Health Affairs

This Article Abstract Submit a response to this article Services E-mail this article to a friend Alert me to new issues of the journal

Click here to view a Webcast of an interview
with Robert Berenson and Steven Lieberman
(available December 10, 2003).

Read related papers by
Steven Lieberman et al. • Robert A. Berenson •
David E. Wennberg and John E. Wennberg

M E D I C A R E : 
R E D U C I N G  S P E N D I N G 
W E B  E X C L U S I V E

10 December 2003

Reducing The Growth Of Medicare Spending:
Geographic Versus Patient-Based Strategies

The evidence suggests that a focus on high-spending patients may
yield the most savings to Medicare.

by Steven M. Lieberman, Julie Lee, Todd Anderson, and Dan L. Crippen


ABSTRACT:

This paper explores the potential of two alternative approaches for reducing the rate of growth in Medicare spending. One strategy would focus on reducing the expenditures of high-spending individuals. Given that a large share of Medicare spending is consumed by relatively few beneficiaries, this approach targets the small group responsible for most of the spending. The other strategy would focus on reducing expenditures in high-spending regions. Because either approach would have to overcome major hurdles before lowering Medicare spending, the likely payoff from the alternative strategies is far from clear. Viewed from a budgetary perspective, concentration in Medicare spending suggests the importance of focusing on high-spending patients.

The congressional budget office (CBO) projects that spending on Medicare, Medicaid, and Social Security as a share of the economy will almost double by 2040, growing from 8.4 percent of gross domestic product (GDP) in 2003 to 15.7 percent in 2040. By 2075 these “big three” entitlement programs are projected to consume 21.3 percent of GDP. In 2075 the three programs will claim a bigger share of the economy than today’s entire federal budget does.¹ Changes to current law, such as creating an outpatient prescription drug benefit or increasing Medicare payments to physicians, would raise spending above these projected levels.

Well-known demographic changes—the retirement of the massive baby-boom generation and increasing longevity—are the common factors driving increased costs in Medicare, Medicaid, and Social Security. Surprisingly, demographics account for only 30 percent of Medicare spending growth during the seventy-five-year projection period, when Medicare almost quadruples from 2.5 percent of GDP in 2003 to 9.2 percent in 2075. The other 70 percent of projected Medicare growth is associated with per capita growth in excess of GDP growth.²

Limiting Medicare’s future size by reducing the rate of growth in health care spending requires innovative strategies.³ One potential approach involves reducing Medicare expenditures in high-spending regions. Another involves reducing Medicare expenditures on high-spending individuals. The leading proponents of the approach focusing on regional variations have argued that lowering Medicare expenditures in high-spending regions to their level in low-spending regions would reduce total expenditures by as much as 29 percent, without reducing quality.⁴ Similarly, leading proponents of the approach focusing on high-spending individuals maintain that increasing compliance with evidence-based medicine will lower spending and improve quality. While other approaches to identifying high-cost beneficiaries exist (and more could emerge), one such approach, disease management, is widely available in the commercial (under age sixty-five) market.⁵

This paper explores the potential of these two approaches to lowering Medicare spending, or growth in spending if the former proves unrealistic. We examine the concentration of Medicare spending, analyzing whether high-spending beneficiaries remain persistently high spenders over a five-year period. After summarizing data on chronic conditions common to high-spending beneficiaries and describing disease management, we discuss some of the hurdles to converting a strategy that focuses on high-spending beneficiaries to a policy that would reduce Medicare spending.

We then analyze the concentration of expenditures within 306 Hospital Referral Regions (HRRs). Average per capita Medicare expenditures for various high- and low-spending groups within HRRs are highly correlated with the end-of-life expenditure index (EOL-EI), even where beneficiaries have no inpatient use and low physician spending. We also analyze the concentration of spending within HRRs and show that there is no regional variation in the distribution of spending across HRRs. Finally, we compare the shares of total Medicare expenditures associated with high-spending regions and beneficiaries. We conclude by assessing some of the challenges associated with adopting either strategy to achieve lower Medicare spending. We suggest some more immediate steps that might be undertaken.

High-Spending Medicare Beneficiaries

Spending distributions. Medicare spending is highly concentrated, with a small number of beneficiaries accounting for a large proportion of annual expenditures.⁶ During 1995–1999 the most costly 5 percent of beneficiaries in each year accounted for 47 percent of total Medicare spending, while the most costly 20 percent accounted for 84 percent of spending. By contrast, the least costly 40 percent of beneficiaries accounted for 1 percent of spending (Exhibit 1).

Given such an unequal distribution in spending, we would expect dramatic differences in the level of spending between the highest- and lowest-spending beneficiaries; the data in Exhibit 1 bear this out. Not surprisingly, spending per beneficiary correlates strongly with the use of inpatient hospital services. While almost all high-spending beneficiaries incurred inpatient hospital spending, no beneficiaries in either the lowest two or third quintiles had any inpatient spending (Exhibit 2).

Description of the high spenders. Who are these high-spending beneficiaries, and why do they have such high spending? As their inpatient use indicates, they account for tens of thousands of Medicare dollars because they are more likely to be sick. The prevalence of serious chronic conditions is higher among high-spending beneficiaries than low-spending beneficiaries, for example (Exhibit 2).⁷ Almost 90 percent of beneficiaries in the top 5 percent of annual spending had at least one of the seven chronic conditions analyzed in this paper, compared with less than 30 percent of those in the bottom 40 percent (not shown).

An important consideration is whether high-spending beneficiaries are expensive because they are in the last year of life. If their high-spending designation reflected the typically high spending at the end of life, we would expect a sizable turnover in the composition of this group from one year to the next. In other words, those near death would receive a lot of expensive care at the end of life and would be included in the high-spending group that year but would die soon afterward. The dynamics of this process suggest that the next year a different group of beneficiaries at the end of life would constitute the high-spending beneficiaries. If this were true, the opportunity to intervene successfully with high-spending beneficiaries and reduce their Medicare spending would be limited.

Our data show that mortality is indeed higher in the top spending groups. During 1995–1999 a beneficiary ranked in the most expensive 5 percent was five times more likely to die than the average beneficiary. However, only one-fifth of the people in that group died by the end of that year; these decedents accounted for 11 percent of total Medicare spending in that year. Survivors accounted for 36 percent.

If a large fraction of total Medicare expenditures by high-spending beneficiaries is not incurred at the end of life, another important consideration is the nature of medical conditions responsible for high spending. In other words, do people who are not in their last year of life have high spending because they have an acute and expensive episode in one year but subsequently recover, or do they have chronic and persistently expensive conditions year after year? In the first case we would expect a high turnover in the composition of the high-spending group, whereas in the second case we would not. More opportunities for successful intervention exist in this second group.

Five-year spending patterns. We have analyzed the persistence of Medicare spending over a five-year period (1995–1999). Among the beneficiaries enrolled in traditional fee-for-service (FFS) Medicare at the beginning of 1995, 27 percent accounted for 75 percent of five-year cumulative spending, whereas 73 percent accounted for 25 percent of cumulative spending. Of the 27 percent of beneficiaries accounting for the majority of cumulative spending, two-thirds (18 percent of all beneficiaries) were included in the top quartile of spenders in each year for at least two consecutive years. This group of persistently high-spending beneficiaries accounted for 57 percent of cumulative spending. The remaining third (9 percent of all beneficiaries) accounted for 18 percent of cumulative spending. Among persistently high-spending beneficiaries, 60 percent were alive at the end of five years, which suggests that the majority of spending associated with very costly beneficiaries is used by those who continue to live.

Disease management to reduce spending. To lower Medicare spending, we can focus on this small group of very costly patients, at least in theory. Translating our insights on the patterns in health care spending into a workable program, however, is difficult. Disease management is one potential strategy that focuses on beneficiaries with high need for medical care. Disease management attempts to address two limitations in current medical practice.⁸ First, patients may lack coordinated care because they receive care from many different physicians or providers and might be limited in their ability to coordinate care themselves. Second, as reported by the Institute of Medicine (IOM), there exists a large gap between evidence-based treatment guidelines (what medical research has shown to be the most effective protocols for treating specific diseases) and current practice.⁹ Disease management, by coordinating care across providers and encouraging adherence to evidence-based treatment guidelines, hopes to lower spending, improve the quality of care, and achieve better health outcomes.

Disease management is now offered as a health benefit by many large employers. Health plans either provide the service directly or subcontract with specialized disease management entities. Many population-based disease management companies have developed complex algorithms and use “data mining” to identify potentially high-spending beneficiaries, such as those with specific chronic conditions. After identifying beneficiaries who are at greatest risk of having costly medical events, disease management companies offer an array of services intended to stabilize or improve the health of a beneficiary and avoid adverse medical events. The interventions might focus primarily on the beneficiary or his or her physician, seeking to educate, improve self-care, or increase adherence to evidence-based medicine. Proponents of disease management frequently claim savings, as well as improved quality or outcomes.

Disease management must overcome major challenges before its ability to lower Medicare spending can be determined. The effectiveness of the predictive modeling algorithms developed by disease management companies to identify high-spending beneficiaries remains unclear. Interventions developed for workers and their families in employer-sponsored insurance could be inappropriate or infeasible for elderly or disabled Medicare beneficiaries, especially given the prevalence of dementia and multiple chronic conditions. Our ongoing survey of the peer-reviewed literature suggests, at best, weak empirical evidence for long-term savings resulting from existing disease management programs. Finally, absent rigorous experimental designs with randomized control groups, evaluating any claims of cost savings requires capturing the full cost of the intervention, including both administrative and benefit costs.

Focusing on high-spending beneficiaries is conceptually straightforward: To save money, go where the money is. Success in this endeavor, however, depends on two propositions. First, we need to identify beneficiaries who are going to account for high spending. And second, to realize savings, we need to intervene effectively before they become high spenders. Difficulty in identifying these beneficiaries and implementing cost-saving interventions remain the key challenges in the strategy targeting high-spending individuals.

Concentration Of Spending Within Regions

Research on regional variations in spending has advanced an alternative strategy: Organize analysis around geographic areas, rather than individuals.
Reasons for geographic variations. There are many reasons why health care spending varies geographically. First of all, populations vary across regions. For example, Florida is likely to have higher per capita spending than Utah because Florida has a higher concentration of older residents. Other demographic and health characteristics also cause geographical spending differences. Prices of medical services also vary because of differences in wages and other resource prices, resulting in differences in spending.
However, the literature on regional variations shows that there are remaining area-level differences not explained by differences in the prices of medical services or in the levels of illness. Rather, they are attributable to differences in the quantity of medical services consumed. For example, Elliot Fisher and colleagues reported that patients in higher-spending regions receive approximately 60 percent more care. They concluded that “regional differences in Medicare spending are largely explained by the more inpatient-based and specialist-oriented patterns of practice observed in high-spending regions.”¹⁰ Moreover, higher use in these regions does not seem to improve quality of or access to care. Therefore, if higher spending does not bring better outcomes, total Medicare spending can be lowered by reducing higher spending to the level prevailing in low-spending regions.

Analysis of geographic variations. In a simple analysis, we explore the relationship between geographic variations and patterns of Medicare spending. We have borrowed two key methodological elements from the literature. First, the 306 HRRs, as defined by the Dartmouth Atlas of Health Care, are the geographic unit of analysis.¹¹ Each HRR contains at least one hospital that performed major cardiovascular procedures and neurosurgery. Each ZIP code of Medicare beneficiaries’ residence is assigned to an HRR based on the hospital in which the majority of beneficiaries receive care. Therefore, HRRs represent regional health care markets for tertiary medical care.

Second, in identifying high-spending regions, we want an indicator of spending correlated with overall Medicare expenditures but controlled for health status. Fisher and colleagues used an end-of-life expenditure index (EOL-EI) as a measure of regional variation in Medicare spending attributable to differences in practice patterns, not in illness or price.¹² The EOL-EI is the average per capita spending on hospital and physician services, adjusted for age, sex, race, and price, provided to Medicare enrollees in their last six months of life in each of the 306 HRRs.

To explore the concentration of spending, we ranked Medicare FFS beneficiaries by expenditures within each HRR, from most to least spending, and divided them into various spending groups within each HRR (Exhibit 3). By construction, the EOL-EI should be highly correlated with Medicare spending. Fisher and colleagues reported a correlation coefficient of 0.81 between the index and the average per capita Medicare spending (adjusted for age, sex, race, and price). We estimated a similar overall level of correlation (r = 0.84) between the EOL-EI and unadjusted Medicare expenditures by HRR.¹³ Exhibit 3 shows that the correlations between the EOL-EI and mean spending at the HRR level remained quite high across spending groups. The correlation coefficient was highest for the fourth quintile of spenders at 0.82, whereas it was lowest for the bottom two quintiles at 0.50. However, the relationship between the EOL-EI and mean spending remained strong even in the absence of inpatient services (in the third quintile), or even without much spending (in the bottom two quintiles).

The concentration of expenditures within HRRs, as reported in Exhibit 3, is similar to that in Exhibit 1. Concentration at the HRR level is slightly different for two reasons. First, approximately 9 percent of beneficiaries were excluded from the analysis reported in Exhibit 3 because they could not be assigned to an HRR. Second, because some HRRs have lower spending overall, the top 5 percent spending group, for example, includes those who would not qualify for the top 5 percent in the national sample and vice versa. Exhibit 4 shows that membership in these two types of spending groups—the overall groups corresponding to Exhibit 1 (but excluding those who could not be assigned to an HRR) and the within-HRR groups corresponding to Exhibit 3—overlaps for the most part.

Exhibit 5 compares the concentration of Medicare spending among regions. The forty-one HRRs in the top quintile are those with the highest EOL-EI, where each HRR is weighted by the number of beneficiaries. In contrast, the ninety-six HRRs in the bottom quintile are those with the lowest EOL-EI. These HRR quintiles are analogous to those in the analysis by Fisher and colleagues. In spite of regional differences in the index and the level of spending, there is almost no variation in the concentration of spending across HRRs grouped by the index.

Finally, we compare the distribution of Medicare spending corresponding to the two approaches discussed above, high-spending regions and high-spending individuals. First, we sort all beneficiaries in two ways: by beneficiaries’ annual Medicare spending from highest to lowest (as in Exhibit 1) and by beneficiaries’ EOL-EI (which is defined by the HRR a beneficiary belongs to and is shared by everyone belonging to the same HRR) from highest to lowest. Second, we calculate the share of total spending accounted for by various groups of beneficiaries. Exhibit 6 contrasts the concentration of spending resulting from the two methods. The exhibit indicates, not surprisingly, a much higher payoff for targeting individuals than for targeting regions.
Targeting beneficiaries based on high-spending regions results in a relatively constant share of spending. In contrast, targeting beneficiaries based on individual spending results in a disproportionate share of spending. For example, while the 20 percent of beneficiaries living in the highest-intensity regions account for 24 percent of Medicare spending, the 20 percent of highest-spending individuals account for 84 percent of Medicare spending.

Concluding Comments

Fisher and colleagues concluded that Medicare beneficiaries in high-spending regions receive more care than those in low-spending regions but do not have better health outcomes or higher satisfaction with care.¹⁴ In other words, beneficiaries in high-spending regions do not reap health benefits from their additional use or spending. This key result suggests that if high-spending regions could be made to behave more like low-spending regions, there could be big savings without adverse health outcomes. In their recent Health Affairs paper, John Wennberg and colleagues proposed potential ways to accomplish this goal, including the creation of Comprehensive Centers for Medical Excellence (CCMEs), a public/private partnership that would attempt to encourage better medical practices (and fewer errors) and discourage the use of unnecessary services, particularly by specialists and in hospitals.¹⁵

Nonetheless, as in the case of high-spending beneficiaries, the challenges involved in changing spending patterns in high-spending regions appear daunting. The statistically important correlations established in the recent paper in the Annals of Internal Medicine do not address issues of causation, which may very well be impossible to ascertain.¹⁶ For instance, exactly what about these regions causes them to develop different practice patterns? Even if we could identify such factors, could they be minimized or avoided elsewhere? Without understanding the causal process, we could be limited in our ability to reduce Medicare spending by targeting regions. Moreover, HRRs, although intellectually appealing, remain abstractions. In contrast, successful public policy interventions typically require concrete solutions and actors.

Most of the plans for reducing spending cited in the literature could also apply to high-spending individuals (as opposed to high-spending regions). From a budgetary perspective, a strategy centered on high-spending individuals could hold the promise of greater “bang for the buck.” Simply, there’s more money concentrated in fewer individuals. However, such a focus presents an imposing initial challenge: identifying high-spending beneficiaries—or those who are at high risk of becoming high spenders—while there is still time to affect treatment, outcomes, and costs. A second challenge requires developing and implementing effective interventions to improve outcomes and quality of care, avoid preventable complications and deteriorating health status (in progressive but treatable conditions), and lower costs. A third challenge entails designing and implementing an appropriate payment system. (Some observers might conclude that these challenges reasonably equate to a search for the Holy Grail.)

High-spending beneficiaries may present a compelling target if something is to be done before more sophisticated tools become available. Demonstrations might focus on rigorously testing interventions targeted on such beneficiaries. Simple identification strategies could be adopted, such as flagging beneficiaries upon admission to a hospital. Differing strategies for “case management,” in which treatment is tailored to the individual and administered by a team of providers with someone responsible for care coordination (if not care decisions), could be tried and rigorously evaluated. Further analyses might inform the potential payoff from focusing on high-spending individuals within high-spending regions versus focusing exclusively on either high-spending patients or high-spending regions.

This analysis of high-spending regions and individuals suggests different, but not mutually exclusive, interventions such as the more systematic practice of evidence-based medicine and a further refinement of protocols for the elderly with multiple conditions. The elderly and taxpayers alike would benefit from fewer hospitalizations and better coordination of care. Ultimately, physicians will need to enlist (or be drafted) in the struggle to find the least costly treatments with the best outcomes for very sick elderly and disabled beneficiaries.

This paper is based on a paper presented at the Princeton Conference on Regional Variations in Health Spending: Implications for the Private Market and Medicare, in May 2003. The views expressed in this paper are those of the authors and should not be interpreted as those of the Congressional Budget Office. Helpful comments from Amber Barnato, James Baumgardner, W. Philip Ellis, Jeffrey Kelman, and Bruce Vavrichek are gratefully acknowledged.

NOTES

1. In 2002 the total federal budget was 19.5 percent of GDP. See Congressional Budget Office, “An Analysis of the President’s Budgetary Proposals for Fiscal Year 2004” (Washington: CBO, 2003)
2. For the past fifty years health care costs have grown much faster than the economy. On average, after the age and sex of the population are adjusted for, Medicare spending per beneficiary grew annually 2.9 percent faster than per capita GDP between 1970 and 2002. Because legislation limiting payments to providers repeatedly lowered Medicare reimbursements in the 1980s and 1990s, growth during this period would have exceeded 2.9 percent without these changes in the program. For decades, long-range projections had assumed that the forces causing health care costs to grow faster than the overall economy would disappear. Starting in 2000, however, long-range projections assume that per capita Medicare spending will grow one percentage point faster than per capita GDP.
3. For purposes of this paper, we distinguish initiatives to reform Medicare from efforts to identify mechanisms that lower costs. While identifying a viable approach to lowering spending may not be sufficient to constrain the future size of Medicare, it certainly is a necessary condition.
4. J.E. Wennberg, E.S. Fisher, and J.S. Skinner, “Geography and the Debate over Medicare Reform,” 13 February 2002, www.healthaffairs.org/WebExclusives/Wennberg_Web_Excl_021302.htm (5 November 2003); E.S. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 1: The Content, Quality, and Accessibility of Care,” Annals of Internal Medicine (18 February 2003): 273–287; and E.S. Fisher et al., “The Implications of Regional Variations in Medicare Spending, Part 2: Health Outcomes and Satisfaction with Care,” Annals of Internal Medicine (18 February 2003): 288–298.
5. Recent survey data from Hewitt Associates indicate that 76 percent of large employers offer some form of disease management. See L. Sprague, “Disease Management to Population-Based Health: Steps in the Right Direction?” National Health Policy Forum Issue Brief no. 791 (Washington: George Washington University, 16 May 2003). Disease management is generally intended both to improve quality and to lower spending. Because of the focus of this paper, we explore the potential to lower spending. Clearly, disease management retains the potential to improve quality or lower mortality.
6. For a detailed discussion of the analysis presented in this section, see CBO, “Concentration and Persistence of Expenditures among Medicare Beneficiaries” (Washington: CBO, forthcoming).
7. We analyzed seven chronic conditions: coronary artery disease, chronic obstructive pulmonary disease (COPD), congestive heart failure (CHF), diabetes, cognitive impairment, asthma, and end-stage renal disease (ESRD). Beneficiaries were defined as having a chronic condition if they had the given diagnosis for any two separate months during the sixty-month period 1995–1999.
8. Case management generally represents an alternative approach. Case management programs typically enroll patients with complex combinations of medical conditions, not just those with specific diseases. They frequently identify high-risk patients and design a care plan tailored for the individual patient’s needs, emphasizing the use of social support services and engaging family and caregivers.
9. Institute of Medicine, Crossing the Quality Chasm: A Health System for the Twenty-first Century (Washington: National Academies Press, 2001). See also E. McGlynn et al., “The Quality of Health Care Delivered to Adults in the United States,” New England Journal of Medicine (26 June 2003): 2635–2645.
10. Fisher et al., “The Implications of Regional Variations in Medicare Spending,” Parts 1 and 2.
11. J.E. Wennberg and M.M. Cooper, eds., Dartmouth Atlas of Health Care 1998 (Chicago: American Hospital Association Press, 1998).
12. Fisher et al., “The Implications of Regional Variations in Medicare Spending,” Parts 1 and 2. We thank Daniel Gottlieb for providing us with the index and for his assistance.
13. Unlike Fisher and colleagues, our Medicare spending numbers are adjusted only for inflation, converting them into 1999 dollars. Also, the correlation they reported is between the EOL-EI and per capita Medicare spending in 1996, whereas ours is between the EOL-EI and annual per capita Medicare spending from 1995 to 1999. See Fisher et al., “The Implications of Regional Variations in Medicare Spending,” Parts 1 and 2.
14. Ibid.
15. Wennberg et al., “Geography and the Debate over Medicare Reform.”
16. Fisher et al., “The Implications of Regional Variations in Medicare Spending,” Parts 1 and 2.

Steven Lieberman is assistant director, Health and Human Resources Division, at the Congressional Budget Office (CBO) in Washington, D.C. Julie Lee, julie.lee@cbo.gov, is an analyst in that division, as is Todd Anderson. Dan Crippen is former CBO director.

Read related papers by Steven Lieberman et al. • Robert A. Berenson • David E. Wennberg and John E. Wennberg.

©2003 Project HOPE–The People-to-People Health Foundation, Inc.