Health Policy Analysis| Volume 16, ISSUE 4, P629-638, June 01, 2013

# Illustrating Potential Efficiency Gains from Using Cost-Effectiveness Evidence to Reallocate Medicare Expenditures

Open ArchivePublished:May 13, 2013

## Abstract

### Objectives

The Centers for Medicare & Medicaid Services does not explicitly use cost-effectiveness information in national coverage determinations. The objective of this study was to illustrate potential efficiency gains from reallocating Medicare expenditures by using cost-effectiveness information, and the consequences for health gains among Medicare beneficiaries.

### Methods

We included national coverage determinations from 1999 through 2007. Estimates of cost-effectiveness were identified through a literature review. For coverage decisions with an associated cost-effectiveness estimate, we estimated utilization and size of the “unserved” eligible population by using a Medicare claims database (2007) and diagnostic and reimbursement codes. Technology costs originated from the cost-effectiveness literature or were estimated by using reimbursement codes. We illustrated potential aggregate health gains from increasing utilization of dominant interventions (i.e., cost saving and health increasing) and from reallocating expenditures by decreasing investment in cost-ineffective interventions and increasing investment in relatively cost-effective interventions.

Complete information was available for 36 interventions. Increasing investment in dominant interventions alone led to an increase of 270,000 quality-adjusted life-years (QALYs) and savings of $12.9 billion. Reallocation of a broader array of interventions yielded an additional 1.8 million QALYs, approximately 0.17 QALYs per affected Medicare beneficiary. Compared with the distribution of resources prior to reallocation, following reallocation a greater proportion was directed to oncology, diagnostic imaging/tests, and the most prevalent diseases. A smaller proportion of resources went to cardiology, treatments (including drugs, surgeries, and medical devices, as opposed to nontreatments such as preventive services), and the least prevalent diseases. ### Conclusions Using cost-effectiveness information has the potential to increase the aggregate health of Medicare beneficiaries while maintaining existing spending levels. ## Keywords ## Introduction It is well documented that US health care spending growth is unsustainable [ • Keehan S.P. • Sisko A.M. • Truffer C.J. • et al. National health spending projections through 2020: economic recovery and reform drive faster spending growth. , Organisation for Economic Co-operation and Development. OECD Health Data 2010: How Does the United States Compare? Paris, France: OECD Publishing ]. Compared with other developed countries, return on health care spending in the United States is poor, with a significant proportion of the American population lacking health insurance and the health care system performing poorly across key metrics such as life expectancy and infant mortality [ Davis K, Schoen C, Stremikis K. Mirror, mirror on the wall. How the performance of the US health care system compares internationally. 2010 update. 2010. Available from: http://www.commonwealthfund.org/∼/media/Files/Publications/Fund%20Report/2010/Jun/1400_Davis_Mirror_Mirror_on_the_wall_2010.pdf. [Accessed June 8, 2012]. ]. In many countries, cost-effectiveness analysis is used to prioritize scarce health care resources among competing interventions. Despite the immediate need to increase the value of health care spending, however, decision makers in the United States have resisted this approach [ • Bryan S. • Sofaer S. • Siegelberg T. • Gold M. Has the time come for cost-effectiveness analysis in US health care?. ]. More than 46 million Americans, including those 65 years and older and those with certain disabilities, receive health insurance through Medicare. The Centers for Medicare & Medicaid Services (CMS) does not operate with a fixed budget, and program cost has increased annually at a relatively rapid rate. The program’s current annual cost is estimated at upwards of$600 billion, approximately 3.5% of the gross domestic product, and may reach $1 trillion by 2020 [ Centers for Medicare & Medicaid Services. National health expenditure data. Available from: https://www.cms.gov/NationalHealthExpendData/downloads/tables.pdf. [Accessed June 8, 2012]. ]. Research indicates that approximately 30% of Medicare spending may be inappropriate or unnecessary [ • Bentley T.G. • Effros R.M. • Palar K. • Keeler E.B. Waste in the U.S. health care system: a conceptual framework. , • Fisher E.S. • Wennberg D.E. • Stukel T.A. • et al. The implications of regional variations in Medicare spending, part 1: the content, quality, and accessibility of care. , • Garber A. • Goldman D.P. • Jena A.B. The promise of health care cost containment. , United States General Accounting Office. Report to the Chairman, Subcommittee on Health, Committee on Ways and Means, House of Representatives. Medicare: Divided Authority for Policies on Coverage of Procedures and Devices Results in Inequities. Apr 11, 2003, GAO-03-175. Available from: http://www.gao.gov/assets/240/237841.pdf. [Accessed June 8, 2012]. , • Berwick D.M. • Hackbarth A.D. Eliminating waste in US health care. ]. The CMS issues approximately 10 to 15 national coverage determinations (NCDs) each year for interventions deemed to have a significant impact on the Medicare program [ Medicare Program; Revised Process for Making Medicare National Coverage Determinations; Notice of proposed rulemaking, 68 Federal Register 187 (26 September 2003), pp 55634–41 ]. With respect to cost-effectiveness evidence, CMS states that it “is not a factor CMS considers in making NCDs” [ Centers for Medicare & Medicaid Services. Guidance for the public, industry and CMS staff: factors CMS considers in opening a national coverage determination. Available from: http://www.cms.gov/medicare-coverage-database/details/medicare-coverage-document-details.aspx?MCDId=6&McdName=Factors+CMS+Considers+in+Opening+a+National+Coverage+Determination&mcdtypename=Guidance+Documents&MCDIndexType=1&bc=BAAIAAAAAAAA&. [Accessed June 8, 2012]. ]. While research suggests that coverage decisions made in NCDs are broadly consistent with cost-effectiveness evidence—that is, technologies associated with favorable cost-effectiveness estimates tend to be covered—a number of covered interventions are not cost-effective by traditional standards, with incremental cost-effectiveness ratios (ICERs) greater than$250k per quality-adjusted life-year (QALY) gained [
• Chambers J.D.
• Neumann P.J.
• Buxton M.J.
Does Medicare have an implicit cost-effectiveness threshold?.
]. Thus, efficiency gains are possible through disinvestments in cost-ineffective interventions and investments in relatively cost-effective interventions.
The objective of this research was to estimate potential aggregate health gains from increasing utilization of dominant (i.e., cost-saving and health increasing) interventions, and from a hypothetical reallocation of expenditures among interventions subject to NCDs, through the use of a cost-effectiveness decision rule. We also sought to estimate the impact of reallocation on the distribution of expenditures across diseases and types of intervention. We acknowledge that this is an illustrative exercise, but we believe that it is important as the first of its kind to demonstrate the consequences of using cost-effectiveness information to inform resource allocation. We highlight the research challenges, particularly with regards to data limitations.

## Methods

### National Coverage Determinations

We created a database of NCDs issued by the CMS from 1999 through 2007. We excluded incomplete NCDs or those pertaining to minor coding or language changes, as well as those pertaining to off-label treatments, coverage in clinical trials, coverage with evidence development policies, or treatment facilities. Frequently, NCDs include multiple coverage decisions, often for different interventions or patient populations. Furthermore, on occasion coverage is permitted only for patient subgroups that meet certain conditions and restrictions. An entry was made in the database for each separate coverage decision implied within each NCD. We have previously used this database to evaluate factors that predict positive CMS coverage decisions for interventions [
• Chambers J.D.
• Morris S.
• Neumann P.J.
• Buxton M.J.
Factors predicting Medicare national coverage: an empirical analysis.
].

### Reallocation of Expenditures

To facilitate our analysis, we limited our sample to NCDs in which we could find available estimates of: cost-effectiveness; incremental cost; cost of intervention and comparator in the first year of use; incremental health gain; number of Medicare beneficiaries currently receiving the intervention; and the size of the unserved patient population, that is, Medicare beneficiaries who were eligible for the intervention but did not receive it. Each parameter will be discussed further.

### Cost-Effectiveness

On occasion we were able to identify the cost-effectiveness estimate from CMS’s decision memo, which comprises the agency’s public communication about the NCD, including the evidence featured in its review. In the majority of the cases, we identified cost-effectiveness evidence through a literature search by using the PubMed database, the Tufts Medical Center Cost-Effectiveness Analysis Registry, the Health Economic Evaluations Database, and the National Health Service Economic Evaluation Database [

U.S. National Library of Medicine. PubMed: U.S. National Library of Medicine, National Institutes of Health. Available from: http://www.ncbi.nlm.nih.gov/pubmed/. [Accessed June 8, 2012].

,

CEA Registry. Tufts Medical Center Cost-Effectiveness Analysis (CEA) Registry. Available from: http://www.cearegistry.org. [Accessed June 8, 2012].

,

Health Economic Evaluation Database. Health Economic Evaluation Database. Available from: http://www3.interscience.wiley.com/cgi-bin/mrwhome/114130635/HOME. [Accessed June 8, 2012].

,

Centre for Reviews and Dissemination. National Health Service-Economic Evaluation Database. Available from: http://www.crd.york.ac.uk/crdweb/. [Accessed June 8, 2012].

]. The findings of the literature search have been published elsewhere [
• Chambers J.D.
• Neumann P.J.
• Buxton M.J.
Does Medicare have an implicit cost-effectiveness threshold?.
]. Most frequently, the reported ICER was in the form of a cost per QALY gained. On occasion, the ICER was presented in the form of a cost per life-year (LY) gained, and we adjusted incremental survival gain with a utility weight for Americans aged 65 to 69 years to create an estimate of incremental QALY gained [

Erickson P, Wilson R, Shannon I. Years of healthy life. Healthy People 2000 Stat Notes 1995;7:1–15.

]. This adjustment may underestimate the incremental QALY gain as only the years of life extended by the treatment (incremental LYs gained) are accounted for when adjusting for quality of life, not prior years of treatment during which patient quality of life may have been improved. In sensitivity analyses, we included cost-effectiveness studies that estimated the intervention to be “dominant”—that is, more effective and less costly than the comparator—even if the study reported health outcome using disease-specific units, for example, tumors detected, rather than QALYs or LYs. The majority of cost-effectiveness studies were performed in a US health care system setting (26 of 34, 67%), and of those 63% (15 of 26) incorporated Medicare costs. Occasionally, a US study was unavailable and we included a non-US study. In these instances, we converted the ICER into US dollars by using purchasing power parities, and indexed to the year the coverage decision was made by using the health component of the consumer price index [

Organisation for Economic Co-operation and Development. Purchasing power parity (PPP) statistics. 2010. Available from: http://www.oecd.org/dataoecd/61/56/39653523.xls. [Accessed February 27, 2013].

,

U.S. Bureau of Labor Statistics. Measuring Price Change for Medical Care in the CPI. Available from: http://www.bls.gov/cpi/cpifact4.htm. [Accessed June 8, 2012].

].

### Utilization Rate—Served and Unserved Population

We estimated intervention utilization rates by using a database of Medicare inpatient and outpatient claims [

Centers for Medicare & Medicaid Services . Research, statistics, data and systems: limited data sets. Available from: http://www.cms.gov/LimitedDataSets/01_Overview.asp#TopOfPage. [Accessed June 8, 2012].

]. We used International Classification of Diseases, Ninth Revision (ICD-9) diagnostic codes reported in the database to identify Medicare beneficiaries eligible for an intervention, as defined by the parameters of the NCD. The database also includes Common Procedural Terminology (CPT) codes used for physician reimbursement. We estimated utilization rates by calculating the number of beneficiaries who had matching relevant ICD-9 diagnostic and CPT codes. We estimated the size of the unserved eligible population by calculating the difference between the number of beneficiaries who were a match for both ICD-9 diagnostic and CPT codes and those who were a match solely with ICD-9 diagnostic codes.

### Incremental Cost Data

We extracted incremental cost data, that is, the net present value of future expenditures (the numerator of the ICER), from the included cost-effectiveness study, and when necessary adjusted it to 2007 USDs.

### Cost of Intervention and Comparator in First Year of Use

We included the cost of the intervention and the comparator in the first year of use when it was reported in the cost-effectiveness study (64% of cases). When not reported, we estimated the cost of the intervention and the comparator in the year following first use from Medicare and physician reimbursement codes (36% of cases). Pertinent reimbursement codes were identified from Medicare documentation, the included cost-effectiveness study, or the manufacturer’s website. For interventions subject to noncoverage decisions, we obtained the relevant information from the cost-effectiveness study.

### Categorization of Interventions

To analyze the effect of the reallocation exercise on the distribution of expenditures, we categorized interventions with respect to disease (cardiology, oncology, and other), type of intervention (treatment, diagnostic, and other, i.e., education, preventative care, and mobility assistive equipment), and size of the eligible population (>1 million beneficiaries, 50,000–1 million beneficiaries, and<50,000 beneficiaries).

### Reallocation of Expenditures

In the first analysis, we illustrated the effects of increasing the utilization of dominant interventions, while maintaining the existing utilization of nondominant interventions. That is, for dominant interventions we decreased by 50% the size of the unserved population, that is, Medicare beneficiaries who were eligible for the intervention but did not receive it. We assumed a 50% shift for the reallocation, reasoning that shifting all beneficiaries from one intervention to another would be infeasible in practice. To illustrate the possible range of aggregate health gains, we repeated this analysis by adjusting utilization by 10% and 90%, respectively.
In the second analysis, we reallocated existing resources by using an iterative process. First, we ranked interventions in order of cost-effectiveness. Second, to generate resources for investment in more cost-effective interventions, we “disinvested” in the least cost-effective intervention. Disinvestment was achieved by reducing the existing utilization of the least cost-effective intervention by 50%. Third, with the resources generated from the disinvestment, we increased investment in the most cost-effective intervention. This was achieved by decreasing the size of the unserved eligible population by up to 50%. We continued this process by repeating the second and third steps, that is, disinvesting in the next least cost-effective intervention and investing in the next most cost-effective intervention, until no further reallocation of expenditures was possible and there was no net change in expenditure. We also repeated this analysis by adjusting utilization by 10% and 90%, respectively. Finally, we compared the existing and “ideal” distributions of resources across indications, types of technology, and conditions of different prevalence.

### Assumptions

The reallocation exercise adhered to the assumptions necessary for the league table approach described by Johannesson and Weinstein [
• Johannesson M.
• Weinstein M.C.
On the decision rules of cost-effectiveness analysis.
]: perfect divisibility; that is, a partially implemented health care program will maintain the characteristics of the entire program; and constant returns to scale; that is, costs and effects are proportional to the scale of implementation. Our analysis required additional assumptions. First, all Medicare beneficiaries eligible for care, as defined by the parameters of the NCD, received an intervention. Second, the comparator included in the cost-effectiveness study was the alternative intervention received in all cases. Third, we ordered interventions by using the net present value of their total costs (including downstream costs) and assumed that for interventions with a high upfront cost, for example, surgeries, resources were available to fund that “initial investment.”

## Results

Figure 1 shows the process by which we arrived at the final sample of interventions. Thirty-six of 64 interventions associated with an estimate of cost-effectiveness were included (Table 1). Twenty-six of the 28 excluded interventions were removed because of incomplete data, for which the most common reason was our inability to accurately identify the utilization rate for the intervention in the indicated patient population. For example, for ultrasound stimulation for nonunion fracture healing, no beneficiaries had a combination of the necessary ICD-9 codes, that is, ICD-9 code for nonunion fracture healing (733.82) and for fractures of the relevant bones (tibial [823], scaphoid [814], and radius [813]). We excluded two interventions because they were dominated by another intervention in the sample.
Table 1Interventions included in reallocation of expenditures.
InterventionPopulationCoverage statusUtilization (per annum)Cost-effectivenessCosts in year following first useSpending on intervention
Received Tx for diagnosisEligible but did NOT receive txInc. cost ($)Inc. QALYICER ($)Cost of intervention ($)Cost of comparator ($)Existing spending (million) (3.s.f) ($)Reallocated spending Reallocation through disinvestment in less cost-effective interventions by 50%, and investment in relatively cost-effective interventions by decreasing the size of the unserved eligible population by up to 50%. (million) (3.s.f) ($)
Ventricular assist devicesDestination therapy—patients with chronic end-stage heart failure who meet specified criteriaCovered201,474,400416,5450.42986,630331,87865,1779.634.82
Transmyocardial revascularizationPatients with severe angina (stable or unstable), which has been found refractory to standard medical therapyCovered40143,14019,7770.04489,41718,1234,0861.020.511
Liver transplantationPatients suffering from hepatitis BCovered4014,280150,9670.74204,186117,6248,55810.95.45
Ocular photodynamic therapy with verteporfinMacular degeneration— predominately classic subfoveal CNV lesionsCovered1,20072,20014,5040.03195,5669,570018.79.36
Lung volume reduction surgerySevere upper-lobe emphysemaCovered120109,06060,2430.50120,46087,90528,72716.38.17
Implantable cardioverter defibrillators (ICDs)Patients with documented familial or inherited conditions with a high risk of life-threatening ventricular tachyarrhythmiasCovered28,1801,276,88021,1020.16
Reallocation through disinvestment in less cost-effective interventions by 50%, and investment in relatively cost-effective interventions by decreasing the size of the unserved eligible population by up to 50%.
99,78292,78365,8463,6101,800
Pancreas transplantationPancreas transplants—patients who meet the specified criteria (type 1 diabetes etc.)Covered72067,200198,3512.2090,159227,7884,218267134
Positron emission tomographyEsophageal cancerCovered20080,2005,5980.0781,4854,1921,43812.06.00
ICDsNIDCM, documented prior MI, class II and III heart failureCovered03,24077,1131.0176,24437,4747,09000
Deep brain stimulationParkinson’s diseaseCovered39,860687,94047,1210.7265,97053,8535,98824,30012,100
ICDsDocumented sustained ventricular tachyarrhythmiaCovered28,040931,02034,3750.65
Reallocation through disinvestment in less cost-effective interventions by 50%, and investment in relatively cost-effective interventions by decreasing the size of the unserved eligible population by up to 50%.
39,971101,31073,9124,3602,180
Autologous stem cell transplantation (AuSCT)Patients suffering from multiple myelomaCovered801,52083,1231.69
Reallocation through disinvestment in less cost-effective interventions by 50%, and investment in relatively cost-effective interventions by decreasing the size of the unserved eligible population by up to 50%.
37,2752,3961066.963.48
AcupunctureOsteoarthritisNoncovered0744,8605360.0220,38397000
Lumbar artificial disc replacementBack painNoncovered0140,7007,6250.3918,93925,98616,54700
Laparoscopic adjustable gastric banding (LAGB)—bariatric surgeryTreatment of morbid obesityCovered6,6005,976,9008,1000.4518,0283,36614253.510,300
Cochlear implantationPostlingually hearing impaired patientsCovered1,12031,22041,5203.8011,65326,748046.5695
Hyperbaric oxygen therapyHypoxic wounds and diabetic wounds of the lower extremities—diabetic wounds of the lower extremitiesCovered43,8001,196,8001,7710.276,64952403945,770
Electrical bioimpedance for cardiac output monitoringHypertensionNoncovered01,429,0603140.056,408628515017,000
External counterpulsation (ECP) therapyVarious cardiac conditionsNoncovered05,018,5008200.263,2645,343002,060
Positron emission tomographyHead and neck cancersCovered800575,2001,4250.443,2246,0224,5974.821,740
Screening immunoassay fecal-occult blood test—hemoccult IIScreening for colon cancerCovered56,400476,8004000.13
Reallocation through disinvestment in less cost-effective interventions by 50%, and investment in relatively cost-effective interventions by decreasing the size of the unserved eligible population by up to 50%.
1,3185022.6118
Foot careDiabetic peripheral neuropathy with loss of protective sensationCovered400473,200−3860.05Dominates20702.261,340
Cardiac rehabilitation programsAcute MICovered46,400153,800−4700.60Dominates6908032,130
Cardiac rehabilitation programsPercutaneous transluminal coronary angioplastyCovered152,400479,000−4700.60Dominates6902,5606,580
Positron emission tomography (FDG)Breast cancer—initial staging of axillary lymph nodesNoncovered01,257,240609NADominates901002,680
Positron emission tomographyLung cancer (non–small cell)Covered3, 000835,400−698NADominates2,0382,7366.11857
Positron emission tomography (FDG)Breast cancer—staging and restagingCovered2,4001,948,800−759NADominates953022.29,050
Cryosurgery ablationPrimary treatment for clinically localized prostate cancer (stages T1–T3)Covered5,0001,383,600−2,189NADominates6,0178,20630.14,190
Warm-up wound therapy aka noncontact normothermic wound therapy (NNWT)Stage III and IV ulcersNoncovered01,119,120−14,7060.12Dominates5,7538,43103,810
BP, blood pressure; CNV, choroidal neovascularisation; FDG, fludeoxyglucose (18F); ICER, incremental cost-effectiveness ratio; MI, myocardial infarction; NA, not applicable/available; NIDCM, nonischemic dilated cardiomyopathy; QALY, quality-adjusted life-years; Tx, most effective therapeutic option.
* Reallocation through disinvestment in less cost-effective interventions by 50%, and investment in relatively cost-effective interventions by decreasing the size of the unserved eligible population by up to 50%.
We included both positive and noncoverage decisions included in NCDs made from 1999 through 2007. Of the 36 coverage decisions, 29 (81%) were positive and 7 (19%) were noncoverage decisions. We included noncoverage decisions to maximize the sample size and to evaluate the consequences of reallocation when using cost-effectiveness evidence as the sole criterion for resource allocation. While our previous research has demonstrated that covered technologies tend to be more cost-effective than noncovered ones, this was not necessarily the case for the technologies included here [
• Chambers J.D.
• Neumann P.J.
• Buxton M.J.
Does Medicare have an implicit cost-effectiveness threshold?.
]. Indeed, the least cost-effective technologies in our sample were covered, while noncovered technologies were typically associated with favorable ICERs (e.g., the noncovered intervention acupuncture for osteoarthritis had a reported ICER of $18,383) (Table 1). Prior to reallocation, 470,000 beneficiaries received the most effective of the interventions included in the cost-effectiveness analysis, at a cost of approximately$8 billion. Notably, the most cost-ineffective interventions were already in general used least frequently, with interventions associated with ICERs greater than $100,000 per QALY associated with negligible utilization rates (Table 1). For the first analysis, increasing utilization of dominant interventions had a substantial impact on aggregate health gain and cost-savings. When we considered the net present value of future commitments, and included dominant interventions without an estimate of incremental QALY gain, increasing the utilization of dominant interventions while maintaining the existing utilization of interventions associated with a positive ICER resulted in an additional 5.5 million beneficiaries receiving the most effective intervention, an additional 0.27 million QALYs gained, and approximately$12.9 billion of cost-savings (Table 2).
Table 2Estimated gains in aggregate health and cost-savings.
Net present value of future commitmentsYear following first use of the intervention
ReallocationAdditional beneficiaries receiving most effective intervention (millions) (3.s.f)Cost savings (millions) (3.s.f) ($)QALY gain (millions) (3.s.f)Additional beneficiaries receiving most effective intervention (millions) (3.s.f)Cost savings (millions) (3.s.f) ($)QALY gain (millions) (3.s.f)
All interventions (all health outcome units)
Increased utilization of dominant interventions
Increased utilization of dominant interventions through decreasing the size of the unserved population, i.e., Medicare beneficiaries who were eligible for the intervention but who did not receive it, by 50% (10%–90%).
5.54 (1.11– 9.96)13,000 (2,590– 23,300)0.269 (0.0538–0.484)5.54 (1.11–9.96)2,540 (507–4,560)0.269 (0.0538–0.484)
Reallocation of expenditures
Reallocation through disinvestment in less cost-effective interventions, achieved by decreasing the utilization of the intervention by 50% (10%–90%) and investment in relatively cost-effective interventions, achieved by decreasing the size of the unserved eligible population, i.e., Medicare beneficiaries who were eligible for the intervention but who did not receive it, by up to 50% (10%–90%).
11.1 (2.22–20.0)NA1.86 (0.373–3.35)6.73 (1.35–12.1)NA0.580 (0.116–1.04)
Interventions with QALY data
Increased utilization of dominant interventions
Increased utilization of dominant interventions through decreasing the size of the unserved population, i.e., Medicare beneficiaries who were eligible for the intervention but who did not receive it, by 50% (10%–90%).
1.11 (0.223–2.00) 8,470 (1,690– 15,200)0.269 (0.0538– 0.484)1.11 (0.222–2.00)1,430 (286– 2,570)0.269 (0.0538–0.484)
Reallocation of expenditures
Reallocation through disinvestment in less cost-effective interventions, achieved by decreasing the utilization of the intervention by 50% (10%–90%) and investment in relatively cost-effective interventions, achieved by decreasing the size of the unserved eligible population, i.e., Medicare beneficiaries who were eligible for the intervention but who did not receive it, by up to 50% (10%–90%).
6.14 (1.23–11.1)NA1.61 (0.323–2.91)2.10 (0.420–3.78)NA0.527 (0.105–0.949)
NA, not applicable/available; QALY, quality-adjusted life-year; 3.s.f, three significant figures.
* Increased utilization of dominant interventions through decreasing the size of the unserved population, i.e., Medicare beneficiaries who were eligible for the intervention but who did not receive it, by 50% (10%–90%).
Reallocation through disinvestment in less cost-effective interventions, achieved by decreasing the utilization of the intervention by 50% (10%–90%) and investment in relatively cost-effective interventions, achieved by decreasing the size of the unserved eligible population, i.e., Medicare beneficiaries who were eligible for the intervention but who did not receive it, by up to 50% (10%–90%).
For the second analysis, when we considered the net present value of future commitments and included dominant interventions without an estimate of incremental QALY gain, reallocation of expenditures resulted in an additional 11.1 million beneficiaries receiving the most effective intervention (Table 2). This corresponded to an additional 1.86 million QALYs, approximately 0.17 QALYs per beneficiary affected by the reallocation. The ICER of the marginal technology, that is, the least cost-effective intervention for which utilization was increased, was approximately $18,000 per QALY (bariatric surgery for the treatment of morbid obesity). When we reallocated expenditures in the year following the first use of the intervention, the findings were similar, although the magnitude of changes was smaller. The ICER of the marginal technology was approximately$3,300 per QALY (external counterpulsation therapy).
When we considered only interventions with an available estimate of incremental QALY gain, 25 interventions were included in the reallocation. Findings were broadly consistent with the analysis including interventions without a QALY gain estimate, although the magnitude of gain was smaller (Table 2).
The reallocation had a notable impact on the distribution of expenditures across diseases (Table 3). Following reallocation, a greater proportion of expenditures was directed to beneficiaries receiving an oncology-related intervention (approximately 43%), for example, positron emission tomography for various cancers. In contrast, a decreased proportion was directed to those receiving a cardiology-related intervention (approximately 34%), for example, implantable cardioverter defibrillators. Furthermore, a decreased proportion was directed to interventions categorized as other (approximately 24%), for example, warm-up wound therapy for ulcers. With respect to intervention type, following reallocation the proportion of beneficiaries receiving an intervention categorized as treatment, for example, ocular photodynamic therapy with verteporfin for macular degeneration, or diagnostic, for example, screening immunoassay fecal-occult blood test for colorectal cancer, increased (approximately 50% and 44%, respectively), while the proportion of beneficiaries receiving interventions categorized as other, for example, foot care for diabetic peripheral neuropathy with loss of protective sensation, decreased (approximately 7%). Following reallocation, a much greater proportion of expenditures was directed to beneficiaries receiving interventions with an eligible population of more than 1 million beneficiaries (approximately 78%), for example, external counterpulsation therapy, while a decreased proportion was directed to beneficiaries receiving an intervention with an eligible population of 50,000 to 1 million beneficiaries (approximately 22%), for example, cardiac rehabilitation programs for acute myocardial infarction and percutaneous transluminal coronary angioplasty, and less than 50,000 beneficiaries (0.2%), for example, intravenous immune globulin for bullous pemphigoid, respectively.
Table 3Distribution of resources before and after reallocation
Reallocation through disinvestment in less cost-effective interventions, achieved by decreasing the utilization of the intervention by 50% (10%–90%), and investment in relatively cost-effective interventions, achieved by decreasing the size of the unserved eligible population, i.e., Medicare beneficiaries who were eligible for the intervention but who did not receive it, by up to 50% (10%–90%).
of expenditures.
Prior to reallocationFollowing reallocation
Beneficiaries receiving most effective therapeutic option (3.s.f)Distribution (%)Beneficiaries receiving most effective therapeutic option (millions) (3.s.f)Distribution (%)
Disease area
Cardiology256,88054.73.8933.6
Oncology118,88025.54.9442.6
Other disease area94,26020.12.7623.8
Type of intervention
Treatment155,22033.05.7549.7
Diagnostic115,60024.65.0843.8
Other type of intervention199,20042.40.752 6.5
Size of untreated patient population
Large135,60028.89.0378.0
Medium332,78070.82.5321.8
Small1,640 0.30.0230 0.2
3.s.f, three significant figures.
* Reallocation through disinvestment in less cost-effective interventions, achieved by decreasing the utilization of the intervention by 50% (10%–90%), and investment in relatively cost-effective interventions, achieved by decreasing the size of the unserved eligible population, i.e., Medicare beneficiaries who were eligible for the intervention but who did not receive it, by up to 50% (10%–90%).

## Discussion

Coverage of cost-ineffective interventions generates relatively little health gain for the expenditure and suggests that existing resources could provide greater benefits if directed toward more cost-effective alternative interventions. As others have shown, using cost-effectiveness evidence can lead to more efficient resource allocation [
• Lord J.
• Laking G.
• Fischer A.
Health care resource allocation: is the threshold rule good enough?.
]. For this research we used an empirical approach to estimate efficiency gains by reallocating expenditures among interventions considered in NCDs by using a cost-effectiveness decision rule. This approach differs from studies that highlight inefficiencies by comparing expenditures across jurisdictions when adjusting for differences in populations [
• Matlock D.D.
• Peterson P.N.
• Heidenreich P.A.
• et al.
Regional variation in the use of implantable cardioverter-defibrillators for primary prevention: results from the National Cardiovascular Data Registry.
,
• Song Y.
• Skinner J.S.
• Bynum J.P.W.
• et al.
Regional variations in diagnostic practices.
]. While these studies highlight opportunities for efficiency gains, they do not suggest an approach to making care more efficient. In contrast, this study illustrates the potential to increase the aggregate health of Medicare beneficiaries while maintaining existing spending levels.
Our findings suggest that substantial efficiency gains are achievable by reallocating expenditures in accordance with cost-effectiveness evidence. Simply increasing the utilization of dominant interventions increases aggregate health gain, while generating addition resources, an approach in which no patients would receive a less effective therapeutic option. Reallocating expenditures in accordance with a broader definition of the cost-effectiveness evidence increases the number of beneficiaries receiving the most effective therapeutic option—though inevitably some patients would then receive a less effective option than their current therapy—and results in sizeable aggregate health gains. As we included dominant interventions without an estimate of incremental QALY gain in the analysis, aggregate population health gain estimates are likely conservative.
Reallocation also affects the distribution of expenditures across diseases. Following reallocation, a greater proportion of resources was directed to beneficiaries receiving oncology-related interventions and a lesser proportion to beneficiaries receiving cardiology-related interventions and diseases categorized as other. Interestingly, many cost-effective, oncology-related interventions in our sample were diagnostic imaging modalities (e.g., positron emission tomography for various cancers) and tests, rather than chemotherapies, which are often associated with high ICERs [
• Greenberg D.
• Earle C.
• Fang C.H.
• et al.
When is cancer care cost-effective? A systematic overview of cost-utility analyses in oncology.
].

### Limitations and Challenges

Many of the challenges of this research pertain to data limitations. Unlike other health technology assessment agencies—for example, the National Institute of Health and Clinical Excellence in the United Kingdom—the CMS does not independently perform cost-effectiveness analyses, nor does it require submission of cost-effectiveness evidence. Consequently, few of the cost-effectiveness estimates we used originated from CMS decision memos, and for the most part we relied on estimates identified in the cost-effectiveness literature. As a result, there may have been a lack of consistency among cost-effectiveness studies with respect to methodology, perspective, costing, country of study, and so on.
While necessary, we used ICD-9 diagnostic codes to identify eligible beneficiaries, though this approach was somewhat crude, as such codes do not sufficiently capture all factors that inform patient management in practice, for example, patient preference. The Medicare claims database provides a “snapshot” of interventions received by Medicare beneficiaries and does not distinguish between incident and prevalent cases. Without this information, identifying eligible beneficiaries is imprecise; for example, identifying a beneficiary with Parkinson’s disease is insufficient to confirm his or her eligibility for deep brain stimulation, as this treatment is indicated only once pharmaceutical management is no longer effective. For other interventions, this is less problematic (e.g., foot care for diabetic patients suffering from diabetic peripheral neuropathy with loss of protective sensation), as the ICD-9 diagnostic codes were sufficient to identify this patient population.
An important assumption was that eligible patients who did not receive the intervention instead received the study comparator. While we attempted to ensure that CMS deemed the comparator relevant by affirming that CMS discussed it in the decision memo, this assumption introduces potential bias. In many cases, despite eligibility, Medicare beneficiaries are likely to receive no therapy or an alternative therapy not included in the cost-effectiveness analysis. Thus, there is potential for the ICER to be either overestimated or underestimated. In future research, we plan to explore and try to validate this assumption, potentially utilizing clinical opinion.
The requirement to integrate data from various sources limits the accuracy of study findings. Indeed, our study highlights the challenges of using currently available data for this type of analysis. While US-based recommendations for the performance cost-effectiveness analyses exist, evidence suggests that they are followed inconsistently [
• Phillips K.A.
• Chen J.L.
Impact of the U.S. panel on cost-effectiveness in health and medicine.
].

### Policy Significance and Next Steps

In the United States, cost-effectiveness is often conflated with “rationing” and a reduction in health care provision. This research shows that rather than using it to reduce spending, it can identify how to increase aggregate population health while maintaining existing spending levels. Despite the limited number of considered interventions, our analysis illustrated that an additional 1.86 million QALYs (0.17 QALYs per affected beneficiary) were achievable, suggesting substantial gains in aggregate health if the policy were implemented on a larger scale.
The least cost-effective intervention to which more expenditure was allocated after reallocation was associated with an ICER of $18,000 per QALY (bariatric surgery for the treatment of morbid obesity), less than the often used benchmarks of value in the United States, for example,$50,000 to \$100,000 per QALY [
• Greenberg D.
• Neumann P.J.
Is it cost-effective? It depends on who you ask.
]. This finding suggests that there is much potential for Medicare to increase spending on interventions associated with ICERs judged highly cost-effective by traditional standards, while reducing spending on interventions deemed comparatively cost-ineffective [
• Weinstein M.C.
• Skinner J.A.
Comparative effectiveness and health care spending--implications for reform.
].
Our research is not offered as a precise accounting exercise, but rather as an illustration of the potential of this approach, which, while incorporating a number of assumptions, underscores the substantial aggregate health gains potentially achievable from using cost-effectiveness evidence to inform resource allocation. Indeed, gains may be even greater if one could better account for patient heterogeneity and if beneficiaries who would benefit most were prioritized for treatment.
For a number of included dominant interventions, positive expenditure was required in their first year of use, with aggregate cost-savings achieved in subsequent years, for example, foot care for diabetic patients with neuropathy. This finding emphasizes that considering interventions over a short time horizon may not adequately account for the potential positive financial impact on the entire health care system. The apparent underutilization of dominant interventions provides an opportunity for policymakers. Research suggests that underutilization may be due to a lack of physician referral, insufficient physician reimbursement, and perceived clinical benefits of the intervention, among other factors [
• Daly J.
• Sindone A.P.
• Thompson D.R.
• et al.
Barriers to participation in and adherence to cardiac rehabilitation programs: a critical literature review.
,
• Parkosewich J.A.
Cardiac rehabilitation barriers and opportunities among women with cardiovascular disease.
,
• Thomas R.J.
Cardiac rehabilitation/secondary prevention programs: a raft for the rapids: why have we missed the boat?.
].
Ideally, data available for the research in this article would be more abundant, and of higher quality and greater consistency. Preferably, available cost-effectiveness studies would include Medicare-specific direct costs and account for all relevant competing interventions. Costs would meet the specifications necessary to facilitate potential legislative action, for example, meet the standards of the Congressional Budget Office and CMS’s Office of the Actuary [

Congressional Budget Office. Our processes. Available from: http://www.cbo.gov/about/our-processes [Accessed June 8, 2012]

,

Centers for Medicare & Medicaid Services. CMS's Office of the Actuary. Available from: http://www.cms.gov/About-CMS/Agency-Information/CMSLeadership/Office_OACT.html [Accessed June 8, 2012].

]. Technologies should be evaluated over a time horizon for which costs and consequences, in terms of QALYs, are likely to differ, and reported on an annual basis. Ideally, studies would account for patient heterogeneity, with data available on the effectiveness of the intervention across beneficiaries in the indicated population.
Using a cost-effectiveness decision rule for resource allocation will impact the distribution of resources across Medicare beneficiaries. Despite the small number of included coverage decisions, the findings underscore the trade-offs inherent in resource allocation decisions, for example, the opportunity cost of prioritizing resources to a particular patient group. Indeed, the objective of maximizing aggregate health subject to a budget constraint is not entirely consistent with CMS’s decision-making criteria. Other factors including the nature, strength, and uncertainty of the available evidence; the availability of alternative interventions; potential impact of the decision on access to health care; statutory mandates to cover certain services; and political considerations are important as well [
• Keenan P.S.
• Neumann P.J.
• Phillips K.A.
Biotechnology and Medicare’s new technology policy: lessons from three case studies.
,
• Griffin S.
• Claxton K.
• Palmer S.
• Sculpher M.J.
Dangerous omissions: the consequences of ignoring decision uncertainty.
].
It is important to highlight that a minority of CMS’s coverage decisions for interventions are made at the national level in NCDs and that we included only coverage decisions associated with cost-effectiveness estimates. Consequently, this research is limited to a relatively small, and potentially unrepresentative, selection of interventions. Included interventions may not be those for which a reallocation of expenditures would yield the greatest efficiency gains. Indeed, targeting interventions with the largest potential health gains may be impractical if they reflect widely used and accepted services, despite poor cost-effectiveness. Rather, targeting interventions for which a change in therapeutic management is more feasible may be a more appropriate approach. We chose arbitrarily to use a 50% (range 10%–90%) change in utilization to simulate resource allocation. Accounting for the relative ease of investing/disinvesting in interventions would be one approach for advancing this research and may produce more realistic efficiency gain estimates.
As an illustrative exercise this research could be furthered by focusing on a select group of interventions for which high-quality data are available, potentially interventions from the same class or for the same indication. Having data of sufficient quality would allow the adoption of alternative approaches, for example, integer programming, or a stochastic process to account for uncertainty in the parameter estimates and for disease incidence.
A larger sample would allow a more comprehensive categorization of interventions and facilitate a more thorough examination of the consequences of using cost-effectiveness evidence to guide resource allocation. For example, as research has shown that society has a preference for the treatment of severe diseases, a variable to capture disease severity would be valuable [
• Coast J.
Is economic evaluation in touch with society’s health values?.
,
• Dolan P.
• Cookson R.
A qualitative study of the extent to which health gain matters when choosing between groups of patients.
,
• Nord E.
• Richardson J.
• Street A.
• et al.
Maximizing health benefits vs egalitarianism: an Australian survey of health issues.
,
• Ubel P.A.
Pricing Life: Why It’s Time for Health Care Rationing.
]. By better understanding Medicare beneficiaries’ resource allocation preferences, the appropriateness of alternative resource use patterns could be evaluated.

## Conclusions

While the US health care system has not embraced the use of cost-effectiveness evidence to inform health care resource allocation, this research illustrates the potential value and consequences of such an approach. While it is apparent that available data present challenges with this methodology, our research illustrates that substantial health gains are achievable from a reallocation of expenditures within existing spending levels.
Source of financial support: This research was funded in part by a grant from the Commonwealth Fund (Grant No. 20090203).

## References

• Keehan S.P.
• Sisko A.M.
• Truffer C.J.
• et al.
National health spending projections through 2020: economic recovery and reform drive faster spending growth.
Health Aff (Millwood). 2011; 30: 1594-1605
1. Organisation for Economic Co-operation and Development. OECD Health Data 2010: How Does the United States Compare? Paris, France: OECD Publishing

2. Davis K, Schoen C, Stremikis K. Mirror, mirror on the wall. How the performance of the US health care system compares internationally. 2010 update. 2010. Available from: http://www.commonwealthfund.org/∼/media/Files/Publications/Fund%20Report/2010/Jun/1400_Davis_Mirror_Mirror_on_the_wall_2010.pdf. [Accessed June 8, 2012].

• Bryan S.
• Sofaer S.
• Siegelberg T.
• Gold M.
Has the time come for cost-effectiveness analysis in US health care?.
Health Econ Policy Law. 2009; 4: 425-443
3. Centers for Medicare & Medicaid Services. National health expenditure data. Available from: https://www.cms.gov/NationalHealthExpendData/downloads/tables.pdf. [Accessed June 8, 2012].

• Bentley T.G.
• Effros R.M.
• Palar K.
• Keeler E.B.
Waste in the U.S. health care system: a conceptual framework.
Milbank Q. 2008; 86: 629-659
• Fisher E.S.
• Wennberg D.E.
• Stukel T.A.
• et al.
The implications of regional variations in Medicare spending, part 1: the content, quality, and accessibility of care.
Ann Intern Med. 2003; 138: 273-287
• Garber A.
• Goldman D.P.
• Jena A.B.
The promise of health care cost containment.
Health Aff (Millwood). 2007; 26: 1545-1547
4. United States General Accounting Office. Report to the Chairman, Subcommittee on Health, Committee on Ways and Means, House of Representatives. Medicare: Divided Authority for Policies on Coverage of Procedures and Devices Results in Inequities. Apr 11, 2003, GAO-03-175. Available from: http://www.gao.gov/assets/240/237841.pdf. [Accessed June 8, 2012].

• Berwick D.M.
• Hackbarth A.D.
Eliminating waste in US health care.
JAMA. 2012; 307: 1513-1516
5. Medicare Program; Revised Process for Making Medicare National Coverage Determinations; Notice of proposed rulemaking, 68 Federal Register 187 (26 September 2003), pp 55634–41

6. Centers for Medicare & Medicaid Services. Guidance for the public, industry and CMS staff: factors CMS considers in opening a national coverage determination. Available from: http://www.cms.gov/medicare-coverage-database/details/medicare-coverage-document-details.aspx?MCDId=6&McdName=Factors+CMS+Considers+in+Opening+a+National+Coverage+Determination&mcdtypename=Guidance+Documents&MCDIndexType=1&bc=BAAIAAAAAAAA&. [Accessed June 8, 2012].

• Chambers J.D.
• Neumann P.J.
• Buxton M.J.
Does Medicare have an implicit cost-effectiveness threshold?.
Med Decis Making. 2010; 30: E14-E27
• Chambers J.D.
• Morris S.
• Neumann P.J.
• Buxton M.J.
Factors predicting Medicare national coverage: an empirical analysis.
Med Care. 2012; 50: 249-256
7. U.S. National Library of Medicine. PubMed: U.S. National Library of Medicine, National Institutes of Health. Available from: http://www.ncbi.nlm.nih.gov/pubmed/. [Accessed June 8, 2012].

8. CEA Registry. Tufts Medical Center Cost-Effectiveness Analysis (CEA) Registry. Available from: http://www.cearegistry.org. [Accessed June 8, 2012].

9. Health Economic Evaluation Database. Health Economic Evaluation Database. Available from: http://www3.interscience.wiley.com/cgi-bin/mrwhome/114130635/HOME. [Accessed June 8, 2012].

10. Centre for Reviews and Dissemination. National Health Service-Economic Evaluation Database. Available from: http://www.crd.york.ac.uk/crdweb/. [Accessed June 8, 2012].

11. Erickson P, Wilson R, Shannon I. Years of healthy life. Healthy People 2000 Stat Notes 1995;7:1–15.

12. Organisation for Economic Co-operation and Development. Purchasing power parity (PPP) statistics. 2010. Available from: http://www.oecd.org/dataoecd/61/56/39653523.xls. [Accessed February 27, 2013].

13. U.S. Bureau of Labor Statistics. Measuring Price Change for Medical Care in the CPI. Available from: http://www.bls.gov/cpi/cpifact4.htm. [Accessed June 8, 2012].

14. Centers for Medicare & Medicaid Services . Research, statistics, data and systems: limited data sets. Available from: http://www.cms.gov/LimitedDataSets/01_Overview.asp#TopOfPage. [Accessed June 8, 2012].

• Johannesson M.
• Weinstein M.C.
On the decision rules of cost-effectiveness analysis.
J Health Econ. 1993; 12: 459-467
• Lord J.
• Laking G.
• Fischer A.
Health care resource allocation: is the threshold rule good enough?.
J Health Serv Res Policy. 2004; 9: 237-245
• Matlock D.D.
• Peterson P.N.
• Heidenreich P.A.
• et al.
Regional variation in the use of implantable cardioverter-defibrillators for primary prevention: results from the National Cardiovascular Data Registry.
Circ Cardiovasc Qual Outcomes. 2011; 4: 114-121
• Song Y.
• Skinner J.S.
• Bynum J.P.W.
• et al.
Regional variations in diagnostic practices.
New Engl J Med. 2010; 363: 45-53
• Greenberg D.
• Earle C.
• Fang C.H.
• et al.
When is cancer care cost-effective? A systematic overview of cost-utility analyses in oncology.
J Natl Cancer Inst. 2010; 102: 82-88
• Phillips K.A.
• Chen J.L.
Impact of the U.S. panel on cost-effectiveness in health and medicine.
Am J Prev Med. 2002; 22: 98-105
• Greenberg D.
• Neumann P.J.
Is it cost-effective? It depends on who you ask.
in: Rosen B. Israeli A. Shortell S. Improving Health and Healthcare. Who is Responsible? Who is Accountable? Tel Hashomer, Israel: The Israel National Institute for Health Policy Research. 2011
• Weinstein M.C.
• Skinner J.A.
Comparative effectiveness and health care spending--implications for reform.
N Engl J Med. 2010; 362: 460-465
• Daly J.
• Sindone A.P.
• Thompson D.R.
• et al.
Barriers to participation in and adherence to cardiac rehabilitation programs: a critical literature review.
Prog Cardiovasc Nurs. 2002; 17: 8-17
• Parkosewich J.A.
Cardiac rehabilitation barriers and opportunities among women with cardiovascular disease.
Cardiol Rev. 2008; 16: 36-52
• Thomas R.J.
Cardiac rehabilitation/secondary prevention programs: a raft for the rapids: why have we missed the boat?.
Circulation. 2007; 116: 1644-1646
15. Congressional Budget Office. Our processes. Available from: http://www.cbo.gov/about/our-processes [Accessed June 8, 2012]

16. Centers for Medicare & Medicaid Services. CMS's Office of the Actuary. Available from: http://www.cms.gov/About-CMS/Agency-Information/CMSLeadership/Office_OACT.html [Accessed June 8, 2012].

• Keenan P.S.
• Neumann P.J.
• Phillips K.A.
Biotechnology and Medicare’s new technology policy: lessons from three case studies.
Health Aff (Millwood). 2006; 25: 1260-1269
• Griffin S.
• Claxton K.
• Palmer S.
• Sculpher M.J.
Dangerous omissions: the consequences of ignoring decision uncertainty.
Health Econ. 2011; 29: 212-224
• Coast J.
Is economic evaluation in touch with society’s health values?.
BMJ. 2004; 329: 1233-1236
• Dolan P.
• Cookson R.
A qualitative study of the extent to which health gain matters when choosing between groups of patients.
Health Policy. 2000; 51: 19-30
• Nord E.
• Richardson J.
• Street A.
• et al.
Maximizing health benefits vs egalitarianism: an Australian survey of health issues.
Soc Sci Med. 1995; 41: 1429-1437
• Ubel P.A.
Pricing Life: Why It’s Time for Health Care Rationing.
Cambridge, MA: MIT Press, 2001