Geographic Variation in Health Care Spending
Per capita health care spending varies widely across the United States. In 2004, as an example, per capita spending ranged from roughly $4,000 in Utah to $6,700 in Massachusetts (see Figure 1). The variation is even greater among smaller geographic units and among individual medical providers (see Figure 2). Among large hospitals in California from 1999 to 2003, Medicare spending per patient in the last two years of life ranged more than fourfold, from less than $20,000 to almost $90,000 (Wennberg and others 2005). Researchers affiliated with the Dartmouth Atlas of Health Care estimate that among groups of Medicare beneficiaries who are otherwise similar, individuals who live in high-spending areas receive approximately 60 percent more in services than do those who live in low-spending areas (Fisher and others 2003a).1 The amount of spending involved is quite large—one report indicated that Medicare spending would fall by 29 percent if spending in medium- and high-spending regions were the same as that in low-spending regions (Wennberg and others 2002).
Health Care Spending per Capita, 2004
Source: Congressional Budget Office based on data from the Centers for Medicare and Medicaid Services.
Medicare Spending per Beneficiary, by Hospital Referral Region, 2005
Source: Congressional Budget Office based on data from the Centers for Medicare and Medicaid Services.
Note: The data are for Medicare spending per beneficiary in the fee-for-service program on the basis of beneficiaries’ residences and adjusted for age, sex, and race. The geographic unit is the hospital referral region, as defined by the Dartmouth Atlas of Health Care. Areas labeled "Not Populated" include places without residents, such as national parks, forests, lakes, and islands.
Large differences across the country in spending for the care of similar patients could indicate a health care system that is not as efficient as it could be, particularly if that higher spending does not produce commensurately better care or improved health outcomes. Given the importance of health care spending in the nation’s long-term fiscal outlook, identifying and encouraging patterns of care that are more efficient is clearly important. Because Medicare is paid for in part by federal taxes, high spending in one area is, in effect, funded to some extent by taxpayers in other areas, raising additional concerns.
This paper by the Congressional Budget Office (CBO) examines the amount of and trends in geographic variation in health care spending, and the root causes of that variation. It also examines the relationship between spending and quality of care, and it discusses what those findings imply about how health care is produced in the United States and how it could be made more efficient. The paper focuses primarily on spending in the Medicare program because there are more data available about the cost of providing health care to Medicare beneficiaries than there are for other populations. Also, as the largest federal health care program, Medicare is highly relevant to and directly influenced by federal policy.
The results of current research indicate the following:
|
■ |
Two factors—the prices of health care services and severity of illness—are important in explaining geographic variation in health care spending. Different studies support different conclusions about the relative importance of those two factors, but most concur that together they account for less than half (and possibly much less than half) of the geographic variation in spending. |
|
■ |
Income and the preferences of individuals for specific types of care appear to explain little of the variation in spending. |
■ | A substantial portion of the variation remains unexplained after those factors are considered. Unmeasured differences in demand for care could be important, but some of the variation in medical practice probably is attributable to regional differences in the supply of medical resources (specialist physicians or health care facilities, for example) and the propensity to take advantage of the financial incentives provided by Medicare or other payers in developing and using those resources. |
■ | Some regions appear more prone to adopt low-cost, highly effective patterns of care whereas others are more prone to adopt high-cost patterns of care and to deliver treatments that provide little benefit or are even harmful. |
■ | Geographic variation in total health care spending per capita has shown an upward trend in recent years; over the past three decades, in contrast, variation in Medicare spending has narrowed sharply. That reduction could be the result of changes in Medicare’s reimbursement policies. |
The evidence suggests that efficiency gains in the health care system are possible: Spending in high-spending regions could be reduced without producing worse outcomes, on average, or reductions in the quality of care. But policies that reduce spending in high-spending areas would not necessarily lead to increased efficiency—and could result in worse health outcomes—unless the reductions targeted ineffective or harmful treatments. Reforms that are designed to increase efficiency in the health care sector generally could, as a side effect, reduce geographic variation. Some of those proposals are discussed briefly toward the end of this paper. (CBO is undertaking an expanded effort to examine options for modifying the health care system in the United States. The discussion of those options will include their potential impact on geographic variation.)
Measuring Geographic Variation
Researchers generally use relatively large areas, such as states, as the unit of analysis to examine geographic variation in health care spending. Doing so allows them to identify factors that vary systematically across areas and that affect regional patterns of care. Researchers also have analyzed variation among smaller geographic units, such as hospital referral regions (HRRs), counties, and metropolitan statistical areas (MSAs), and among individual hospitals or medical centers.2
Much of the research on geographic variation in health care spending has focused on Medicare spending, at least in part because of the availability of billing records for Medicare’s fee-for-service patients. The data provide details about where beneficiaries live (state, county, and zip code) and on the use of and spending for services covered by Medicare. There is no comparable data source for the privately insured population that both covers a large segment of the population and includes detailed information on geography and spending. Variation among states in Medicare spending per beneficiary in 2004 was similar to the variation in total personal health care spending: It ranged from $5,600 per beneficiary in South Dakota to $8,700 in Louisiana.
The coefficient of variation (COV) is a commonly used statistic for quantifying the degree of variation in a variable. The COV is the standard deviation divided by the mean.3 (As a simple illustration, the mean height of men in the United States is about 69 inches, and the standard deviation is about 3 inches—a fairly narrow distribution—so the COV is 3 divided by 69, or 0.04.) Geographic variation in Medicare spending per beneficiary is substantially larger: In 2005, the COV in state-level Medicare spending per beneficiary was 0.11.
Another way to measure geographic variation is to calculate the amount by which total spending would be reduced if spending in high-spending areas were reduced to that of low-spending areas. The Dartmouth Atlas researchers undertook such an analysis using Medicare data (Wennberg and others 2002). They calculated that Medicare spending would fall by 29 percent if spending in medium- and high-spending regions were the same as in their benchmark regions, defined as those with spending in the lowest decile.
Geographic Variation in Context
As a first step in explaining the significance of geographic variation in health care spending, it is useful to analyze how that variation has changed since the 1970s, how variation in Medicare spending compares with variation in total health care spending, and how variation in health care spending compares with variation in spending on other goods and services. It also is useful to examine how the United States compares with other countries and how Medicare compares with the Department of Veterans Affairs (VA) health care system.
The following conclusions can be drawn:
■ | Geographic variation in total health care spending per capita has been growing in recent years. Geographic variation in Medicare spending, in contrast, has dropped sharply over the past three decades and recently has been slightly lower than the variation in total health care spending. |
■ | There also is geographic variation in per capita spending on non-health care items, such as housing, food, and transportation (see Box 1). The degree of variation in Medicare spending per beneficiary is relatively high compared with those other spending categories, but it is not completely out of line. |
Geographic Variation in Medicare Spending Compared with Spending on Food, Housing, and Transportation
|
The Congressional Budget Office (CBO) used published results from the Bureau of Labor Statistics’ 2004–2005 Consumer Expenditure Survey to compare geographic variation in Medicare spending with variation in spending for food, housing, and transportation.1 The survey’s sampling strategy prevents its use for generating estimates at the state level, but it can be used to measure per capita spending in 24 large metropolitan areas.2
CBO measured geographic variation for spending on food, housing, and transportation because, like health care, those categories represent substantial shares of the economy. To make the comparison with Medicare spending valid, Medicare spending per beneficiary was measured separately for each metropolitan area, and a coefficient of variation (COV) for Medicare spending was calculated from data only from those areas. Geographic variation was fairly similar among the spending categories examined. Per capita spending on food ranged from $1,880 in Baltimore to $3,010 in Boston, with a metropolitan COV of 0.120 (see the table below). There was greater variation in per capita spending on housing: Pittsburgh was lowest, at $5,231, and San Francisco was highest, at $8,802 (COV, 0.143). Per capita spending for transportation ranged from $2,416 in Miami to $5,038 in Anchorage (COV, 0.143). The analogous COV in Medicare spending per beneficiary was 0.148, which is slightly higher than the COVs for housing and transportation. Analysis of Geographic Variation in Spending on Medicare and Other Goods and Services, 2004 to 2005
Sources: Bureau of Labor Statistics, 2004–2005 Consumer Expenditure Survey, Current Metropolitan Statistical Area Tables; and Congressional Budget Office calculations based on sample data from the Centers for Medicare and Medicaid Services’ Continuous Medicare History Sample data. Notes: R2 = estimated share of metropolitan-level variation in spending that is explained by income. Spending per person is measured for 24 selected metropolitan statistical areas (MSAs). Medicare spending is measured per beneficiary in 2005, for fee-for-service beneficiaries only. The unadjusted coefficient of variation (COV) is calculated using unadjusted spending per person and is weighted by population. The estimated coefficient on income is from a weighted ordinary least squares (OLS) regression of MSA-level spending per person on MSA-level pretax income per capita; it represents the estimated change in spending per person (in dollars) for goods in the indicated category in response to a one-dollar increase in pretax income per capita. The MSA-level income elasticity is calculated at the overall mean spending and income; it equals the estimated coefficient on income divided by the ratio of mean spending per person to mean income. The income-adjusted COV is calculated using adjusted spending (overall mean plus the residual from the OLS model) and is weighted by population. a. p < 0.01. Differences in income may account for some of this variation. CBO used a regression analysis to examine the extent to which geographic variation in income accounts for variation in spending per person on Medicare and on other goods and services (see the table). A COV was calculated to measure the degree of geographic variation, after controlling for income. Spending per capita on food and on housing are strongly and positively correlated with income. For Medicare spending (and for transportation), in contrast, there was little or no relationship between per capita income and per capita spending, so the income-adjusted and unadjusted COVs are similar for those spending categories. (Medicare spending could be less tied to income because the program’s spending is financed largely by federal taxes, rather than purchased by the individual.) After controlling for income, geographic variation in Medicare spending appears even greater relative to variations in spending for food and housing; it remains similar to geographic variation in spending for transportation. Although the variation in Medicare spending is not completely out of line with that observed in other sectors of the economy, it does warrant closer examination. It is reasonable to assume that the value of the goods and services consumed in most other sectors is readily apparent. For example, if people in a given area spend a relatively large amount on housing, it is reasonable to assume that they are aware of the attendant benefits (for example, the area might provide better job opportunities, have a mild climate, offer access to cultural amenities, or have good public schools) and they choose to spend more as a result. That assumption does not necessarily hold for health care. Health care providers usually have a strong influence on the choice of treatment, and the quality or value of the benefits received from higher spending is much more difficult for patients to discern. 1. Bureau of Labor Statistics, Current Metropolitan Statistical Areas Tables, 2004–2005, www.bls.gov/cex/home.htm. Data limitations prevent an analogous analysis of total per capita health care spending by metropolitan area. 2. Anchorage, Alaska; Atlanta, Georgia; Baltimore, Maryland; Boston, Massachusetts; Chicago, Illinois; Cleveland, Ohio; Dallas-Fort Worth, Texas; Denver, Colorado; Detroit, Michigan; Honolulu, Hawaii; Houston, Texas; Los Angeles, California; Miami, Florida; Minneapolis-St. Paul, Minnesota; New York, New York; Philadelphia, Pennsylvania; Phoenix, Arizona; Pittsburgh, Pennsylvania; Portland, Oregon; San Diego and San Francisco, California; Seattle, Washington; St. Louis, Missouri; and Washington, D.C. |
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
■ | In recent years, geographic variation in health care spending has been much higher in the United States than in Canada, and somewhat higher than in the United Kingdom. Financing of health care in those countries is more centralized than it is in the United States. |
■ | In recent years, geographic variation in spending in the VA health care system has been similar to that in Medicare, despite the fact that the VA system uses an explicit allocation formula to distribute funds to regions. |
Trends in Geographic Variation
To examine changes in geographic variation over time, CBO analyzed two state-level measures of health care spending: total health care spending per capita and Medicare spending per Medicare beneficiary. Total spending per capita was calculated from data for 1991 through 2004 published by the Centers for Medicare and Medicaid Services, or CMS (2007). Medicare spending per beneficiary was calculated using the Continuous Medicare History Sample, a dataset created by CMS that includes spending data for a sample of 5 percent of Medicare’s beneficiaries from 1974 through 2005.
The COV for total health care spending per capita rose gradually during the 1990s and early 2000s, from a low of 0.112 in 1991 to a high of 0.123 in 2004 (see Figure 3). In contrast, with the exception of the early- to mid-1990s, the COV for Medicare spending showed a dramatic downward trend. From a peak of 0.200 in 1976 it fell sharply, to 0.125 by 1991, and then rebounded in the early 1990s before resuming a sharp decline, ending at 0.110 in 2005.
Variation in State-Level Medicare and Overall Health Care Spending per Capita
Source: Congressional Budget Office based on data from the Centers for Medicare and Medicaid Services.
To examine the reasons for the trends in Medicare, CBO decomposed the geographic variation in spending into four components representing major categories: spending on care at "short-stay" hospitals (as opposed to longer-term care in another kind of facility), spending for services provided by physicians and laboratories, spending on post–acute care (provided in skilled nursing or long-term care facilities, as home health care, or as hospice care), and outpatient spending (for "in-and-out surgery," for example; see Figure 4). Each category’s contribution to the overall COV for Medicare depends both on its share of total spending and on the degree of geographic variation within the category. That decomposition reveals several trends:
■ | In the 1970s, spending on inpatient hospital care accounted for a substantial amount of geographic variation in Medicare spending. But the amount of variation attributable to hospital spending declined throughout the 1970s and 1980s, leveled off in the 1990s, then declined again in the early 2000s. |
■ | The large spike in geographic variation in Medicare spending that occurred in the early 1990s was attributable to increasing variation in post–acute care spending. |
■ | In the early 2000s, geographic variation in Medicare spending declined, and each major spending component contributed to that decline. |
Contributions of Major Service Categories to State-Level Variation in Medicare Spending per Beneficiary
Source: Congressional Budget Office based on data from the Centers for Medicare and Medicaid Services.
Why did geographic variation in Medicare spending change so dramatically? One hypothesis is that revisions in Medicare’s payment policies contributed substantially.4 In the 1980s, Medicare gradually began to phase out cost- and charge-based reimbursement and to implement a set of formula-based prospective payment systems. In a cost-based system, Medicare payments to providers are determined by the costs incurred, including labor and capital costs. In a charge-based reimbursement system, payments are determined by the charges or fees submitted. Under cost- and charge-based systems, providers had considerable influence over payment rates, which in turn allowed for substantial geographic variation in payments.
CBO tested the hypothesis concerning reimbursement methods by examining average payment rates for hospital stays and for physicians’ and laboratory services.5 Beginning in 1983, hospitals were switched by Medicare to a formula-based payment system that uses the discharge as the unit for payment. Since 1983, there has been much less state-to-state dispersion in average Medicare payment rates for hospitals (see Figure 5). However, the dispersion in Medicare’s hospital payment rates already was declining sharply before then, so other forces (such as the anticipation of the policy change or a stricter review of hospital charges leading up to the policy enactment) could have been at work. In 1992, physicians’ reimbursement changed from the cost-based system to Medicare’s "resource-based relative value scale" (see Figure 6). The dispersion of average payment rates for physicians’ and laboratory services was increasing in the years leading up to 1992, but it declined sharply after the new system’s introduction. (Note that the dispersion of overall spending for physicians’ and laboratory services, shown in Figure 4, did not decline as much.)
Dispersion in State-Level Mean Medicare Payments per Hospital Stay
Source: Congressional Budget Office based on data from the Centers for Medicare and Medicaid Services.
Dispersion in State-Level Mean Medicare Payments per Physician or Laboratory Service
Source: Congressional Budget Office based on data from the Centers for Medicare and Medicaid Services.
Geographic Variation in Canada and the United Kingdom
Geographic variation in health care spending in the United States could be related to idiosyncrasies in the nation’s system of health care financing and delivery. The United States differs from most other high-income countries in having a relatively decentralized system with a relatively large role for private insurers. The share of the population with health insurance varies from region to region, as do the type and the comprehensiveness of that insurance coverage. It has been hypothesized that, relative to other countries, the United States might therefore exhibit a high degree of geographic variation in health care use and spending.
To test that hypothesis, CBO used publicly available data to compare variation in health care spending per capita among states in the United States, among provinces in Canada, and among regions in the United Kingdom (see Figure 7).6 Those countries were chosen because they are similar to the United States in many respects (they have comparable per capita income and systems of governance, for example) and because regional data on health care spending are available for all three countries.
Geographic Variation in Health Care Spending per Capita in Selected Countries
Source: Congressional Budget Office based on data from the Centers for Medicare and Medicaid Services, HM Treasury (for United Kingdom data), and the Canadian Institute for Health Information.
Geographic variation in health care spending has consistently been much higher in the United States than in Canada and somewhat higher than in the United Kingdom in the years for which data are available. From 1991 through 2004, the COV in state-level health care spending per capita in the United States varied between 0.112 and 0.123. Over the same period, the COV in per capita spending by province in Canada (for public and private spending) varied between 0.059 and 0.088, with an increase in recent years. In the United Kingdom, the COV by region has varied in recent years between 0.091 and 0.107.
The greater variation within the United States is not surprising given that the health care systems in Canada and the United Kingdom are explicitly designed to distribute funds from the central governments to the province or region according to "needs-based" formulas. In Canada, health care is financed jointly by the federal, provincial, and territorial governments and other sources. Funds are explicitly allocated from the federal government, partly on a uniform per capita basis through the Canada Health Transfer and partly through the Equalization Program, which is designed to counteract disparities among provinces in the capacity to provide comparable health services.
In the early decades of its National Health Service, the United Kingdom allocated funds to different regions on the basis of historical spending in each region, updated for inflation. Beginning in the 1970s, researchers began to link that approach to financing with unequal regional distributions of funds (Culyer and others 1981, European Observatory on Health Systems and Policies 1999). Those investigations culminated in a plan developed in 1976 by the Resource Allocation Working Group, which laid out a formula for regional health care financing that was based on health care needs and local differences in practice costs. Over the next decade the formula was adjusted to reduce regional disparities.
Variation in Health Care Spending by the Department of Veterans Affairs
The VA health care system is an example of centrally budgeted health care in the United States. It is a large, integrated system that typifies managed care, particularly of the kind practiced by health maintenance organizations, or HMOs. Services are provided primarily through a limited network of staff physicians and hospitals owned and operated directly by the department.
VA’s financing structure differs from Medicare’s in important ways, allowing for an interesting comparison both with Medicare and with the rest of the U.S. health care system. First, VA operates under a global budget that is determined by Congressional appropriations. Medicare benefits, in contrast, are paid through an entitlement program that does not require a specific appropriation each year. Second, VA funds are allocated to 21 geographically defined units (called Veterans Integrated Services Networks, or VISNs) on the basis of the number of veterans served and their health care needs. The flow of Medicare funds, in contrast, is determined on the basis of the volume of health care services provided in each region. (Both systems adjust for local input costs.) Because of those differences in financing systems, a reasonable hypothesis is that VA health care would exhibit less variation in spending per capita than Medicare.
CBO used VA data for fiscal years 2001 and 2007 (Department of Veterans Affairs 2001, 2007) and data from the General Accounting Office (2002) to measure geographic variation in VA spending. Variation was measured for two separate years to identify changes over time. The COV by VISN-level allocations per patient in the VA system was 0.085 in 2001 and 0.104 in 2007.7 To allow a valid comparison with Medicare, CBO grouped Medicare beneficiaries, based on residence, into geographic areas matching those of the VISNs and measured VISN-level Medicare spending per beneficiary. The result was a Medicare COV of 0.141 for 2001 and a COV of 0.116 for 2005 (the most recent year for which data are available).
The calculations show that, in 2001, Medicare exhibited substantially more geographic variation in health care spending per person than the VA system did. Since then, however, the gap appears to have been largely eliminated, as variation in spending in the Medicare program fell while that in the VA program increased. It appears, therefore, that the centrally budgeted VA system does not display much less geographic variation in spending than is exhibited in the unbudgeted Medicare program.
Why did geographic variation in VA spending increase between 2001 and 2007? One possibility is the introduction of a more complex methodology for adjusting allocations based on case mix. (Case mix refers to health needs of the population served.) The 2001 VA allocations were based on a case mix system that had only three patient groups and three payment rates. By 2007, the system differentiated among 20 patient groups, each with a separate payment rate. Between 2001 and 2007, the VA methodology also was refined to include an allocation adjustment for treatment of unusually high-cost patients ("outliers"). Both refinements have been described as significant improvements, and both might have contributed to the increase in geographic variation in VA spending.
In addition to exhibiting geographic variation in spending, the VA system shows substantial variation in patterns of clinical practice despite the fact that VA’s management tracks providers’ compliance with national guidelines for the treatment of many medical conditions. Several studies have documented wide geographic differences within the system in patterns of treatment for several medical conditions: acute myocardial infarction (heart attack), upper respiratory infection, depression, and prostate cancer (Aspinall and others 2005, Fortney and others 1996, Subramanian and others 2002, Wilt and others 1999). The implication is that local norms can influence practice patterns, even in a relatively centralized system that places a strong institutional emphasis on adherence to clinical guidelines for care.
The evolution of the regional financing system for VA health care has strong parallels with the development of the regional health financing formula in the United Kingdom. In each case, funds initially were allocated to regions, primarily on the basis of historical costs that had been adjusted for inflation. Each system’s administrators recognized later that the result was an inequitable distribution of funds; that conclusion in turn led to the implementation of regional allocation formulas based on population, health status, and local practice costs. Iglehart (1996) has reviewed and described changes in the VA financing system.
Explaining Geographic Variation in Health Care Spending
Several researchers have examined explanations for geographic variation in per capita health care spending (see Table 1); most of their studies focus on the Medicare fee-for-service program, largely because better data are available for Medicare than for the private sector. The typical approach has been to measure geographic variation in unadjusted spending per capita and then to measure variation in spending per capita after adjusting for various factors that are believed to affect spending. The contribution of a given factor to geographic variation is measured by the degree to which variation is reduced after adjusting for that factor.
Those factors can be divided into four broad categories, each discussed in detail in the following sections:
■ |
■ |
■ | Regional preferences about the use of health care services (and the determinants of those preferences, such as income), and |
■ |
Research on Geographic Variation in Health Care Spending
|
Medicare physician spending per beneficiary, 1989 |
Inpatient hospital admission rate, physicians per capita, proportion of physicians engaged in primary care |
|||
|
Health risk behaviors, mortality rates, race, income, education, HMO market share, supply of medical providers |
||||
|
Share of beneficiary population under age 65, share over age 85 |
||||
|
Center for the Evaluative Clinical Sciences (1999) |
Age, sex, race, illness, prices, HMO market share, supply of medical providers |
|||
|
Utilization of Medicare-covered services per beneficiary, 1989–1991 |
Education, income, cigarette sales, obesity, air pollution, race, region, urbanization |
|||
|
Medicare Payment Advisory Commission (2003) |
Prices, health status, Medicare Part A and Part B participation rates, special hospital payments |
|||
|
Health status, local practice costs, special payments to hospitals, managed care enrollment, intensity of care |
||||
|
Population characteristics, health care needs, prices, intensity of care |
||||