New York, NY, USA: Commonwealth Fund.
Division of Health Policy Research and Education, Harvard University, Boston, MA, USA
Almost all observers agree in principle that the Medicare program needs better risk adjusters than the adjusted average per capita cost (AAPCC) methodology it has historically used in calculating payments to at-risk health care plans. Indeed, Congress mandated changes in risk adjustment methods as part of the 1997 Balanced Budget Act.
The lack of adequate risk adjustment results in favorable risk selection of beneficiaries into health plans, which in turn causes three problems: potential barriers to access for beneficiaries seen as bad risks; incentives for plans to spend resources toward ends that are not socially productive; and increased costs to the Medicare program. Strong evidence suggests that, as a group, at-risk plans have benefited from favorable risk selection; that is, they have disproportionately attracted low-cost enrollees. Additionally, payment of a lump sum, independent of use, leaves a financial incentive to stint on services.
Adequate risk adjustment would cause payment to match expected individual spending closely. The adjusters in the current AAPCC formula explain only about 1 percent of the variance in actual spending, while an ideal adjuster might explain 20 times more variance. Fortunately diagnostic-based adjustment techniques such as hierarchical coexisting conditions(HCCs), diagnostic cost groups (DCGs), and ambulatory care groups (ACGs) have advanced to the point of making a substantial improvement over the AAPCC. In terms of ability to explain variance prospectively using Medicare data, the HCC model appears to do better, explaining about 8 to 9 percent (Ellis et al., 1996).1 By contrast, ADGs explain about 6 percent (Weiner et al., 1996), as do PIP-DCGs. Although these figures are still a fair distance from the 20-2 percent lower bound, each is clearly a substantial improvement on the AAPCC.
In addition to explained variance, both Ellis et al. and Weiner et al. (1996) present data on how these risk adjusters affect profitability, thereby suggesting both the improvement that they make and the magnitude of the remaining problem. For example, Ellis and colleagues calculate that plans would experience substantial average profits and losses in 1992 from enrolling those in the lowest and highest quintiles of 1991 expenditures, assuming that 1992 spending by the plan would be the same as spending in traditional Medicare (table 1).
Table 1. Effect of the AAPCC and HCC Adjusters on Average Estimated Profit and Loss
|Subgroup||Profit or Loss in 1992: AAPCC Adjusters||Profit or Loss in 1992: HCC Adjusters|
|Top 20% of Spenders
Bottom 20% of Spenders
Source: Ellis et al., 1996. Assumes plan was paid 100% of the AAPCC.
As table 1 shows, the HCC method yields a noteworthy reduction in incentives to cream and dump, cutting profits and losses by more than a factor of three relative to the AAPCC. Nonetheless, HCCs leave nontrivial levels of profits and, especially, losses.2 (For purposes of scaling these figures, mean spending in 1992 was $3,800.)
The Health Care Financing Administration (HCFA) has announced its intention to proceed with a variant of the DCG method beginning January 1, 2000. Even this method, however, leaves two potential problems: there is still an incentive, though substantially reduced, to select good risks; and there is still an incentive to stint on care. Moreover, the incentive to avoid bad risks is increased by HCFA’s inability in the short term to include diagnostic information from outpatient settings in the risk adjustment formula.
Recent research has also documented substantial underpayment for frail elderly beneficiaries, even after diagnosis-based risk adjustment. To deal with this problem, the Medicare program would need to adopt risk adjustment based on functional status—which in turn poses difficult implementation problems. Similarly, once Medicare moves to an annual rather than a monthly enrollment period in 2002, payment for new nursing home entrants who are also eligible for Medicaid appears much too low.
All these problems would be mitigated by moving policy away from exclusive reliance on a lump-sum, risk adjusted payment. Partial capitation—under which payment for an individual enrollee would combine capitation methods and some reflection of that person’s actual use of services—would address both selection and stinting incentives. A somewhat analogous but less desirable approach would be to require that capitation payments carry a reinsurance or a formal outlier policy, under which very expensive cases would be reimbursed an additional amount. Another proposal, potentially useful for terminally ill patients, would allow plans to name in advance a percentage of high-cost cases for which they would be partially or wholly reimbursed using traditional Medicare.
As the Medicare at-risk program grows, lack of satisfactory risk adjustment is causing serious problems. Better risk adjusters now exist. For valid reasons, HCFA is proposing to introduce them into the program slowly. Even when fully implemented, however, it is highly optimistic to think that current problems of selection will fall to negligible levels. Taking some account of the actual use of services will likely be necessary to keep risk selection negligible.
Please log in/register to access.