotiated Prices for Medicare Part D Sample Drugs ....................122 Table A 2 Average Risk Score by PDP Region .............................................................123 Table A 3 Medicare Population by PDP Region ...........................................................124 Table A 4 Company Information by Contract ...............................................................125 Table A 5 Average Plan Bid and Member Premium by Year .......................................128 Table A 6 Average Plan Bid and Member Premium by PDP Region ...........................129 Table A 7 Average Plan Bid and Member Premium by Contract .................................130
vi Table A 8 Regression Results : Firm Fixed Effects Model Dependent Va riable: Ln (Bid), Plans with Risk Scores < 1.0 ......................132 Table A 9 Regression Results : Firm Fixed Effects Model Dependent Vari able: Ln (Bid), Plans with Risk Scores >= 1.0 ...................133 Table A 10 Regression Results: OLS, Dependent Variable: Ln (Bid) ...........................134 Table A 11 Regression Results : Firm Fixed Effects Model Dependent Variable: Ln (Premium) ............................................................135 Table A 12 Regression Results : Firm Fixed Effects Model Dependent Variable: Square Root (Bid) .....................................................136
vii List of Figures Figure 1 Medicare Revenue ................................................................................................2 Figure 2 Medicare Benefits .................................................................................................3 Figure 3 Medicare Spending ...............................................................................................9 Figure 4 Part D Standard Plan ...........................................................................................12 Figure 5 National Average Bid and Member Premium ....................................................17 Figure 6 Part D Reimbursement Method ...........................................................................19 Figure B 1 Normal Probability Plot Ln(Bid) ................................................................138 Figure B 2 Medicare Population Characteristics .............................................................139 Figure B 3 Data File Layouts ...........................................................................................140 Figure B 4 2006-2007 Part D Plan Standard Benefits .....................................................141 Figure B 5 2007-2008 Part D Standard Benefits .............................................................142 Figure B 6 2006 Part D Risk Co rridor Reconciliation Amount ......................................143 Figure B 7 2007 Part D Risk Co rridor Reconciliation Amount ......................................144 Figure B 8 2006-2007 Risk Corridors .............................................................................145 Figure B 9 2008-2011 Risk Corridors .............................................................................146
viii Medicare Part D: Prescription Drug Plan Copayment Structure and Premium Sensitivity Rui Dai ABTRACT Since January 2006 Medicare beneficiar ies have the option to purchase prescription drug benefits from Medicare unde r the Part D program. The addition of outpatient drugs to the Medicare progra ms reflects Congress recognition of the fundamental change in recent ye ars in how medical care is delivered in the U.S. It recognizes the vital role of prescription drugs in the health care delivery system and the need to modernize Medicare to assure their av ailability to Medicare beneficiaries. The Medicare Prescription Drug Improvement and Modernization Act of 2003 (MMA) created the Medicare drug benefit and specified a standard plan. The law also enables plans to offer alternative benefit packages that are either actuarially equivalent or provide enhanced benefits above the basic benefits. A majority of these alternative plans offer multitiered formulary where different medications have different patient copayments. Different from traditional Me dicare, Part D benefits are provided by private sector plans through a competitive bidding process. Firms submit a bid to the Center for Medicare and Medicaid Services (CMS) which represents the expected cost to the firm for providing basic benefits to an individual of average health. The competition between plans was expected to drive premiums down toward marginal cost, ensuring that the beneficiaries receive maximum benefits for a gi ven public expenditure (Biles et al. 2004).
ix This dissertation examines the standalone Medicare Prescription Drug Plans (PDPs) bid and premium from the following perspectives using th e 2006-2008 PDP data. First, we examine the use of multiple-tier copayment structures. In particular, we tend to discover the relationship between enrollee cost sharing at each tier and prescription drug plan (PDP) bids. Bids are equivalent to the total premiums charged by an insurer. This includes the premium paid by the consumer and the portion paid by the federal government. Further, we decompose plan bid and pr emium changes between 2006 and 2008 into two components, the proportion due to changes in plan characteristics and the proportion due to changes in marginal price. By doing so, we estimate whether the actuarial methods used to price those characteristics play a ro le in explaining the plan bid and premium difference across years. Finally, we measure the Medicare benefici aries sensitivity to price in the PDP market, specifically the elasticity and semi-ela sticity of enrollment with respect to PDP premium.
1 Chapter One Introduction This chapter consists of two sections. Section 1.1 introduces the background of the Medicare program, its current status and challenges faced. Section 1.2 discusses the Medicare Part D program a nd some specific issues. 1.1 Medicare Medicare, the social insurance program in the United States, was signed into law in 1965 by President Johnson as amendments to Social Security legislation. It provides health insurance coverage to the people w ho are aged 65 or older, or people under 65 with permanent disabilities, ESRD (End Stage Renal Disease), or Lou Gehrigs disease. 1.1.1 Eligibility To be eligible for Medicare, people need to have made payroll tax contributions for at least 10 or more years. Their spouses if not working, are only eligible for Part A. 1.1.2 Administration and Financing Medicare is administered by the Centers for Medicare and Medicaid Services (CMS). As illustrated in Figure 1, it is par tially financed by payroll taxes (41% in 2009) imposed by the Federal Insurance Contribu tions Act (FICA) and Self-Employment Contributions Act of 1954. Other financing sources include general revenue (39% in 2009), beneficiary premiums (12% in 2009), interest, and others.
Figure 1 Medicare Revenue Data Source: Kaiser Family Foundation, Medicare at a Glance, November 2008. The original data is from 2008 Annual Report of th e Boards of the Federal Hospital Insurance and Federal Supplemental Medical Insurance Trust Funds. 1.1.3 Medicare Benefits: Medicare benefits are categor ized as Part A, Part B, Part C, and Part D as illustrated in Figure 2. Part A (Hospital Insurance) and Part B (Medical Insurance) are the two parts in the original Medicare program. Part A covers inpatient hospital, skilled 2
nursing care, home health (also under Part B) and hospice care. Part A accounts for 36% of benefit spending in 2009 according to the Congressional Budget Office (CBO Medicare Baseline, March 2008). Figure 2 Medicare Benefits Notes: Doesnt include administrative expenses su ch spending to administer the Medicare Drug benefits and the Medicare Advantage program. Data Source: Kaiser Family Foundation, Medicare at a Glance, November 2008. The original data is from CBO Medicare Baseline, March 2008. 3
4 Part B coverage includes services and pr oducts not covered by Part A, generally on an outpatient basis, such as physician a nd nursing services, x-rays, laboratory and diagnostic tests, durable medical equipment, etc. Part B accounts for 29% of benefit spending in 2009 (CBO). Part B coverage is optional and is allowed to be deferred if the Medicare beneficiary or their spouse is still actively working. Part C refers to the Medicare + Choice program, which was passed by Balanced Budget Act of 1997. This program allows the Medicare beneficiarie s to receive their Medicare benefits thr ough private health insurance plans, instead of through the original Medicare program. The Medicare + C hoice program was renamed as Medicare Advantage since the inception of the Medi care Part D program in 2006, but is still referred to as Part C. Most Medicare Advant age (MA) plans offer coverage that meet or exceed the standards set by the original Medi care program. Due to the flexibility of benefits they offer, Medicare Advantage plans have gained popularity since their inception. Medicare Advantage plans that offer prescription drug c overage are called Medicare Advantage Prescription Drug plan (MAPD). In recent years, Congress has increased payments to Medicare private plans to encourage plan participa tion throughout the country. As a result, the average Medica re payment to Medicar e Advantage plans is 113% of the cost of similar benefits in the original fee-for-service (FFS) program (MedPAC, 2008). Now, Part C accounts for 24% of benefit spending. Medicare Part D program started in January 1, 2006, providing the prescription drug coverage. Currently, more than 25 milli on beneficiaries are enrolled in Medicare Part D plans and Part D account s for 11% of Medicare bene fit spending in 2009 (CBO). Detailed discussion on Medicare Pa rt D is presented in section 1.2.
5 1.1.4 Medicare Supplemental Coverage Medicare has a high member cost-sharing requirement, no limit on the out-ofpocket spending and coverage gap in the Part D benefits. Therefore, most Medicare beneficiaries have some other forms of supplemental insurance, such as employersponsored retiree health plans, Medicaid a nd Medigap (supplemental private insurance for medical expenses that are not covered or partially covered by Medicare). Only 11% of Medicare beneficiaries had no supplemental coverage in 2006. 1.1.5 Reimbursement Metho d and Risk Scores The 1997 Balanced Budget Act modified the Medicare Managed Care plans and pays private plans participati ng in the Medicare + Choice mark et a monthly capitated rate to provide health care services to enrolled Medicare benefi ciaries (Pope et al. 2004). Historically the capita tion payments were linked to th e FFS expenditures and set at 95% of an enrollees countys adjusted average per capita cost (AAPCC). The AAPCC rates were defined by age, sex, Medicaid enrollme nt status, institutiona l status, and working age status. Separate county factors were calculated for th e aged and non-aged disabled, and at the state level only for ESRD entitled beneficiaries. The AAPCC rates only account for 1% of the variation in Medicare beneficiaries expenditures and do not pay more for sicker people. Thus it caused the Managed Care Organizations to select healthier members a nd as a result, the ove rall Medicare program expenditure increased. The Medicare + Choice program fundamentally changed the Medicare managed care capitation method in 2000 and implemented the Medicare risk adjustment CMS HCC (Hierarchical Conditi on Categories) model in 2004. During the
6 transitional period, the PIP-DCG (Princip al Inpatient Diagnostic Cost Grouping) model was used as a health based payment adjuster (Pope et al. 2004). The HCC diagnostic classification syst em first classifies each of over 15,000 ICD-9-CM (international statistical classificati on of diseases and rela ted health problems) codes into 804 diagnostic groups or DxGroups, which are furt her, aggregated into 189 Condition Categories, or CCs. CCs describe a broader set of similar diseases. Hierarchies are imposed among similar CCs. Some non-significant HCCs were excluded and only 70 HCCs were included in the final CMS HCC model. The CMS HCC model also relies on dem ographic factors, Medicaid status, originally disabled status, a nd institutional status. These factors and the 70 HCCs are assigned coefficients which are estimated from clinical data. Individual Medicare beneficiarys Medical risk scores are calcula ted based his or her age, gender, Medicaid status, originally disabled or not, instituti onal status and HCCs. The coefficients are updated annually to account for changes. The nationwide overall risk scores are normalized at 1.0. A higher risk score indicates a worse health status while a lower risk score means a better health status. The capitation payments using the CM S HCC model are proportional to the Medicare beneficiaries risk scores. Managed Care Organi zations enrolling healthier members with lower risks scores receive le ss payment from CMS. On the other hand, they are compensated for enrolling sicker memb ers. Thus, favorable selection or cherrypicking problem in the traditional ma naged care industry is alleviated.
7 1.1.6 Medicare Advantage Bidding Process Starting from 2006, a competitive bidding process has replaced the Adjusted Community Rate Proposal filings required in 2005 and prior years (The Actuary Magazine, Oct, 2005). The insurance companies that want to participate in the Medicare Advantage market are required to submit th eir bids to CMS by the end of the first Monday of June prior to the contract year on a plan base. Each bid is associated with a unique contract ID and plan ID. Most in surance companies offer one contract but multiple plans each year. Some big insurance companies may offer multiple contracts. For Part A and B benefits, Medicare Adva ntage plans bid on traditional Medicare benefits including traditional Me dicare cost sharing levels. Lower cost sharing levels and Medicare non-covered benefits are optional. The projected claim costs for each line of the benefits, projected admini stration costs, and profits based on the projected enrollment are inputs required in the CMS bid forms. MMA declared plan bids would be based on a national profile population. In other words, each plans bid is normalized at risk score of 1.0. For Part D, a separate bid form has to be submitted. The Part D competitive bids are based on a national profile population as well If a plan bid is higher than the national average bid, its member premium for Part D is increased by the difference. Similarly, if a plan bid is lower than the national averag e bid, it will have a lower Part D member premium. The payments each plan receives from CM S are directly determined by the bids and adjusted by the risk scores. For sicker members who incur more claims, the plan will receive more payment from CMS. Similarly, for healthier members, the plan receives
8 less payment from CMS. This process alle viates the anti-selection problem that the Managed Care Organizations tend to enroll healthier members. The competitive bidding process gives the plans little incentive to under or over bid because they are only compensated up to the benchmark payment set by CMS. If a plan bid is lower than the benchmark, they wi ll receive 75% of the difference between the benchmark and the plan bid as a rebate in addition to its bid amount. On the other hand, if a plan bid is higher than the benchm ark, the amount above the benchmark will be passed to its members in terms of a higher member premium. 1.1.7 Current Status and Challenges In 2007, Medicare provided health care coverage for 43 million Americans and currently covers 45 million Americans. Enrollment is expected to reach 77 million by 2031 when the baby boom generation is fully enrolled. Medicare benefit outlays are expected to total $477 billion in 2009, accounting for 13% of the federal budget and 22% of persona l health care expenditu re (CBO). It is projected to reach $871 billion in 2018 accordin g to CBO. Two main factors influencing the annual growth of Medicare spending are th e increasing volume of services and rising prices. CBO estimates that a larger share of future growth in Medicare spending as a share of the Gross Domestic Product will result from growth in health care cost rather than from growth in enrollment. Efforts to control rising health costs would help mitigate Medicares future funding shor tfall (Kaiser, Medicare Nov 2008). The greatest challenge for Medicare is the financing. According to the Medicare Trustees, Part A Trust Fund is projected to be depleted in 2019, with insufficient funds to
pay benefits (Kaiser, Medicar e Nov 2008). Figure 3 shows the financial burden of health spending among Medicare beneficiaries from 1997 to 2005. While the spending is increasing, the speed has slowed down in recent years. Figure 3 Medicare Spending Data Source: Kaiser Family Foundation, Medicare at a Glance, November 2008. The original data is from Kaiser/UCLA analysis of Medicare Current Beneficiary Survey Cost and Use files, 1997-2005. Other critical issues that Medicare fa ces include the management of care for chronically ill high-cost benefi ciaries, fairness of payments to providers and plans, aging population, etc. For reference, Appendix B1 shows the characteristics of the Medicare population. 9
10 1.2 Medicare Part D Medicare Part D refers to the Medica re Prescription Drug Program, which was established by section 101 of the Medicare Prescrip tion Drug, Improvement, and Modernization Act of 2003 and went into effect in January 2006. The new Part D benefits constitute perhaps the most significant change to the Medicare program since its inception in 1965. The prescription drug benefit is not part of the original Medicare program. The addition of outpatient drugs to the Medicar e programs reflects Congress recognition of the fundamental change in recent years in how medical care is delivered in the U.S. It recognizes the vital role of prescription drugs in the health care delivery system and the need to modernize Medicare to assure their availability to Medica re beneficiaries. Effective January 1, 2006, the Part D program established an optional prescription drug benefit for individuals who are entitled to Medicare Part A and/ or enrolled in Part B. 1.2.1 Eligibility and Enrollment Process Individuals who are entitled to Medicare Part A (whether actually enrolled or not) or currently enrolled in Part B are eligible for Medicare Part D benefits. Enrollment in Part D is voluntary except for individuals who are dual eligibles (those also in Medicaid). Individuals who are first eligib le for Medicare are required to enroll three months before or three months after they turn 65. If they fa il to enroll in that 6month period, they have to pay a penalty in the form of a higher premium. Individuals who are already in Medicare can enroll in a Part D plan during the open enroll ment period which starts on November 15 and lasts until the end of Decemb er of the year. During this period, they
11 can choose to enroll or switch plans. Afte r this period, they must affirmatively stay enrolled in a Part D plan. CMS will auto-enroll or facilitate enrollment for Medicare beneficiaries who are eligible for Low-income subsidy (LIS). Du al eligible LIS beneficiaries who stay in traditional FFS Medicare or enrolled in an MA only plan are randomly enrolled into one of benchmark PDPs. Dual eligib les enrolled in a MA only plan can also be auto assigned to a MAPD benchmark plan. The benchmark pl ans are those that offer defined standard benefits with a premium below the benchmark in each region set by CMS. Facilitated enrollment is the process for other LIS eligible s. The process is essentially the same as auto-enrollment, but the timing of the first round assignments differs. Furthermore, all LIS beneficiaries can switch plans anytime during the contract year whereas other beneficiaries can only switch plans during the annual open enrollment period. 1.2.2 MAPD vs. PDP Different from traditional Medicare, Medicare Part D is provided through private companies or entities approved by CMS. Bene ficiaries can obtain drug benefits through two types of private plans, the stand-alone PDPs or MAPDs which cover both medical service and prescription drugs. Individuals enrolled in PDPs receive their medical benefits from traditional FFS Medicare or MA only plans. Different from the MAPDs, which are offered at the county level, the PDPs operates at the PDP region level. Defined by CMS, there are 34 PDP regions in the United States, each of which cover one or more states (see Appendix A, Table A2).
1.2.3 Part D Standard Benefits The MMA established a standard Medicar e Part D benefit package which is defined in terms of benefit structure, not in terms of the drugs that must be covered. As illustrated in Figure 4, in 2007,the standard benefits are $265 annual deductible, 25% coinsurance, $2,400 initial coverage limit (ICL), $3,850 member out of pocket threshold (OOP max). After meeting the $265 annual dedu ctible, the beneficiary pays 25% of the cost of a covered Part D prescription drug up to an ICL of $2,400. Once the ICL is reached, the beneficiary is liable for the full drug cost, which is called the coverage gap or more commonly known as the donut-hole. Figure 4 Part D Standard Plan Data Source: CMS. 12
13 When the beneficiarys total out-of-p ocket cost (including the deductible, copayments, and spending in the coverage gap, but not the monthly premium) for the year reaches $3,850, he or she reaches the catast rophic coverage, in which he or she pays $2.15 for a generic or preferred drug and $5.35 for other drugs, or 5% coinsurance, whichever is greater. Federa l government pays 80% of the drug cost with the remaining 15% paid by the private insurance plans. The deductible, ICL, OOP max and catas trophic copayments are updated every year to account for the inflation and increasi ng drug costs. Table 1 shows the standard benefits from 2006 to 2009. Table 1 Medicare Part D Defi ned Standard Benefits Part D Standard Benefit Design 2006 2007 2008 2009 Deductible $250 $265 $275 $295 Coinsurance (all tiers) 25% 25% 25% 25% Initial Coverage Limit $2,250 $2,400 $2,510 $2,700 Out-of-Pocket Threshold $3,600 $3,850 $4,050 $4,350 1.2.4 Part D Alternative Benefits The defined standard benefits are not the most common benefits offered by Part D plans. Only 10 percent of plans offer the defined standard benef its. Many plans have used the flexibility allowed by MMA to vary their benefit designs. A majority of plans eliminated at least part of the standard deductible, substituted flat copayments for coinsurance, and adopted tie red cost-sharing where benefi ciaries pay different amounts for different types of drugs. The most comm on approach was to use three or four tiers
14 with different copayment amounts for generi c drugs, preferred brand-name drugs, nonpreferred brand-name drugs and sometimes specialty drugs (e.g. biotechnology products or injectable drugs) (Hoadley, 2006; Duggan, Healy, and Morton, 2008). These alternative plans are categorized as actuarial equivalent, basic alternative, or enhanced alternative dependi ng on benefit structure. Ac tuarial equivalent plans and basic alternative plans are actua rially equivalent to the defined standard plans. The difference lies in how the bene fit structure is adjusted. Ac tuarial equivalent plans can only adjust the coinsurance and are not allowed to change the standard deductible. On the other hand, basic alternative plans can ad just both the deductible and coinsurance. Enhanced alternative plans offer richer benefi ts than defined standard plans, such as lower deductibles and copayments, and partial or full gap coverage. For approval, these alternative bids need to pass certain tests specified by CMS. These tests include Test 1: The total coverage is equal to or greater than that of the defined standard benefit. Test 2: The unsubsidized value is equal to or greater than that of the defined standard benefit. Test 3: The average cost at the ICL is equal to or greater than that of the defined standard benefit. Test 4: Deductible is equal to or less th an that of the defined standard benefit. Test 5: Average catastrophic cost sharing is equal to or less than that of the defined standard benefit.
15 Actuarial equivalent bids only need to pass test 3 and test 5, while basic and enhanced alternative plans are required to pass all five tests. 1.2.5 Plan Formularies One reason for an insurer to offer an alternative plan is to incorporate utilization controls, such as multi-tiered form ularies, into benefit structure. Formulary is a list of drugs covered by the plans. Different from the benefits, there is no such a standard formulary although CMS releases a list of Pa rt D covered drugs. Plans are not required to pay for all Part D covered drugs. Instead, plans can establish their own formularies as long as the formulary and benefit stru cture are not found by CMS to discourage enrollment by certain Medicare beneficiaries. In addition, plans can change drugs on their formulary during the course of the year with a 60-day notice to affected parties. Generally, each plans formulary is organized into tiers, and each tier is associated with a set of copayment amounts. Lower tiers are associated with lower copayments. Most plans offered four-tier formularies. Tier 1 is generic drugs, tie r 2 is preferred brand drugs, tier 3 is non-preferred brand drugs, and tier 4 is specialty and injectable drugs. Some plans may offer 5 tiers by breaking gene ric drugs into prefe rred generics and nonpreferred generics, while some plans may of fer 3 tiers by combining preferred brand and non-preferred brand drugs. The primary difference between the formular ies of different Part D plans lies in the coverage of brand name drugs. Plans can also offer Part D excluded drugs as supplement benefits. However, plans offeri ng excluded drugs are not allowed to pass on
16 those costs to Medicare, and ar e required to repay CMS if th ey are found to have billed Medicare on these cases. Utilization control tools, such as prio r authorization, quantity limit and step therapy, are used to help manage drug use and total costs (Hoadley, 2006). The application of such tools can be an importa nt way for plans to steer beneficiaries to specific drugs as well as to control the use of certain drugs. Yet enrollees may not know whether these tools might create a real barrie r to getting their medication until they first attempt to fill a prescription for a specific drug under their plans. 1.2.6 Part D Bidding Process and Beneficiary Premium Similar to Medicare Part C, Medicar e Part D premiums and subsidies are determined through a competitive bidding process. Firms submit separate Part D bids to CMS on a plan-by-plan base. These bids repr esent the expected cost to the firm for providing the basic benefits (def ined standard benefits) to an individual of average health (individuals with a risk score equal to 1.0). In addition to the bid amount, the projected low income subsidy and federal reinsurance for catastrophic claims are required to be filled in the bid form. Different from Part C, Part D member pr emium is also affected by the national average bid amount and national average federal reinsurance. Each year, CMS calculates the national average monthly bid amount a nd federal reinsuran ce amount. In 2006, the national average bid and federal reinsurance we re calculated on an equal weighting base. In other words, all plans are given equa l weights no matter how many members they enroll. Enrollment weighting replaced the e qual weights in contra ct year 2009. In the
transitional years 2007 and 2008, the national average bid amount was a composite of the two approaches. For example, in 2008, 40% of the national average bid amount was based on the uniform weighted average and 60% was based on the enrollment weighted average (CMS, Apr, 2007). Figure 5 National Average Bid and Member Premium Data Source: Medpac, Part D Payment System (October, 2008) Once the national average bid amount is determined, the national average member premium is calculated as 25.5% of the sum of national average bid and national average federal reinsurance. However, plans may bi d higher or lower than the national average 17
18 bid, the difference becomes (or reduces) the members liability. The members must pay the national average premium plus (or minus) any difference between the plans bid and the national average bid. Figure 5 illustrates how the national average bid and member premium are calculated and how each plans member prem ium is determined. In this example, members who choose plan 2 which is equal to the national average bid pay the national average member premium. Members who choose plan with a higher bid have to pay a higher premium than the national average pr emium. On the other hand, members who choose plan 1 with lower bid pay lower premiums. 1.2.7 Part D Reimbursement Method and Risk Scores Using an approach similar to the medical CMS HCC model, the part D capitation payments are calculated using the CMS RxHCC (prescription drug) model. Different from the medical CMS HCC model, CMS Rx HCC model uses the low-income status instead of Medicaid status. The low inco me beneficiaries include not only Medicaid beneficiaries, but also Medica re beneficiaries whose family income is below the 150% of the poverty line. In addition, the RxHCC m odel used different ICD-9-CM codes and aggregated them into RxHCCs. Similar to the medical CMS HCC model, a Medicare beneficiarys Part D risk score is determined by his or her age, gende r, low-income status, institutional status, disabled status, and RxHCCs in the CMS RxHCC model. A higher risk score indicates a poorer health status. In a ddition, the coefficients of th e above factors in the RxHCC model are updated annually.
Figure 6 Part D Reimbursement Method Data Source: Medpac, Part D Payment System (October, 2008) As illustrated in Figure 6, the plan capitation payments is risk-adjusted. Specifically, the capitation paymen ts are proportional to the Pa rt D risk scores produced by the CMS RxHCC model. Plans are paid more for enrolling sicker members (with higher risk scores) while they are paid less if they enroll healthier members (with lower 19
20 risk scores). The plans are also compensate d by enrolling high risk members in terms of federal reinsurance subsidy and low income members in terms of low income subsidy. The Part D risks scores are not comparable to the medical risk scores since the CMS HCC model and RxHCC mode l are built based on different diagnostic codes. In other words, it is not necessary that Medica re beneficiaries with higher medical risk scores have higher Part D ri sk scores. Therefore, the capitation payments for medical service and for Part D coverage are independent of each other. 1.2.8 Government Subsidy For each Medicare beneficiary enrolled in a MAPD or a PDP plan, Medicare provides plans with a subsidy that averages 74.5 percent of standa rd coverage for all types of beneficiaries (MedPac, Sep 2006). Or, with the exception of low income subsidy plans, the consumer premium is 25.5% of the sum of the bid and federal reinsurance on average. The subsidy take s two forms: direct subsidy and federal reinsurance subsidy. Direct subsidy a capitated payment to plan s calculated as a share of the adjusted national average of plan bids. The direct su bsidy is calculated as the difference between the risk-adjusted bid and the fi xed member basic premium. Federal reinsurance Medicare subsidiz es 80 percent of drug spending above an enrollees catastrophic threshol d. Reinsurance acts as a form of risk adjustment by providing greater federal subsidies for higher cost enrollees. In addition, Medicare establ ishes symmetric risk corrido rs separately for each plan to limit the plans overall losses or pr ofits. Under risk corri dors, Medicare limits a
21 plans potential losses (or gains) by financing some of the higher-than-expected costs (or recouping excessive profits). These corridors are scheduled to widen, meaning that plans should bear more insurance risk over time. (MedPac, Sep 2006) Since 2006, Medicare Part D replaced Medicaid as the primary source of prescription drug coverage for individuals who are dually eligible for Medicare and Medicaid. Special consideration has been give n to the low-income beneficiaries in terms of providing them very rich benefits. Sp ecifically, qualifying low-income beneficiaries are eligible for the special need plans that have no premiums, deduc tibles, or coverage gaps and limited cost sharing ($1 to $5 per prescription). The low income member cost sharing by year is provided in Appendix B, Figure B3 and Figure B4. Plans enrolling LIS members receive additional subsidies from the federal government to cover the beneficiarys premium and additional benefits. 1.2.9 Current Status In 2006, about 65 organizations chose to participate in the Medicare Part D market offering 1,314 MAPD plans and 1,429 PDPs. In 2007 and 2008, the number of organizations and plans increased moderatel y. In 2009, a total of 1,689 PDPs are offered nationwide, down from 1,824 PDPs in 2008. Th ese PDPs are provided by PDP region. In other words, a PDP is required to be ope n to all Medicare beneficiaries in the PDP region that it chooses to enter. In each of the 34 PDP regions defined by CMS, a total of 40-60 PDPs are available to the beneficiaries. In 2009, the number of PDPs per region ranges from a low of 45 PDPs in Alaska region to a high of 57 PDPs in the Pennsylvania/West Virginia re gion. These numbers are down slightly from a range of 47
22 PDPs (Alaska region) to 63 PDPs (Pennsylva nia/West Virginia re gion) in 2008 (Kaiser, Nov, 2008). The average monthly PDP premium in 2009 (unweighted by enrollment) is $45.45. This is a 14% increase from the unw eighted average mont hly premium of $41.02 in 2008, up from $37.43 in 2006. PDP premiums will vary widely by region, ranging from a low of $10.30 per month for a PDP in New Mexico to a high of $136.80 per month for a PDP in New York. (Kaiser, N ov, 2008). This premium variation by region may reflect heath difference beyond those capture d by risk adjusters, variations in the prescribing practices of physicians, and th e extent of expected competition from Medicare Advantage plans. The market share of each organization is rela tively stable, for instance, nine out of the top ten organizations with the highest enrollment in 2006 were also among the top ten organizations by enrollment in 2007. No significant change was found in the market share in 2008. United Health Group, Huma na and Universal American Financial Corporation remain the top three in terms of total enrollment from 2006 to 2008. In August 2008, CMS estimated that the 10-year cost of the Part D program would be $395 billion, down from the orig inal estimated $634 billion. One factor contributing to the lower cost is the increas ed use of generic drugs. This trend is expected to continue as many brand drugs lost their patents recently. As of November 2008, 17.5 million Medicare beneficiaries enrolled in PDPs and 8.6 million Medicare beneficiarie s enrolled in MAPDs. The PDP penetration rate is 39% and the MAPD penetration is 19%. Of these 26 million members, 9.4 million are enrolled as low income members including 6.2 million as full-benefit dual eligibles.
23 Other Medicare beneficiaries have other sour ces of creditable coverage, such as employer group health plans, Veterans Administration, etc. However, based on Department of Health and Human Services (HHS) estimat es, approximately 10% of the Medicare beneficiaries lacked creditable drug coverage in 2007 (Kaiser, Nov, 2008). 1.2.10 Challenges The main challenges for the federa l government are budgeting and financing. CMS, as the administrator of Medicare, has to deal with many issues, such as monitoring PDP and MAPD plan enrollment, market st ability, cost sharing and formulary, lowincome subsidy participation, and the impact of Part D on total drug expenditures and on out-of-pocket spending by Me dicare beneficiaries. The insurance companies face new challenges in addition to the risks in the regular insurance market. The MAPD plans a nd PDPs that choose to enter the Medicare Part D market have to determine what premiu ms to charge, whether to offer alternative benefits, and/or special need plans, and how to structure copayments and the formulary files in order to survive and succeed in the ma rket. Similar to the commercial insurance market, adverse selection and moral hazard ma y exist in the Medicare Part D market. Adverse selection arises if onl y those most likely to have cl aims enroll in the plans while those least likely do not. Thus, part D could fail to meet fina ncial targets if healthy fails to enroll. Adverse selection could also ar ise from consumer shoppi ng across plans to find formularies that include drugs they need; this can cause plans with broad formularies to selectively attract consumers with expensive drug needs, making them unprofitable. The
24 plans also have to face potential moral hazar d, in which the Part D coverage encourages doctors and patients to opt for more medicati ons, and be less selective in keeping down drug costs and insurers respond by making the approval process for branded drugs burdensome (Winter, 2006). In order to overcome the hurdles of the adverse selection and moral hazard, private insurance plan may choose use the utilization control tools such as prior authorization (.i.e., plan a pproval of a particular drug be fore the prescription can be filled), step therapy (i.e., requirement that a less expensive drug be used before the originally prescribed drug can be obtained), or quantity limits (i.e., restrictions on how many pills can be obtained at one time) (Hoadley, 2006). Early evidence has suggested that some plans are flagging a substantial num ber of drugs with these restrictions, while other plans use them far more sparingly, (H oadley, 2006). These utilization control tools are expected to control the enrollees pres cription drug utilizati on and hence lower the plans claim costs. Implementation of the Medicare Part D program brought new challenges not only to the federal government and insurance companies, but also to Medica re beneficiaries. In order to receive Part D bene fits, beneficiaries need to ac tively enroll in either standalone PDPs or MA-PDs during the open en rollment period. Online enrollment is available and encouraged. CMS provides conve nient tools to help Medicare beneficiaries to choose the plans that best meet their needs. For exampl e, Medicare beneficiaries can easily find the plans that cover their medi cations and compare the plan premiums on CMS website. However, some Medicare beneficiar ies fail to enroll in a Part D plan due
25 to the lack of computer knowledge, access to computers or Part D information while some others complain about the complicated Part D benefits and enrollment process. Effectively educating Medicare beneficiaries is a critical issue for the successful implementation of the Medicare Part D program.
26 Chapter Two Literature Review This chapter consists of two sections. Section 1 provides an overview of existing literatures related to the Medicare Part D program. Section 2 sp ecifically reviews the studies focusing on the impacts of insurance characteristics. 2.1 Medicare Part D Program The Medicare Part D program received ex tensive attention from researchers and policy makers even before its inception in 2006. Criticisms were heard frequently as well. For example, the donut-hole made many, especially for those who need drug benefits most, without drug coverage for much of the plan year. Past studies have covered many different aspects of the Part D program, such as program costs, implementation, impacts, benefits, enrollment, etc. To avoid an exhaustive list, we have selected some representative studies, summarized as follows. From the policy makers perspective, Ho adley (2006) discussed the governments challenges in implementing the new Medicare prescription benefits, such as overseeing the enrollment, plan formularies and benefits He mentioned that the programs success would be judged by whether beneficiaries en roll in plans that meet their needs and whether the programs costs are held within reasonable limits. Researchers are more interested in the impacts that the new Medicare Part D program has brought. Lucarelli (2006) found th at the Medicare Part D program has a positive effect on health status and life expect ancy. Blum (2005) measured the impact of
27 enrollment assumptions in the Medicare prescription drug benefit on premiums and federal costs. In 2005, the Congressional B udget Office (CBO) in CMS projected that a significant proportion of Medicare beneficiaries would enroll in the new Medicare Part D program starting in 2006. Blums analysis s howed that the average premiums and total costs could be significantly high er than CBO projections if enrollment is significantly concentrated among beneficiaries who have higher expected drug spending. Medicare Part Ds specific benefits struct ures are also of interest, especially the donut-hole. Stuart (2005) assessed the imp act of coverage gaps (donut-hole) in the Medicare Part D benefits. Th e author found that the discon tinuities in drug benefits resulted in sizable reductions in medicati on use and spending, which is magnified in people with common chronic illness. Individual s with chronic illnesses that result in very high medication use are particularly likely to reach the donut-hole. For example, Patel and Davis (2006) found that the Medicare benefi ciaries with ESRD f ace substantial total expenditure and most of them will reach th e donut-hole. Gold (2006) described the premiums and cost-sharing characteristics of the Medicare Part D benefits offered by all PDPs and MAPD plans in 2006. Hoadley (2006) compared the benefit design and formularies offered by plans in 2006 and 2007. Hoadley (2006) also gave an in-depth examination of formularies and other feat ures of Medicare Pa rt D plans and found significant variation across plans with respect to formularies, cost-sharing and utilization control tools. As the consumer, Medicare beneficiaries have received much attention as well. Winter et al. (2006) found that a majority of the Medicare beneficiaries had information about the program and planned to enroll before the open enrollment began. They
28 expected that enrollees would benefit from the program and showed concern that elderly with poor health or cognitive impairment w ould make poor enrollment and plan choice due to complexity of the competing plan s. Heiss, McFadden and Winter (2006) investigated why some Medicare beneficiaries failed to enroll in the Medicare Part D and found that majorities of the senior are tr oubled by the deductible and gap provision and the stability of the plan formularies. Dual eligibles are not only given extra help from government, but also received extra attention from researchers. Bu chsbaum, Varon and Kagel (2007) gathered information on the ongoing successes and challenges that dual eligibles faced. The dual eligibles reported problems with formulary, utilization control, enrollment, spend-down issues, communication with Part D plans and payment issues. Simon and Lucarelli (2006) are the pioneer researchers who used the econometric models to measure the impacts of the Medicare Part D program. Using the 2006 (the first year of Medicare Part D program) PDP data, they tested how insurers set premiums in the Part D market. Particularly, th ey found that (1) the number of insurers in a market is big enough that it does not appear to affect the premium. (2) the full drug prices are listed appear to be reflected to some degree in the premium charged. (3) weak relationship between premiums and out-of-pocket payments for different set of drugs. (4) the institutional setting and the regional mark et characteristics affect the firms bidding behavior and the resulting prem iums. However, while premiums are clearly important to beneficiaries, given the substantial govern ment subsidies, prem iums may not reflect insurers expected costs for offering a specific benefit package. The premium for a plan reflects the enrollee share of the bid, the difference between the firms bid and the
29 national average bid, plus the full value of any enhanced benefits. The proportion of expected costs covered by the government subsidy can vary widely across plans. 2.2 Effects of Insurance a nd Plan Characteristics This section specifically reviews the st udies focusing on the impacts of insurance characteristics (cost sharing a nd utilization control tools) on the demand and utilization of medical services and prescription drugs, and the impact of plan characteristics on premium setting. 2.2.1 Cost Sharing Many researchers have studied the effect s of insurance characteristics on the demand and consumption of health services. For example, one focus of research has been examining how cost sharing affects the us e of services. Low cost sharing is often linked to higher, potentially inefficient utilizat ion, referred to as moral hazard. On the other hand, higher cost sharing (deductible and coinsurance) reduces the demand for medical service and hence th e total health car e expenditures (Manning, 1987). Such findings exist for total healthcare use as well as for specific services such as preventive services (Solanki, 2000). Cost-sharing also affects prescription drug use. An in crease in the prescription copayment is associated with a drop in the num ber of prescriptions filled (Harris, 1990). Such a reduction may enhance efficiency if the low cost sharing resulted in inefficient utilization. However, such a reduction may ha ve negative consequen ces if the original utilization levels we re not inefficient. Gibson, Oz minkowski, and Goetzel (2005) found
30 that cost-sharing reduces the c onsumption of prescription drug s, and suggests that such reductions have unintended effects on the process and outcomes of therapy. Such unintended effects were f ound by Tamblyn (2001) with increased cost-sharing for prescription drugs in elderly pe rsons and welfare recipients leading to a reduction in drug utilization and a higher ra te of adverse events. Many studies have shown that a tiered cost -sharing structure is an effective tool for insurance companies to control costs. Hu skamp et al. (2005) examined the change in demand behavior after the introduction of a third tier for non-preferred brand drugs. They found that adding a third tier induces a shift to lower tiered drugs and strengthens the plans negotiating power over drug prices. The introduction of a third tier caused individuals to shift from non-preferred brand medications to preferred brand name medications, however, the effect of a tier 2 copayment increase has not been consistently found to cause a shift towards generics (Gibson, Ozminkowski, and Goetzel, 2005). Overall, Joyce et al. (2002) found that plans w ith more tiers have le ss total plan spending. Motheral and Fairman (2001) showed that th ree-tier prescription copayments controls drug costs without changing the use of other medical resources. Gilman and Kautter (2007) focused on Medi care beneficiaries. They found that higher tiered drug plans reduce overall expenditures and the number of prescriptions purchased by Medicare benefici aries. However, they also showed that beneficiaries are less responsive (i.e., demand is less elastic) to cost sharing incentives when using drugs that treat chronic conditions. There are a few studies that measure the relationship between plan benefit structure and premiums, but none of these ar e specific to drug plans. Jensen and
31 Morrisey (1990) measured the relationship be tween group health insurance premiums and policy characteristics including plan benefits, cost-sharing and out-of-pocket expense limits. They found that the member cost-sharing, especially for hospital care, significantly lowers fee-for-service premiu ms. Robst (2006) examined the Medigap insurance premiums and estimated the marginal prices for Medigap benefits. His study showed that the Medigap plans are generally priced in accordance with the actuarial value of the benefits. Some studies focused on the impact of tiered copayments on the enrollees demand behaviors. Overall, cost sharing has been found to reduce consumer demand. Most insurance products in th ese studies were priced usi ng experience rating and thus reflect the expected costs of providing bene fits to enrollees. C onversely, Part D plans started using experience rating in 2008, and bids reflect the expected cost of providing the standard benefits to a person of averag e health. Thus, a relationship between cost sharing and plan bids ma y be less apparent. There are at least two reasons to expect a relationship between cost sharing and firm bids. First, plan bids vary from the na tional average bid, and also vary within each region. Thus, firms have different expectations within a region. In part, expected costs will differ based on the utilization management level of a firm. Given that Part D plans are required to price their products using a ppropriate actuarial methods, plan bids are expected to be lower for plans with lower e xpected costs that results from higher member cost-sharing. In addition, util ization management allows insu rers to better control costs and reduce the degree of uncertainty. A reduction in uncertainty normally leads to a reduction in the risk spread that an insurer builds into the bid. It is, however, difficult to
32 predict how the effects will vary across tiers. For example, cost sharing may have a greater effect on brand name medications than generics if individuals respond to a tier 2 cost share by switching to cheaper generics However, research results have not consistently shown that cost sharing i nduces a shift toward generics (Gibson, Ozminkowski, and Goetzel, 2005). 2.2.2 Utilization Control Tools Plans may also face moral hazard, in which the coverage encourages doctors and patients to opt for more medical service, pe rhaps to the point where the marginal cost exceeds the marginal benefit. In order to reduce moral hazard, plans may choose to use cost management tools. These tools ha ve been used widely by managed care organizations to control the costs. The e ffects have been confirmed by researchers. Feldstein, Wickizer and Wheeler (1988) showed that utilization review program by private insurance companies effectively control th e health service utilization and costs. The most commonly used utilization contro l tools include prior authorization, step therapy, and quantity limits (Hoadley, 2006). Some researchers have conducted the clinical analysis to examine th e impacts of these tools on the u tilization of certain drugs. For example, Goldfarb et al. (1999) showed that implementation of a monthly limit (four tablets or injections) on sumatriptan (a treatment for migraines) decreased an HMO's pharmacy costs. Smalley et al. (1995) found that th e PA requirements may be highly cost effective with regard to expe nditures for drugs that have very similar efficacy and safety, but substantial variation in costs. MacKi non and Kumar (2001) did a critical review of the literature of prio r authorization programs. They f ound that the overall effect of PA
33 programs in controlling drug costs is efficient. Yokoyama et al. (2007) demonstrated that a step-therapy intervention for ARBs that required prior use of an ACEI or an ARB was associated with an approximately 13% lower drug cost per day compared with a health plan with no step-therapy intervention. On the other hand, some researchers hold different views. Panzer (2005) showed that implementing a generic step therapy formulary for selective serotonin reuptake in hibitors (SSRIs) in patients with anxiety disorders may be associated with an incr eased amount of therapy change and early treatment discontinuation, resulting in an overall cost increase to a health plan. Since the inception of the Medicare Part D program, pharmacy utilization control tools including prior authori zation, quantity limit and step therapy have been used by insurance companies to manage drug utilization and total co sts. According to Hoadley (2006), plans varied significantl y in the type of utilization co ntrol tools used to restrict enrollees access to specific drugs, and in th e frequency these tool s were applied. In addition, plans were more likely to apply quant ity limits for covered drugs than to require step therapy, which was applied slightly more often than prior authorization requirements. He also mentioned that at least half of the plans used one or more utilization control tools on fi ve of the top 10 brand-name drugs. Conversely, quantity limit restrictions were far less commonl y used for the top 10 generic drugs. 2.2.3 Premium and Premium Elasticity Insurance premium is one of the favorable research subjects as well. McLaughlin (2002) showed that Medigap premiums vary considerably among geographic markets. They also found a strong positive relations hip between Medigap premiums and HMO
34 participation. Atherly (2004) demonstrated th at premiums have a significant effect on plan selection in the Medicare program. As introduced in the previous sections, Jenson and Morrisey (1990) measured group hea lth insurance premiums and Robst (2006) measured Medigap premium using hedonic pric ing models. In 2007, Robst measured the market structure, regulations and adverse selection as the dete rminants of Medigap supplemental insurance premiums. Simon and Lucarelli (2006) have examined the determinants of premiums in the Part D program. They found that premiums in 2006 were weakly related to beneficiary out-of-pocket costs, and reflected regional characteristics to a greater degree. The price sensitivity of Medicare beneficiaries is of interest to policy makers and researchers. The question of whether Medicare beneficiaries are sensitive to price in the PDP market pertains directly to the justification for pr ivate drug coverage under Medicare (Frakt and Pizer, 2009). However, limited studies have been done to measure Medicare beneficiaries pr emium elasticities. Town and Liu (2003) estimated the monthly semi-elasticity to be -0.009 for a typical Medicare HMO using a mean utility logit model, while the median plan elastic ity is -0.33 conditional on charging a positive premium. Frakt and Pizer (2009) estimated pr ice elasticity in the PDP market using 2007 PDP enrollment data. The authors found a price elasticity of 1.45 with the elastic demand indicating that PDP premiums are clos er to marginal cost than Medicare HMO premiums. This dissertation reexamines price elasticity in the PDP market. There are at least two reasons to revisit this question. Firs t, in 2006 and 2007, plans submitted bids using manual rating due to a lack of experience in the market. In other words, plans used
35 market characteristics to genera te bids, which limited variability in pricing for similar products. In 2008, plans were required to use experience rating to price their products. Experience rating generates greater variability in bids and premiums for similar products than manual rating (Cutler, 1994). Such varia tion is expected to l ead to greater price sensitivity among Medicare PDP enrollees. Second, Frakt and Pizer (2009) assumed that individuals not enrolled in the PDPs purchased a composite outside good, whos e characteristics are not included in the utility function. However, individuals who ar e not enrolled in the PDPs are more likely to enroll in MAPDs, rather than an unknown outside good. In this dissertation, we define MAPD plans as the outside good a nd include MAPD premiums into the utility function. Consistent with Town and Lius ( 2003) analysis of HMOs, the price is defined as the difference in PDP and MAPD premiums. 2.3 Summary Correctly pricing the Part D bid is critic al for the successful implementation of the Medicare Part D program. As we know, an overpriced plan requires enrollees pay higher premiums and represent an inefficient use of the government subsidy. On the other hand, an underpriced plan drives the plan out of bus iness in the long run. According to CMSs guidance, all plan bids should be priced us ing actuarial assumptions. In other words, correctly priced plan bids should be a functi on of the plan characteristics, such as the annual deductible, member cost sharing, drugs on the formulary, etc. Medicare beneficiaries are expected to enroll in plan s that best meet their needs in terms of premium and coverage. The question whether Medicare beneficiarie s are sensitive to
36 price in the PDP market pertai ns directly to the justification for private drug coverage under Medicare (Frakt and Pizer, 2009). After reviewing the existing literature, we found that little res earch has been done to measure the relationship between Medicare Part D plan characteri stics and the Part D bids/premiums, and premium elasticity.
37 Chapter Three Research Design This chapter consists of four sections. Section 3.1 outlines the objectives and hypotheses to be carried out in this dissert ation. Section 3.2 first presents the data sources and information contained in each source, and then discuss briefly the compilation of the data, including the data cl eansing and merging processes. Section 3.3 discusses, in detail, the va riables included in our mode l specifications. Section 3.4 presents our methodology applied and the econometric models derived. 3.1 Objectives and Hypotheses This dissertation examines the stand-alone PDP bids and premiums from different perspectives using 2006-2008 PDP data. First, we consider how the plan characteris tics affect the bids. Bids are equivalent to the total premiums charged by an insurer. This includes the premium paid by the consumer and the portion paid by the federal government. Specifically, we examine the effect of multiple-tiers copayment structure on the PDP bids. We also measure how the relationship between the copayment structure a nd the plan bids varies by tier. As such, we can assess the copayment elasticity across tiers. Further, we decompose plan bid and pr emium changes between 2006 and 2008 into two components, the proportion due to changing plan characteristics and the proportion due to changes in the marginal pric es associated with plan characteristics. While plan characteristics are an importan t determinant of bids and premiums, the
38 actuarial methods used to price those charac teristics are also important. Since 2006 was the first year of the Medicare Part D program, insurers were unable to base their bids on experience and all plans submitted manual ra ted bids. Starting in 2008, plans were required to submit experience rated bids. Each plans 2006 experience was required to be used to develop the 2008 bids. Due to different pricing methods, the relationship between plan characteristics a nd plan bids is likely to differ between 2006 and 2008. Finally, we measure the Medicare benefici aries sensitivity to price in the PDP market. Specifically, we will combine the approaches by Town and Liu (2003) and Frakt and Pizer (2009) to estimate th e elasticity and semi-elasticity of enrollment with respect to PDP premiums. The hypotheses to be tested in this dissertation include Hypothesis 1: The tiered copayments are c onsistent with their actuarial values. Hypothesis 2: The utilization cont rol tools lower the plan bids. Hypothesis 3: Actuarial pricing methods pl ay an important role in explaining the premium and bid difference between 2006 and 2008. Hypothesis 4: Medicare beneficiarie s are sensitive to PDP premiums. 3.2 Description of Data The data used in this dissertation comes from several sources. The major source is the CMS Prescription Drug Plan and Pharma cy Network Files. Other sources include CMS Landscape Source Data, CMS Part D Risk Score by County Data, CMS PDP Penetration Files and CMS monthly Enrollment Files. Some Kaiser Family Foundation data is used, such as 2006-2007 Medicare Be neficiaries by State File. Sections 3.2.1
39 through 3.2.3 describe in detail each of thes e data sources while section 3.2.4 describes the construction of the datasets utilized in this dissertation. 3.2.1 Prescription Drug Plan and Pharmacy Network Files The major data source for this disse rtation is the 2006-2008 CMS prescription drug plan and pharmacy network files. Thes e data are public-use files available to researchers for a fee. It contains formul ary and pharmacy network data for Medicare PDPs and MAPD plans with the exception of employer and PACE plans. These files are updated monthly with updates being available at the end of the first week of each month. This public file is composed of the fo llowing sub-files: Plan Information File, Formulary File, Geographic Locat or File, Beneficiary Cost File, and Pharmacy Network File. These files contain a unique plan identi fier and a formulary identifier that can be used to combine information in these files. Figure B2 in Appendix B shows the diagram of how these files are related. Two supporting crosswalk files are need ed to interpret the codes for the identifiers in these files. 126.96.36.199 Plan Information File The plan information file includes organization contract number assigned by CMS, plan identifier assigned by CMS, unique identifier assigned to the formulary, monthly premium amount, a nnual deductible amount, annual ICL, regional Medicare Advantage plan service area, PDP plan service area, state and county codes.
40 The unique contract number, plan identif ier and formulary identifier allow us to link the plan information file to other file s. Plans service area, state, and county indicators were used to link w ith geographic information files. 188.8.131.52 Formulary File The formulary file provided detailed formulary information including a unique formulary identifier, the 11-digi t NDC (national drug code), the tier level associated with the NDC, indicators for quantity limits, prior au thorization requirement s and step therapy requirement for each NDC. The unique formulary identifier in this file was used to link the plan information file. 184.108.40.206 Beneficiary Cost File Beneficiary cost file contains plan level co st-sharing details by tier. This file also contains contract number and plan number that can be used to link with the plan information file to obtain the characteristics of each plan. 220.127.116.11 Pharmacy Network File The pharmacy network file contains Nationa l Association of Boards of Pharmacy (NABP) numbers for each network pharmacy. It includes indicators for preferred, retail, and mail order. NABP is the independent, international, and impartial association that assists in developing and main taining the standards for the purpose of protecting public
41 health. NABP assigns a unique seven-digit code for each licensed pharmacy in the United States. The Pharmacy network file also contains the common contract number and plan number that can be used to link to the other files. 18.104.22.168 Geographic Locator File The geographic locator file contains c ounty code and name, state name, MA and PDP region codes and description. CMS esta blished 26 MA regions and 34 PDP regions for the administration. MA regional plans and PDPs operate at the regional level. They are required to be open to all the Medicare bene ficiaries in each region they enter. The county code, MA and PDP region codes can be used to link with the plan information file to provide the description of service area for each plan. 22.214.171.124 Supporting Files Two supporting files are needed to interpre t the codes. One is national council for prescription drug programs (NCPDP) data that crosswalk the unique NABP pharmacy number to pharmacy names and addresses in the pharmacy network file. The other one is the MediSpan or First Data Bank data to crosswalk NDCs to drug names in the formulary file. 3.2.2 Other CMS Data Other CMS data used include CMS Part D Risk Score by County, plan enrollment data, PDP Penetration data, and PDP landscape file, etc. These data are updated on either
42 a monthly or an annual basis. All of these data are open to public and can be downloaded from the CMS website. 126.96.36.199 Part D Risk Score File CMS Part D Risk Score by C ounty provide the county leve l Part D risk scores. Only 2006 Part D risk scores were released by CMS. CMS released county level risk score data to help insurance companies pr epare for the 2006 Part D bids because 2006 was the first year of the Medicare Part D program and all plans lacked Medicare beneficiaries Part D risk scores. After 2006, the Part D plans obtained their members risk scores and CMS no longer released the ri sk score information. In this dissertation, the 2006 PDP level risk scores were weighted by over 65 popul ations in each county at the end of each year (2005-2007) to derive the 2006-2008 PDP region level risk scores. 188.8.131.52 PDP Penetration File CMS started releasing the MA and PDP state-county penetration data on its website since May 2008. These files provi de information on the number of Medicare beneficiaries, the number of enrolled, and penetr ation rate by county. In this dissertation, we converted this county level information to PDP region level information. Since the number of Medicare bene ficiaries varies slightly by month, the 2008 Medicare beneficiaries in each PDP region were represented by the monthly average of the Medicare beneficiaries from May 2008 to Sept ember 2008 (the latest information when building the models).
43 184.108.40.206 CMS PDP Monthly Enrollment File CMS has been releasing the plan enro llment data for MAPDs and PDPs on its website since 2006. The plan level enrollment information was updated in 2006 and 2007. Unfortunately, only July enrollment data are available for 2006 and 2007. Since May 2008, this information has been updated on a monthly basis. For consistency, July 2006, July 2007 and July 2008 plan enrollment data were used. 220.127.116.11 CMS Landscape File Since 2006 CMS has been releasing the CMS MAPD Landscape Source Data and PDP Landscape Source Data on an annual basis. These files are generally released two or three months before the calendar year starts. Starting in 2008, the special need plans for dual eligibles or institutional members have been released separately. These files provides the basic plan information, such as contract ID, plan ID, annual deductible, plan type, plan member premium, service area, et c. The service area in the MAPD files and special need plan files is shown by county while the service area in the PDP files is shown by state. 3.2.3 Other Data The 2006-2007 Medicare beneficiary count data were origin ally released by CMS, but are no longer available on the CMS we bsite. These data were obtained from Kaiser Family Foundation. Kaiser Family Foundation is a US based non-profit private operating foundation focusing on the major hea lth care issues facing the nations. It
44 provides summarized updated health data, polic y and other healthcare related information obtained from CMS, states and othe r sources in a timely manner. In this dissertation, the state level information in th ese files was converted to PDP region level information in orde r to merge with other files. 3.2.4 Data Compilation The focus of this dissertation is on the st and-alone PDPs. The premiums (bids) of the MAPDs are mainly determined by the medical benefits, such as inpatient, outpatient, and physician services. Although these plans also cover prescription drugs, the portion of bids for providing drug benef its cannot be separated. Therefore, this dissertation excluded MAPDs and measures PD Ps only. In addition, we study the PDPs in the Unites States only and the PDPs in the territories of the United States, such as Puerto Rico were excluded. By examining the data more carefully, we found and removed some outliers. For example, there is one plan in 2006 (contract ID S5585, plan ID 001) which charged an unreasonably high premium for providing the define d standard benefits. As a result, this plan failed to enroll any members. This plan was likely priced incorrectly and therefore was excluded. Sixteen plans that offered defi ned standard benefits had only one tier on their formulary files with 25% coinsurance. It is likely these plans put all the drugs (both generic drugs and brand name drugs) on one tier. Since the focus of this dissertation is on the tiered copayment structure, these plans were excluded. Each contact has a unique contract ID approved by CMS and each plan under the same contract has a unique plan ID. Each formulary file also has a unique formulary ID.
45 These IDs together with the geographic identif ier were used to merge the files described above. For example, formulary IDs were used to combine the formulary file and the plan information file. Contract ID and plan ID were used to combine the plan information file, beneficiary cost shari ng file, CMS enrollment data, a nd CMS landscape source data. The PDP region number was used to combine th e plan information file with the Part D risk score file and Medi care beneficiary file. Most plans covered medications in four tiers, including tier 1 for generic drugs, tier 2 for preferred brand drugs, tier 3 fo r non-preferred brand drugs, and tier 4 for specialty and injectable drugs. Some plans choo se not to offer tier 3 or tier 4. In this case, tier 3 or tier 4 are coded as uncovere d. Some plans do not offer the typical four tiers. For example, some plans may offer 5 tie rs by breaking tier 1 into preferred generic and non-preferred generics. In this case, we converted it in to the typical four tiers by combining the preferred generic and non-preferre d generic tiers into one tier of generics. Some plans switched the tier orders, for exampl e, they cover specialty drugs on tier 3 and non-preferred brand drugs on tier 4. In this case the tiers are reconstructed to the typical four tier structure. 3.3 Description of Variables This section describes the variables from a modeling perspective, i.e., dependent and independent variables.
46 3.3.1 Dependent Variables Three dependent variables are selected depending on the modeling purposes and needs in this dissertation. To test Hypothesis 1 and Hypothe sis 2, plan bid was chosen as the dependent variable. As introduced in Ch apter One, the Medicare member premium is only 25.5 % of the total plan cost on average. The remaining is paid by the federal government in terms of subsidies. Similar to the member premiums in the commercial insurance market, the plan bids of the PDPs capture the tota l plan cost of providing the prescription drug coverage. Therefore, we sel ected the plan bid as the dependent variable instead of the member premiums. Table 2 National Average Part D Numbers Year Bid Basic Premium Direct Subsidy 2006 $97.00 $33.00 $64.00 2007 $80.43 $27.35 $53.08 2008 $80.52 $27.93 $52.59 The bid each plan submitted to CMS is co mposed of two parts, the basic member premium and government direct subsidy. Th ese amounts are required to be submitted to CMS at a normalized risk score (1.0) base to facilitate the calculation of risk adjusted payments. As introduced in Chapter One, th e basic member premium is also determined by the national average bid, which is also normalized at the risk score of 1.0. For reference, the national average bid, national average member basic premium, and national average government direct subsidy from year 2006 to 2008 are summarized in Table 2.
47 The federal reinsurance which is used together with national average bid to determine the national average premium is not included. The difference between plan bid and nati onal average bid becomes the members liability. In other words, the members must pay the national average premium plus any difference between the plans bid and the nati onal average bid. For the defined standard, actuarial equivalent, and basic alternative plans, members are only required to pay a basic premium while members enrolled in the e nhanced alternative plans have to pay a supplemental premium in addition to the basi c premium. The supplemental premium is not part of, but in addition to the plan bid. Different from the bid, it is based on the projected risk score, not the normalized risk score of 1.0. However, the actual plan bid submitted to CMS is not directly obtainable. The available data only contains the informa tion of total member premiums. For the enhanced alternative plans, the split of the premium (basic vs. supplemental) is unobtainable either. Fortunately, using the national average bi ds and national average member basic premiums in Table 2, we were able to rec onstruct the bids using the following steps. (1) Calculate the national average direct s ubsidy as the difference of the national average bid and the national average member basic premium. (2) Add the national average direct subsidy by year to the member total premiums of each plan. In summary, we computed the plan bi d as the sum of member premium and government direct subsidy for the basic benefit package assuming a risk score of 1.0. For the standard, actuarially equiva lent, and basic alternative plans, the plan bid is simply
48 calculated as the sum of member premium a nd the national average direct subsidy, which equals the actual bids that each firm submitted to CMS. For the enhanced alternative plans, the actual bids submitted by the firm cannot be calculated with available data. Only total member premiums were reported which represents the beneficiary share of standard benefits and the total cost of the enha nced benefits. As with the other plans, the bid is computed as the sum of the member premium and government subsidy, but the computed bid differs from the actual bid submitted to CMS. The computed bid represents the cost of providi ng the basic benefits (at risk score equal to 1.0) plus the actual expected cost of providing the enhanced benefits, not s imply the expected cost of providing the basic benefits. The per member per month bid is transfor med into the natural logarithm due to the skewed distribution of the variable. Using the transformed variable, Whites (1980) test for heterskedasticity did not reject the null hypothesis of homoskedasticity. We also did the normality testing for the log transformed bid. As shown in Appendix B Figure 1, it is approximately normally distributed. Other functional forms of the dependen t variable were attempted too. For reference, we have provided, in Appendix A, the estimation results of using the square root transformation. Instead of plan bid, we also attempted to use the member premium as the dependent variable. Relevant resu lts are presented in the Appendix A for the purpose of comparison. To test Hypothesis 3, we used both th e plan bid and plan premiums as the dependent variables. Logarithm transf ormation was applied to both variables.
49 For Hypothesis 4, each plans market share was used as the dependent variable. The market share is calculated as the ratio of each plans enrollment divided by the total number of Medicare benefi ciaries in each PDP region. 3.3.2 Explanatory Variables For clarity, we categorized the explanat ory variables into six groups, including plan benefit variables, plan ch aracteristic variables, formul ary variables, time variables, and market characteristics variables. The plan benefit variables used in this dissertation include an nual deductible, tier 1 copayment, tier 2 copayment, tier 3 coinsuranc e, and tier 4 coinsurance. In the dataset, some plans offer flat copayments while some plans offer coinsurance (as a percentage of the total drug cost). In order to measure the benefits on the same base, tier 1 and tier 2 coinsurance were converted to copayments while tier 3 and tier 4 copayments were converted to coinsurance using the national median drug costs on each tier (Appendix A, Table A1). For tier 3 and tier 4, we used coinsurance instead of copayment because coinsurance can capture the fact that some plans dont cover tier 3 or tier 4 drugs. According to Kaiser Family Foundations in -depth examination on the formularies of Medicare drug plans in 2006, the median price of generic drugs is $18.11 per script and the median price of brand name drugs is $92.16. For plans that offe r tier 1 coinsurance, the tier 1 copayment is calculated as the product of tier 1 coinsu rance and the average generic drug cost of $18. Similarly, for plan s that offer tier 2 coin surance, the tier 2 copayment is calculated as the product of tie r 2 coinsurance and the average brand name drug cost of $92. For plans that offer tier 3 copayments, the tier 3 coinsurance is
50 calculated as tier 3 copayments divided by the average brand name drug cost of $92. For plans that offer tier 4 copaym ents, the tier 4 coinsurance is calculated as tier 4 copayments divided by $600 which is the minimal specialty drug cost per script defined by CMS. For plans that do not cover tier 3 or tier 4 drugs, the coinsu rance is set to be 100%. Three dummy variables capturing the plan characteristics are included. The first one is whether the plan charges $0 premium to members eligible for full LIS. In other words, these plans can be treated as benchmar k plans which aim to enroll the low income people and their main revenue source is th e government. The second one is whether the plan offers generic drug coverage in the ga p or donut hole. The third one is whether the plan offers all drugs coverage (both generic drugs and brand name drugs) in the donut hole. The coverage ga p or donut hole as a specia l feature of the Medicare standard plans aimed to control total drug spending. Some enhanced alternative plans (approximately 25% of the plans in the samp le) choose to cover generic drugs or all drugs in the donut hole to attract Medicare beneficiaries to enroll. The formulary variables selected include the numbers of drugs on tier 1 to tier 4. The number of drugs on tier 1 or tier 2 wa s transformed by natural logarithm function while the number of drugs on tier 3 or tier 4 was kept at the level due to fact that some plans do not cover tier 3 or tier 4 drugs. In addition, we also in cluded the utilization control tool variables, including the number s of drugs subject to quantity limit, prior authorization, and step therapy.
51 The PDPs are offered by contract year, whic h coincide with the calendar year. As the data used in this dissertation contains the PDPs from 2006 to 2008, two year dummy variables were used to captur e the time effects. They ar e Year07 and Year08 with year 2006 as the reference year. The PDPs are offered at a regional leve l, and a PDP is required to open to all Medicare beneficiaries in th e region. Market characteristic data include beneficiary health status, market size, and the number of competing plans in each PDP region. Beneficiary health status is measured usi ng the average 2006 Part D risk score in the region. The risk score is derived from a prospective model designed to predict medication needs in next year based on observe d diagnoses in the prio r year. Interested readers can refer to Robst, Levy, and Ingber ( 2007) for a detailed desc ription of the Part D risk adjustment model. Only the 2006 count y level risk score data were available from CMS. Thus the county level risk scores we re assumed constant from 2006 to 2008. The calculated risk scores by PDP region from 2006 to 2008 are presented in Appendix A, Table A2. Market size is defined as the number of Medicare beneficiaries in each PDP region, which is presented in Appendix A, Table A3. Another market characteristic variable is the number of competing PDPs within each PDP region. This variable captures the competition level with in each PDP market. 3.4 Methodology This section describes the methodology us ed to test our hypotheses. Hedonic pricing model is used to test Hypothe ses 1 and 2. The decomposition method by
52 Neumark is used to test Hypothesis 3. A m ean utility logit mode l is used to test Hypothesis 4. In addition, we also discuss so me empirical problems and the strategies we used to construct our models. 3.4.1 Hedonic Pricing Model The term hedonics is de rived from Greek word hedonikos which means related to pleasure. The term is frequen tly used by both economists and scientists in other fields. It simply means that one item or measure is judged better than another. In the economic context, hedonics refers to th e utility or satisfaction one derives through the consumption of goods or services. The essence of hedonic pricing is that the price of good is related to the attributes of the pr oduct. Hedonic pricing models examine the relationship between the observed prices and the attri butes of the product. In this sense, it estimates the implicit price of each attri butes the product has, or the consumers willingness to pay for certain attributes a ssociated with the product of interest. Two researchers have made major contri butions to the theoretical work on Hedonic pricing. Lancaster (1966) developed a new approach to consumer theory. He broke away from the traditional approach that goods are the direct objects of utility. Instead, he supposed that it is the properties or characteristics fr om which utility is derived, or the consumers preferences are ex ercised. Rosen (1974) formulated a theory of hedonic prices as a problem in the economy of spatial equilibrium in which the entire set of implicit prices guide both consum er and producer locational decisions in characteristics space. Both approaches linked the observed product prices and the
53 specific amounts of characteristic s associated with each good de fining a set of implicit or hedonic prices. Rosen also advanced the hedonic prici ng theory by identifying the inverse demand curve and examined both consumer and supplier decisions in a perfectly competitive market. Specifically he built the hedonic pricing model through two distinct stages. In the first stage, the marginal or implicit price function was estimated using the regression of the product price on the characteristics. In the second stage, the inverse demand curve or the marginal willingness-to-p ay function was derived by taking the first derivative of the implicit price function estimated in stage one. Other researchers also made considerable contributions to the development of hedonic pricing theory, such as relaxing th e assumptions of perfect competition in hedonic pricing models. Lucas (1977) include d buyer characteristics and Berndt (1995) added firm effects. Recently, the hedonic pricing has been used in the health insurance market. Using a hedonic pricing model, Jensen and Morrisey (1990 ) measured the relationship between group health insurance premiums and policy characterist ics including plan benefits, cost-sharing and out-of-pocket expe nse limits. They also considered other group (buyer) characteristics, such as locati on and industry of the enrollee, and plan (supplier) characteristics, such as whether it is a self-ins ured plan or a commercial plan. More recently, Robst (2006) used a hedoni c pricing model to examine the Medigap insurance premiums and estimated the marg inal prices for Medigap benefits. He considered both product attributes, and buyer/supplier ch aracteristics.
In this dissertation, we proposes to us e a hedonic pricing model to estimate the bid (price) of PDPs as a function of plan characteristics, the ch aracteristics of PDP regions, and the characteristics of insurance companies (see Equation (3.1)). ),_,(k j i ijkInsurer region PDPPlanfBid (3.1) where i indexes PDP plans, j indexes PDP regions and k indexes insurers. Bidijk is the monthly bid for plan i offere d in region j by insurer k.; Plani represents a vector of plan characteristics including cost -sharing, formulary etc; PDP_regionj represents a vector of CMS defined PDP region (one or more states) characteristics; Insurerk represents a vector of insurance company characteristics. 18.104.22.168 Missing Variable Problem Assuming a linear specificat ion in parameters and using the natural logarithm transformation of the PDP bid, Equations ( 3.1) can be more specifically written as: it k j i ijkuYear Insurer region PDP Plan BidLn 4 3 2 10_ )( (3.2) To account for time effects, we added a vector of year dummy variables ( Yeart). ui represents the error term. Assuming that the model specification in Equation 4.2 is correct, we cannot directly estimate this model due to so me missing variables. Many firm level characteristics such as discounts negotia ted with drug companies are not public 54
information. However, these variables are likely to be correlated with the plan benefit variables. For example, plans that recei ve higher discounts from their PBM (pharmacy benefit manager) are likely to offer richer bene fits or lower member cost sharing. Simply excluding these variables will make the model suffer from the omitted variable problem and cause the estimation to be biased. 22.214.171.124 Firm Fixed Effects Model Specification In order to account for the missing firm le vel characteristics, a firm fixed effects model is proposed under the assumption that firm level variables are time-invariant. This is not an unreasonable assumption as most in surance companies keep the same PBM over years and the PBM contracts are not likely to change significantly over years. Use of the firm fixed effects model will remove insu rer characteristics and produce consistent estimates for the plan characteristic variab les and market characteristic variables. The fixed effects transformation, also called within transformation, is obtained by first averaging equation (3.2) fo r all plans offered by the same contract at year t for all contracts, resulting in the following equation: ktk k jk k kuYear Insurer region PDP plan BidLn _______ 4 3 2 _____ 10 _________ )( (3.3) where is the average plan bid and is the averaged plan characteristics in the same contract k,_________)(kBidLnkplan_____ kInsurer is the averaged insurer characteristics, 55
jkregion PDP is the averaged PDP characteristics, and ku is the average error for contract k Next, to erase the insure r characteristics, Equation ( 3.3) is subtracted from Equation (3.2), resulting in Equation (3.4). )() ()_ _() ( )() () ( )_ _() ()()(_______ 4 _______ __________ 2 _____ 1 ______ 4 3 2 _____ 1 _____ k i tk t jk j k i k i tk t jk k jk j k i k iuuyearYear region PDP region PDP planplan uuYearYear Insurer Insurer region PDP region PDP planplan BidLnBidLn Or, we can simply write: it j i iuYear region PDP Plan BidLn 4 2 1_ )( (3.4) where k i iBidLnBidLnBidLn )()()( is the contract-demeaned data on the plan bids, and similarly for , and iPlanjregion PDP_tYear iuThe fixed effect model assumes strict er ogeneity of the explanatory variables on the unobserved effects, which can be expressed as Equation (3.5). 0),_ |(, k j i iInsurer region PDPPlanuE. (3.5) For the fixed effect analysis, is allowed to be any functions of the explanatory variables. )_,|(j i kregion PDPPlan InsurerE56
Equation (3.4) can be estimated using standard econometric methods, such as Ordinary Least Square (OLS), given E quation (3.5) is satisfied and no unobserved heterogeneity. However, the interpreta tion of the estimated coefficients of s is based on Equation (3.2). 3.4.2 Decomposition Model Oaxaca (1973) developed empirical techniques to decompose the wage difference between men and women into two components. The first component is the proportion of the wage gap due to difference in charac teristics between men and women while the second component is the proportion due to difference in the returns to those characteristics. Neumark (1988) built on Oaxacas method to develop a general theoretical model of employer discriminatory behavior. Here we follow Neumarks approach to decompose the plan bid and premium difference between 2006 and 2008. Let )(2006BidLnand )(2008BidLn be the mean of the natural logarithm transformed plan bids for 2006 and 2008, respectively. The average difference in 2006 bids and 2008 bids can be expressed as: )](')('[')()(2006 2006 2008 2008 2006 2008 X XX BidLn BidLn (3.6) where 2006' X and 2008' X are vectors containing the means of the explanatory variables for 2006 and 2008 samples respectively, while 2006 2008''' XXX 2006 and2008 are the estimated coefficients from estimating equa tion (3.2) separately for each year, and is estimated coefficients using combined data from both years. The coefficients represent 57
the marginal price of the associated plan characteristics. The first term on the right hand side of Equation (3.6) is the proportion of bid difference that is due to changes in plan characteristics while the second term is the proportion of the difference due to changes in pricing associated with plan characteristics. 3.4.3 Premium Elasticities This section discusses the methodology for estimating the premium elasticities. Section 126.96.36.199 introduces Berrys mean utility function. Section 188.8.131.52 discusses, in detail, the instrument variables and 2SLS specification. Finally, Section 184.108.40.206 presents the premium elasticity definition. 220.127.116.11 Mean Utility Function Berry (1994) developed a discrete ch oice model to measure the endogenously determined price by price-setting firms. Sp ecifically, a utility logit model was used to estimate demand parameters under imperf ect competition in markets with product differentiation. Berrys approach is well su ited to the PDP mark et (Frakt and Pizer, 2009). This study follows Berrys (1994) appro ach by assuming the consumer indirect utility function as: ijrfr jr jr ijrMarket Plan emium U Pr (3.7) where, i indexes individual, j indexes the plan, f denotes firms, and r indexes PDP regions. is a scalar for plan premium; is a vector of plan characteristics; jremium PrjrPlan58
rMarketis a vector of market characteristics; f indicates unobservable firm characteristics; and ijr denotes the random error. According to the utility theory, a Medicare beneficiary chooses the plan that maximizes his or her ut ility. Utility is a function of the plan premium and known plan characteristics incl uding member cost sharing, drugs on the formulary, and coverage in the gap. Mark et characteristics (regional risk scores and the number of compe ting PDPs) were also included in the utility function assuming that the utility derived may be a function of health, and that individuals benefit from competition bot h directly (through lower premiums) and indirectly (due to better custom er service, more choices). Assuming the random error ijr is independently and identically distributed across individuals, regions and produc ts, the individuals choice of PDPs can be modeled using a conditional logit model (Berry, 1994). Equa tion (3.7) can be rewritten as the following linear marker share equation (Town and Liu, 2003): fr jr jrMarket emium LnjrPlanrLn Pr))(Pr (Pr0 (3.8) where Prjr is the probability of an individua l in region r choosing plan j. Pr0r is the probability of an individual in the same region not choosi ng a PDP, instead choosing an outside good. The outside good is defined using two diffe rent approaches. First, Frakt and Pizers approach is used by defining a com posite good that is consumed by Medicare beneficiaries who are not enrolled in any PDPs. Second, MAPDs are explicitly defined as the outside good, which is similar to Town and Lius approach of defining the 59
Medigap policies as the outsi de good of Medicare HMOs. Medigap plans are viewed as the alternative to MA covera ge because the majority of Medicare FFS members who are not enrolled in Medicare HMOs supplement thei r coverage with Medigap policies. This reasoning also applies to the Medicare Part D market since Medicare beneficiaries who do not enroll in PDPs are most likely to en roll in MAPDs. In th e second approach, the premium in Equation (3.8) becomes the di fference between the PDP premium and the average MAPD premium in the same PDP region. MAPDs are responsible for medical care and prescription drug cove rage. Premiums for medical and drug coverage are not reported separately. Thus, the Part D pr emium for MAPD plans in each region is calculated as the difference between the av erage premium for MAPD plans and MA-only plans. While MAPDs are the most common altern ative to PDPs, there is not a direct correlation between service areas of manage d care plans and PDPs. PDPs must offer products in an entire region, while managed care plans can offer products in specific counties. Given that managed care plans te nd to focus on urban areas, individuals in some rural areas may not have a MAPD option. However, while acknowledging this shortcoming, most enrollees have a MAPD op tion and thus the effect on the estimated price elasticity is examined by explicitly incl uding this option in the utility function. Using market shares as an empirical measure of the probability of enrollee choices, Equation (3.8) can be rewritten as: fr jr jr r jrMarket Plan emium MSMSLn Pr)/(0 (3.9) 60
where MSjr is the market share of plan j in region r and MSor is the market share of the outside good in region r. In order to rem ove company-specific unobserved characteristics from the error terms in Equation (3.9), fi rm fixed effects models are estimated by including categorical variables (f ) for each firm in the specification. 18.104.22.168 Instrumental Variables OLS estimation of Equation (3.9) genera tes biased results because the plan premium is likely to be correlated withf It is standard to assume plan characteristics to be exogenous leaving only the possibility of endogenous premiums (Frakt and Pizer, 2009). Thus, two-stage least squares (2SLS) is used to obtain unbiased estimates. Valid instruments must be correlated with the plan premium but not with unobservable factors that aff ect utility. Town and Lius approach is followed by selecting the maximum, minimum, and mean premiums of the plans offered by the same insurance company in other PDP regions as in struments. These premiums are suitable for instruments because shocks to the marginal co st are reflected in changes in premiums in other regions, holding the charac teristics in other regions co nstant, and those shocks are uncorrelated with the change in plan quality (Town and Liu, 2003). The mean number of competing MAPDs and PDPs in those region s are also included among the instruments leading to a total of five inst ruments for one endogenous variable. 22.214.171.124 Premium Elasticity Definition 61 The premium elasticity and semi-elas ticity of demand are calculated for PDP enrollees using definitions by Dowd et al (2003). The estimated coefficient on the
relative premium ( in Equation (3.9)) can be transfor med into the average plan-level premium elasticity of demand, using Equation (3.10). The percent change in market share due to $1 change in premiu m is given by the semi-elasticity, using Equation (3.11). k ___ __________ ___Pr)1( emium MS (3.10) )1(___MS k (3.11) where, and are the sample average market share and premium across all regions. ___MS___ __________Pr emium62
63 Chapter Four Research Results This chapter presents our research re sults. Section 4.1 pr esents descriptive statistics of the variables, and the results of the hedonic pr icing model with firm fixed effects. Section 4.2 discusses the decompos ition model results. Section 4.3 describes statistics of the variables used in the OL S and 2SLS models and presents the model estimates, together with the PDP premium elasticities. 4.1 Hedonic Pricing Model Results The following section describes the summary statistics of the final dataset used in the hedonic pricing models to test Hypothe sis 1 and Hypothesis 2. Statistical data analysis and fixed effects model estimation resu lts are also discussed in detail in this section. 4.1.1 Descriptive Statistics The sample used for estimating the firm fixed effects model includes 5,101 stand alone PDPs with 1,414 in 2006, 1,865 in 2007, and 1,822 in 2008. 89% of the plans are alternative plans and 25% of the plans offe r some coverage in the donut hole. The descriptive statistics of the variables in this sample are shown in Table 3. As shown in Table 3, the dependent variab le varies significantly from the lowest plan bid of $62.39 offered by Well Point, In c. in 2008 to the highest plan bid of $188.78 offered by United Health Group in 2007. The average plan bid is $94.08. Consistent
64 with the plan bid, the member premium varies significantly too, from the lowest member premium of $1.87 offered by Humana in 2006 to the highest member premium of $135.70 offered by United Health Care Group in 2007. It is interesting to see that both the highest bid and the lowe st bid are offered by large insurance companies. Significant variations were also found for ma jority of the explan atory variables in the sample. For example, tier 1 copayment for generic drugs ranges from $0 to $25 with an average of $5. Tier 2 copayment for pr eferred brand name drugs ranges from $10 to $73 with an average of $28. Instead of using fixed copaymen ts, coinsurance is used for tier 3 and tier 4 member cost sharing. Aver age coinsurance is 67% for tier 3 medications (drugs) and 36% for tier 4 medi cations. The maximum coinsurance in tier 3 and tier 4 is 100%, which indicates the plan does not o ffer medications in these tiers. Such medications may be covered in a lower tier or not covered at all. The minimum values of tier 3 and tier 4 are 25% and 4% coinsuran ce respectively, which indicates that members enrolled in these plans only pay 25% or 4% of the to tal drug cost. Most firms offer a considerable number of medications on their formulary. For the purposes of this study, each NDC is consider ed to be a medication. NDC refers to the National Drug Code, which is a unique 11-digit, 3-segment number assigned to each medication listed under the Section 510 of the U.S. Federal Food, Drug and Cosmetic Act. The first segment iden tifies the manufactures; the second segment identifies a specific strength, dosage form and formulation for a particular firm; the third segment identifies the package size.
65 Table 3 Descriptive Statistics of Variables for the Firm Fixed Effects Model (n=5,101) Variables Mean STD Max Min Bid 94.08 17.14 188.78 62.39 Premium 38.15 16.53 135.7 1.87 Cost sharing Tier 1copayment 5.16 3.3 25 0 Tier 2 copayment 28 7.44 73 10 Tier 3 coinsurance 0.67 0.24 1 0.25 Tier 4 coinsurance 0.36 0.23 1 0.04 # drugs on each tier (in thousands) # drugs on tier 1 4,162 5,623 106,958 599 # drugs on tier 2 1,136 581 10,910 410 # drugs on tier 3 1,093 1,652 20,863 0 # drugs on tier 4 365 393 4,559 0 Utilization controls (in thousands) Quantity limits 756 2,414 37,001 4 Prior authorization 525 412 3,829 13 Step therapy 76 220 3,687 0 Other population and plan characteristics Risk score 0.99 0.04 1.05 0.91 Medicare population (in millions) 1.31 0.98 4.47 0.05 LIS_0prem 0.3 0.46 1 0 Deductible 96.92 122.79 275 0 Gap coverage (Generics only) 0.25 0.43 1 0 Gap coverage (All Drugs) 0.01 0.11 1 0 Year 2007 0.37 0.48 1 0 Year 2008 0.36 0.48 1 0 In this dataset, the average number of generics on tier 1 is over 4,000 and there are over 1,000 preferred brand name medications on an average formulary. The tier with fewest medications is tier 4 (specialty dr ugs), which has 365 medications on an average
66 formulary. The number of medications on each tie r varies significantl y. Specifically, tier 1 medications range from 599 to over 100,000 a nd tier 2 medications range from 410 to over 10,000. Some plans do not offer tier 3 or tier 4 while some other plans cover over 20,000 non-preferred drugs and thousands of specialty drugs. The drugs that have utilization control re present a fairly small percentage of the sample, approximately 11% with quantity limits 8% required for prior authorization, and 1.1% required for step therapy on an average formulary. As seen, quantity limits are most commonly used and step therapy is leas t commonly used. Overall, some plans have a few medications subject to utilization c ontrol while other plans put thousands of medications under utilization control. Not surprisingly, the average risk score (.99) is clos e to the intended national average of 1.0. The budget neutrality require s the national average risk score to be normalized at 1.0 every year. Medicare advantage plans actively seek coding improvements to increase their members risk scores in order to receive more money from CMS. However, Medicare payments come from a fixed pool of money. If increase in risk scores causes the to tal Medicare spending to increase from previous year, CMS applies an adjustment factor to compensate this fluctuation. Another market characteristic variab le, the average number of Medicare beneficiaries in each PDP region is about 1.31 million. Only a small percentage of plans offer the defined standard benefit (11%), of which most offer alternative plans (42%) or enhanced benefit plans (47%). The annual deductible ranges from $0 to $275 with many firms covering a portion or all of the deductible. The mean value of the annual deductible is $96.92. Approximately 25% of
67 plans offer generic drug coverage and 1.2% of plans offer both generic and brand name drug coverage in the donut hole. 30% of the plans in this sample enrolled qualified lowincome Medicare beneficiaries with $0 member premium. The mean values of plan bid amounts and member premiums by year are shown in Appendix A (Table A5). The average plan bid is $94.10 and the average member premium is $38.16 over the three-year study pe riod. The average plan bid in 2006 is highest at $101.48 while the average plan bid in 2007 is the lowest at $89.89. Consistently, the average member premium is the lowest at $36.81. However, the highest average member premium ($40.04) was found in 2008. The mean values of plan bid amounts and member premiums are also shown by PDP region in Appendix A (Table A6). These mean values across PDP regions are relatively stable. The highest average bid ($97.21) and member premiums ($41.23) were found in PDP region 15 (Indiana and Kentucky) However, the regions with lowest average bid and lowest memb er premium ($89.87) differ. PDP region 26 (New Mexico) has the lowest average bid at $89.87 and the PDP region 32 (Californi a) has the lowest member premium at $33.89. Given all other f actors constant higher risk regions are expected to have higher bids for taking mo re risk. However these unadjusted mean values are not consistent with the average Part D risk score in each PDP region. As shown in Appendix A Table A 2, region 11 (Florida) has the highest risk score while region 24 (Alaska) has the lowest risk score. This indicates that the plan characteristics vary across regions. In addition, the average PDP bid and me mber premium across contracts vary considerably as shown in Appendix A (Table A7). The number of plans each contract
68 offers differs, from 1 to 306. There are four contracts that have the highest number of plans (306) for the three-year study period, which are offered by Cigna, Universal American Corporation, Aetna, and United Health Group. Contract S5932 offered by Healthspring, Inc. had the lowest average bid ($75.56) and the lowest member premium ($22.26). Contract S4231 offered by United Heal th Care, Inc. has th e highest average bid ($139.44) and member premium ($75.44). However, by simply looking at these unadj usted average plan bids and member premiums, we cannot draw any conclusions about the relationship between the firm/market characteristics and plan bids. 4.1.2 Statistical Analysis To identify the relationship between the plan bids and member cost sharing, the average plan bids across cost sharing rates are summarized in Table 4. As expected, tier 1 copayments tend to be low in order to encourage the use of generic medications. Tier 2 copayments are much higher for preferred brand medications. Approximately 20% of the plan s do not offer medications in tier 3. Of those that do, coinsurance rates are quite high with more than half of the plans requiring over 50% of the cost to be borne by the consumer. The high rates are intended to encourage enrollees to use preferred brand na me medications. Specialty medications are typically covered in tie r 4. Most plans offer coverage of some specialty medications, of which coinsurance rates are lower than the coinsurance of the non-preferred brand medications. This is not surprising becau se CMS requires that the maximum member coinsurance of specialty dr ugs shall not exceed 33%.
69 However, by looking at the average bids we cannot identify a consistent trend. While bids are expected to decline as cost sharing increased, none of the tiers exhibit such a monotonic relationship. For tier 1, pl ans with medium level costing sharing has the lowest average bid, but for tier 2 plans w ith the highest level cost sharing has the lowest average bid. For tier 3, although the plans without coverage on tier 3 drugs have the lowest average bid, the medium level co st sharing is associated with the higher average bid than the low level cost sharing. In deed, in tier 4 the av erage bid increased as enrollee cost sharing increased. Table 4 Average Bids by Cost Sharing at Each Tier Observations Average bid Tier 1 copayment $0-$4.14 1,503 $97.64 $4.5-$6 2,207 $91.14 $6.5-$25 1,391 $94.91 Tier 2 copayment $10-$24.5 1,842 $90.99 $25-$30 2,069 $98.79 $30.36-$73 1,190 $90.68 Tier 3 coinsurance 50% 1,283 $95.32 >50% 2,745 $95.39 Not Covered 1,073 $89.25 Tier 4 coinsurance 25% 2,586 $92.11 >25% 1,969 $95.26 Not Covered 546 $99.23
70 Of course, there are numerous potential re asons for this unexpected relationship. In Table 5, the relationship between cost sharing and other plan and market characteristics is explored. For example, firms with lower cost sharing may have other plan characteristics that are associated with lower or higher bids. For each tier, the sample is divided based on enrollee cost shar ing (low, medium, and high) and the average numbers of medications availa ble on each tier, and the percentage of medications subject to quantity limits, prior authorization, and step therapy are examined. The enrollee cost sharing levels (low, medium, and high) are consistent with those in Table 4. These variables examined are treated as plan charac teristics, and thus the percentages are not specific to the medications in the tier. Each tier is treated separate ly, thus plans in the lowest group for tier 1 are not necessarily in the lowest grou p for tier 2, tier 3 or tier 4 and vice versa. For tier 1 the clearest finding is that plans covering more medications and imposing fewer utilization controls tend to ha ve higher bids. The medium cost sharing group has the lowest average number of medicat ions covered and also the lowest bids. The medium cost share group also has the gr eatest proportion of me dications subject to quantity limits and prior author ization, which also contribut es to the lower bids. In tier 2, there is little difference in th e number of brand name drugs covered as enrollee cost sharing increases. However, th e number of drugs subj ect to the utilization control tools differs across leve ls of cost sharing. For example, quantity limits are most common among plans with lower cost sharing. Interestingly, firms with the highest cost sharing are the most likely to require prior au thorization and step th erapy. Plans with low level and high level cost sharing have approxi mately the same percentage of total drugs
71 subject to the utiliza tion control tools. These plans al so the approximately same average bid. Table 5 Statistics by Tier Member Cost Sharing Cost sharing Low Medium High T1 Copayment $0-$4.14 $4.5-$6 $6.5-$25 Bid $97.64 $91.14 $94.91 Avg. # drugs on tier 1 4,950 3,612 4,185 Avg. # drugs on tier 2 1,175 1,138 1,090 Avg. # drugs on tier 3 1,113 1,101 1,058 Avg. # drugs on tier 4 266 374 457 % of drugs subject to: Quantity limits 8.90% 14.20% 9.50% Prior authorization 7.50% 8.20% 7.40% Step therapy 1.40% 1.10% 0.80% Observations 1,503 2,207 1,391 T2 Copayment $10-$24.5 $25-$30 $30.36-$73 Bid $90.99 $98.79 $90.68 Avg. # drugs on tier 1 4,425 4,441 3,271 Avg. # drugs on tier 2 1,100 1,174 1,125 Avg. # drugs on tier 3 1,278 1,275 490 Avg. # drugs on tier 4 271 474 320 % of drugs subject to: Quantity limits 14.50% 9.30% 9.00% Prior authorization 7.10% 7.40% 10.00% Step therapy 1.00% 1.00% 1.60% Observations 1,842 2,069 1,190
72 Table 5 Statistics by Tier Mem ber Cost Sharing (Continued) Cost sharing Low Medium High T3 Coinsurance <=50% >50% Not Covered Bid $95.32 $95.39 $89.25 Avg. # drugs on tier 1 3,270 4,261 5,027 Avg. # drugs on tier 2 1,037 1,133 1,271 Avg. # drugs on tier 3 1,480 1,306 0 Avg. # drugs on tier 4 294 374 430 % of drugs subject to: Quantity limits 16.60% 8.70% 12.40% Prior authorization 7.60% 8.20% 6.70% Step therapy 1.50% 1.10% 0.70% Observations 1,283 2,745 1,073 T4 Coinsurance <=25% >25% Not Covered Bid $92.11 $95.26 $99.23 Avg. # drugs on tier 1 3,844 4,074 5,983 Avg. # drugs on tier 2 1,056 1,118 1,580 Avg. # drugs on tier 3 1,033 1,141 1,203 Avg. # drugs on tier 4 401 418 0 % of drugs subject to Quantity limits 14.90% 7.90% 7.80% Prior authorization 7.00% 8.50% 8.50% Step therapy 0.90% 0.90% 2.50% Observations 2,586 1,969 546 A similar relationship exists between cost sharing and quantity limits for tier 3. Plans with the lowest cost sharing are more likely to have quantity limits. Those plans with the low and medium tier 3 cost shari ng cover tier 3 medications, but also have higher bids than plans not covering non-prefer red brand name medications. The majority of plans (2,745) charge substantial coinsurance (>50%) fo r non-preferred brand name drugs, although most plans (4,028) do have so me coverage for such medications.
73 Similarly, most plans (4,555) offer coverage for specialty drugs (tier 4). There is little difference in the number of specialty drugs covered between the plans charging lower coinsurance and the plans charging high er coinsurance. These plans with lower coinsurance have more medications subject to the utilization controls but also have higher bids than the plans with higher coinsurance. However, plans without coverage on tier 4 drugs tend to have the highest bids, which is counterintuitive. One possibility is that these plans placed specialty drugs on lower ti ers, which resulted in higher costs to the plans and thus the higher bids. 4.1.3 Firm Fixed Effects Model Results Through the data discussion in the precedi ng section, it is difficult to draw any quantitative conclusions on the relationship unde rlying the data. To further explore the data variation, we used a firm fixed e ffects model. The estimation results are summarized in Table 6, with th e natural logarithm of the PM PM (per member per month) bid as the dependent variable. Three different specifications are attempted. The first specification includes limited utilization control measures, namely the enrollee cost sharing variables. The second specification adds the number of medi cations covered at each tier, and the third specification adds additional ut ilization controls (n umbers of medications subject to quantity limits, prior authoriz ation, and step therapy). Note that the grouping of explanatory variables is consistent with those in Chapter Three.
74 Table 6 Regression Results: Firm Fixed Effects Model Dependent Variable: Ln (Bid) Base + # Drugs + Utilization Specification Covered Controls Cost sharing Tier 1 copayment -0.0037*** -0.0035*** -0.0036*** (0.0005) (0.0005) (0.0005) Tier 2 copayment -0.0019*** -0.0020*** -0.0023*** (0.0002) (0.0002) (0.0002) Tier 3 coinsurance -0.0018 -0.0085 -0.0237** (0.0085) (0.0089) (0.0093) Tier 4 coinsurance -0.0175** -0.0289*** -0.0223** (0.0082) (0.0097) (0.0099) # drugs covered ln(# drugs on tier 1) --0.0210*** -0.0227*** (0.0042) (0.0042) ln(# drugs on tier 2) -0.0731*** 0.0651*** (0.0091) (0.0092) # drugs on tier 3 (in thousands) -0.0018* -0.0045*** (0.0010) (0.0016) # drugs on tier 4 -0.00004*** 0.00005*** (0.0000) (0.0000) # drugs subject to: Quantity limits (in thousands) --0.0048*** (0.0010) Prior authorization (in thousands) ---0.0373*** (0.0065) Step therapy (in thousands) --0.0457*** --(0.0075)
75 Table 6 Regression Results: Firm Fixed Effects Model Dependent Variable: Ln (Bid) (Continued) Base + # Drugs + Utilization Specification Covered Controls Other plan characteristics Deductible -0.0003*** -0.0003*** -0.0003*** (0.0000) (0.0000) (0.0000) Gap coverage for generics 0.1767*** 0.1752*** 0.1780*** (0.0037) (0.0037) (0.0037) Gap coverage for all drugs 0.1675*** 0.1824*** 0.1872*** (0.0135) (0.0137) (0.0136) LIS_0prem -0.0551*** -0.0560*** -0.0527*** (0.0038) (0.0038) (0.0038) Year 2007 -0.1535*** -0.1510*** -0.1603*** (0.0039) (0.0052) (0.0054) Year 2008 -0.1182*** -0.1134*** -0.1250*** (0.0040) (0.0060) (0.0062) Regional risk score 0.2052*** 0.2035*** 0.2028*** (0.0405) (0.0400) (0.0399) Regional Medicare population -0.0118*** -0.0119*** -0.0118*** (0.0016) (0.0010) (0.0015) N 5,101 5,101 5,101 R squared 0.73 0.73 0.74 Notes: (1) *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level. (2) The specification also includes a categorical variable for each firm. Among the plan benefit variables, hi gher copayments for tier 1 and tier 2 medications lower bids by insurers in a ll three specifications. Similarly, higher coinsurance for specialty medications lowers bids. We also find a negative relationship between tier 3 coinsurance and plan bids in th e third specifications. Overall, there is a
76 negative relationship between enrollee cost shar ing and the plan bid which is consistent with actuarial principles. Enrollee cost sharing affects plan bids in two ways. First, lower enrollee cost sharing means the plan is responsible for la rger portion of the drug cost on a per script base. Second, lower cost sharing encourag es enrollees to use more scripts of the prescription drugs, which is ca lled induced utilization. Thro ugh these two different ways, lower enrollee cost sharing results in higher plan liability (claim costs). In order to cover these claim costs and survive in the Medicare Part D market, plans need to charge higher bids. While the coefficients are statistically significant, the magnitude of the effects is rather small. For example, a $1 increase in the tier 1 copayment reduces the bid by a mere 0.36%. The marginal effect is also small for tier 2, with a $1 increase in the copayment leading to a 0.2% reduction in bi d. However, $1 represents a far larger proportion of the median cost of a generic me dication compared to the median cost of a brand name medication. Hoadley (2006) examin ed the prices of th e top 150 medications in the Part D program. Based on his results, the median costs are $18.11 and $92.16 for generic medications and brand name medications respectively. This implies that plans pay about $13 ($18 minus $5) for a generic medication and $64 ($92 minus $28) for a brand name medication. In terms of elasticity, a 10% decline in the price of a tier 1 medication (from $13 to $11.70) reduces the bid by 0.4%, while a 10% decrease in the median tier 2 medication price (from $64 to $57.60) decreases bids by 1.4%. Despite th e bid being quite inelastic, the effect is larger for pref erred brand name medications th an generics. Our finding is
77 consistent with the study performed by Sim on and Lucarelly (2006). They found a weak relationship between the PDP premiums and the simulated out-of-pocket payments for different sets of drugs. Overall, the sma ll effects suggest that firms do not expect consumers to substantially reduce their quantit y demanded in response to a change in cost sharing. In addition, the re lative higher elasticity of the preferred brand name medications indicates that firm s expect Medicare enrollees to switch to low cost generic drugs if these generic drugs are the substitutes for the preferred brand name drugs. Similarly, small effects exist for tier 3 and tier 4 enrollee costing sharing. For example, as shown in the third specification in Table 6, one percentage point increase in tier 3 enrollee cost sharing w ould result in a 0.02% decrease in plan bid. A plan going from no coinsurance in tier 3 to 100% coinsu rance would reduce plan bid by about 2.3%. Similarly, the third specification in Table 6 in dicates that one percentage increase in tier 4 enrollee coinsurance would reduce plan bid by approximately 0.02%. Plans going from 0% to 100% coinsurance are expected to have 2.2% lower bids. The formulary variables that measure th e number of covered medications in each tier are significantly related to th e plan bids. The number of drugs in tier 1 is inversely related to the bid. Specifically, if the plans increase the number of generic drugs on tier 1 by 1%, the plan bid would be reduced by a pproximately 2.3% indicated by the third specification in Table 6. On the other hand, th e numbers of medicati ons in tiers 2, 3, and 4 are positively related to the plan bid. The third specification in Table 6 also shows that 1% increase in the number medications in tier 2 would increase the plan bid by approximately 6.5%. The second specification indicates an even higher increase in the plan bid. In the third specif ication, the number of medicati on covered in tier 3 is not
78 statistically significant. However, the sec ond specification shows a positive relationship between the number of medicati ons in tier 3 and the plan bi d. Specifically, if a plan increases the number of non-pr eferred brand name drugs by 1,000, the plan bid would increase by approximately 0.18%. Similar results are found for the number of tier 4 medications. If a plan covered 1,000 more specialty drugs, the plan bid would increase by approximately 4% in the second specifica tion and 5% in the th ird specification in Table 6. In conclusion, the more generic medicat ions covered by the plan, the lower the expected costs and the lower the plan bid. In contrast, the more brand name medications covered by the plan, the higher the expected costs and the higher the plan bid. More importantly, the number of medication in tier 2 shows the highest elasticity across the four tiers, which indicates the tier shifting from the preferred brand name drugs to the low cost generic drugs. This fi nding further confirms the tier shifting effect found in the copayment elasticities. Firms employ additional utilization contro l tools to control dr ug spending. Such utilization controls ( quantity limits, prior authorization, a nd step therapy) are expected to reduce expected costs and lower plan bids. Ho wever, the results in Table 6 indicate that only prior authorization is a ssociated with lower bids. If a plan required prior authorization for 1,000 more medications the plan bid would be reduced by approximately 3.7%. The numbers of drugs with quantity limits or re quiring step therapy are positively related to the plan bids. C ontrary to expectations, adding one thousand more medications subject to quantity limit or step therapy would increase the plan bid by approximately 0.48% and 4.6%, respectively, accord ing to the model results in Table 6.
79 There are many possible reasons for this in consistency. First, quantity limits and step therapy may be put into place when in surers include very high cost medications on their formulary. Second, quantity limits and step therapy may be difficult for plans to actually control. For example, when the quant ity limit is reached or step therapy limits access to certain medications, Medicare enro llees may be able to switch to other medications that have equivalent therapeutic effects. These altern ative medications may enable enrollees to work around some utiliza tion controls. Third, quantity limits and step therapy require approval from the insuran ce company, which increases administrative costs and thus increase the plan bids. Finally and most likel y, since the Medicare Part D plans are still at their early age, the firms may not be able to sophisticatedly utilize these complicated utilization control tools to control the drug costs as they are intended to. Or they may not have reflected the potential savings of these tools in the plan bids. Among the other plan characteristics, plan s waiving part or all of the deductible have higher bids than plans that require higher deductibles. According to the model results in specification 3 in Table 6, a $100 deduction (increase) in the annual deductible would increase (decrease) the plan bid by approximately 3%. Consistent results are found in the first and second specifications in Table 6. The signs and magnitudes of the other plan characteristic variables (LIS_0prem, gap coverage of generics, and gap coverage of brand name drugs) are also expected. Bids for plans that offer $0 premium with full low income subsidy are 5.1% lower (e-1 using the beta from specification #3). CM S randomly auto-assigns the new dual eligible enrollees to the Part D plans that are belo w the regional low income subsidy benchmark.
80 In order to get the auto-assigned members, th ese plans generally bid lower than the plans that do not intend to enroll low income members. Covering medications in th e gap also increases plan bids. Plans with gap coverage of generics have a 19% higher bid as indicated by specification 3 in Table 6. Plans that cover brand name drugs in the gap tend to bid 20% higher than the plans that only cover generic drugs in the gap. In total, plans covering both generic and brand name drugs in the gap are approximately 39% higher than plans without any gap coverage. All three specifications show consistent results in terms of sign and magnitude. Our finding shows the huge impact of gap coverage to the plan bids. The ga p or the donut hole plays an important role in controlling the total drug spending as expected. We also measured the effects of market characteristic variables. The results in Table 6 also show that the plan bid is posit ively related to the PD P region Part D risk scores. In other words, plans in high risk regions tend to bid highe r for bearing higher financial risks. Specifica lly regions with 10% higher risk scores tend to bid approximately 2% higher in a ll three specifications. This is what we expected because for the enhanced alternative plans, the risk scores are directly reflected in the member supplemental premium. For the other types of plans, plans may view regions with less healthy beneficiaries as riskier and put more ma rgins in the bids. Cost controls may also be deemed less effective in high risk regions. On the other hand, the number of Medica re beneficiaries in a PDP region is negatively correlated with plan bids. According to the actuarial pricing principles, large population pools mitigate the plans potential risks. The plan bids which capture the plan expected claim costs are expected to be lower. In addition, large population pool may
81 lower per person administrative costs due to economies of scale. Specifically, an increase of one million Medicare beneficiaries in a PDP region results in an approximately 1.2% lower bid as illustrated in Table 6. All th e three specifications show consistent results in term s of sign and magnitude. The results in Table 6 also show that the bids vary across years. Year 2007 bids are found to be 15% lower and year 2008 bids are found to be 12% lower than year 2006 bids by specification 3 in Table 6. The first two specifications in Table 6 show consistent results. These results are not surprising. Since 2006 was the first year of the Part D program, most plans priced thei r bids conservatively due to the lack of any historical information. In 2007, plans tended to price competitively after le arning that the 2006 bids were overpriced and there was the potential for substantial risk corridor payments to CMS. Plans also priced aggressively in or der to increase market share. In 2008, plans are more mature after two y ears of experience in the Medicare Part D market and CMS required plans to develop 2008 bids based on the plans 2006 claim experience if they had any. Thus, 2008 bids are expected to be more stable, which is consistent with our results. 4.1.4 Low and High Risk Region Analysis We further tested whether the relatio nship between the bids and the tiered copayments differ for the plans in high risk regions versus plans in low risk regions. Gilman and Kautter (2007) found that Medicare beneficiaries with chronic conditions are less responsive to the cost sharing incentives of prescription drugs. In this dissertation, we used the Part D risk scores as the proxy variable of chroni c conditions or health status.
82 If the insurers are sophisticat ed enough, they would anticipate that the cost sharing would have a smaller marginal effect on the demand of enrollees in areas wi th higher risk scores and consider this marginal effect when pricing their plan bids. Table 7 Differentiating between Low and High Risk Regions Dependent Variable: Ln (Bid) Regions with risk 1 Regions with risk < 1 Coef Std err Coef Std err Diff Std err Tier 1 copayment -0.0041*** 0.0007 -0.0031*** 0.0007 0.0010 0.0010 Tier 2 copayment -0.0022*** 0.0003 -0.0025*** 0.0004 -0.0003 0.0005 Tier 3 coinsurance -0.0240* 0.0128 -0.0237* 0.0134 0.0003 0.0185 Tier 4 coinsurance -0.0235* 0.0132 -0.0191 0.0146 0.0044 0.0197 N 2,584 2,517 R squared 0.757 0.73 Notes: (1) *** Significant at 1% level; ** Significant at 5% level; Significant at 10% level. In order to explore the effect of cost sh aring in high risk ar eas versus low risk areas, we separate the sample based on the av erage Part D risk in the region. The full specification including all formulary and uti lization control variables in Table 6 was estimated separately for regions with risk sc ores less than 1.0 and for regions with risk scores greater than or equal to 1.0. The re sults are provided in Ta ble 7. We found that cost sharing does not have signi ficantly different effects on pl an bids in PDP regions with healthier versus less healthy residents. While the coefficients differ, none of the differences are statistically significant. T hus, while insurers overa ll price plans higher
83 when the residents of a region are less healthy, the marginal effect of cost sharing on plan bids is not found to differ based on the health of residents. While Gilman and Kautter (2007) found le ss elastic demand of enrollees with chronic conditions using prescrip tion drug claims data, we di d not find such evidence in the pricing of plan bids in our study. The model results of the full specification are provided in Appendix A (Table A8 and Table A9). 4.1.5 Other Model Forms, Function Forms and Variables As introduced in Chapter Three, to meas ure the impact of copayment structure on the plan bids, we also attempted different set of explanatory variables and model forms. First, the inclusion of ex planatory variables denoting whether the plan was an actuarial equivalent plan, basic alternative plan, or enhanced benefit plan was considered. However, the characteristics that differentiate these plans are already included in the plan benefit variables in the specification. In Ch apter One, we introduced the five tests that the alternative plans have to pass in order to get the bids approved. All the tests are directly related to the plan benefits. In othe r words, the type of plan is determined by the plan benefits including annual deductible and member cost sharing. Thus, the addition of these variables did not add expl anatory power to the model. For comparison purposes, we ran ordinary l east square (OLS) re gressions with the results summarized in the Appendix (Table A10). A majority of the explanatory variables in the OLS model have the same signs as those in the firm fixed effects model.
84 However, OLS was abandoned due to the omitted variable problem described in Chapter Three. Different dependent variables were attemp ted too. The firm fixed effects model results using natural log transformed me mber monthly premium as the dependent variable are provided in Appe ndix (Table A11). All the va riables except for the year dummy variables have the same sign as thos e in Table 6. However, the magnitudes are significantly different. In this dissertation, we used the plan bid as the dependent variable because it captures the total expected claim costs of the plan. The importance of using the bid (particularly when transformed) can be seen with a simple example. Assuming a national average bid of $90 and federal re insurance is $10, a plan bidding $115 would have a premium of $50.50 (25.5% of $100 plus the $25 difference between the plan bid and national average bid). A plan bidding $90 would have a premium of $25.50. Thus, a 28 percent difference in bids leads to almo st a 100 percent increas e in the premium. Finally, different functional forms of the dependent variable were attempted. The firm fixed effects model results using square root transformed plan bid on a per-member per-month base as the dependent variable is provided in Appendix A (Table A12). All estimated coefficients have the same sign as those in specification 3 in Table 6. The log transformation is finally chosen because it is more likely to resemble the relationship between the plan bids and the explanatory variables. For example, the percentage change in member cost sharing, number of medications and risk scores are likely to impact the plan bids by certain percentage rather than fixed amounts.
85 4.2 Decomposition Model Results Different actuarial methods were used by firms to price plan bids in 2006 and 2008 (manual rating vs. experience rating). Th is section tests Hypothesis 3 whether the pricing methods play an important role in determining the plan bids and premiums. First, we compare the variable statistics in three datasets (sampl e of 2006 data, sample of 2008 data, and the full sample of combined 2006 and 2008 data). Following this, regression results using the three datasets are presente d. Meaningful decom position results showing whether the bid/premium change can be attributed to changes in plan characteristics or marginal price associated with plan charac teristics (or different pricing methods) are discussed. 4.2.1 Descriptive Statistics Table 8 presents descriptive statistics of three samples previously described, including 1,414 and 1,822 standalone PDPs in 2006 and 2008, respectively. The average bid declined between 2006 and 2008 from $101 to $93, while the average premium increased from $37 to $40. No significant change was found for the member cost sharing variables. Cost sharing changed with insure rs reducing tier 1 copayments for generics, and increasing tier 2 copayments for pref erred brand name medications. Tier 3 coinsurance stayed almost constant at .69 in 2006 and .68 in 2008. Coinsurance for tier 4 specialty medications declined from .50 to .31. Most firms entered the program in 2006 offering a considerable number of medications on their formulary. Firms have covered fewer medications over time as the average number of generics on tier 1 declined from 9,375 in 2006 to 1,860 in 2008.
86 Similarly, the average number of preferred brand name medications has declined from 1,508 to 937. Only tier 4 has seen an increase in the number of medications, which likely represents some of the brand name medications no longer covered in tiers 2 and 3. Table 8 Descriptive Statistics, (2006 and 2008 Data) Variables 2006 Means 2008 Means Difference# Full Sample (Std err) (Std err) (p value) (Std err) (N=1,414) (N=1,822) (N=3,236) Bid 101.48 92.63 -8.85 96.5 (12.80) (19.95) <.0001 (17.74) Premium 37.48 40.04 2.56 38.92 (12.80) (19.95) 0.0479 (17.24) Cost sharing Tier 1 copayment 5.52 5.24 -0.28 5.36 (3.20) (3.44) 0.0002 (3.34) Tier 2 copayment 26.69 29.78 3.10 28.43 (8.35) (7.22) <.0001 (7.88) Tier 3 coinsurance 0.69 0.68 -0.01 0.68 (0.24) (0.24) 0.6964 (0.24) Tier 4 coinsurance 0.498 0.309 -0.19 0.392 (0.34) (0.13) <.0001 (0.26) # drugs on each tier # drugs on tier 1 9,375 1,860 -7,516 5,144 (8,699) (282) <.0001 (6856) # drugs on tier 2 1,508 937 -571 1,186 (786) (357) <.0001 (650) # drugs on tier 3 1,515 834 -680 1,131 (2,707) (793) 0.0336 (1916) # drugs on tier 4 284 400 116 349 (426) (355) <.0001 (392)
87 Table 8 Continued Variables 2006 Means 2008 Means Difference# Full Sample (Std err) (Std err) (p value) (Std err) Utilization controls Quantity limits 1,414 552 -862 929 (4,482) (361) 0.1283 (3,005) Prior authorization 662 473 -189 555 (547) (242) <.0001 (415) Step therapy required 103 79 -24 90 (383) (113) <.0001 (267) Other population and plan characteristics Risk score 0.99 0.99 0.00 0.99 (0.037) (0.037) 0.400 (0.037) Medicare Population (in millions) 1.30 1.33 0.03 1.31 (0.976) (0.998) 0.126 (0.988) LIS_0prem 0.28 0.27 -0.01 0.28 (0.449) (0.444) 0.520 (0.446) Deductible 90.58 104.85 14.28 98.61 (115.2) (128.8) <.0001 (123.3) Gap coverage (Generics) 0.16 0.29 0.13 0.23 (0.362) (0.454) <.0001 (0.421) Gap coverage (All drugs) 0.02 0.00 -0.02 0.01 (0.151) (0.023) <.0001 (0.102) Note: # p-value from two tailed Mann Whitney U test. A minority of covered medications had utiliz ation controls such as quantity limits, prior authorization requireme nts, or step therapy requirements in both years. Approximately 11% of covered medications were subject to quantity limits in 2006. Despite an insignificant change in the number of medications subject to quantity limits, given the decline in the number of covere d medications the percentage of covered
88 medications subject to quantity limits increased to 13.7%. While the number of medications subject to prior au thorization and step therapy de clined, they also comprised a higher percentage of covered medications in 2008. Five percent of covered medications were subject to prior author ization in 2006 and 12% in 2008, while .8% were subject to step therapy in 2006 and 2.0% in 2008. In 2006, approximately 16% of the plans c overed generic drugs in the coverage gap and 2% of the plans covered brand name drugs. In 2008, 29% of the plans covered generic drugs while less than 0.1% of the plans covered brand name drugs in the coverage gap. Covering brand names drugs in the gap increased plan liability while covering generic drugs in the gap encouraged enrollees to use more low-cost generic drugs. The average risk score was close to the national average of 1.0 in both years. The number of plans offering $0 premium to qualified low-income people (LIS_0prem) was consistent between the two years. The av erage deductible increas ed as CMS updated the standard deductible amount over time. 4.2.2 Firm Fixed Effects Model Results Table 9 presents the results from the firm fixed effects regressions using the natural log of the per member per month bid as the dependent variable The results from three regression models are reported. The fi rst uses data from 2006, the second uses data from 2008, while the third uses the combined sample.
89 In 2006, higher cost sharing was associated with lower bids. The relationship between bids and the number of drugs cove red was not strong with only the number of tier 4 specialty medications was associated wi th higher bids. The only other utilization control related to bids was the number of dr ugs subject to prior authorization which was associated with lower bids. Plans with gap coverage or offered in regions with higher risk scores had higher bids, while plans offe ring $0 premium low-income plans or offered in regions with more Medicare residents had lower bids. By 2008, the relationship between plan bids and cost sharing in tiers 1, 2, and 3 declined in magnitude. Only the tier 4 coinsurance rates be came more strongly associated with plan bids, although unexpected ly, higher cost sharing was associated with higher plan bids. The marginal bid associated with the numbe r of covered medications in tiers 1 and 2 increased in magnitude. The number of covered generics had a more negative effect on plan bids, while marginal pr ice associated with the number of preferred brand name medications increase d. Tier 3 and tier 4 medicat ions are not found to have a significant impact to 2008 plan bids, although they are larger in magnitude in 2008 than in 2006. Among the utilization control variables, only the marginal importance of step therapy changed significantly, although this utilization control became significantly related to higher bids. The relationship between other plan and population characteristics changed differently from the 2006 sample to the 2008 sample. For example, the importance of regional characteristics including regional Part D risk scores and number of Medicare
beneficiaries, declined between 2006 and 2008, while the relative importance of generic gap coverage increased. Table 9 Firm fixed effects Model Estimates, Dependent Variable: Ln(Bid) (2006 and 2008 Data) Variables2006 Sample2008 SampleDifferenceFull Sample Cost sharing Tier 1 Copay -0.0107***-0.0036***0.0071***-0.0023*** (0.0009)(0.0008)(0.0012)(0.0007) Tier 2 Copay -0.0021***-0.0009*0.0012**-0.0034*** (0.0002)(0.0006)(0.0006)(0.0003) Tier 3 Coinsurance -0.2099***-0.01650.1934***-0.1316*** (0.0139)(0.0222)(0.0262)(0.0131) Tier 4 Coinsurance 0.11040.5267***0.4163***-0.0142 (0.1103)(0.0962)(0.1463)(0.0125) # drugs on each tier LN (# drugs on tier 1) -0.0893-0.5449***-0.4556**0.0155*** (0.1100)(0.1632)(0.1968)(0.0044) LN(# drugs on tier 2) 0.02310.104**0.0809*0.111*** (0.0148)(0.0436)(0.0460)(0.0110) # drugs on tier 3 (in 1,000s)-0.0071-0.0285-0.0214-0.0185*** (0.0048)(0.0258)(0.0263)(0.0020) # drugs on tier 4 0.0018***0.0011***-0.0008 0** (0.0006)(0.0002)(0.0006)0.0000 Utilization controls Quantity limits (in 1,000s) 0.001-0.016-0.0170.0156*** (0.0061)(0.1065)(0.1066)(0.0013) Prior authorization (in 1,000s)-0.1064**-0.04330.0631-0.0301*** (0.0500)(0.1350)(0.1439)(0.0075) Step therapy required (in 1,000s)0.01141.9315***1.9201***0.0483*** (0.0202)(0.2413)(0.2422)(0.0089) 90
Table 9: Continued Variables2006 Sample2008 SampleDifferenceFull Sample Other population and plan characteristics Risk score 0.3751***0.053-0.3221***0.1959*** (0.0355)(0.0550)(0.0654)(0.0513) Medicare Population (in millions)-0.0136***-0.0065***0.0071***-0.0101*** (0.0014)(0.0021)(0.0025)(0.0020) LIS_0prem -0.0709***-0.0147**0.0562***-0.0517*** (0.0038)(0.0066)(0.0076)(0.0049) Deductible -0.0005***-0.0002***0.0002***-0.0004*** (0.0000)(0.0000)(0.0000)(0.0000) Gap coverage (Generics only)0.0533***0.287***0.2336***0.1766*** (0.0052)(0.0055)(0.0076)(0.0048) Gap coverage (All Drugs)0.3748***-0.0412-0.416***0.213*** (0.0103)(0.1045)(0.1050)(0.0182) (N=1,414)(N=1,822) (N=3,236) R-Square 0.89810.8679 0.743 Percent due to characteristics change 72% Percent due to coefficients change 28% *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level At the bottom of Table 9, the coefficien ts along with the variable means from Table 8 are used to estimate the percentage difference in bids due to changes in plan characteristics and the proportion due to change s in the marginal prices associated with the plan characteristics. Using Neumarks (1988) approach, 72% of the difference in plan bids is due to changes in plan characteristics while 28% of the difference is due the marginal prices associated with the plan char acteristics. Thus, the ma jority of the change in bids is due to changes in the plan characteristics. 91
Using the same approach, plan premium change between 2006 and 2008 is also decomposed into characteristic change and ma rginal price change. Table 10 presents the results from the firm fixed effects regressi ons using the natural log of member monthly premium as the dependent variable. Table 10 Firm fixed effects Model Estimates, Dependent Variable: Ln(Premium) (2006 and 2008 Data) Variables2006 Sample2008 SampleDifferenceFull Sample Cost sharing Tier 1 Copay -0.0263***-0.0112***0.015***-0.0138*** (0.0029)(0.0020)(0.0035)(0.0018) Tier 2 Copay -0.0076***-0.0010.0066***-0.0084*** (0.0007)(0.0013)(0.0015)(0.0008) Tier 3 Coinsurance -0.2513***-0.03180.2196***-0.2877*** (0.0478)(0.0539)(0.0720)(0.0324) Tier 4 Coinsurance 0.3351.8003***1.4654***-0.0961*** (0.3800)(0.2330)(0.4458)(0.0309) # drugs on each tier LN (# drugs on tier 1) -0.9931***-2.5928***-1.5997***-0.0846*** (0.3792)(0.3953)(0.5478)(0.0108) LN(# drugs on tier 2) 0.2799***0.6227***0.3428***0.2665*** (0.0510)(0.1056)(0.1173)(0.0273) # drugs on tier 3 (in 1,000s)0.023-0.0715-0.0945-0.0522*** (0.0166)(0.0625)(0.0647)(0.0050) # drugs on tier 4 0.0051***0.0041***-0.0010.0001*** (0.0020)(0.0005)(0.0020)0.0000 Utilization controls Quantity limits (in 1,000s)-0.01680.02990.04670.0425*** (0.0209)(0.2579)(0.2587)(0.0031) Prior authorization (in 1,000s)0.19180.44760.2558-0.1791*** (0.1723)(0.3269)(0.3696)(0.0185) Step therapy required (in 1,000s)0.03244.857***4.8246***0.1696*** (0.0697)(0.5846)(0.5888)(0.0220) 92
Table 10: Continued Variables2006 Sample2008 SampleDifferenceFull Sample Other population and plan characteristics Risk score 1.2604***0.2877**-0.9727***0.7065*** (0.1222)(0.1331)(0.1807)(0.1270) Medicare Population (in millions)-0.045***-0.0195***0.0255***-0.0322*** (0.0047)(0.0050)(0.0069)(0.0048) LIS_0prem -0.2275***-0.0576***0.1699***-0.1594*** (0.0131)(0.0161)(0.0207)(0.0122) Deductible -0.0013***-0.0004***0.0009***-0.0011*** (0.0001)(0.0001)(0.0001)(0.0001) Gap coverage (Generics only)0.1185***0.613***0.4945***0.4333*** (0.0180)(0.0133)(0.0224)(0.0120) Gap coverage (All Drugs)1.2687***0.0147-1.254***0.6226*** (0.0356)(0.2531)(0.2556)(0.0452) (N=1,414)(N=1,822) (N=3,236) R-Square 0.87720.8551 0.732 Percent due to characteristics change 99% Percent due to coefficients change 1% *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level As shown in Table 10, three regression results using different samples are presented. The results for premiums are consis tent with the bid results. Cost sharing is associated with lower premiums, while the number of covered brand name medications and gap coverage are associated with higher premiums. The estimates of other plan and population variables are consistent with the bi d results too. For example, coverage of a greater number of generic medications is a ssociated with lower premiums. Medicare beneficiaries in regions with higher risk scores had higher premiums, while Medicare 93
94 beneficiaries in regions w ith a greater number of Medicare enrollees had lower premiums. The marginal effect of cost sharing (in tiers 1, 2, and 3) and the regional characteristics declined between 2006 and 2008. On the other hand, the marginal effect of the quantity of covered medications (in ti ers 1 and 2) and the availability of gap coverage increased. The utilization controls became more important in determining the plan premiums in 2008 than in 2006. Despite changes in a number of coefficients, the effects largely offset. Surprisingly, nearly all of the premium difference was due to changes in plan characteristics between the two years while only 1% of the difference is due to changes in the marginal price of th e plan characteristics. In conclusion, changes in the average bids and premiums are primarily due to changes in plan characteristics between year 2006 and year 2008. 72% of the change in bid and 99% of the change in premium can be attributed to changes in plan characteristics. Different act uarial pricing methods are not found to be the key factor in explaining the bid and premium difference between 2006 and 2008. 4.3 Premium Elasticities As introduced in section 1.2.1, CMS auto-enrol ls or facilitate-enr olls for Medicare beneficiaries who are eligible for LIS. The LIS beneficiaries pay no or little premium and cost sharing; therefore, plan premiums will not be related to their demand for prescription drug coverage. They must be excl uded from the analysis of enrollment with respect to premium (Frakt and Pizer, 2009) Although LIS beneficiaries can choose to
95 enroll in any plans, the vast majority of th em remain in the plans to which they were auto-assigned (Neuman et al., 2007). Given this fact and the fact that Medicare non-LIS beneficiaries are not allowe d to enroll in the benchmark plans, we excluded the benchmark plans from our analysis. Specifi cally, a subset of 2008 PDP data containing only non-benchmark plans is used to measure the price sensitivity of Medicare beneficiaries. Following the descriptive statistics in Subsection 4.3.1, Subsection 4.3.2 presents the OLS and 2SLS regression results together with the estimated premium elasticity and semi-elasticity. 4.3.1 Descriptive Statistics Table 11 presents the descriptive statis tics of the sample data comprising 1,296 non-benchmark PDPs in the United Stated fo r year 2008. As indicated in Table 11, the data indicate reasonable variation across mo st of the variables. For example, the minimum tier 1 copayment is $0 in contrast to the maximum of $18. Some plans choose to cover over 2,000 medications on tier 1 wh ile some other plans covered only a few hundred on tier 1. Also indicated in Ta ble 11, the minimum PDP premium is $12.90 per month while the highest is over $100. The average PDP pr emium of $45.72 is $20 higher than the premium for MAPDs. The differential pa rtly reflects the fact that the majority of PDP enrollees are Medicare fee-for-service me mbers, who are generally less healthy than MAPD enrollees. For market share, the m ean is 0.48% with a maximum of 10.7%. This indicates that the PDP market was domina ted by a few large insurance companies in 2008.
96 Table 11 Descriptive Statistics, 2008 Non-benchmark Plans (n=1,296) Variables Mean Std. Dev Max Min PDP premium 45.72 20.48 107.5 12.9 Premium avg. MAPD premium 20.15 21.49 85.79 -28.58 Market ahare 0.48% 1.20% 10.70% 0.00% Cost sharing Tier 1 copayment 5.28 3.16 18 0 Tier 2 copayment 30.29 7.1 45 15 Tier 3 coinsurance 68% 21% 100% 25% Tier 4 coinsurance 30% 8% 100% 8% # of drugs on each tier # of drugs on tier 1 1,885 271 2,282 599 # of drugs on tier 2 933 352 3,360 468 # of drugs on tier 3 897 782 3,007 0 # of drugs on tier 4 357 288 1,359 0 Utilization controls Quantity limits 565 365 1,860 4 Prior authorization 471 233 2,961 71 Step therapy 84 114 424 0 Other plan characteristics Deductible 56.29 104.12 275 0 Gap coverage 0.4 0.49 1 0 Market characteristics Risk score 0.99 0.04 1.05 0.91 Number of Competing PDPs 53.84 3.08 63 47 Instrument variables Mean premium 41.63 12.46 82.86 19.9 Max premium 69.9 22.88 107 0 Min premium 19.94 10.77 63 9.8 Number of PDPs 51 1 56 50 Number of MAPDs 56 4 105 25
97 We also introduced a new variable of market characteristics, the number of competing PDPs, and 5 instrument variables for the endogenous premium, including the mean, maximum, and minimum premiums, th e number of PDPs and the number of MAPD plans in other service re gions. The number of competing PDPs varies moderately from 47 to 63. The mean, maximum and mini mum premiums in other service regions show considerable variation in the range of $53 to $107. The number of PDPs in other service regions is relativel y stable ranging from 50 to 56. Conversely the number of MAPD plans in other service regions starts as low as 25 while ends as high as 105. Compared to the full sample of 2008 in Table 8, the non-benchmark plans have slightly higher member cost-sharing except for tier 4. In addition, these non-benchmark plans cover approximately the same amount of medications on each tier and utilization controls as the full sample. No signifi cant difference was found for other plan and market characteristics variables between the two samples. 4.3.2 OLS and 2SLS Model Results Equation (3.9) using the composite outsi de good was estimated by OLS and 2SLS with the firm fixed effects model. The regression results, elasticities ( e ) and semielasticities ( k ) are presented in Table 12. The OLS-estimated elasticity and semielasticity are -0.5 and -0.01, respectively. However, give n that plan premiums are endogenous, OLS estimates are biased. Consiste nt estimates are obtained via 2SLS. The 2SLS-estimated elasticity and semi-elasticity are -1.80 and -0.04 are over three times the magnitude of the OLS-estimated elasticities. Frakt and Pizer (2009) also found that the 2SLS-estimated elasticities were greate r in magnitude than the OLS ones.
Table 12 Regression Results Assu ming Composite Outside Goods Variables Estimates Estimates Estimates PDP Premium --0.04(0.0047)***-0.011(0.0032)*** Intercept -346.849(44.5126)***1.848(3.2797)9.811(2.9980)*** Cost Sharing Tier 1 Copayment -0.482(0.0976)***-0.011(0.0135)-0.025(0.0140)* Tier 2 Copayment 0.446(0.0567)***0.013(0.0076)*0.007(0.0080) Tier 3 Coinsurance -4.695(1.9313)**-0.238(0.2564)0.378(0.2516) Tier 4 Coinsurance 9.805(4.3833)**1.64(0.6362)**1.148(0.4746)** # of Drugs on Each Tier Ln (# of Drugs on Tier 1) 8.617(2.4408)***0.924(0.3689)**0.061(0.3432) Ln (# of Drugs on Tier 2)-3.613(0.9110)***-1.327(0.1382)***-1.387(0.1354)*** # of Drugs on Tier 3 -0.001(0.0006)0.001(0.0001)***0.001(0.0001)*** # of Drugs on Tier 4 0.003(0.0012)**0.001(0.0002)***0.001(0.0001)*** Utilization Controls Quantity Limits 0.005(0.0011)***0.001(0.0001)***0.001(0.0001)*** Prior Authorization 0(0.0017)-0.002(0.0002)***-0.002(0.0002)*** Step Therapy -0.002(0.0035)-0.004(0.0005)***-0.005(0.0005)*** Other Plan Characteristics Annual Deductible -0.024(0.0034)***-0.007(0.0005)***-0.006(0.0005)*** Gap Coverage 27.712(0.5924)***-0.718(0.1316)***-1.169(0.1321)*** Market Characteristics Risk Score -1.693(7.0763) -2.7(1.0461)**-3.225(1.0745)*** Number of Competing PDPs-0.054(0.0923)-0.056(0.0129)***-0.073(0.0133)*** Instrument Variables Mean Premium 0.345(0.1698)**--Max Premium 0.138(0.0476)***--Min Premium 0.467(0.1425)***--Number of PDPs 5.949(0.7579)***--Number of MAPDs -0.069(0.0667)--R-square =0.84R-Square =0.54R-Square =0.48 e =-1.804 e =-0.496 k =-0.04 k =-0.011 Standard Error Standard Error Standard Error Two-Stage Least Square OLS First Stage CoefficientSecond Stage Coefficient *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level 98
99 These estimates are greater in magnitude than Frakt and Pizer (2009) found for 2007 ( e = -1.45 and k = -0.039). As expected, with the implementation of experience rating, the estimated elasticity and semi-elast icity for 2008 are greater in magnitude than the estimated elasticity in 2007. The results ma y also suggest that with another year of knowledge on the PDP products, Medicare benefi ciaries are more informed to choose the plans that best fit their needs and thus are more sensitive to plan premiums. Although the welfare study of Medicare HM Os by Town and Liu (2003) does not include the Part D market, it is worth comparing our elasticity estimated with theirs. They estimated a premium elasticity of de mand of -0.33 in the Medicare HMOs between 1993 and 2000, which is significantly lower than our 2SLS estimates. Frakt and Pizer (2009) gave explanation for the higher PDP premium elasticity. The PDP market has a large number of entrants due to the low fixed costs of entry. PDPs do not have to establish pr ovider networks as Medicare HMOs or employer sponsored plans do. Medicare beneficiaries have a larg e number of PDPs av ailable to choose from and hence are more sensitive to price change. Equation (3.9) is also estimated by defining MAPDs as the outside good. Both OLS and 2SLS estimates, together with the premium elasticity and semi-elasticity, are presented in Table 13. It is important to point out that the plan premium in Equation (3.9) now becomes the difference between the PDP premium and the average MAPD premium in each PDP region. The market share of the outside good, MS0, is the market share of aggregate MAPDs in each PDP region.
Table 13 Regression Results Assu ming MAPDs as Outside Goods Variables Estimates Estimates Estimates PDP Premium MAPD Premium --0.039(0.0052)***-0.002(0.0031) Intercept -389.525(57.0340)***6.39(3.6245)*16.791(3.2397)*** Cost Sharing Tier 1 Copayment -0.443(0.1251)***-0.016(0.0149)-0.034(0.0152)** Tier 2 Copayment 0.412(0.0727)***0.012(0.0084) 0(0.0086) Tier 3 Coinsurance -5.618(2.4745)**-0.336(0.2833)0.435(0.2722) Tier 4 Coinsurance 16.814(5.6163)***1.909(0.7031)***1.252(0.5172)** # of Drugs on Each Tier Ln (# of Drugs on Tier 1)8.198(3.1273)***0.976(0.4077)**-0.235(0.3683) Ln (# of Drugs on Tier 2)-3.854(1.1673)***-1.334(0.1527)***-1.396(0.1477)*** # of Drugs on Tier 3 -0.001(0.0008)0.001(0.0001)***0.001(0.0001)*** # of Drugs on Tier 4 0.004(0.0016)***0.001(0.0002)***0.001(0.0002)*** Utilization Controls Quantity Limits 0.005(0.0014)***0.001(0.0001)***0.001(0.0001)*** Prior Authorization 0(0.0021)-0.002(0.0003)***-0.002(0.0002)*** Step Therapy -0.004(0.0045)-0.004(0.0006)***-0.005(0.0005)*** Other Plan Characteristics Annual Deductible -0.021(0.0044)***-0.008(0.0005)***-0.007(0.0005)*** Gap Coverage 27.597(0.7591)***-0.734(0.1454)***-1.429(0.1345)*** Market Characteristics Risk Score -36.208(9.0668)***-1.904(1.1560)*-2.606(1.1721)** Number of Competing PDPs0.224(0.1182)*-0.136(0.0143)***-0.151(0.0145)*** Instrument Variables Mean Premium 0.488(0.2176)**--Max Premium 0.13(0.0610)**--Min Premium 0.307(0.1826)*--Number of PDPs 6.775(0.9711)***--Number of MAPDs -0.136(0.0855)--R-square =0.75R-Square =0.51R-Square =0.46 e =-0.778 e =-0.043 k =-0.039 k =-0.002 Standard Error Standard Error Standard Error Two-Stage Least Square OLS First Stage CoefficientsSecond Stage Coefficients *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level 100
101 In Table 13, the OLS-estimated elasticity (e = -0.04) and semi-elasticity (k = 0.002) are much lower than those for the com posite outside good in Table 12. The 2SLSestimated elasticity ( e = -0.78) is less than half of that in Table 12 while the semielasticity ( k = -0.04) is similar to that in Table 12. The estimated price elasticity is much smaller when explicitly including an outside good. The reason for this is straightforward. Given the average PDP premium of $45, a ten percent price increase would be a $4.50 premium increase. However, when following Town and Liu (2003) and defining the price as the difference between the premium and the premium of the outside good, ceteris paribus that 10% increase in premium results in a 22% increase in the premium differe nce (the $4.50 increase in the $20 average difference). Thus, consumers app ear much more price sensitive when the outside good is not explicitly included in the analysis. Including the MAPD product as the outside good indicates that c onsumers are less sensitive to price. The fact that both methods found a similar semi-elasticity in Tabl e 12 and Table 13 is consistent with this argument. In Table 12 and Table 13, the relationsh ips between most variables and market share are as expected. Among plan char acteristics, higher premiums and annual deductibles are associated with lower mark et share. Inclusion of more drugs on formulary tiers tends to encourage enrollment except for Tier 2 brand name medications. Enrollees are responsive to the number of ge neric drugs (tier 1 drugs). This is not surprising given the fact th at the generic drugs comprise over 60% of overall drug utilization. Thus, consumers are sensitive to access to medications. Among the
102 utilization controls, prior au thorization and step therapy tend to lower enrollment while quantity limits do not. Market characteristics also affect enrollmen t significantly. Intuitively, the greater the number of competing plans, the lower the enrollment in each plan. As more plans enter the PDP market, each can only get a sma ll slice of the market given a fixed number of Medicare beneficiaries. Not all results are as expected. First, tier 1 copayment and tier 3 coinsurance have negative signs while rates of cost sharing in tiers 2 and tier 4 have positive signs. However, only tier 4 coinsurance is statistically significant. This finding along with the positive relationship between the number of covered medications and market share suggests that individuals are mo re concerned with coverage of medications than the level of copayment. In addition, whether medica tions are covered, and the overall premium and deductible are relatively transparent to consumers when deciding which plan to purchase. The implications are levels of copayments across tiers may be less clear to consumers. Second, controlling for the number of PDPs in the region, plans tend to have a smaller market share in regions with sicker Medicare beneficiaries. It was expected that enrollees with poorer health would de rive greater utility from prescription drug coverage. Another counterintuitive observation is th e sign associated with gap coverage. Gap coverage is expected to attract indivi duals to enroll. However, in Table 12 and Table 13, offering gap coverage was associated with lower market share. The results in Section 4.1.3 showed that cove ring generic drugs in the ga p increases the plan bid by approximately 19% and covering brand names drugs increases the plan bid by an
103 additional 20%. Since gap coverage is a supplemental benefit, the cost of providing gap coverage is completely passed on to Medicare beneficiaries. In other words, plans offering gap coverage charge significantly higher premiums than plans that do not because the premium will cover the expect ed cost of providing the benefit plus administrative costs. While gap coverage is quite useful for individuals that are high users of medications, most individuals do not have sufficient drug spending to reach the donut hole. Thus, such consumers may not be willing to pay extra out-of-pocket cost to get gap coverage that may not be necessary to them. 4.4. Summary Given the relative short history of th e Medicare Part D program, not much research has been done to examine the Part D plans. Serving as one of the pioneer studies, this dissertation has ta ken a three-step approach to test four hypotheses in the Medicare PDP market. First, using Hedonic pricing models with firm fixed effects, we found evidence to support Hypothesis 1 that the tiered copayments are consistent with their actuarial values. Our results show that higher copayment on each tier is associated with lower plan bid. However, no evidence was found for Hypothesi s 2 that utilization control tools lower plan bids. Second, adopting the decomposition method by Neumark (1988), we decomposed the difference in bid and premium between 2006 and 2008 into two parts: changes in plan characteristics and changes in marginal price. We found that the di fference was primarily caused by the difference in plan characteristics. As a result Hypothesis 3 that actuarial
104 pricing methods play an important role in explaining the premium and bid difference between 2006 and 2008 was rejected. Finally, we estimated the premium elastici ty and semi-elasticity of demand using a mean utility logit model. The estimated elasticity of -1.804 and semi-elasticity of -0.04 supports Hypothesis 4, that Me dicare beneficiaries are sensitive to PDP premiums.
105 Chapter Five Discussion 5.1 Conclusions The Medicare Part D program represents the largest expansi on of the Medicare program in Medicare history. While the MMA provided a basic bene fit structure by law, most firms have chosen to provide alternative benefit structures that included the use of tiered cost sharing. The firms are also given the flexibility to esta blish formularies and apply utilization control tools to covered drugs, such as prior authorization, quantity limits, and step therapy. The plan design is subject to the approval of CMS, the agency that administers the Medicare program. This dissertation is one of th e pioneer studies in the field that measure the PDPs in the context of the highly regulated Medicare Pa rt D market. Specifica lly, this dissertation tested four hypotheses related to PDPs from different perspectives, including benefit structure, pricing method and sensitivity of enrollees to premium, using 2006-2008 Prescription Drug Plan (PDP) data. We found that the tiered copayments are c onsistent with their actuarial values. The results of the firm fixed effects model show that plan bid is inversely related to enrollees cost sharing. However, despite be ing statistically significant, the marginal effects are quite small. The effects were larger for preferred brand name medications than generic medications, suggesti ng that insurers expect an in crease in tier 2 cost sharing to induce a small shift toward generic drugs. However, we did not find evidence on tier shifting from non-preferred brand name drugs and specialty drugs to preferred name
106 drugs. The effect of cost sharing on consum er demand for medications and plan bids has important policy implications. One of the primary goals of prescription drug coverage was to increase access to medications for th e elderly. Reductions in consumer demand due to high cost sharing would need to be monitored to ensure this goal is not compromised. In addition, the Part D plan pri ce elasticities can be informative to CMS in monitoring plans and consumer behavior. Among the utilization control tools, we f ound that only prior authorization lowers plan bid. Although counterintuitive, step therapy and quantity limits were found to increase plan bid. These utilization contro l tools are designed to lower the expected claim cost which is positively related to plan bid. But we did not fi nd consistent evidence to support the hypothesis that these tools lower plan bid. However, this does not necessarily imply that these utilization control tools fail to function as they were designed to. The insurers may have failed to reflect the potential savings in the plan bids. Considerable changes in average plan characteristics had occurred between 2006 and 2008. Many plans have been adjusted to cover fewer medications and encourage beneficiaries to use generic medications ove r brand name medications. In addition, the bids in 2006 were based on manual rates due to the lack of experien ce. Starting in 2008, experience-based bid has been required by CMS as plans accumulate Part D experience. This would lead to considerable changes in the marginal prices associated with plan characteristics. Overall, changes in average bids and pr emiums were primarily due to changes in plan characteristics. Nearly th ree quarters of the change in bid and virtually all of the change in premium was attributed to plan ch aracteristics. Thus, th e move to increasing
107 cost sharing for brand name medications a nd covering fewer medications has led to a reduction in plan bid. A number of additional results are worthy of discussion. The average bid declined in 2008 compared to 2006 while the average beneficiary premium increased. This is likely due to the weighting method used by CMS to arrive at the national average bid. As discussed earlier, all bids were we ighted equally in 2006 due to the lack of enrollment history. By 2008, the national average bid was a blended average of the unweighted average bid and an enrollment weighted average bid. The beneficiary premium is a percentage of the national average bid, sugges ting that in 2008 weighted average bid was greater than the unweighted av erage. This led to an increase in the national average bid used to calculate the beneficiary premium. Many of the changes in plan characteris tics likely reflect the lack of knowledge insurers had in serving this market. Given this lack of experience, many plans covered all the drugs on CMS formulary file although they were not require d to in 2006. After gaining two years experience in the Medicare Part D market, in 2008, plans are more sophisticated in benefit design and formulary c ontrols. Similarly, in 2006 the plan bid is positively related to the PDP region Part D ri sk scores and negatively related to the Medicare population in the region. However, the importance of the regional variables declined by 2008. As insurers gained mo re knowledge and began to use experience rating, regional characteristics become less important. One interesting aspect of the part D progr am is the generous benefits offered to low-income beneficiaries. Individuals mee ting certain income requirements have their premiums and deductibles covered by the fede ral government. In addition, cost sharing
108 is quite limited. CMS randomly auto-assigns th e new dual eligible enrollees to the PDPs that are below the regional low income subs idy benchmark. In order to get the autoassigned members, these plans generally bid lower than the plans that do not intend to enroll low income members. Enrollment is very important in meas uring the success of implementing the Medicare Part D program. Another goal of the program was to encourage competition between plans in order to maximize consumer benefits and minimize costs. However, in order to achieve this goal, enrollees must be responsive to price differences between plans. Consequently, enrollee pr ice sensitivity to plan premiu ms is of great interest to many researchers and policy makers. We estimated the elasticity of Medicare PDP enrollment with respect to plan premiu m (-1.80 using a composite outside good and -0.78 using MAPDs as the outside good). Such es timates are higher in magnitude than prior research on enrollee price sensitivity in the Medicare HMO market. According to Frakt and Pizer (2009), the higher PDP premium elasticity is consiste nt with the nature of the PDP market. Due to the lower fixed cost of entry, PDPs can easily enter the Medicare Part D market. In each PDP region, Medicare beneficiaries generally have over 50 PDPs to choose from. These PDPs are more simila r than plans than t hose of the Medicare HMO market. In addition, PDPs do not requ ire restrictive provider networks that Medicare HMOs have. Therefore, PDP enrollees are more sensitive to premiums than Medicare HMO enrollees. The estimated premium elas ticity using a composite outside good is larger in magnitude than Frakt and Pizers estimates (1.804 vs. -1.475). The increased sensitivity to price was expected with the change to experience rating. Experience rating was
109 expected to result in greater premium va riation among similar plans (Cutler, 1994), and thus results in consumers being more price sens itive. The results may also indicate that with one more year experience in the Part D market, PDP enrollees are more knowledgeable about the PDP products. As such, they are more responsive to the PDP premiums. Plan premium is an important factor in determining the plans market share. This study also expanded on Frakt and Pi zers (2009) paper by including MAPD premiums and enrollment as an outside good. The estimated price elasticity is much smaller when explicitly including an outside good. Thus, consumers appear more price sensitive when the outside good is not explici tly included in the anal ysis. Including the MAPD products as an outside good resulted in consumers appearing much less sensitive to price. However, the semi-elasticity whic h is measures the consumers response to a $1 change in premiums indicates little diffe rence between the two methods. Insurance companies aimed to attract Medi care beneficiaries to enroll by offering tiered copayments instead of fixed member cost-sharing. However, our results showed that lower copayments do not necessarily a ffect market share. Gap coverage was associated with lower market share. The relationship may reflect the higher premiums associated with gap coverage and the fact that most enrollees do not use sufficient medications to reach the gap. Consumers may not be willing to pay a known higher price for benefits that are unlikely to be used. 5.2 Limitations Although a systematic method has been employed to explore the relationship between plan bids and plan characteristics and premium elasticities, this dissertation
110 does have some limitations. Firs t, the results need to be interpreted carefully. The results cannot be applied to the MAPD plans as the enrollees in the MAPD plans may have different utilization patterns th an the PDP enrollees. A major ity of the PDP enrollees are Medicare FFS members who are usually less healthy than the MAPD enrollees. Another possible concern is that the use of a fixed effect mode l specification may not be appropriate. If insurers do not vary their cost sharing st ructure across plans or geographic regions, then the fixed effects spec ification will not be able to accurately estimate the relationship between cost shari ng and bids, and a specification without firm fixed effects may be more appropriate. For this purpose, we estimated an OLS model without the firm fixed effects to determine whether the results differ or not. The results are qualitatively similar, alt hough the effects for tier 1 and ti er 2 cost sharing are even weaker without the firm fixed effects. Thus, it does not appear that the use of a firm fixed effects specification leads to the small measured relationship between cost sharing and plan bids. No evidence was found on the differentia l effects of low risk and high risk regions. This result is cons istent with a number of previous studies that have not found medications for chronic conditions to be more sensitive to cost sharing. However, it is important to note that this di ssertation used a market level variable to measure health status within the region. Firms may attract di fferent risks across regions, and the market level variable may not be strongly correlated with a firms experience. Given the limitation imposed by the data used, we could not examine specific drugs. While CMS set the minimum standards fo r the formulary files, they also give the firm latitude to modify their formulary, subject to review and approval. As a result, some
111 plans may cover some brand drugs on tier 2, some plans may cover the same drugs on the tier 3 while others may not cover them at all. The impact of specific drugs to the bids and the effect of adding the same drug on different tiers are not provided in this dissertation. Further, we used the median negotiated prices for generic drugs and brand name drugs to convert the tier 1 and tier 2 coinsurance in to copayments, and convert tier 3 and tier 4 copayments into coinsurance. However, the median drug cost of each tier may differ by plan due to the different numb er and type of drugs covered. The coefficient estimates associated with the tier cost sharing variab les in the firm fixed effects model cannot be used to predict individu al plans behavior. More importantly, since the Medicare Part D program started in 2006, both the 2006 and the 2007 bids were developed using manual rates. Although the 2008 bids were supposed to be experience based, some plans continued to use manual rates in lack of creditable experience. Some of these early ag e bids used in this dissertation may not be mature enough to accurately capture actual util ization patterns and cl aim costs associated with each plan. In other words, these proj ected plan costs may differ from the actual costs. The significant risk corridor reconc iliation amounts in Appendix B (Figure B5 and Figure B6) in the end of years 2006 and 2007 suppor t this point of view As more claims experience becomes available and the plans have more creditable experience in the Medicare Part D market, the plan bids in the future will be fully experience based, which may have different copayments elasticities than the plan bids documented in this dissertation. Similarly, these early age bids used in th is dissertation have not gained expertise to effectively use the utilization control tools. The manual rates used in the bids may not
112 correctly reflect the potential savings. That may explain the counterintuitive results produced by the models in this dissertation. We expect the future fully experience based bids would more accurately capture the savings caused by these utilization control tools. From the enrollees perspective, PDPs are new products that they have little knowledge about. Given the short history of the Part D program, they may not be knowledgeable enough to choose the plans that be st meet their needs. In other words, enrollment behavior may occur that would not appear to maximize utility because of enrollees incomplete informati on. If the lack of information or the ability to use the information resulted in a large number of seemingly irrational plan enrollments, the premium elasticity estimated in this study may no t be valid. This problem is expected to be alleviated as the level of information in creases with the greater experience gained by enrollees in the future. In addition, due to data limitation, we did not consider the government subsidies, such as low income subsidy and federal reinsurance subsidy. Although the bid amount reflects the plans po rtion of potential claim liabiliti es, the government subsidies do impact enrollees overall utilization patterns, especially for the low income enrollees. Thus, the plans liability ma y be impacted indirectly. For the same reason, this dissertation did not take into account of the risk corridor reconciliation payments si nce the plan level data were not available. The significant risk corridor payments at the end of years 2006 a nd 2007 also indicate that not all plan bids were priced accurately to refl ect the actual claim costs incu rred by each plan. In other words, not all plan bids captured the plans expected claim costs correctly. Thus, the
113 copayment elasticities estimated may not reflect the enrollees demand for prescription drugs. The final limitation is the missing insurer characteristic data, such as rebates received from pharmaceutical companies and/or AWP (average wholesale price) discounts with the PBM, underwriting and admini strative costs, etc. Inclusion of these data, when they are available, would be e xpected to improve the model accuracy and reduce the variation. 5.3 Future Research Since the inception of the Medicare Part D program, criticisms have been frequently heard, such as, the limited access to medical care service due to the specific design of donut hole, the complicated bene fit structure design, the governments lack of negotiating power on drugs with pharmaceu tical companies, premium hikes, etc. These criticisms and concerns should be addressed using pr escription drug claims data when they become available in the future. Other limitations mentioned in the preceding section should also be addressed in future research. Fortunately, CMS has recently planned to initiate a phased schedule to release the Medicare Part D experience data (detailed claims data by enrollee) to researchers. With these experience data be ing available, most limitations discussed can be addressed. For example, with information on cost per script, dispensing fees, and plan paid amount becoming available, the average drug cost can be measured more accurately.
114 Moreover, following the work from this dissertation, future research could be directed to measure the impact of government subsidies using accumulated Part D experience data. Risk corridor reconciliation is another interesting topic for future research. Starting in 2008, the risk co rridor threshold band has wide ned and the risk sharing percentage has changed as mandated by the MMA (Appendix B Figure B7 and Figure B8). Eventually the plans will bear more ri sk. Whether and how the participating firms will change their pricing strategies is one of th e potential research directions in this field. Currently, a few large insurance companie s are dominant in the PDP market. Benefiting from the economy of scale, they are more likely to charge lower premiums than the small firms in the future. Our results show that PDP enrollees are very sensitive to plan premiums. As such, these large insurance companies are likely to further expand their market share and put the small firms in an even worse situation. Will the Medicare Part D market eventually become a monopoly or an oligopoly market ? Or instead, will the government play the provider role like Canada ? These concerns ar e also of interest to us and should be addressed in future research.
115 References Atherly, Adam, Dowd, Bryan E. and Feldman, Roger (2004). The Effect of Benefits, Premiums, and Health Risk on Health Plan Choice in the Medicar e Program. Health Service Research, Vol.39, Issue 4p1, pp 847-864 Bartik, Timothy J. (1987), The Estimati on of Demand Parameters in Hedonic Price Models. The Journal of Polit ical Economy, Vol.95, No.1. pp.81-88 Berndt, Ernst R; Griliches, Z; Rappaport, Neal J. (1995). Econo metric Estimates of Price Indexes for Personal Computers in the 1990s. Journal of Econometrics. 1995;68:243. Berry S. (1994). Estimating Discrete C hoice Models of Product differentiation. The RAND Journal of Economics 25(2):242-262. Biles, B., Dallek, G., Hersch Nicholas, L. (2004) Medicare Advantage: Dj Vu All Over Again? Health Affairs Web Exclusives W4:586-597. Blum, Jonathan, Bowman, Jennifer and White Chiquita (2005). The Impact of Enrollment in the Medicare Prescription Dr ug Benefit on Premiums. The Medicare Rx Drug Law, October 2005. Kaiser Family Foundation. Buchsbaum, L., Varon, J., Kagel,E., Boyle, R. and McGarvey, K. (2007). Perspectives on Medicare Part D and Dual Eligibles: Ke y Informants Views from Three States. Kaiser Commission on Medicaid and Uninsured. May 2007. CMS, CMS Announces 2008 Medicare Advantag e Payment Rates and Part D Payment Updates. http://www.cms.hhs.gov/apps/media/ April, 2007. Culter, D. (1994). A Guide to Health Care Reform. Journal of Economic Perspectives Vol.8, No.3, p13-29. Dowd, B., Feldman, R., Coulam, R. (2003). The Effect of Health Pl an Characteristics on Medicare + Choice Enrollment. Health Services Research 38 (1):1. Duggan, M., Healy, P., & Morton, F.S. (2008). Providing prescription drug coverage for the elderly: Americas experiment with Medicare Part D. Journal of Economic Perspectives, 22, 69-92.
116 Fairman, Kathleen A., Motheral, Brenda R., and Henderson, Rochelle (2003). Retrospective, Long-term Follow-up Study of th e Effect of A Threetier Prescription Drug Copayment System on Pharmaceutical and Other Medical Utilization and Costs. Clinical Therapeutics, Vol. 25; Issue 12; pp3147-3161. Feldstein, PJ., Wickizer, TM. And Wheeler, JR. (1988). Private Cost Containment: The Effects of Utilization Review Programs on Health Care Us e and Expenditures. The New England Journal of Medicine, Vol.318:1310-1314. Frakt A, Pizer S (2009) Beneficiary price se nsitivity in the Medi care prescription drug plan market. Health Economics, DOI:10,1002/hec.4151. Gibson, Terasa B., Ozminkowski, Ronald J. and Goetzel, Ron Z. (2005). The Effects of Prescription Drug Cost Sharing: A Review of the Evidence. American Journal of Managed Care, November 2005. Gilman, B., & Kautter, J. (2007). Impact of multitiered copayments on the use and cost of prescription drugs among Medicare beneficiaries. Health Services Research, 43, 478495. Gold, Marsha (2006). Premiums and Cost Sharing Features in Medicare New Prescription Drug Program, 2006. Kaiser Family Foundation. http://www.kaiserfamilyfoundation.org/medicare/upload/7517.pdf Goldfarb, Scott D., Duncan, Babette S., Pete r E, Dans and Sloan, Alice, S. (1999). HMO Direct Costs and Health Care Resource Use after Implementation of a Monthly Limit on Sumatriptan. American Journal of Hea lth System Pharmacy. Volume 56(21), 1 November 1999, pp 2206-2210 Harris, Brian L. Stergachis, Andy and Reid, LD (1990). The Effect of Drug Copayments on Utilization and Cost of Pharmaceuticals in a Health Maintenance Organization. Medical Care, Vol.28, No.10. pp.907-917. Heiss, Florian, McGadden, Daniel and Winter Joachim (2006). Who Failed to Enroll in Medicare Part D, And Why? Ea rly Results. Health Affairs 2006. Hoadley, Jack (2006). Medicares New Advent ure: The Part D Drug Benefit. Report for the Commonwealth Fund/Alliance for Heal th Reform 2006 Bipartisan Congressional Health Policy Conference. Huskamp, H.A., Frank, R.G., McGuigan, K.A ., & Zhang, Y. (2005). The impact of a three-tier formulary on demand response for prescription drugs. Journal of Economics & Management Strategy, 14, 729-753.
117 Jensen, Gail A. and Morrisey, Michael A (1990). Group Health Insurance: A Hedonic Approach. Review of Economics and Statistics, Vol.72, No.1. pp38-44. Joyce, G.F., Escarce, J.J., Solomon, M.D., & Goldman, D. (2002). Employer drug benefit plans and spending on prescription dr ugs. Journal of the American Medical Association, 288, 1733-1739. Kaiser Family Foundation, Medicare at a Glance (2008). Medicare, November 2008. Lancaster, Kelvin J. (1966). A New Appro ach to Consumer Theory. The Journal of Political Economy, Vol.74 No.2. pp.132-157 Lancaster, Kelvin J. (1966). A New Appro ach to Consumer Theory. The Journal of Political Economy, Vol.74 No.2. pp.132-157 Lucarelli, Claudio (2006). An Analysis of the Medicare Prescr iption Drug Benefit. PARC Working Paper Series, WPS 06-09. Lucas, R. 1977. Hedonic Wage Equations and Psychic Wages in the Returns to Schooling. American Economic Review 67: 549-58. Mackinnon, Neil J. and Kumar, R. (2001). P rior Authorization Programs: A Critical Review of the Literature. Managed Care Pharmacy 2001, p 297-302. Manning, Willard G., Newhouse, Joseph P, Duan, Naihua, Keeler, Emmett B. and Leibowitz, Arleen (1987). Health Insura nce and the Demand for Medical Care: Evidence from a Randomized Experiment. American Economic Review, Vol.77, No.3. pp 251-277. McLaughlin, Catherine G., Cher new, Michael and Erin Frie s Taylor (2002). Medigap Premiums and Medicare HMO Enrollment. Health Service Research, 37(6),1445-1468 MedPac, Part D Payment System. (2006). www.medpac.gov September 2006, revised October 2008. Motheral, Brenda and Fairman, Katheleen (200 1). Effect of Three-Tier Prescription Copay and Other Medical Utiliza tion. Medical Care. 39(12):1293-1304. Neuman, P, et al. (2007). Medicare Prescr iption Drug Benefit Pr ogress Report: Finding from a 2006 National Survey of Seni ors. Health Affairs 26(5):w630-w634. Neumark, David (1988). Employers Discriminatory Behavior and the Estimation of Wage Discrimination. The Journal of Human Resources. 23 (3):279-295 Oaxaca, Ronald (1973). Male-Female Wage Differentials in Urban Labor Markets. International Economic Review 14(3):693-709.
118 Panzer, P.E. et al. (2005) Implementations of an SSRI Generic Step Therapy Pharmacy Benefit Design: An Economic Model in Anxi ety Disorders. The American Journal of Managed Care. S370-S379. October 2005. Patel, Uptal D. and Davis, Matthew M. (2006). Falling into the Doughnut Hole: Drug Spending among Beneficiaries with End-St age Renal Disease under Medicare Part D Plans. Journal of the American Society of Nephrology. Doi:10.1681/ASN.2005121385. Pope, G. et al Risk Adjustment of Medi care Capitation Payments Using the CMS-HCC Model. Health Care Finance Review, Summer 2004, Vol 25, No. 4. Robst, John (2006). Estimation of a Hedoni c Pricing Model for Medigap Insurance Health Services Research 2006. Robst. J, Levy JM and Ingber MJ (2007). D iagnosis-based Risk Adjustment for Medicare Prescription Drug Plan Payments. Health Care Financing Review, 2007 Summer;28(4):15-30 Robst, J.(2007). Market Structure, Regula tion, and Adverse Selection as Determinants of Medigap Supplemental Insurance Premiums. Rosen, Sherwin. (1974). Hedoni c Prices and Implicit Mark ets: Product Differentiation in Pure Competition. The Journal of Political Economy, Vol.82, No.1. (Jan-Feb., 1974), pp.34-35 Smalley, W. et al. (1995). Effect of a Prio r-Authorization Requirement on the Use of Nonsteroidal Antiinflammatory Drugs by Medicaid Patients. The New England Journal of Medicine, Vol. 332, No.24. Simon, K and Lucarelli, C (2006), What Drove First Year Premiums in Stand-Alone Medicare Drug Plans? NBER Working Paper Series, http://www.nber.org/papers/w12595 Solanki G, Shcauffler HH and Miller LS (2000) The Direct and Indirect Effects of Cost-Sharing on the Use of Preventive Services. Health Services Research, 2000 Feb;34(6):1331-50. Stuart, Bruce, Simoni-Wastila, Linda and Chauncey, Danielle (2005). Assessing the Impact of Coverage Gaps in the Medicare Pa rt D Drug Benefit. Health Affairs, 19April 2005. Tamblyn, R., Larprise, R. Hanley, J., Abra hamowicz, M., Scott, S., et al. (2001). Adverse Events Associated With Prescr iption Drug Cost-Sharing Among Poor and Elderly Persons. JAMA.2001;285;421-429.
119 Town R, Liu S (2003) The welfare impact of Medicare HMOs. Rand Journal of Economics, 34(4):719-736. Winter, J. et al. (2006). Medicare Prescr iption Drug Coverage: Consumer Information and Preferences. PNAS Vol.103. No.20. 7929-7934. Yokoyama et al. (2007). Effects of a Step -Therapy Program for Angiotensin Receptor Blockers on Antihypertensive Medication Utilization Patt erns and Cost of Drug Therapy. Journal of Managed Care Pharmacy. Vol. 13,No.3.
120 Websites http://www.cms.hhs.gov/MCRAdvPartDEnrolData/ http://www.cms.hhs.gov/NonIdentifiableDataFiles/09_PrescriptionDrugPlanFormularyan dPharmacyNetworkFiles.asp#TopOfPage http://www.cms.hhs.gov/MCRAdvPartDEnrolD ata/PDPSCPen/list.asp#TopOfPage http://www.cms.hhs.gov/MCRAdvPartDEn rolData/EP/list.asp#TopOfPage http://www.cms.hhs.gov/MCRAdvPartDEnro lData/PDPPD/list.asp#TopOfPage http://www.cms.hhs.gov/PrescriptionDrugCovGenIn/ http://www.cms.hhs.gov/PrescriptionDr ugCovGenIn/08_PartDData.asp#TopOfPage http://www.kff.org/medicare/
121 Appendix A: Tables
122 Table A 1 Median Negotiated Prices for Medicare Part D Sample Drugs All Drugs $49.82 Generic Drugs $18.11 Brand Name Drugs $92.16 Notes: The median price is from Ho adley (2006) who examined prices of the top 150 medications in the Part D program.
123 Table A 2 Average Risk Score by PDP Region Average Risk Score PDP Region States 2005 2006 2007 1 ME/NH 0.9707 0.9707 0.9707 2 CT/MA/RI/VT 1.0141 1.0140 1.0138 3 NY 1.0402 1.0403 1.0402 4 NJ 1.0335 1.0335 1.0334 5 DE/DC/MD 1.0132 1.0131 1.0130 6 PA/WV 1.0272 1.0270 1.0268 7 VA 0.9954 0.9950 0.9946 8 NC 1.0120 1.0119 1.0119 9 SC 1.0217 1.0217 1.0216 10 GA 1.0215 1.0212 1.0210 11 FL 1.0503 1.0502 1.0501 12 AL/TN 1.0323 1.0323 1.0324 13 MI 1.0057 1.0055 1.0052 14 OH 1.0228 1.0227 1.0225 15 IN/KY 1.0038 1.0039 1.0039 16 WI 0.9514 0.9512 0.9509 17 IL 0.9699 0.9698 0.9697 18 MO 1.0063 1.0063 1.0063 19 AR 0.9833 0.9833 0.9832 20 MS 1.0004 1.0004 1.0004 21 LA 1.0243 1.0229 1.0230 22 TX 0.9979 0.9978 0.9978 23 OK 0.9842 0.9843 0.9843 24 KS 0.9616 0.9618 0.9621 25 IA/MN/MT/NE/ND/SD/WY 0.9268 0.9268 0.9268 26 NM 0.9374 0.9373 0.9372 27 CO 0.9328 0.9325 0.9322 28 AZ 0.9548 0.9548 0.9547 29 NV 0.9583 0.9581 0.9580 30 OR/WA 0.9349 0.9349 0.9348 31 ID/UT 0.9211 0.9212 0.9214 32 CA 1.0026 1.0025 1.0025 33 HI 0.9627 0.9627 0.9627 34 AK 0.9067 0.9070 0.9073 Notes: Source file: CMS 2006 Part D risk score by county file Avg risk Part D.xls. Weighed by census data 2005-2007 over age 65 population by county.
124 Table A 3 Medicare Population by PDP Region Medicare Eligibles PDP Region States 2006 2007 2008 1 ME/NH 239,424 243,190 251,595 2 CT/MA/RI/VT 2,002,074 2,020,204 2,044,099 3 NY 2,858,747 2,879,429 2,882,739 4 NJ 1,261,180 1,270,110 1,276,946 5 DE/DC/MD 914,799 928,255 953,905 6 PA/WV 2,537,956 2,556,932 2,583,239 7 VA 1,002,150 1,023,400 1,071,683 8 NC 1,288,827 1,318,782 1,390,313 9 SC 654,600 673,878 714,218 10 GA 1,045,818 1,076,986 1,144,013 11 FL 3,094,899 3,135,438 3,189,991 12 AL/TN 1,698,204 1,736,672 1,796,704 13 MI 1,519,223 1,537,840 1,569,168 14 OH 1,797,320 1,811,669 1,827,984 15 IN/KY 1,613,801 1,639,637 1,680,069 16 WI 844,212 854,772 869,604 17 IL 1,734,572 1,749,064 1,766,839 18 MO 930,083 942,794 959,988 19 AR 479,834 489,388 504,941 20 MS 465,962 471,940 475,855 21 LA 659,249 642,618 652,137 22 TX 2,570,082 2,641,789 2,779,572 23 OK 550,500 559,862 574,386 24 KS 408,800 412,026 434,408 25 IA/MN/MT/NE/ND/SD/WY 1,933,426 1,953,686 2,018,057 26 NM 270,105 277,591 291,894 27 CO 529,442 542,294 574,368 28 AZ 797,108 818,639 861,625 29 NV 302,537 308,802 327,742 30 OR/WA 1,377,990 1,409,270 1,474,709 31 ID/UT 431,107 443,820 473,591 32 CA 4,325,861 4,386,037 4,466,044 33 HI 186,157 189,271 193,033 34 AK 53,218 55,058 59,324 Notes: (1) 2006 and 2007 data are from Kaiser Family Foundation, 2006 Med Beneficiary.pdf and 2007 Med Beneficiary.pdf. The state level data are summarized by PDP region. (2) 2008 data is taken from CMS monthly penetration files (May2008-Sep 2008). The average members by county are calculate d and then summarized by PDP region.
125 Table A 4 Company Information by Contract Contract Number Start Date Tax Status Paren t Company S0043* 1/1/2006 For Profit Aveta, LLC. S0197 1/1/2006 For Profit Coventry Health Care Inc. S1030 1/1/2007 Not-for-Profit/Non-P rofit BCBS OF AL & BCBS OF TN S1516* 1/1/2008 Not-for-Profit/Non-Profit Mennonite General Hospital, Inc S1566 1/1/2006 For Profit Bravo Health, Inc. S2321 1/1/2006 Not-for-Profit/Non-Profit Independence Blue Cross S2468 1/1/2006 Not-for-Profit/Non-Pro fit Blue Shield of California S2505 1/1/2007 For Profit Windsor Health Group S2770 1/1/2006 Not-for-Profit/Non-Profit Independence Blue Cross S2874* 1/1/2007 For Profit Humana Inc. S2893 1/1/2006 For Profit Wellpoint, Inc. S3230 1/1/2007 For Profit MEDICAL MUTUAL OF OHIO S3389 1/1/2006 For Profit University of Pittsburgh Medical Center S3440 1/1/2008 For Profit H ealth Alliance Plan (HAP) S3521 1/1/2006 Not-for-Profit/Non-Pro fit Lifetime Healthcare, Inc. S3994 1/1/2007 Not-for-Profit/Non-Profit Hawaii Medical Service Association S4231 1/1/2006 For Profit Universal Health Care, Inc. S4248 1/1/2007 For Profit Geisinger Health System S4496 1/1/2007 For Profit Independence Blue Cross S4749* 1/1/2008 For Profit Preferred Health Inc S4802 1/1/2006 For Profit Munich American Holding Corporation S4877* 1/1/2007 Not-for-Profit/Non-Profit Cooperativa de Seguros de Vida de Puerto Rico S5540 1/1/2006 Not-for-Profit/Non-Profit Blue Cross and Blue Shield of North Carolina S5552 1/1/2006 For Profit Humana Inc. S5555* 1/1/2006 For Profit Medical Card System, Inc. S5557 1/1/2006 For Profit Fox Rx Inc. S5566 1/1/2006 For Profit Health Care Service Corporation S5569 1/1/2006 For Profit Coventry Health Care Inc. S5578 1/1/2006 For Profit HealthSpring S5580 1/1/2006 For Profit Torchmark Corporation S5581 1/1/2006 For Profit Universal American Financial Corporation S5584 1/1/2006 Not-for-Profit/Non-Profit Blue Cross Blue Shield of Michigan S5585 1/1/2006 Not-for-Profit/Non-Profit HealthNow New York Inc. S5588 1/1/2006 For Profit Promedica Health System S5593 1/1/2006 For Profit Highmark Inc. S5596 1/1/2006 For Profit Wellpoint, Inc.
126 Table A4 Continued Company Information by Contract Contract Number Start Date Tax Status Paren t Company S5597 1/1/2006 For Profit Universal American Corp. S5601 1/1/2006 For Profit CVS Caremark Corporation S5609 1/1/2006 Not-for-Profit/Non-Profit The Regence Group S5617 1/1/2006 For Profit CIGNA S5644 1/1/2006 For Profit Longs Drug Stores Corporation S5650 1/1/2006 For Profit AmeriHealth Mercy Health Plan S5660 1/1/2006 For Profit Medco Health Solutions, Inc. S5670 1/1/2006 For Profit Coventry Health Care Inc. S5674 1/1/2006 For Profit Coventry Health Care Inc. S5678 1/1/2006 For Profit Health Net, Inc. S5704 1/1/2006 For Profit GlobalHealth Incorporated S5715 1/1/2006 For Profit Health Care Service Corporation S5726 1/1/2006 For Profit Blue Cross Blue Shield of Kansas S5740 1/1/2006 For Profit NewQuest Health Solutions LLC S5741 1/1/2006 Not-for-Profit/Non-Profit EmblemHealth Inc. S5743 1/1/2006 Not-for-Profit/Non-Profit Blue Cross and Blue Shield of Minnesota S5753 1/1/2006 Not-for-Profit/Non-Profit Wisconsin Physicians Service Ins Corporation. S5755 1/1/2006 For Profit Torchmark Corporation S5766 1/1/2006 Not-for-Profit/N on-Profit CareFirst, Inc. S5768 1/1/2006 For Profit Coventry Health Care Inc. S5775* 1/1/2006 For Profit Pharmacy Insurance Corporation of America S5783 1/1/2006 For Profit QCC Insurance Company S5795 1/1/2006 Not-for-Profit/Non-Profit Arkansas Blue Cross Blue Shield S5803 1/1/2006 For Profit Universal American Corp. S5805 1/1/2006 For Profit UnitedHealth Group, Inc. S5810 1/1/2006 For Profit Aetna Inc. S5815 1/1/2006 For Profit NewQuest Health Solutions LLC S5820 1/1/2006 For Profit UnitedHealth Group, Inc. S5822 1/1/2006 For Profit Bravo Health, Inc. S5825 1/1/2006 For Profit Universal American Corp. S5840* 1/1/2007 For Profit First Medical Health Plan S5857 1/1/2006 For Profit Spectrum Health System S5860 1/1/2006 Not-for-Profit/Non-Profit Rocky Mountain Health Maintenance Inc. S5877 1/1/2006 Not-for-Profit/Non-Profit Educators Mutual Insurance Association S5884 1/1/2006 For Profit Humana Inc. S5902 1/1/2006 For Profit Presbyterian Healthcare Services
127 Table A4 Continued Company Information by Contract Contract Number Start Date Tax Status Parent Company S5904 1/1/2006 Not-for-Profit/Non-Profit Bl ue Cross and Blue Shield of Florida S5915 1/1/2006 Not-for-Profit/Non-Profit Scott and White S5916 1/1/2006 Not-for-Profit/Non-Profit The Regence Group S5907* 1/1/2006 For Profit Blue Shield of Puerto Rico S5917 1/1/2006 For Profit UnitedHealth Group, Inc. S5921 1/1/2006 For Profit UnitedHealth Group, Inc. S5932 1/1/2006 For Profit HealthSpring, Inc. S5937 1/1/2006 Not-for-Profit/Non-Profit BlueCross BlueShield of Louisiana S5946 1/1/2006 For Profit BlueCross BlueShield of South Carolina (BCBSSC) S5953 1/1/2006 For Profit BlueCross BlueShield of South Carolina (BCBSSC) S5954 1/1/2006 For Profit Dean Health Systems Inc. S5960 1/1/2006 For Profit Wellpoint, Inc. S5966 1/1/2006 Not-for-Profit/Non-Profit EmblemHealth Inc. S5967 1/1/2006 For Profit WellCare Health Plans, Inc. S5975 1/1/2006 For Profit The ODS Companies (ODS) S5983 1/1/2006 For Profit Medco Health Solutions, Inc. S5993 1/1/2006 Not-for-Profit/Non-Profit Horizon Blue Cross Blue Shield of New Jersey, Inc. S5998 1/1/2007 For Profit Bravo Health, Inc. S6874* 1/1/2008 For Profit Capital BlueCross S6875* 1/1/2006 Not-for-Profit/Non-Profit Independence Blue Cross S7694 1/1/2007 For Profit Envision Insurance Company S7950 1/1/2007 For Profit Express Scripts, Inc. S8067 1/1/2006 For Profit Capital BlueCross S8201 1/1/2007 For Profit University of Pittsburgh Medical Center S8277 1/1/2007 For Profit Carolina Care Plan, Inc S8465 1/1/2008 For Profit Citrus Health Care, Inc. S8475 1/1/2007 For Profit Quality Health Plans, Inc. S8841 1/1/2007 For Profit National Medical Health Card Systems, Inc. S9086 1/1/2006 For Profit America's Health Choice Medical Plans, Inc S9176 1/1/2007 Not-for-Profit/Non-Profit Capital District Physicians' Health Plan, Inc. Notes: (1) are the contracts in US territori es, which are excluded in this study. (2) For contracts that changed company names, used 2008 company information.
128 Table A 5 Average Plan Bid and Member Premium by Year Year Obs. Avg. PDP Bi d Avg. Member Premium 2006 1,414 $101.48 $37.48 2007 1,865 $89.89 $36.81 2008 1,822 $92.63 $40.04 Total 5,101 $94.10 $38.16
129 Table A 6 Average Plan Bid and Member Premium by PDP Region PDP Region States Obs. Avg. Plan Bid Avg. Member Premium 1 ME/NH 147 $94.53 $38.58 2 CT/MA/RI/VT 146 $93.24 $37.04 3 NY 161 $90.38 $34.42 4 NJ 158 $93.89 $37.94 5 DE/DC/MD 153 $94.37 $38.17 6 PA/WV 180 $93.49 $37.48 7 VA 146 $94.81 $38.84 8 NC 140 $96.41 $40.62 9 SC 155 $95.40 $39.53 10 GA 148 $94.91 $39.06 11 FL 156 $92.69 $36.93 12 AL/TN 149 $96.05 $40.21 13 MI 149 $94.40 $38.57 14 OH 160 $93.96 $38.19 15 IN/KY 146 $97.21 $41.23 16 WI 156 $94.48 $38.43 17 IL 151 $94.30 $38.36 18 MO 146 $95.11 $39.14 19 AR 153 $94.24 $38.48 20 MS 139 $96.15 $40.26 21 LA 140 $96.14 $40.28 22 TX 163 $93.35 $37.29 23 OK 150 $96.41 $40.44 24 KS 145 $95.39 $39.47 25 IA/MN/MT/NE/ND/SD/WY 146 $94.35 $38.38 26 NM 155 $89.87 $33.93 27 CO 153 $93.57 $37.60 28 AZ 147 $90.72 $34.62 29 NV 151 $91.41 $35.32 30 OR/WA 157 $94.35 $38.32 31 ID/UT 154 $97.11 $41.08 32 CA 158 $90.04 $33.89 33 HI 124 $90.92 $35.48 34 AK 119 $96.48 $41.12
130 Table A 7 Average Plan Bid and Member Premium by Contract Contract Number of Plans Avg. Plan Bid Avg. Member Premium S5617 306 $96.14 $39.58 S5803 306 $94.57 $38.02 S5810 306 $104.90 $48.34 S5921 306 $94.65 $38.10 S5884 288 $94.70 $38.37 S5597 285 $93.48 $37.00 S5601 272 $92.91 $37.28 S5967 272 $88.98 $31.93 S5960 258 $88.50 $31.32 S5644 210 $85.70 $30.78 S5820 201 $86.75 $30.08 S5670 189 $91.81 $35.25 S5678 182 $83.28 $29.67 S4802 166 $105.90 $50.91 S5660 165 $94.81 $39.84 S5755 163 $93.69 $38.73 S5768 159 $84.38 $30.76 S7694 136 $121.45 $68.61 S5674 108 $91.10 $34.55 S5917 96 $101.08 $47.25 S5596 90 $93.05 $36.49 S5932 69 $75.56 $22.56 S5581 66 $116.89 $52.89 S5557 34 $88.33 $34.26 S7950 34 $100.93 $47.85 S8841 34 $84.82 $31.74 S5825 30 $107.98 $46.22 S5715 27 $93.47 $36.91 S5998 21 $77.88 $24.97 S5946 12 $90.17 $33.61 S2505 10 $77.94 $25.06 S0197 9 $90.33 $33.77 S2893 9 $96.25 $39.69 S5552 9 $95.48 $38.93 S5566 9 $91.36 $34.80 S5593 9 $91.11 $34.55 S5726 9 $91.55 $35.00 S5743 9 $113.92 $57.36 S5795 9 $95.32 $38.77 S5877 9 $105.91 $49.35 S3521 8 $89.89 $32.83 S5904 8 $101.65 $46.03 S5954 8 $98.83 $41.84 S2321 7 $103.81 $46.19 S5753 7 $98.76 $42.77 S5783 7 $87.82 $23.82
131 Table 7 Continued Contract Number of Plans Avg. Plan Bid Avg. Member Premium S5805 7 $86.52 $28.90 S5822 7 $80.26 $25.76 S5915 7 $100.00 $43.94 S5993 7 $99.26 $43.27 S8067 7 $88.91 $34.48 S1030 6 $99.42 $46.58 S2468 6 $90.86 $34.31 S5540 6 $115.50 $58.94 S5569 6 $83.64 $28.99 S5584 6 $94.32 $37.76 S5766 6 $98.98 $42.42 S5902 6 $86.63 $30.08 S5953 6 $94.68 $38.12 S3389 5 $89.89 $34.82 S5580 5 $89.14 $34.08 S5588 5 $90.07 $35.00 S5609 5 $104.54 $49.48 S5650 5 $83.90 $28.83 S5741 5 $83.79 $28.72 S5860 5 $101.38 $44.03 S5916 5 $106.17 $51.11 S5975 5 $106.11 $51.04 S5983 5 $94.62 $39.65 S1566 4 $82.58 $26.89 S3230 4 $95.29 $42.45 S4248 4 $78.96 $26.13 S4496 4 $92.66 $39.83 S5704 4 $110.05 $51.51 S8475 4 $85.46 $32.63 S2770 3 $83.55 $26.99 S3440 3 $93.26 $40.67 S5585 2 $86.49 $33.65 S5857 3 $89.54 $32.99 S5937 3 $97.65 $41.09 S5966 3 $79.64 $23.08 S9176 3 $86.81 $33.73 S4231 2 $139.44 $75.44 S8201 2 $85.09 $32.25 S8277 2 $96.58 $43.50 S8465 2 $89.94 $37.35 S3994 1 $82.08 $29.00 S5578 1 $88.78 $24.78 S5740 1 $88.78 $24.78 S5815 1 $89.73 $25.73 S9086 1 $77.98 $24.90
132 Table A 8 Regression Results: Firm Fixed Effects Model, Dependent Variable: Ln (Bid), Plans with Risk Scores < 1.0 Explanatory Variables Es timates Standard Error Cost sharing Tier 1 copayment -0.0031*** 0.0007 Tier 2 copayment -0.0025*** 0.0004 Tier 3 coinsurance -0.0237* 0.0134 Tier 4 coinsurance -0.0191 0.0146 # drugs covered ln(# drugs on tier 1) -0.0142** 0.0060 ln(# drugs on tier 2) 0.0573*** 0.0134 # drugs on tier 3 (in thousands) -0.0036* 0.0022 # drugs on tier 4 0.00005*** 0.0000 # drugs subject to: Quantity limits (in thousands) 0.0036*** 0.0013 Prior authorization (in thousands) -0.0487*** 0.0102 Step therapy (in thousands) 0.0574*** 0.0120 Other plan characteristics Deductible -0.0004*** 0.0000 Gap coverage for generics 0.1791*** 0.0053 Gap coverage for brands 0.1763*** 0.0198 LIS_0prem -0.0449*** 0.0055 Other Variables Year 2007 -0.1540*** 0.0079 Year 2008 -0.1180*** 0.0092 Regional risk score 0.0331 0.0778 Regional Medicare population 0.0000 0.0028 N 2,517 R squared 0.73 Notes: (1) *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level. (2) The specification also includes a categorical variable for each firm.
133 Table A 9 Regression Results: Firm Fixed Effects Model, Dependent Variable: Ln (Bid), Plans with Risk Scores >= 1.0 Estimates Standard Error Cost sharing -0.0041*** 0.0007 Tier 1 copayment -0.0022*** 0.0003 Tier 2 copayment -0.0240* 0.0128 Tier 3 coinsurance -0.0235* 0.0132 Tier 4 coinsurance # drugs covered ln(# drugs on tier 1) -0.0338*** 0.0058 ln(# drugs on tier 2) 0.0811*** 0.0126 # drugs on tier 3 (in thousands) -0.0056** 0.0022 # drugs on tier 4 0.00005*** 0.0000 # drugs subject to: Quantity limits (in thousands) 0.0061*** 0.0014 Prior authorization (in thousands) -0.0292*** 0.0084 Step therapy (in thousands) 0.0356*** 0.0095 Other plan characteristics Deductible -0.0003*** 0.0000 Gap coverage for generics 0.1769*** 0.0050 Gap coverage for brands 0.1921*** 0.0184 LIS_0prem *** -0.0617*** 0.0052 Other Variables Year 2007 -0.1665*** 0.0074 Year 2008 -0.1311*** 0.0082 Regional risk score -0.2804** 0.1341 Regional Medicare population -0.0178*** 0.0018 N 2,584 R squared 0.76 Notes: (1)*** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level. (2) The specification also includes a categorical variable for each firm.
134 Table A 10 Regression Results: OL S, Dependent Variable: Ln (Bid) Explanatory Variables Es timates Standard Error Intercept 4.2882*** 0.0610 Cost sharing Tier 1 copayment -0.0022*** 0.0005 Tier 2 copayment -0.0004* 0.0002 Tier 3 coinsurance -0.0676*** 0.0078 Tier 4 coinsurance -0.0337*** 0.0082 # drugs covered ln(# drugs on tier 1) 0.0123*** 0.0035 ln(# drugs on tier 2) 0.0274*** 0.0046 # drugs on tier 3 (in thousands) -0.0071*** 0.0015 # drugs on tier 4 -0.00002*** 0.0000 # drugs subject to: Quantity limits (in thousands) 0.0019** 0.0009 Prior authorization (in thousands) -0.0290*** 0.0046 Step therapy (in thousands) 0.0248*** 0.0074 Other plan characteristics Deductible -0.0001*** 0.0000 Gap coverage for generics 0.1954*** 0.0041 Gap coverage for brands 0.1811*** 0.0151 LIS_0prem -0.1042*** 0.0042 Other Variables Year 2007 -0.1374*** 0.0058 Year 2008 -0.1104*** 0.0065 Regional risk score 0.1549*** 0.0464 Regional Medicare population -0.0116*** 0.0018 N 5,101 R-square 0.61 Adj. R-square 0.61 Notes: *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level.
135 Table A 11 Regression Results: Firm Fixed Effects Model Dependent Variable: Ln (Premium) Explanatory Variables Estimates Standard Error Cost sharing Tier 1 copayment -0.0103*** 0.0013 Tier 2 copayment -0.0075*** 0.0006 Tier 3 coinsurance 0.0014 0.0240 Tier 4 coinsurance -0.0381 0.0254 # drugs covered ln(# drugs on tier 1) -0.0606*** 0.0108 ln(# drugs on tier 2) 0.2270*** 0.0237 # drugs on tier 3 (in thousands) -0.0113*** 0.0040 # drugs on tier 4 0.0001*** 0.0000 # drugs subject to: Quantity limits (in thousands) 0.0131*** 0.0025 Prior authorization (in thousands) -0.0817*** 0.0169 Step therapy (in thousands) 0.0809*** 0.0194 Other plan characteristics Deductible -0.0009*** 0.0000 Gap coverage for generics 0.3903*** 0.0095 Gap coverage for brands 0.5182*** 0.0351 LIS_0prem *** -0.1745*** 0.0098 Other Variables Year 2007 -0.0653*** 0.0140 Year 2008 0.0249 0.0159 Regional risk score 0.6553*** 0.1030 Regional Medicare population -0.0353*** 0.0039 N 5,101 R-square 0.69 Notes: (1) *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level. (2) The specification also includes a categorical variable for each firm.
136 Table A 12 Regression Results: Firm Fixed Effects Model Dependent Variable: Square Root (Bid) Explanatory Variables Es timates Standard Error Cost sharing Tier 1 copayment -0.0152*** 0.0026 Tier 2 copayment -0.0113*** 0.0012 Tier 3 coinsurance -0.1683*** 0.0469 Tier 4 coinsurance -0.0718 0.0497 # drugs covered ln(# drugs on tier 1) -0.1040*** 0.0211 ln(# drugs on tier 2) 0.2867*** 0.0463 # drugs on tier 3 (in thousands) -0.0268*** 0.0079 # drugs on tier 4 (in thousands) 0.0003*** 0.0001 # drugs subject to: Quantity limits (in thousands) 0.0254*** 0.0049 Prior authorization (in thousands) -0.1983*** 0.0330 Step therapy (in thousands) 0.2359*** 0.0379 Other plan characteristics Deductible -0.0017*** 0.0001 Gap coverage for generics 0.8904*** 0.0185 Gap coverage for brands 1.0138*** 0.0685 LIS_0prem *** -0.2362*** 0.0192 Other Variables Year 2007 -0.7749*** 0.0274 Year 2008 -0.5866*** 0.0311 Regional risk score 0.9644*** 0.2011 Regional Medicare population -0.0560*** 0.0077 N 5,101 R-square 0.73 Notes: (1) *** Significant at 1% level; ** Significant at 5% level; Significant at the 10% level. (2) The specification also includes a categorical variable for each firm.
137 Appendix B: Figures
138 Figure B 1 Normal Probability Plot Ln(Bid) The SAS System The UNIVARIATE Procedure Variable: Ln(Bid) Normal Probability Plot 5.225+ | | | | *** | *** + | ***+++ | ***++ | ****+ | ***+ | *** 4.675+ *** | **** | **** | **** | +*** | +**** | ****** | ***** | ******+ |******+++ |* ++ 4.125+*+++ + + + + + + + + + + + 2 1 0 +1 +2
Figure B 2 Medicare Population Characteristics Data Source: Kaiser Family Foundation, Medicare at a Glance, November 2008. The original data is from Income data from US Census Bureau, Current Population Survey published on statehealthfact.org; all other data from Kaiser Fa mily Foundation analysis of the Medicare Current Beneficiary Survey 2006 Access to Care file. 139
Figure B 3 Data File Layouts Data Source: CMS Prescription Drug Plan Fo rmulary and Pharmacy Network File layouts. 140
Figure B 4 2006-2007 Part D Plan Standard Benefits Data Source: CMS 2007 Part D Parameter Update 5_30_2006.pdf. 141
Figure B 5 2007-2008 Part D Standard Benefits Data Source: CMS PartDannouncement2008.pdf. 142
Figure B 6 2006 Part D Risk Corridor Reconciliation Amount Note: The totals include all MAPDs and PDPs in 2006. Data Source: CMS 2006_Part_D_Payment_Recon.pdf. 143
Figure B 7 2007 Part D Risk Corridor Reconciliation Amount Note: The totals include all MAPDs and PDPs in 2007. Data Source: CMS Part_D_2007_Reconciliation.pdf. 144
Figure B 8 2006-2007 Risk Corridors Data Source: Department of Health and Hu man Services, Office of Inspector General. Medicare Part D Sponsors: Estima ted Reconciliation Amounts for 2006. 145
Figure B 9 2008-2011 Risk Corridors Data Source: Department of Health and Hu man Services, Office of Inspector General. Medicare Part D Sponsors: Estima ted Reconciliation Amounts for 2006. 146
147 About the Author Rui Dai obtained her undergraduate degr ee in Japanese Foreign Affairs from China Foreign Affairs University in 1999. After earning her MA in International Relations at Waseda University in Japan in 2001, she entered the PhD program in the Department of Economics at the University of South Florida. Her main area of research is Health Economics.