Unlike ACOs or P4P, implementation of bundled payment for inpatient and post acute care in Medicare would modestly reduce geographic variation in spending.
Objectives: The Institute of Medicine has recently argued against a value index as a mechanism to address geographic variation in spending and instead promoted payment reform targeted at individual providers. It is unknown whether such provider-focused payment reform reduces geographic variation in spending.
Study Design: We estimated the potential impact of 3 Medicare provider-focused payment policies—pay-for-performance, bundled payment, and accountable care organizations—on geographic variation in Medicare spending across Hospital Referral Regions (HRRs). We compared geographic variation in spending, measured using the coefficient of variation (CV) across HRRs, between the baseline case and a simulation of each of the 3 policies.
Methods: Policy simulation based on 2008 national Medicare data combined with other publicly available data.
Results: Compared with the baseline (CV, 0.171), neither pay-for-performance nor accountable care organizations would change geographic variation in spending (CV, 0.171), while bundled payment would modestly reduce geographic variation (CV, 0.165).
Conclusions: In our models, the bundled payment for inpatient and post acute care services in Medicare would modestly reduce geographic variation in spending, but neither accountable care organizations nor pay-for-performance appear to have an impact.
Am J Manag Care. 2015;21(6):e390-e398
Medicare spending per fee-for-service beneficiary in 2008 varied from $6000 in Rapid City, South Dakota, to more than $18,000 in Miami, Florida. While some policy makers favor directly reducing payments in high-cost areas, the Institute of Medicine favors policies focused on provider inefficiency. We investigated whether several such policies would have the virtue of also reducing variation in spending. If inefficiency is concentrated in high-cost areas, then this may be the case.
Medicare spending per fee-for-service (FFS) beneficiary in 2008 varied widely across the 306 Hospital Referral Regions (HRRs) of the United States: it ranged from $6000 in Rapid City, South Dakota, to more than $18,000 in Miami, Florida. Some of this geographic variation is driven by differing health status and other demographic differences among residents of these regions.1 However, research has also suggested that a large component of this geographic variation is due to differences in medical practice that do not appear to be associated with better healthcare quality or outcomes.2-4 These concerns have raised interest in policies aimed toward reducing geographic variation in spending.
As part of the debate over the Affordable Care Act, Congress considered implementing a “value index,” which would directly address this variation by reducing the rates Medicare pays providers in high-cost regions.5 However, the Institute of Medicine (IOM) has argued against such policies, deeming them an overly blunt instrument.6 Varying payment rates in this way would not account for the substantial differences within regions in provider efficiency, and thus, could penalize low-cost providers in high-cost regions.
Instead, the IOM favors policies that aim to reduce inefficiency at the provider level. Indeed, if inefficiency is more prevalent in high-cost areas, then policies that reduce inefficiency might also reduce geographic variation in spending as a beneficial accompanying effect. To assess the effects of such policies, the IOM asked us to model the impact of provider-focused interventions on geographic variation in spending.
In this article, we estimate the impact of 3 provider-focused policies on geographic variation in Medicare spending: 1) bundled payment, 2) pay-for-performance (P4P), and 3) accountable care organizations (ACOs). We chose these 3 policies as prominent, realistic interventions that are currently being implemented or piloted in Medicare, as well as in the private sector. Generally, these policies aim to improve upon the incentive inherent in FFS payment to increase volume of care without necessarily improving outcomes.
For each policy, we generated a number of scenarios representing realistic but robust implementations of the policy and then estimated geographic variation in Medicare spending under each scenario. To ensure that the scenarios would be realistic, we based their design on policies that Medicare has already implemented (either as pilots or full programs). To ensure that they are robust (ie, illustrative of the potential impacts of a large-scale program), we modified the existing programs in many key ways. For example, instead of a voluntary bundled payment program, as is currently being piloted in Medicare, we modeled a mandatory program. We provide an overview of the policies and scenarios below with additional details in the eAppendix (available at www.ajmc.com).
How the 3 Policies Theoretically Could Decrease Geographic Variation
In this section, we explain potential mechanisms by which these policies could reduce geographic variation in spending. Under a P4P program, providers such as hospitals, medical groups, and nursing homes receive higher payments if they attain a high level of performance on quality measures, or improve their performance on quality measures (some related recent policies also reward performance on cost measures, but we did not include such scenarios). For P4P to decrease geographic variation in spending, there must be a cost-quality relationship. If high-quality providers are clustered in regions with lower spending, then the P4P program would shift money from high-cost areas to low-cost areas, thereby reducing geographic variation in spending.
“Bundled payment” is a payment method in which providers receive a single payment for so-called bundles of healthcare services related to a patient’s medical condition or a medical procedure. For bundled payment to decrease geographic variation in Medicare spending, there would need to be a national payment rate for each bundle and high-cost providers of bundles clustered geographically.
ACOs were initiated in Medicare as part of the Affordable Care Act. Organizations assume responsibility for the total costs of care for a designated population of Medicare beneficiaries, and if Medicare payments for assigned beneficiaries fall below a target, Medicare pays the provider organizations a fraction of the difference as bonus payments (if quality standards are met), and thus both Medicare and the organization benefit financially. In some models, provider organizations may also lose money if Medicare payments for assigned beneficiaries exceed the target. If ACOs do save money for Medicare on net and are clustered in higher-cost areas—and/or if ACOs save more in high-cost areas than low-cost areas—then geographic variation in spending could be reduced.
To evaluate whether the 3 policies would decrease geographic variation in Medicare spending, we compared 2008 Medicare spending for each HRR under the baseline case with scenarios in which the policy was implemented. We compared the degree of geographic variation in the baseline case with that under the policies. Given space limitations, we provide an overview of our work and only 1 scenario per policy. The online eAppendix includes a detailed description of methods and data as well as results of other scenarios (sensitivity analyses) for each policy.
We compared all scenarios with unadjusted total Medicare spending for FFS full-year Part A and Part B enrollees 65 years or older in 2008, as reported by the IOM.7 The underlying data derives from the CMS Chronic Conditions Warehouse,8 which contains all Medicare claims for FFS beneficiaries. Methods, key assumptions, and data unique to each of our policy scenarios are described below. We present each policy independently, though we recognize potential policy interaction with simultaneous implementation.
We analyzed the impact of Medicare P4P programs targeting hospitals, nursing homes, and home health agencies (we report the effects of all programs combined). We based scenarios on existing or pilot Medicare P4P initiatives—specifically, the Hospital Value-Based Purchasing Program, the Nursing Home Quality-Based Purchasing Demonstration, and the Home Health Pay-for-Performance Demonstration.
Reflecting the design of existing and prior P4P programs, we measured each provider’s performance on quality measures in terms of both achievement and improvement based on publicly available quality scores, with the latter based on changes in scores over 2 years. Using the 2008 total Medicare spending baseline, we estimated the effects of transferring 15% of total provider payments to an incentive pool. We then allocated pool funds to providers based on a linear exchange curve method in which the provider with the worst performance received no incentive payments, and providers received larger incentive payments with increased performance. Compared with the nursing home and home health Medicare programs currently being implemented, our scenarios were larger in scope (a national program vs regional pilots) and devoted a much larger amount of money to incentive payments.
We estimated the effects of a hypothetical mandatory Medicare bundled payment program. The main features of the hypothetical program were fashioned after the original design of the Medicare Bundled Payment for Care Improvement Initiative, a voluntary bundled payment program currently being implemented by CMS. (Note: CMS has recently been changing some elements of the design.) Consistent with the Medicare initiative, defined bundles of care include all Medicare Part A and Part B services provided to hospitalized beneficiaries from admission through 30 days post discharge. We created bundles for 10 high-volume, high-cost conditions as defined by 27 Medicare Severity — Diagnosis-Related Groups (MS-DRGs): acute myocardial infarction, congestive heart failure, chronic obstructive pulmonary disease, gastrointestinal bleed, hip fracture, kidney/urinary tract infection, lower extremity joint replacement, pneumonia, septicemia, and stroke. Together, the services included in bundles for these 27 MS-DRGs accounted for 15% of total Medicare Part A and Part B spending in 2008. We set a national base payment rate for each bundle that was adjusted for area-level input prices, similar to the Medicare Inpatient Prospective Payment System. The base payment rate for each bundle was set such that national spending on the bundled services remained unchanged from the baseline (revenue-neutral). In our policy scenario, all providers meeting a minimum volume threshold of 10 bundles per year would receive the bundled payment; providers below the volume threshold would continue to receive payments under status quo policies.
Accountable Care Organizations
We estimated the impact of ACOs on geographic variation in Medicare spending by estimating current ACO enrollees’ locations and assuming a reduction in spending for each enrollee. We identified beneficiaries associated with 148 ACOs participating in Medicare programs as of the end of 2012, plus 77 private sector ACOs that currently participate in various initiatives or pilots. We included private sector ACOs as “proxies” for where future Medicare ACOs might form, allowing us to model a more robust ACO program. In order to define which areas are affected by ACO enrollment, we assigned Medicare beneficiaries in these ACOs to an HRR using data directly from Medicare (when we had such data available). For other ACOs, we assigned beneficiaries to HRRs based on where the primary care physicians in the ACO were located, or if lacking that information, based on the location of associated hospitals or the ACO headquarters. Together, we estimated that the ACOs in our policy scenario would serve roughly 10% of Medicare FFS beneficiaries (7% in Medicare ACOs, 3% in private-sector ACOs). We employed an estimate of per beneficiary savings to Medicare of 3% to 5% depending on the type of ACO—rates somewhat higher than those implied by the literature.9,10 We also assumed proportionally larger spending reductions for ACOs in areas with higher risk-adjusted spending, since these ACOs have potentially greater opportunity for improvement. We made these assumptions in keeping with the objective of simulating the potential impacts on geographic variation of robust versions of the policies in question.
Impact of 3 Policies on Geographic Variation in Spending
The effect of each policy on geographic variation in Medicare spending is illustrated in Figure 1, which separates the 306 HRRs in the United States into quintiles in their initial level of spending (lowest-spending HRRs are to the left) and displays the average change in spending in each quintile, under the 3 policies. Under P4P (the first cluster of columns), it is apparent that the effects on spending are very small and do not show a strong geographic pattern of impacts by spending quintile. The coefficient of variation (CV) of Medicare spending remains at 0.171 both in the baseline and under the policy scenario. Similar patterns emerge when each P4P program is analyzed separately: inpatient, home health, and nursing home (results in eAppendix). The ACO scenario reduces spending in all HRRs, but with a relatively weak geographic pattern. Despite the relatively larger reduction in higher-cost HRRs, the CV under this policy is also unaffected and remains at 0.171. Under the bundled payment scenario, there is a clearer pattern of spending increases in the lower quintiles and spending reductions in the higher quintiles, leading to a reduction in the CV of geographic variation in total Medicare spending from 0.171 to 0.165. The reduction would be just over $400 (or 2.3%) in the highest cost HRR: Miami. In analyses focusing on only the portion of Medicare spending contained within the bundles, the CV under the bundled payment policy is 0.131 compared with a baseline of 0.158.
We conducted a number of sensitivity analyses in which we altered key parameters concerning how the policies were implemented (see eAppendix for results). In 1 alteration, we modeled a version of P4P in which incentive payments were allocated “tournament style” (only the top providers received any payments), and in another, we assumed that ACOs proliferated more widely to include 20% of Medicare FFS beneficiaries. Our results were not sensitive to these alternative scenarios. In alternative scenarios using price-standardized Medicare payment rates (eg, omitting disproportionate share hospital, indirect medical education, and area wage and price adjustments), the reduction in geographic variation was slightly smaller, while alterations removing the volume threshold in the bundled payment program resulted in a slightly larger reduction in variation.
What is Driving the Impact (or lack thereof) of the 3 Policies on Geographic Variation in Medicare Spending?
All 3 of the policies examined would have substantial effects on Medicare payments to individual providers, reallocating payments from low-performing to high-performing providers (the definition of good performance differs among the 3 policies). For example, under bundled payment in the case of acute myocardial infarction, the 5% of providers benefiting the most would receive more than a 20% increase in payments, while the 5% faring the worst would face more than a 15% reduction. Under P4P, 5% of home health providers would receive at least a 15% increase in payments and 5% would face more than an 11% reduction.
Nevertheless, we estimated that 2 of the policies (P4P and ACOs) would have no effect on geographic variation in spending, and that the third (bundled payment) would have a modest effect, with the reason varying across the 3 policies.
The lack of effect of P4P on geographic variation is due to the low correlation between quality and spending in a given area. For example, shows little systematic correlation between performance on select inpatient quality measures and inpatient Medicare spending at the HRR level. There is also no consistent relationship between quality and spending for the nursing home and home health quality measures (data not shown).
The bundled payment scenario does exhibit a modest impact on geographic variation in Medicare spending—partly because we assume that Medicare will pay a national base rate (although with geographic adjustments for input price) for bundles of services. This policy would essentially “flatten out” variation in payments per bundle, and those payments represent roughly 15% of Medicare spending in a given year. However, this “flattening” alone would not necessarily reduce geographic variation at the HRR level. For example, imagine that all geographic variation across HRRs was due to variations in outpatient physician visits alone, and that inpatient and post acute care treatment and spending averaged exactly the same in each HRR (though it still would vary among providers inside of each HRR). In that case, the bundled payment scenario would still reduce variation within each HRR but have no impact on geographic variation at the HRR level. This is not the case though. As shown in Figure 2, the same HRRs with high overall spending also tend to have high spending on the bundles of care affected by the policy. Therefore, the policy would result in a reduction in payments to high-spending areas and an increase in payments to low-spending areas, thereby decreasing geographic variation.
There are several reasons the estimated reduction in geographic variation due to bundled payment is modest, however. First, approximately half of the spending on the bundles is related to the facility payment for the hospitalization occurring at the beginning of the bundle. That payment does not change under the bundled payment policy—hospitalizations are currently paid for under the Inpatient Prospective Payment System. Thus, differences in the amount spent on bundles under the new policy are driven by differences in what happens after the inpatient admission—mainly variation in readmission rates and post acute care use by region. Second, only 15% of Medicare spending was captured by the conditions we selected for bundled payment, after excluding providers with a low volume of care for any given bundle from the policy scenario. Lastly, overall spending on bundles in a region is a function of cost per bundle and number of bundles per capita. Bundled payment does not directly address the considerable variation across HRRs in the number of bundles provided. It is possible a bundled payment program could result in changes in the volume of bundles provided, but we lacked a solid evidence base to estimate the direction or magnitude of an expected effect.
Enrollment of Medicare beneficiaries in ACOs would reduce geographic variation in Medicare spending if 2 conditions were met: 1) ACOs do indeed achieve cost savings (assumed in our scenario), and 2) ACOs are more likely to form in higher-cost areas and/or result in larger savings in higher-cost areas. Although we assumed a modest degree of enhanced savings in high-cost areas, the lack of a strong association between area-level ACO formation and Medicare spending ensures the lack of an effect on geographic variation. Figure 3 plots participation in ACOs at the HRR level against HRR-level spending.
Slightly higher ACO penetration exists in higher-cost HRRs but the relationship is weak (r = 0.05; P = .34). As a result, we estimate that ACOs will result in lower Medicare payments to areas with both high and low baseline spending, with little effect on the extent of geographic variation in spending.
Congress is seeking mechanisms to decrease geographic variation in spending, and direct interventions, such as payment rate adjustments to all providers in a region based on spending levels, would certainly be effective in reducing variation. Yet because these may negatively impact low-cost providers in high-spending regions, the IOM Committee on Geographic Variation in Health Care Spending and Promotion of High-Value Care has argued that such changes are an overly blunt instrument. The Committee has argued that provider-focused payment reform policies should be promoted. It is important to understand whether such policies reduce geographic variation in spending.
We estimated the impact on geographic variation in spending of 3 policies (P4P, bundled payment, ACOs) that focus on individual providers, are at the forefront of healthcare payment policy, and could theoretically decrease geographic variation in spending. Each scenario assumed that a sizable fraction of Medicare spending (approximately 10%-15%) would be directly affected by the new payment policy in any given year. The number of beneficiaries affected would be potentially even higher—for example, those not in ACOs may still share physicians with those who are; many beneficiaries are cared for in hospitals or other institutions affected by the bundled payment or P4P policies. The policy scenarios therefore result in substantial reallocations of Medicare payments to providers compared with the status quo. However, we estimated that P4P and ACO scenarios would not change geographic variation at all, and that the bundled payment scenario would only modestly decrease geographic variation—as a point of comparison, the reduction is about half as much as would be achieved from simply removing teaching, outlier, and area wage and input cost adjustments to Medicare payment rates.
As further illustration, the impacts of each of the policies on Medicare spending in selected HRRs from each spending quintile is shown in . In the high-spending Miami, Florida area, the bundled payment policy reduces spending by roughly $400 per beneficiary (from $18,017 to $17,598) and raises spending slightly in low-spending Rapid City, South Dakota. Impacts on other HRRs and of other policies are generally less than $100 per beneficiary.
It may not be surprising that we found little impact of P4P and ACOs on geographic variation in spending. For P4P to decrease geographic variation in spending, there must be a relationship between spending and quality, and prior work has documented no consistent relationship between these two factors. For ACOs to decrease geographic variation in spending, they must preferentially locate in geographic areas with high spending; prior work has provided mixed results as to whether ACOs are preferentially forming in such regions. Nevertheless, given continued uncertainty, our results emphasize that such policies currently being promoted would be unlikely to reduce geographic variation in Medicare spending.
As the IOM itself has noted, it is unclear whether reducing geographic variation across HRRs is a good metric of successful policy interventions or a national priority. Medical practice is not homogeneous within HRRs, and variation in care between providers instead of regions might prove a better target for policy. Also, measures of geographic variation in total Medicare spending does not account for the important distinction between high-value and low-value spending.
That the policies we investigated had limited impact on geographic variation in spending does not mean they would be ineffective—they were not designed primarily to influence variation. The 3 policies would have substantial effects on Medicare payments to providers, however. As a result, P4P may drive quality improvement; bundled payment and ACOs may improve care and reduce costs. They might also reduce variation in spending among providers within HRRs, but we did not focus on variation at that level. Our results should therefore not be interpreted as evidence that these provider-focused policies are not useful.
To the extent that reduction in geographic variation in Medicare spending remains a national priority, our results provide insight on how the policies we investigated could be adjusted to achieve that goal. Instead of the set of measures we employed, a set of P4P quality measures could be identified in which high-cost areas of the United States have particularly low quality (eg, readmission rates); that would ensure a transfer of funds from high-cost regions with poor quality scores to low-cost regions with high scores. Also, policy makers could identify barriers to ACO formation in high-cost areas and consider ways to encourage such ACOs to develop. The reach of bundled payment could be extended by broadening the definition of spending included within the bundle (for example, increasing time period to 90 days) or by applying the policy to additional conditions. Applying bundled payment only to hospitals exceeding a minimum volume of bundles could reduce financial risk, but may reduce the impact on geographic variation as well.
We also acknowledge that other interventions could be employed (or are underway) that could also result in a reduction in geographic variation in Medicare spending. For example, adjustments to the Medicare Physician Fee Schedule that favor primary care relative to specialty care could reduce variation if high-cost areas tend to use more specialty care.11 If high-cost regions have more inefficient or low-value care, then policies that directly target inefficient care such as potentially avoidable hospitalizations may be another mechanism to reduce geographic variation. Whether high-cost regions have much higher prevalence of low-value care is unclear.12
Our estimates have some important limitations. First, the scenarios were designed to represent realistic versions of policies that could be implemented in the near future. We therefore relied upon scenarios that closely resembled current Medicare pilots or programs. However, different implementations of these policies could result in a different impact on spending. We explored some of these alternatives in our sensitivity analyses. Second, our results are limited by the available data. For example, in our ACO analyses we allocated beneficiaries to HRRs based on the location of primary care physicians, which only approximates true beneficiary locations. Also, a new set of ACOs was announced in January 2013, too late for inclusion in our analysis. It is possible that inclusion of these newest ACOs would alter our results.
Third, we focused on geographic variation in spending across HRRs. While HRRs are commonly used to examine geographic variation, we recognize that there is notable heterogeneity in spending within HRRs.6,13 Finally, we made only limited assumptions about provider behavior in response to these policies that we felt had a plausible basis in the literature. For example, in the case of bundled payment, we assumed that providers would react to the payment change by either reducing utilization within bundles of services or accepting reduced margins, but that they would not change the number of bundles provided or utilization of services outside of the bundle. However, we acknowledge that if actual behaviors differ systematically from our assumptions—and in particular, if providers in high-cost regions reacted differently from those in low-cost regions—the impact of these policies on geographic variation in spending could differ. As these policies begin to be implemented in pilot form, there may be evidence forthcoming on behavioral responses that would improve future policy design.
In summary, our results are useful to policy makers seeking solutions to the problem of unwarranted geographic variation in spending. Under a set of reasonable choices for implementing the policies we analyzed, we find that while they would reallocate a substantial portion of Medicare payments, P4P and ACOs are unlikely to reduce geographic variation in spending, and bundled payment would only modestly do so. The policies could be reengineered somewhat to have greater impact on this metric, but it is unclear if reduction in geographic variation in Medicare should be a goal, in and of itself, rather than more efficient delivery of care.
Author Affiliations: RAND Corporation (DA), Boston, MA; Department of Health Care Policy, Harvard Medical School (AM), Boston, MA; RAND Corporation, Arlington, VA (PH), Santa Monica, CA (PJH, CL, VS); Paul Merage School of Business at University of California, Irvine (AA), Irvine, CA.
Source of Funding: The project was funded by the Institute of Medicine (IOM; a part of the umbrella organization, the National Academy of Sciences). Funding for the study ultimately derived from CMS via the Affordable Care Act, which contracted with the IOM.
Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.
Authorship Information: Concept and design (DA, MA, AA, PJH, PH); acquisition of data (DA, MA, AA, PJH, PH); analysis and interpretation of data (DA, CL, MA, AA, PJH, PH, VS); drafting of the manuscript (DA, CL, PJH, VS); critical revision of the manuscript for important intellectual content (DA, CL, MA, PH); statistical analysis (DA, PJH); provision of patients or study materials (DA); obtaining funding (DA, MA, PH); administrative, technical, or logistic support (DA, VS); and supervision (DA).
Address correspondence to: David Auerbach, PhD, MS, RAND Corporation, 20 Park Plz, Ste 920, Boston, MA 02116. E-mail: email@example.com.
1. Zuckerman S, Waidmann T, Berenson R, Hadley J. Clarifying sources of geographic differences in Medicare spending. N Engl J Med. 2010;363(1):54-62.
2. Fisher ES, Wennberg DE, Stukel TA, Gottlieb DJ, Lucas FL, Pinder EL. The implications of regional variations in Medicare spending. part 1: the content, quality, and accessibility of care. Ann Intern Med. 2003;138(4):273-287.
3. Sirovich BE, Gottlieb DJ, Welch HG, Fisher ES. Variation in the tendency of primary care physicians to intervene. Arch Int Med. 2005;165(19):2252-2256.
4. Landrum MB, Meara ER, Chandra A, Guadagnoli E, Keating NL. Is spending more always wasteful? the appropriateness of care and outcomes among colorectal cancer patients. Health Aff (Millwood). 2008;27(1):159-168.
5. Congressional Budget Office. Budget Options Volume 1: Health Care. Washington, DC; 2008. http://www.cbo.gov/sites/default/files/12-18-healthoptions.pdf. Accessed May 2015.
6. Institute of Medicine. Geographic Variation in Health Care Spending and Promotion of High-Value Care—Interim Report. Washington, DC; National Academies Press; 2013.
7. HRR level demographic, cost, utilization, and quality data. CMS website. http://www.cms.gov/Research-Statistics-Data-and-Systems/Statistics-Trends-and-Reports/Medicare-Geographic-Variation/GV_PUF.html. Accessed February 13, 2013.
8. Steinbrook R. The role of the emergency department. N Engl J Med. 1996;334(10):657-658.
9. CMS. Proposed rule versus final rule for accountable care organizations (ACOs) in the Medicare Shared Savings Program. http://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/ACO/downloads/appendix-aco-table.pdf. Accessed May 2015.
10. Song Z, Safran DG, Landon BE, et al. The ‘Alternative Quality Contract,’ based on a global budget, lowered medical spending and improved quality [published online July 2012]. Health Aff (Millwood). 2012;31(8):1885-1894.
11. Baicker K, Chandra A. Medicare spending, the physician workforce, and beneficiaries’ quality of care. Health Aff (Millwood). 2004;23(3):w184-w197.
12. Schwartz AL, Landon BE, Elshaug AG, Chernew ME, McWilliams JM. Measuring low-value care in Medicare. JAMA Intern Med. 2014;174(7):1067-1076.
13. Congressional Budget Office. Geographic Variation in Health Care Spending. Washington, DC: Congressional Budget Office; 2008. https://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/89xx/doc8972/02-15-geoghealth.pdf. Accessed May 2015.