HIGH-COST MEDICARE BENEFICIARIES

Percentage with

Table 4.

of acute health shocks several years in a row than have an episode or two in a given year and then recover. Therefore, high expenditures in one year are likely to decrease over time as expenditures regress to the mean in subsequent years.

An examination of the spending patterns of Medicare beneficiaries reveals a second pattern: the quantitative importance of the subsequent death of high-cost beneficiaries. About 14 percent of beneficiaries with high Medicare expenses in a given year die during that year (see Figure 2). Within four years, that fraction accumulates to 40 percent.

In general, impending mortality greatly increases the probability of an individual's incurring high costs regardless of his or her prior spending. Studies show that about one-quarter of total Medicare payments are for the typically expensive and intensive treatment received in a patient's last year of life, which often postpones death for only a short time. Indeed, the high mortality rate among high-cost beneficiaries reported in Figure 2 confirms that a sizable fraction of spending by high-cost beneficiaries is for people near death. But not all deaths result in high spending, nor do all high-cost beneficiaries die soon thereafter. Different trajectories of functional decline at

8

8. See Christopher Hogan and others, "Medicare Beneficiaries' Costs of Care in the Last Year of Life," Health Affairs, vol. 20, no. 4 (July/August 2001), pp. 188-195; and James D. Lubitz and Gerald F. Riley, "Trends in Medicare Payments in the Last Year of Life," New England Journal of Medicine, vol. 328, no. 15 (April 15, 1993), pp. 1092-1096.

Although patients who die incur no further medical costs, they also offer little potential for cost savings if they had been targeted for an intervention strategy. Taking subsequent mortality into account, however, strengthens the empirical correlation of high spending over time. For high-cost beneficiaries in 1997 who did not die over the next four years, nearly one-half-instead of one-quarter-were high cost at the end of 2001. In Figure 2, the numbers of living high-cost and low-cost beneficiaries were roughly equal in each year from 1998 through 2001. Had there been no persistence in high medical expenses, only one-quarter of those beneficiaries would have been expected to be high cost during those years.

9. Moreover, because a patient's time of death is unpredictable (except perhaps in cases such as advanced cancer), it is only in hindsight that researchers can estimate which costs were associated with care at the end of the patient's life and which costs were associated with attempts to save the patient's life.

10. See June R. Lunney, Joanne Lynn, and Christopher Hogan, "Profiles of Older Medicare Decedents," Journal of the American Geriatrics Society, vol. 50, no. 6 (June 2002), pp. 1108-1112; and June R. Lunney and others. "Patterns of Functional Decline at the End of Life," Journal of the American Medical Association, vol. 289, no. 18 (May 14, 2003), pp. 2387-2392.

The Concentration of Spending
Over a Five-Year Period

Given the presence of high end-of-life expenditures and the regression to the mean following a high-cost year, one might expect Medicare expenditures over a longer period to be less concentrated than annual expenditures tend to be. For the entire 1997 cohort of Medicare beneficiaries, that is indeed the case (see Figure 3).11 Compared with the distribution of annual expenditures reported in Figure 1, that cohort's five-year inflation-adjusted cumulative expenditures are somewhat less concentrated: the top 5 percent of beneficiaries, when ranked by five-year cumulative spending, accounted for 27 percent of total five-year Medicare spending from 1997 to 2001, compared with 43 percent for annual spending. Furthermore, the top 25 percent of beneficiaries accounted for 68 percent of total five-year spending, compared with 85 percent for annual spending.

There is still a great deal of concentration of expenditures over five years, however, in part because a significant group of Medicare beneficiaries incurs high spending over an extended period. For beneficiaries whose cumula

11. That cohort is defined as beneficiaries who enrolled in the Medicare program as of January 1997 and who either remained enrolled for five years (until December 2001) or died. Beneficiaries who subsequently enrolled in a Medicare managed care plan were excluded. There were about 1.4 million beneficiaries in CBO's random sample of that cohort.

tive 1997-2001 spending put them in the top 25 percent of all beneficiaries for that 60-month period, Figure 4 displays the distribution of the number of months in which they were in the top 25 percent of beneficiaries in terms of spending in that month. The median number of months is 22. In other words, about half of cumulatively high-cost beneficiaries had high monthly costs during 22 months or more of the 60-month period. That result could indicate that there may be time and opportunity to intervene to affect the use of Medicare services for a significant number of high-cost beneficiaries because they remain persistently high cost over an extended period.

Prospectively Identifying Future
High-Cost Beneficiaries

Whether a strategy of focusing on high-cost beneficiaries could lead to significant reductions in overall Medicare spending would depend on two factors: the ability to identify individuals who will have high costs in the future, and the ability to mitigate those high costs. The existence of Medicare beneficiaries whose high spending persists over an extended period presents potential opportunities for intervention strategies. However, prospectively identifying such individuals could be difficult.

A basic problem is that although researchers can identify characteristics or conditions that are prevalent among high-cost beneficiaries, many low-cost beneficiaries may also share the same characteristics. For instance, a number of chronic conditions were found to be highly prevalent among high-cost beneficiaries, and considerably less prevalent among low-cost beneficiaries. However, because the number of low-cost beneficiaries in this illustration is three times as large as the number of high-cost beneficiaries, the numbers of high-cost and low-cost beneficiaries with those conditions are much more similar (see Table 6). So while diabetes is nearly twice as prevalent among high-cost beneficiaries as it is among low-cost ones, the actual number of low-cost beneficiaries with diabetes greatly exceeds the number of high-cost beneficiaries with that condition. Therefore, any intervention strategy that focuses simply on beneficiaries with diabetes will include a large number of people who will not incur significant medical expenditures (at least soon thereafter). Even the most successful strategies for identifying highcost individuals will probably include some who will not turn out to be expensive.

Illustrative Strategies for Identifying
High-Cost Beneficiaries

This section briefly considers three simple strategies for prospectively identifying high-cost beneficiaries on the basis of the characteristics of those beneficiaries discussed above. The first strategy is to select beneficiaries who were high cost in the previous year. The spending history shown in Figure 2 demonstrates that expenditures in the previous year are correlated with expenditures in the following year. The second strategy is to select beneficiaries who were hospitalized in the previous year based on the correlation between hospital admission and continued high spending. Both the first and second strategies would delay providing interventions until the disease had progressed and some substantial costs had already been incurred. The third strategy is to select beneficiaries who were diagnosed with two or more of seven chronic conditions: asthma, chronic obstructive pulmonary disease, chronic renal failure, congestive heart failure, coronary artery disease, diabetes, and senility. The resulting samples from the three strategies were compared with a sam

2001

ple of randomly selected Medicare beneficiaries. (The selection criteria for all of those strategies also required that the beneficiaries still be alive in January 1998.)12

How the strategies fared is displayed in Table 7 on page 12. The share of the Medicare population included in each of the three selected groups ranged from 17 percent to 22 percent. To make the subsequent shares of spending by the groups more comparable, CBO adjusted the size of each group (by random assignment) to match the size of the smallest original group, or 17 percent of the overall Medicare FFS population. The group with a hospital admission had the largest average spending in 1997 (at $24,900), followed by the high-cost group (at $23,000) and the group with multiple chronic conditions (at $16,900). The reference had $6,200 in average group spending. The previously hospitalized group also had the

12. The selection criteria further required that beneficiaries be enrolled in Medicare's fee-for-service sector from 1997 to 2001, enabling analysts to track their spending over the entire five-year period

years,

In terms of subsequent spending, beneficiaries in each of the three selection groups used more than $46,000 in Medicare-covered services over the next four years, compared with $27,000 for beneficiaries in the reference group. The previously high-cost group accounted for 29 percent of total Medicare spending over those four compared with 28 percent for beneficiaries with a prior hospitalization and 28 percent for those diagnosed with multiple chronic conditions. Those levels of aggregate spending occurred despite the fact that nearly half of the members in each group died before the end of the fouryear period.

Approaches to Managing Care for High-Cost
Beneficiaries

The three selection strategies considered above are highly stylized and conceptual illustrations, and they do not address the challenges of designing and implementing workable programs to reduce costs. However, they broadly reflect some of the approaches currently being developed and tested by various organizations. For example, the selection strategy focusing on people diagnosed with chronic conditions is similar to the approach taken by some private disease management programs.

Over the past decade, many private health plans and organizations have begun to offer disease management as a model of care for chronically ill patients, in an attempt both to improve the quality of care that enrollees receive and to slow the growth of their health care costs. Disease management programs vary widely in the specific techniques and tools they use, but they share some common components that are designed to address several perceived shortcomings of current medical practice. One component is to educate patients about their disease and how they can better manage it. The goal is to encourage patients to use medication properly, to understand and monitor their symptoms more effectively, and possibly to change their behavior. A second component is to actively monitor patients' clinical symptoms and treatment plans, following evidence-based guidelines. A third component is to coordinate care among providers, including physicians, hospitals, laboratories, and pharmacies. A disease management program can provide feedback and support to physicians about patients' status between office visits as well as up-to-date information on best practices as they apply to the specific patient. Although disease management is a term sometimes used as a catchall that addresses any and all limitations of fee-for-service care, it does not encompass general care coordination or general preventive services, such as flu shots. 13

Various demonstration projects and initiatives by the Centers for Medicare and Medicaid Services also focus on strategies to improve care for beneficiaries who account for large amounts of Medicare spending. For example, the Chronic Care Improvement Program was created by the Medicare Prescription Drug, Improvement, and Modernization Act of 2003 to improve clinical care for

13. See Congressional Budget Office, An Analysis of the Literature on Disease Management Programs (October 2004).