oversight

Medicare HMOs: HCFA Can Promptly Eliminate Hundreds of Millions in Excess Payments

Published by the Government Accountability Office on 1997-04-25.

Below is a raw (and likely hideous) rendition of the original report. (PDF)

                 United States General Accounting Office

GAO              Report to the Chairman, Subcommittee
                 on Health, Committee on Ways and
                 Means, House of Representatives


April 1997
                 MEDICARE HMOs
                 HCFA Can Promptly
                 Eliminate Hundreds of
                 Millions in Excess
                 Payments




GAO/HEHS-97-16
      United States
GAO   General Accounting Office
      Washington, D.C. 20548

      Health, Education, and
      Human Services Division

      B-265996

      April 25, 1997

      The Honorable William M. Thomas
      Chairman, Subcommittee on Health
      Committee on Ways and Means
      House of Representatives

      Dear Mr. Chairman:

      Medicare costs have been growing rapidly during the 1990s, and the
      Congressional Budget Office estimates that costs will increase an average
      of 8.4 percent a year during fiscal years 1998 through 2002. As the
      Congress seeks ways to slow this growth rate, several proposals have been
      made that would encourage beneficiaries to join managed care plans.
      These plans typically have a financial incentive to hold down costs; in fact,
      Medicare’s method for paying risk contract health maintenance
      organizations (HMO)—Medicare’s principal managed care option1—was
      designed to save the program 5 percent of the costs for beneficiaries who
      enroll in HMOs. However, a decade of research has found that enrolled
      beneficiaries would have cost the program less if they had stayed in the
      fee-for-service (FFS) sector. The research shows that Medicare’s
      rate-setting method produces excess payments to HMOs because it
      overstates the costs of HMO enrollees. Recently, the Physician Payment
      Review Commission estimated that annual excess payments to HMOs
      nationwide could total $2 billion.2

      Concerned about the inconsistency between the expectation that HMOs
      would save Medicare money and research findings showing that HMOs
      increase the program’s costs, you asked us to (1) explain under what
      conditions Medicare’s method can yield payment rates that are too high
      and (2) suggest a practical improvement to Medicare’s method directed at
      the problems fostering excess payments.

      To do this work, we reviewed previous research on the HMO rate-setting
      method used by the Health Care Financing Administration (HCFA), the
      Department of Health and Human Services’ (HHS) agency responsible for
      administering Medicare. We also developed a method for estimating
      enrollees’ costs using the data Medicare collects to determine HMO

      1
       Other Medicare managed care plans include cost contract HMOs and health care prepayment plans,
      which together enroll fewer than 2 percent of the total Medicare population. Because Medicare pays
      these plans using methods other than capitation, they are not included in this study.
      2
       This estimate was contained in material presented to the Commissioners for their December 12-13,
      1996, meeting.



      Page 1                                        GAO/HEHS-97-16 Medicare HMO Excess Payments
                   B-265996




                   payments and applied the method to each of the 58 counties in California,
                   a state that has about 36 percent of the total Medicare risk HMO population.
                   Our method and estimates of excess payments to HMOs were reviewed by
                   independent experts on HMO payment issues. We performed this work
                   from August 1995 to December 1996 in accordance with generally
                   accepted government auditing standards.


                   Contrary to the expectations built into Medicare law for paying risk
Results in Brief   contract HMOs, these HMOs have not produced savings for Medicare.
                   Medicare law says that the program should pay HMOs 95 percent of what
                   HCFA estimates would have been paid had enrollees remained in FFS.
                   However, Medicare-sponsored research and other studies have found that
                   the program has actually spent more for HMO enrollees than their costs
                   would have been under FFS. Researchers attribute this outcome to
                   “favorable selection,” or the tendency for healthier-than-average
                   individuals to be enrolled in HMOs. Two 1996 studies, each using different
                   methodologies, produced estimates of lower costs for HMO beneficiaries
                   compared with those of FFS beneficiaries—one, 12 percent lower; the
                   other, 37 percent lower. Both estimates could translate into substantial
                   payments in excess of what Medicare would have spent if the HMO
                   beneficiaries had remained in the FFS sector.

                   We have identified a modification to Medicare’s current HMO rate-setting
                   method that could help reduce excess HMO payments. Central to the
                   current method is an estimate of the average cost, county by county, of
                   serving Medicare beneficiaries in the FFS sector. The actual rates are set by
                   adjusting the county averages up or down on the basis of each enrollee’s
                   likelihood of incurring higher or lower costs, a process known as risk
                   adjustment. Although considerable attention has focused on problems
                   with this process, our work centers on a largely overlooked problem
                   regarding the estimates of average county costs—that is, the county rate,
                   commonly known as the AAPCC (adjusted average per capita cost).

                   HCFA’s method of determining the county rate excludes HMO enrollees’
                   costs in estimating per-beneficiary average cost. The result is that in
                   counties experiencing favorable selection, HCFA’s method overstates the
                   average costs of all Medicare beneficiaries and leads to overpayments.

                   Our proposed modification estimates HMO enrollees’ expected FFS costs
                   using information available to HCFA. Our approach produces a county rate
                   that more accurately represents the costs of all Medicare beneficiaries. In



                   Page 2                             GAO/HEHS-97-16 Medicare HMO Excess Payments
             B-265996




             examining the rates HCFA determined for California’s 58 counties in 1995,
             we found that applying our approach would have reduced excess
             payments by about 25 percent, or $276 million. On a monthly,
             per-beneficiary payment level, the county-rate reductions would have been
             relatively small, ranging from $3 to $38. Substantially better risk
             adjustment, which appears to be years away from implementation, would
             have targeted the remaining 75 percent of excess payments.

             We also found that Medicare’s current method produced a greater
             overstatement of county average costs in counties with higher Medicare
             HMO penetration—up to 39 percent.3 This finding calls into question the
             hypothesis put forth by HMO industry advocates and others that the excess
             payment problem will be mitigated as more beneficiaries enroll in
             Medicare managed care and HMOs contain a more expensive mix of
             beneficiaries.


             Essentially, HCFA’s calculation of its per-enrollee (capitation) rate in each
Background   county can be expressed as follows:




             Medicare pays risk HMOs a fixed amount per enrollee—a capitation
             rate—regardless of what each enrollee’s care actually costs. Medicare law
             stipulates that the capitation rate be set at 95 percent of the costs
             Medicare would have incurred for HMO enrollees if they had remained in
             FFS.4 In implementing the law’s rate-setting provisions, HCFA estimates a




             3
              HMO penetration rates are for 1992. Following HCFA’s method of calculating AAPCC rates, our
             estimates of excess payments in 1995 are derived from beneficiary costs in the base year (1992). See
             app. I for an explanation of our method of determining excess payments.
             4
              Section 1876(a)(4) of the Social Security Act (42 U.S.C. 1395mm(a)(4) (1994)).



             Page 3                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                      B-265996




                      county’s average per-beneficiary cost and multiplies the result by 0.95.5
                      The product is the county adjusted average per capita cost rate.6

                      HCFA  then applies a risk-adjustment factor to the county rate. Under HCFA’s
                      risk-adjustment system, beneficiaries are sorted into groups according to
                      their demographic traits (age; sex; and Medicaid, institutional, and
                      working status). These traits serve as proxy measures of health status.
                      HCFA calculates a risk factor for each group—the group’s average cost in
                      relation to the cost of all beneficiaries nationwide. For example, in 1995
                      the risk factor for younger seniors (65- to 70-year-old males) was .85,
                      whereas for older seniors (85-year-old or older males) it was 1.3. HCFA uses
                      the risk factor to adjust the county rate, thereby raising or lowering
                      Medicare’s per capita payment for each HMO enrollee, depending on the
                      individual’s demographic characteristics.


                      For HCFA’s rate-setting method to produce appropriate rates, the risk
How Medicare’s HMO    adjusters must reliably differentiate among beneficiaries with different
Rate-Setting Method   health status. Much has been written about the inadequacy of Medicare’s
Can Lead to Excess    risk adjuster to account for the tendency of HMOs to experience favorable
                      selection. More than a decade of research has concluded that beneficiaries
Payments              enrolling in HMOs are, on average, healthier than those remaining in FFS.7
                      Studies of pre-1990 data found that Medicare HMO enrollees—in a period
                      just prior to their HMO enrollment—had health care costs that were from
                      20 percent to 42 percent lower than those of FFS beneficiaries with the
                      same demographic characteristics. Studies of post-1990 data also showed




                      5
                       A 5-percent discount is taken on the premise that, compared with FFS care, managed care plans
                      achieve certain efficiencies. For example, HMOs can negotiate with hospitals, physicians, and other
                      providers to obtain discounts on services and supplies. In response to concerns that Medicare’s
                      payment rates to HMOs are too high, the administration has publicly discussed phasing in a reduction
                      in HMO payment rates from the current 95 percent to 90 percent of FFS payments.
                      6
                      Medicare determines four capitation rates for each county, one each for part A aged, part B aged, part
                      A disabled, and part B disabled.
                      7
                       Our study entitled Medicare: Changes to HMO Rate Setting Method Are Needed to Reduce Program
                      Costs (GAO/HEHS-94-119, Sept. 2, 1994) discusses at length the inability of HCFA’s rate-setting method
                      to prevent favorable selection from increasing Medicare costs. It cites and reviews numerous studies
                      on the subject of favorable selection in Medicare HMOs. For a review of recent studies and an analysis
                      concluding that Medicare risk HMOs continue to benefit from favorable selection, see also Center for
                      Studying Health System Change, “Policy Implications of Risk Selection in Medicare HMOs: Is the
                      Federal Payment Rate Too High?” Issue Brief, No. 4 (Washington, D.C.: Center for Studying Health
                      System Change, Nov. 1996).


                      Page 4                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
B-265996




costs of Medicare HMO enrollees ranging from 12 percent8 to 37 percent
lower than those of their FFS counterparts.9

The problem for Medicare posed by favorable selection is that HMO
enrollees are healthier than FFS beneficiaries within the same demographic
group; for example, 70-year-old males in HMOs are, on average, healthier
than 70-year-old males in FFS. Medicare’s risk adjuster is said to be
inadequate because, while making broad distinctions among beneficiaries
of different age, sex, and other demographic characteristics, it does not
account for the significant health differences among demographically
identical beneficiaries. The cost implications of health status differences
can be dramatic for two demographically alike beneficiaries: one may
experience occasional minor ailments while the other may suffer from a
serious chronic condition.

Devising a risk adjuster sensitive enough to capture health status
differences, however, is such a technically complex and difficult task that
years of independent research and HCFA-sponsored research have not yet
produced an ideal risk adjuster.10 In reports issued in 1994 and 1995, we
identified several promising, practical risk adjusters and suggested that
HCFA implement an interim improvement.11




8
 See G. Riley, C. Tudor, Y. Chiang, and M. Ingber, “Health Status of Medicare Enrollees in HMOs and
Fee-for-Service in 1994,” Health Care Financing Review, Vol. 17, No. 4 (summer 1996), pp. 65-76. This
study analyzed 1994 data from the Medicare Current Beneficiary Survey and found that HMO enrollees’
costs, post HMO enrollment, were about 12 percent lower than the costs of comparable beneficiaries
in FFS.
9
 Physician Payment Review Commission, “Risk Selection and Risk Adjustment in Medicare,” Annual
Report to Congress, ch. 15 (Washington, D.C.: Physician Payment Review Commission, 1996). In an
analysis of 1989-94 data, the Commission found that health costs of new HMO enrollees—in the 6
months prior to their enrollment in an HMO—were 37 percent lower than the health costs of
beneficiaries with similar demographic traits who remained in the FFS program.
10
  For example, HCFA announced in January 1997 that it was about to launch a demonstration project
on two sophisticated risk-adjustment methods—the ambulatory care group and diagnostic cost group
systems—that seek to differentiate more and less costly patients on the basis of diagnostic information
from inpatient, outpatient, and physician encounters. HCFA has not announced a schedule for
implementing a better risk adjuster programwide.
11
 GAO/HEHS-94-119, Sept. 2, 1994, and Medicare Managed Care: Growing Enrollment Adds Urgency to
Fixing HMO Payment Problem (GAO/HEHS-96-21, Nov. 8, 1995).


Page 5                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                           B-265996




                           Independent of risk adjustment, modifying the method for calculating
HCFA Could Improve         county rate would help reduce Medicare’s excess HMO payments. HCFA
Its Rate-Setting           currently estimates the average Medicare costs of a county’s beneficiaries
Method by Including        using the costs of only those beneficiaries in Medicare’s FFS sector. This
                           method would be appropriate if the average health cost of FFS
HMO Enrollees in Its       beneficiaries were the same as that of demographically comparable HMO
Calculations of            enrollees.12 However, in counties where there are cost disparities between
                           Medicare’s FFS and HMO enrollee populations, this method can either
County Average Cost        overstate the average costs of all Medicare beneficiaries and lead to
                           overpayment or understate average costs and lead to underpayment.

                           To understand how favorable selection can produce an excessive county
                           rate under HCFA’s method of estimating average costs, consider the
                           following hypothetical example:

                           Suppose a county has 1,000 Medicare beneficiaries with identical demographic
                           characteristics.13 Of these, 800 beneficiaries are in Medicare’s FFS program and cost
                           Medicare on average $100 a month. The remaining 200 beneficiaries are enrolled in HMOs,
                           but these beneficiaries would have cost an average of $75 a month had they remained in
                           the FFS program. For all 1,000 beneficiaries, the county average cost would be $95 a month.
                           HCFA’s method excludes the HMO enrollees with their lower costs from its calculations,
                           producing a county average of $100 a month. Consequently, HCFA overestimates this
                           county’s average monthly cost by $5, producing $1,000 a month in excessive Medicare
                           payments to HMOs (200 beneficiaries times $5).


                           The difficulty in correcting this problem comes from the inability to
                           observe the costs HMO enrollees would have incurred if they had remained
                           in the FFS sector. In the illustration above, HCFA needs a way to estimate
                           that the beneficiaries enrolled in HMOs would have cost $75 a month in the
                           FFS sector rather than $100. Therefore, we developed a method to estimate
                           HMO enrollees’ expected FFS costs using information available to HCFA. Our
                           method consists of two main steps:

                       •   First, we computed the average costs of new HMO enrollees during the year
                           before they enrolled—that is, while they were still in FFS Medicare. These
                           FFS costs are available through HCFA’s claims data.




                           12
                            HCFA’s method would also be appropriate if a risk adjuster were available that could remove the
                           effects of favorable, or unfavorable, selection with far more accuracy than is currently achieved or
                           considered feasible today.
                           13
                             The assumption of equivalent demographic characteristics is made to simplify the illustration.



                           Page 6                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                           B-265996




                       •   Next, we adjusted this amount to reflect the expectation that an enrollee’s
                           use of health services will, over time, rise.14

                           Having completed these steps, we combined the result with an estimate of
                           the average cost of FFS beneficiaries. This new average produced a county
                           rate that reflected the costs of all Medicare beneficiaries. Thus, our
                           method helps prevent biasing HMO payments with either overgenerous
                           estimates of enrollees’ initial health costs or low estimates that fail to
                           compensate for the likelihood of rising health costs over time. The
                           technical details of this approach are discussed in appendix I.


Current County Rates       To illustrate the effect of our approach, we analyzed data for counties with
Produce Substantial        different shares of beneficiaries enrolled in HMOs.15 We found that our
Excess Payments            method could have reduced excess payments by more than 25 percent.
                           Substantially better risk adjustment, which appears to be years away from
                           implementation, would target the remaining 75 percent of excess
                           payments. Specifically, for the counties that we analyzed, we estimated
                           that total excess payments in 1995 amounted to about $1 billion of the
                           roughly $6 billion in total Medicare payments to risk HMOs in the state.
                           (App. III discusses excess payment estimates in further detail.) Applying
                           our method for setting county rates would have reduced the excess by
                           about $276 million.

                           We also found that the excess payments attributable to inflated county
                           rates were concentrated in 12 counties with large HMO enrollment and
                           ranged from less than 1 percent to 6.6 percent of the counties’ total HMO
                           payments, representing between $200,000 and $135.3 million.16 (See table
                           1.) Despite the size of these amounts, the application of our method would
                           have produced relatively small changes in the monthly, per-beneficiary
                           capitation payments, ranging from $3 to $38.




                           14
                             Our analysis adjusts for (1) the tendency after joining an HMO for enrollees’ costs to become more
                           like—or “regress” toward—the FFS cost mean and (2) the costs incurred by HMO enrollees who die
                           while enrolled. A more thorough discussion of how our method accounts for these costs is contained
                           in apps. I and II.
                           15
                             We chose counties within a single state to eliminate variations attributable to state differences and
                           selected California because it included counties that in 1995 had the nation’s highest HMO penetration
                           rates.
                           16
                            For the state’s remaining 46 counties, excess payments attributable to inflated county rates
                           amounted to less than 3 percent of the 58-county total. App. III shows projections of excess HMO
                           payments by county for 1996 and 1997.



                           Page 7                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                                        B-265996




Table 1: Estimates of Potential
Reduction in Excess Payments to                                                                                   County-rate excess
California HMOs in 1995, Based on Our                                            County-rate estimates of payments as a percentage
Method for Calculating the County                                                   excess payments (in     of risk contract program
Rate                                    County                                                   millions)                 payments
                                        Los Angeles                                                     $135.3                      6.56
                                        Orange                                                             38.5                     6.37
                                        San Diego                                                          37.3                     5.12
                                        San Bernardino                                                     23.4                     5.79
                                        Riverside                                                          17.5                     3.70
                                        Ventura                                                             6.6                     4.80
                                        Kern                                                                4.4                     3.74
                                        San Francisco                                                       4.0                     2.44
                                        Sacramento                                                          3.2                     1.62
                                        San Mateo                                                           2.9                     2.25
                                        Santa Clara                                                         2.3                     1.18
                                        Butte                                                               0.2                     0.79
                                                                 a
                                        Total (12 counties)                                             $275.7
                                        a
                                            Numbers may not add because of rounding.



                                        The excess payments shown in table 1 reflect the difference between
                                        Medicare’s county rates and rates calculated by our method.17 As shown in
                                        the table, five counties accounted for more than 90 percent of the state’s
                                        county-rate excess payments.

                                        Our analysis did not support the hypothesis, put forward by the HMO
                                        industry and others, that the excess payment problem will be mitigated as
                                        more beneficiaries enroll in Medicare managed care and HMOs
                                        progressively enroll a more expensive mix of beneficiaries. Our
                                        data—from counties with up to a 39-percent HMO penetration—indicated
                                        that excess payments as a percentage of total HMO payments were higher
                                        in counties with higher Medicare penetration. For example, as seen in
                                        figure 1, the four counties with the highest rates of excess payment,
                                        ranging from 5.1 to 6.6 percent, were also among the counties with the
                                        highest enrollment rates.




                                        17
                                            The technical steps to derive our estimates of excess payments are set out in app. I.



                                        Page 8                                            GAO/HEHS-97-16 Medicare HMO Excess Payments
                                      B-265996




Figure 1: Excess Payments Rise With
HMO Enrollment
                                      Percent of 1995 Excess Payments From County Rate

                                       10




                                        8




                                        6




                                        4




                                        2




                                        0
                                            0               10               20                30                40         50

                                            Percent of Medicare Beneficiaries Enrolled in Managed Care




                                      Note: Each data point represents 1 of the 12 California counties listed in table 1.

                                      Source: GAO analysis of HCFA data.




                                      If the relationship between enrollment and excess payments we found for
                                      California in 1995 persists, excess payments are likely to grow. The recent
                                      trend in Medicare HMO enrollment suggests continued growth in the next
                                      several years. Therefore, some counties with moderate enrollment today
                                      may experience higher enrollment rates in the future, exacerbating the




                                      Page 9                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                          B-265996




                          excess payment problem. (See app. III, table III.1, for estimates of future
                          excess HMO payments in California based on projected enrollment.)


Data Are Available to     Because the data we used to estimate HMO enrollees’ costs come from data
Enable HCFA to Promptly   that HCFA compiles to update HMO rates each year, our method has two
Adjust County Rates       important advantages. First, HCFA’s implementation of our proposal could
                          be achieved in a relatively short time. The time element is important,
                          because the prompt implementation of our method would avoid locking in
                          a current methodological flaw that would persist in any adopted changes
                          to Medicare’s HMO payment method that continued to use either current
                          county rates as a baseline or FFS costs to set future rates. Second, the
                          availability of the data would also make our proposal economical: we
                          believe that the savings to be achieved from reducing county-rate excess
                          payments would be much greater than the administrative costs of
                          implementing our modification.

                          We recognize that for counties with little or no HMO enrollment, HCFA’s
                          current method of estimating the county rate would yield virtually the
                          same result as our method because the small number of HMO enrollees is
                          overwhelmed by the large number of FFS beneficiaries and has only a
                          minimal effect on average FFS costs. Thus, HCFA could decide to use a
                          beneficiary enrollment threshold for computing revised county rates.


                          Medicare’s HMO rate-setting problems have prevented it from realizing the
Conclusions               savings that were anticipated from enrolling beneficiaries in capitated
                          managed care plans. In fact, enrolling more beneficiaries in managed care
                          could increase rather than lower Medicare spending—unless Medicare’s
                          method of setting HMO rates is revised.

                          Our method of calculating the county rate would have the effect of
                          reducing payments more for HMOs in counties with higher excess payments
                          and less for HMOs in counties with lower excess payments. In this way, our
                          method represents a targeted approach to reducing excess payments and
                          could lower Medicare expenditures by at least several hundred million
                          dollars each year. Furthermore, because some proposals to reform
                          Medicare HMO rate-setting rely on current county payment rates as a
                          benchmark, correcting the current county rates would avoid locking in
                          varying degrees of excess payments across counties for years to come.




                          Page 10                            GAO/HEHS-97-16 Medicare HMO Excess Payments
                    B-265996




                    We recommend that the Secretary of Health and Human Services direct
Recommendation to   the HCFA Administrator to incorporate the expected FFS costs of HMO
the Secretary of    enrollees into the methodology for establishing county rates using the
Health and Human    method we explain in this report and adjust Medicare payment rates to
                    risk contract HMOs accordingly.
Services
                    In commenting on a draft of this report, HHS agreed that, because Medicare
Agency Comments     HMO enrollees tend to be healthier than FFS beneficiaries, the current
                    payment methodology may have resulted in Medicare’s overpaying HMOs
                    substantially—according to HHS, by $1 billion in fiscal year 1996. HHS noted
                    that the President’s fiscal year 1998 budget proposes to address the excess
                    payment problem by lowering HMO capitation rates in calendar year 2000
                    and developing a new payment system to be phased in beginning in 2001.
                    However, our recommended rate-setting change could be implemented
                    much sooner and would continue to be useful after HCFA develops a new
                    HMO payment system.


                    Although HHS did not question that our recommended rate-setting change
                    would save hundreds of millions of dollars each year for Medicare and
                    taxpayers, the Department doubted the change would be equitable and
                    relatively easy to implement. However, our approach to reducing excess
                    payments is equitable because it is targeted—in contrast to HHS’ proposed
                    across-the-board cut—and would reduce payments only in those counties
                    where HMOs receive excess payments. Furthermore, our recommended
                    change should require very little additional HCFA staff time and no
                    collection of new data. (See app. IV for the full text of HHS’ comments and
                    our response.)


                    As arranged with your office, unless you publicly announce the contents of
                    this report earlier, we plan no further distribution until 30 days after its
                    issue date. At that time, we will send copies to the Secretary of Health and
                    Human Services; the Director, Office of Management and Budget; the
                    Administrator of the Health Care Financing Administration; and other
                    interested parties. We will also make copies available to others upon
                    request.




                    Page 11                           GAO/HEHS-97-16 Medicare HMO Excess Payments
B-265996




This work was done under the direction of William J. Scanlon, Director,
Health Financing and Systems Issues. If you or your staff have any
questions about this report, please contact Mr. Scanlon at (202) 512-7114.
Other GAO contacts and staff acknowledgments are listed in appendix V.

Sincerely,




Richard L. Hembra
Assistant Comptroller General




Page 12                           GAO/HEHS-97-16 Medicare HMO Excess Payments
Page 13   GAO/HEHS-97-16 Medicare HMO Excess Payments
Contents



Letter                                                                                           1


Appendix I                                                                                      18
                        Method for Reducing Excess HMO Payments by Correcting                   18
Methodology               Medicare’s County Rate
                        Method for Estimating Medicare’s Aggregate Excess Payments              29

Appendix II                                                                                     31
                        Methodology Allows RTM Factor to Vary by Beneficiary Survival           32
Adjustments for           Status
Regression Toward       Method Used to Estimate the RTM Factor for Category I                   34
                          Enrollees
the Mean and            Method Used to Estimate the RTM Factor for Category II                  40
Death-Related Costs       Enrollees
in Estimating Excess    The RTM Factor for Category III Enrollees                               41
                        Summary of Adjustments for RTM                                          42
Payments to Medicare
HMOs
Appendix III                                                                                    44
                        Estimates of County-Rate Excess Payments                                45
Estimates of Medicare   Estimates of Aggregate Excess Payments                                  47
Excess Payments to
HMOs in California
Appendix IV                                                                                     50
                        GAO Comment                                                             54
Comments From the
Department of Health
and Human Services
and Our Evaluation
Appendix V                                                                                      58

GAO Contacts and
Staff
Acknowledgments




                        Page 14                         GAO/HEHS-97-16 Medicare HMO Excess Payments
          Contents




Tables    Table 1: Estimates of Potential Reduction in Excess Payments to           8
            California HMOs in 1995, Based on Our Method for Calculating
            the County Rate
          Table I.1: How HMO Enrollee and FFS Beneficiary Costs Were               25
            Estimated, Sample Year 1992
          Table I.2: Ratios of Monthly Average Costs of New Risk HMO               27
            Enrollees to FFS Beneficiaries’ Costs for Three California
            Counties, 1992-94
          Table II.1: Classification of HMO Enrollees by Survival Status           34
          Table II.2: 1991 Distribution Across Cost Categories of HMO              36
            Joiners and FFS Beneficiaries, 65- to 69-Year-Old Females
          Table II.3: Costs of Proxy HMO Joiners Relative to Those of              39
            Proxy FFS Beneficiaries, 1991-94
          Table II.4: Example of Derivation of Regression-Toward-the-Mean          40
            Adjustment Factor From Cost Ratios
          Table II.5: Death Rates, per 100, of Aged Medicare Beneficiaries         42
            by Demographic Group and Year, 1992-94
          Table III.1: Medicare County-Rate Excess Payments for 20                 45
            California Counties in Dollars and as a Percentage of Program
            Payments, 1995-97
          Table III.2: Aggregate Excess Payments by County for 1995 in             48
            Millions of 1995 Dollars

Figures   Figure 1: Excess Payments Rise With HMO Enrollment                        9
          Equation 1                                                               19
          Equation 2                                                               20
          Equation 3                                                               28
          Equation 4                                                               29
          Equation 5                                                               29
          Equation 6                                                               30
          Figure II.1: Annual Medicare Payments in the Years Preceding             33
            Death
          Figure II.2: Regression-Toward-the-Mean Patterns for 10                  38
            Demographic Groups of Proxy HMO Enrollees




          Page 15                          GAO/HEHS-97-16 Medicare HMO Excess Payments
Contents




Abbreviations

AAPCC      adjusted average per capita cost
EDB        Enrollment Database File
FFS        fee-for-service
HCFA       Health Care Financing Administration
HHS        Department of Health and Human Services
HMO        health maintenance organization
PAEP       percent aggregate excess payment
PPRC       Physician Payment Review Commission
RTM        regression toward the mean
RTMF       regression-toward-the-mean adjustment factor
SAC        standard average cost


Page 16                         GAO/HEHS-97-16 Medicare HMO Excess Payments
Page 17   GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix I

Methodology


                        Despite evidence from a number of studies18 that health maintenance
                        organization (HMO) enrollees tend to be healthier than demographically
                        comparable fee-for-service (FFS) beneficiaries (“favorable selection”), the
                        Health Care Financing Administration (HCFA) rate-setting method
                        implicitly assumes that the health service needs of both groups are the
                        same. To the extent that favorable selection occurs, HCFA’s assumption
                        increases the capitation rates HCFA pays to risk HMOs and results in excess
                        payments. This appendix describes how making more realistic
                        assumptions concerning the health status of HMO enrollees can partially
                        correct the excess payment problem. In essence, our approach determines
                        the extent to which HCFA’s method overestimates average Medicare FFS
                        costs and thus inflates the county rate—one component of HMO capitation
                        payments.19 This appendix also briefly discusses a related method for
                        estimating aggregate excess payments.


                        The basic steps HCFA takes to determine capitation payments can be
Method for Reducing     described as follows.
Excess HMO
Payments by                  Step 1

Correcting Medicare’s   HCFA calculates the per capita costs in Medicare FFS, or standard average
County Rate             cost (SAC). This is done for each county, partly to allow for geographic
                        differences in medical prices.

                             Step 2

                        The basic capitation rate, or county rate, is set at 95 percent of the county
                        per capita cost. That is, COUNTY = 0.95 • SAC.20

                             Step 3

                        Finally, payments for specific individuals are adjusted up or down on the
                        basis of a limited set of demographic factors, or “risk factors.” These risk




                        18
                          See footnotes 7 and 8 for studies that have addressed the issue of favorable selection in HMOs.
                        19
                         If HMOs experienced adverse selection (if they enrolled beneficiaries who, on average, were less
                        healthy than FFS beneficiaries), our method would also determine the extent to which HCFA’s
                        methodology underestimated a county’s average Medicare costs.
                        20
                         More precisely, Medicare determines four such rates for each county: one each for part A aged, part
                        B aged, part A disabled, and part B disabled.



                        Page 18                                        GAO/HEHS-97-16 Medicare HMO Excess Payments
                                  Appendix I
                                  Methodology




                                  factors are intended to partially adjust for differences in expected health
                                  care costs of beneficiaries of different ages, gender, and so on.21

                                  Essentially, the capitation rate formula can be expressed as follows:


Equation 1




Sources of Excess                 Excess payments can occur if HMOs enroll a group of beneficiaries that is
Payments to HMOs                  healthier than the average FFS beneficiary and the capitation rate is not
                                  sufficiently adjusted for the differences in health status. In HCFA’s current
                                  method, favorable selection can cause excess payments, partly because
                                  HCFA’s risk factors inadequately adjust for differences in beneficiaries’
                                  health status and partly because SAC overstates the costs of serving HMO
                                  enrollees.

HCFA’s Risk Factors Are           HCFA’s risk factors adjust for favorable selection using five characteristics
Rough Proxies for Expected        (age, sex, Medicaid eligibility status, institutional status, and working
Health Costs and Do Not Fully     status) that are relatively poor predictors of beneficiaries’ health care
Adjust Payments for Favorable     needs.22 Specifically, the risk factors are a set of weights—intended to
Selection                         reflect the relative health risk of each beneficiary—used to adjust the
                                  basic capitation rate up or down. For example, the weight assigned to 65-
                                  to 70-year-old males was .85 in 1995, implying that they had a greater
                                  health cost risk—higher expected health costs—than 65- to 70-year-old
                                  females, whose weight was .70. Beneficiaries with the same risk factor are
                                  assumed to have the same relative health service needs. However, if
                                  70-year-old males enrolling in HMOs tend to be healthier than the
                                  70-year-old males who remain in FFS, then the risk factor will
                                  overcompensate for the enrollees’ costs and the HMOs are said to have
                                  benefited from favorable selection.

HCFA’s Capitation Rate Is         If HMOs’ enrollees tend to be healthier than the average beneficiary in FFS,
Inflated by Favorable Selection   then HCFA’s method will overestimate the expected cost of serving

                                  21
                                   The risk-adjustment component assigns each enrollee to 1 of 70 risk adjustment cells for aged and
                                  disabled beneficiaries (with different cell weights for part A and part B). Payment rates for
                                  beneficiaries with end-stage kidney disease are computed separately.
                                  22
                                   In 1994, we reported that “the demographic variables HCFA uses [as risk adjusters] are only loosely
                                  associated with health care costs . . ..” See Medicare: Changes to HMO Rate Setting Method Are
                                  Needed to Reduce Program Costs (GAO/HEHS-94-119). For a more recent discussion of the weak
                                  correlation between HCFA’s risk factors and beneficiaries’ health care needs, see Physician Payment
                                  Review Commission (PPRC), Annual Report to Congress (Washington, D.C.: Physician Payment
                                  Review Commission, 1996).



                                  Page 19                                       GAO/HEHS-97-16 Medicare HMO Excess Payments
             Appendix I
             Methodology




             Medicare beneficiaries in FFS. The foundation of the rate-setting formula
             consists of the standard average cost to Medicare of a county’s FFS
             beneficiaries.23 (By standard, we mean this cost measure is normalized for
             differences in each county’s demographic composition, relative to the
             national average).24 HCFA calculates SAC from the costs of FFS program
             beneficiaries alone (SACFFS).25,26 However, to the extent that the health
             care costs of Medicare’s HMO enrollee population are lower, on average,
             than those of beneficiaries in FFS, the exclusion of HMO enrollees’ costs
             (that is, what they would have cost Medicare in FFS) causes SAC and,
             ultimately, the capitation rate, to be too high.27

             A better way to set Medicare HMO rates would be based on a SAC that
             reflected both the costs of beneficiaries in FFS (SACFFS) and what the costs
             of HMO enrollees would have been if they had been in FFS (SACHMO). Setting
             rates this way would lessen the amount of adjustment needed to reflect
             differences in health status because HMO enrollees’ expected FFS costs
             would already be included. The estimated average cost for all beneficiaries
             in the county could be calculated as a weighted average of SACFFS and
             SACHMO, where pFFS and pHMO are the proportions of county beneficiaries
             in FFS and HMOs, respectively. (See equation 2.)


Equation 2




             23
              Section 1876(a)(4) of the Social Security Act (42 U.S.C. 1395mm(a)(4) (1994)) provides that the
             Secretary of Health and Human Services (HHS) estimate the average per capita amount that “would be
             payable . . . if the services were to be furnished by other than an eligible organization . . .”—that is, by
             FFS.
             24
               To normalize (or standardize) the average cost for any beneficiary group, HCFA divides that average
             cost by the average risk-adjustment factor for that beneficiary group. The normalized average is
             representative of a demographically average Medicare beneficiary.
             25
                HCFA’s rate-setting method appropriately discards HMO payments (to arrive at SACFFS) because they
             do not represent what HMO enrollees’ costs would be if measured on an FFS basis.
             26
              HCFA’s computation of the average is actually a forecast of expected costs for the contract year.
             HCFA actuaries develop the forecast using cost experience data from a “base year,” which is usually 3
             years prior to the contract year. In setting county rates for contract year 1995, for example, HCFA used
             1992 (and earlier) data. For a detailed description of HCFA’s rate-setting method, see Office of the
             Actuary, HCFA, Adjusted Average Per Capita Cost Methodology For Risk-Sharing Contracts
             (Baltimore, Md.: HHS).
             27
              A number of studies, summarized in table 15.1 of PPRC’s 1996 Annual Report to Congress, p. 258,
             have found that HMO enrollees’ costs are lower than comparable FFS beneficiary costs.



             Page 20                                          GAO/HEHS-97-16 Medicare HMO Excess Payments
    Appendix I
    Methodology




    However, because HCFA cannot directly observe what the FFS costs would
    have been for beneficiaries currently enrolled in HMOs (SACHMO), the
    agency assumes that the averages for the two groups are equal.

    If relatively healthy beneficiaries enroll in HMOs while less healthy
    beneficiaries remain in Medicare FFS, however, SACHMO will be less than
    SACFFS. By assuming the two costs are equal, HCFA overstates the expected
    cost of serving HMO enrollees under FFS. This overestimate increases as the
    gap between SACFFS and SACHMO widens and can increase as the
    proportion of beneficiaries in HMOs (pHMO) increases. Because SAC forms
    one of the building blocks in the capitation rate formula, overestimating
    SAC leads to excess payments to HMOs.


    The following examples illustrate how, in the presence of favorable
    selection, HCFA’s calculation of SAC and COUNTY results in excess
    payments to HMOs.

•   If a county had 10 demographically identical beneficiaries, 8 of whom cost
    Medicare nothing each year and 2 who cost $2,000 each, the county’s
    average per capita cost, or SACALL, would equal $400 ($4,000 divided by the
    10 beneficiaries). If no beneficiaries were enrolled in HMOs, SACFFS would
    equal SACALL, or $400. In contrast, if two beneficiaries costing Medicare
    nothing had joined HMOs, SACFFS—on the basis of the eight remaining FFS
    beneficiaries—would equal $500 ($4,000 divided by eight).
•   Under HCFA’s method, COUNTY would be $500 • .95—reflecting just the
    average costs of beneficiaries in the FFS sector—instead of $400 • .95.
    Thus, Medicare would pay HMOs $100 • .95 more than if capitation rates
    were based on the actual average expected FFS cost of all beneficiaries in
    the county.

    Furthermore, the enrollment of additional beneficiaries with low costs in
    the county’s HMOs would widen the disparity between SACFFS and SACALL.
    For example, if six beneficiaries costing Medicare nothing had joined
    HMOs, SACFFS would equal $1,000 ($4,000 divided by the four beneficiaries
    still in FFS) or more than double SACALL’s value of $400. In this case,
    Medicare’s payments to HMOs would be based on a COUNTY equal to
    $1,000 • .95 instead of the appropriate $400 • .95.




    Page 21                           GAO/HEHS-97-16 Medicare HMO Excess Payments
                          Appendix I
                          Methodology




Estimating Expected FFS   We developed a method to estimate the potential FFS costs for HMO
Costs for HMO Enrollees   enrollees that allows calculation of average FFS cost estimates based on all
                          beneficiaries living in the county (SACALL).28,29 We identified the FFS cost
                          experience of recent risk HMO enrollees prior to their HMO enrollment.
                          Drawing on these prior-use cost data and data on changes in individuals’
                          health costs over time, we estimated the expected costs (on an FFS basis)
                          of people who had been enrolled in an HMO for different periods of time.
                          Finally, we combined these estimates to calculate SACHMO, which reflected
                          the characteristics of the county’s HMO enrollees, including the length of
                          time they had been HMO enrollees. This “prior-use” cost approach is
                          necessary because no other relevant cost data are currently available to
                          HCFA. After a beneficiary enrolls in an HMO, HCFA receives no information on
                          the health care services provided to the beneficiary or their costs.

                          We made adjustments to respond to two major criticisms of previous
                          studies that employed prior-use costs to estimate expected post
                          enrollment costs.

                          1. Unadjusted prior-use estimates do not allow for the possibility that
                          enrollees’ average expected costs can regress toward the mean cost of FFS
                          beneficiaries. That is, as time passes, enrollees’ average costs can rise and
                          approach the average costs of the FFS beneficiaries, rather than remain at
                          their preenrollment levels. If this happens, the disparity between the
                          prior-use costs of HMO enrollees and the costs of comparable FFS




                          28
                           HCFA’s methodological steps, especially those for updating the 1992 cost estimates to a 1995 basis,
                          are complex. However, our method to estimate excess payments is not sensitive to much of this
                          complexity. In particular, our method improves the estimate of SAC while leaving intact all subsequent
                          calculations that HCFA would make involving SAC. (That is, these later calculations still apply
                          whether our estimate of SAC or HCFA’s is used.) Thus, if our estimate of SAC is less than HCFA’s by
                          10 percent, this amount would be passed directly through all subsequent calculations. As a result,
                          payment rates determined with our method would be 10 percent lower than those determined with
                          HCFA’s method.
                          29
                           We mirrored HCFA’s methodology in developing estimates of SACALL and SACFFS from base-year (or
                          earlier) data. However, we did not follow the HCFA approach of using a 5-year average to estimate
                          SACFFS. On the basis of our comparison of the 5-year Average Geographic Adjuster to the base-year
                          Geographic Adjuster, we concluded that the 5-year averaging had little or no effect during our sample
                          years. Nonetheless, our approach could be modified to incorporate the 5-year average approach.



                          Page 22                                        GAO/HEHS-97-16 Medicare HMO Excess Payments
                                 Appendix I
                                 Methodology




                                 beneficiaries overstates the actual difference in cost that exists in years
                                 following enrollment.30

                                 2. Unadjusted prior-use estimates underrepresent enrollees’ “death costs.”
                                 Unadjusted prior-use cost methodologies cannot take account of the full
                                 costs associated with death for enrollees, because beneficiaries must
                                 survive the prior year to enroll.

                                 Not making these adjustments could result in an overestimate of excess
                                 Medicare HMO payments.

                                 In developing our method to approximate SACHMO, we struck a balance
                                 between two potentially conflicting goals: (1) minimizing the
                                 computational burden and (2) maximizing the accuracy of the enrollees’
                                 expected FFS cost estimate. The particular assumptions and modifications
                                 of our augmented prior-use methodology are detailed below. We
                                 recognize, however, that other approaches to approximating SACHMO could
                                 also result in slightly different, but equally plausible, estimates of
                                 enrollees’ expected FFS costs.31 Once we estimated SACHMO, we used the
                                 proportions of beneficiaries in FFS and HMOs to compute SACALL. (See
                                 equation 2.) Because we also knew actual HMO payments for each county,
                                 we could use our new estimates to compute estimates of county rate
                                 excess payments.

Beneficiaries Classified         Because Medicare allows beneficiaries to switch among specific HMOs or
According to Enrollment Status   between an HMO and FFS monthly, we classified beneficiaries according to
                                 the number of months they spent in a risk HMO or FFS during calendar years




                                 30
                                   As applied in the context of health insurance and HMOs, the statistical concept of regression toward
                                 the mean suggests that beneficiaries join HMOs during periods of unusually good health (low cost) but
                                 at some point after enrollment experience cost increases relative to the FFS mean. This hypothesis of
                                 regression toward the mean is plausible. A beneficiary is apt to be influenced to join or avoid an HMO
                                 by his or her perceived health status. Beneficiaries who have recently experienced poor health—and
                                 incurred higher than average costs—may be reluctant to join HMOs. They may have formed a close
                                 relationship with a physician who is not part of the HMO network, fear that certain medical services
                                 might not be covered by HMOs, or simply prefer having greater choice in selecting a physician. In
                                 contrast, beneficiaries who have previously experienced unusually good health may place a higher
                                 value on the monetary benefits that HMOs often provide: zero premiums, low deductibles, low
                                 copayments, and additional benefits.
                                 31
                                  For example, the prior-use measure of enrollees’ costs could be obtained by combining data from
                                 several prior years rather than just the most recent year. Such cost estimates would still need to be
                                 adjusted to account for RTM and death costs.



                                 Page 23                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                                  Appendix I
                                  Methodology




                                  1991 and 1992.32,33 We defined beneficiaries as enrollees (in risk HMOs) if
                                  they were Medicare eligible in 1991 and were enrolled in a risk contract
                                  HMO at least 7 months in 1992. We assigned beneficiaries who died in 1992
                                  to the enrollee category if (1) they died while enrolled in a risk contract
                                  HMO and (2) it would have been feasible for them to have completed 7
                                  months enrolled in an HMO in 1992 had they lived all 12 months of 1992.34,35

                                  To estimate SACHMO, we needed to develop FFS cost estimates for those
                                  beneficiaries soon to enroll in HMOs. Therefore, we created the category of
                                  joiners, a subset of enrollees. Joiners are beneficiaries who spent at least 6
                                  months in FFS in 1991 and at least 7 months in a risk HMO in 1992.

                                  To estimate SACFFS, we used FFS costs for beneficiaries who spent at least
                                  6 months in FFS in both 1991 and 1992. Beneficiaries who died in 1992 and
                                  did not meet the criteria for inclusion in the enrollee category, but who
                                  were enrolled in FFS for at least 6 months in 1991, were assigned to the FFS
                                  category.

Prior-Year FFS Spending Used      We adjusted prior-year cost data of joiners to approximate average costs
to Estimate Base-Year Costs for   in the base year for enrollees36 because their costs (on an FFS basis) are
Each Beneficiary Category         unobserved while they are HMO enrollees.37 (See table I.1 for a summary of
                                  how we adjusted prior-use costs.) In each case, we constructed average

                                  32
                                   To analyze contract year 1995, we used enrollment data from 1991-92; for the 1996 contract year, we
                                  used 1992-93 enrollment data; for the 1997 contract year, we used 1993-94 enrollment data.
                                  33
                                   Because Medicare cost HMOs do not receive capitated payments, our analysis includes beneficiaries
                                  enrolled in such HMOs and their costs as part of the FFS sector.
                                  34
                                   Because we express the criteria for those who do not die in numbers of months, those who died in
                                  1992 might not meet the enrollment criteria to be assigned a category. However, including
                                  beneficiaries who die is important because they often incur extraordinarily high health care costs.
                                  35
                                    Beneficiaries considered disenrollees were excluded from these groupings and our analysis. We
                                  defined disenrollees as beneficiaries that either (1) were enrolled in an HMO at least 7 months in 1991
                                  and fewer than 7 months (including months deceased) in 1992 or (2) met the criteria for enrollees but
                                  then died in 1992 while not enrolled in an HMO (this is a small percentage of all enrollees who died in
                                  1992). Empirical studies have shown that these beneficiaries, once disenrolled from an HMO, have
                                  higher costs than the FFS average. Therefore, had we accounted for their costs in determining SACFFS,
                                  we would have obtained a larger disparity between the cost of HMO enrollees and FFS beneficiaries,
                                  and consequently larger estimates of excess payments to HMOs.
                                  36
                                    Although costs of FFS beneficiaries during 1992 were available, we used 1991 costs so that the FFS
                                  cost measures would be comparable to the (prior-use) costs for enrollees, which are also obtained
                                  from 1991 data.
                                  37
                                    Base-year (1992) cost data were available for FFS beneficiaries only. To maintain comparability with
                                  the joiners’ cost estimates, we also obtained the costs of FFS beneficiaries from 1991 data. Thus, the
                                  1992 costs of both the joiners and FFS beneficiaries were approximated by their actual FFS costs in
                                  1991. In contrast, Medicare uses cost data from 5 consecutive years, the base year being the most
                                  recent, to approximate FFS costs. The 5-year average approach will minimize the influence of an
                                  outlier year.



                                  Page 24                                        GAO/HEHS-97-16 Medicare HMO Excess Payments
                                      Appendix I
                                      Methodology




                                      monthly costs using total Medicare claims paid and months of FFS
                                      eligibility.38 The assumptions and adjustments we made to assign costs to
                                      the enrollee category of beneficiaries are described in the following
                                      sections.

Table I.1: How HMO Enrollee and FFS
Beneficiary Costs Were Estimated,                                                                 Cost estimate
Sample Year 1992                                                                                           Adjustment to cost
                                      Beneficiary group                  Cost measure                      measure
                                      HMO enrollees                      1991 costs of people who          Costs increased to account
                                                                         joined an HMO in 1992             for RTM effect
                                                                         (joiners)
                                      FFS beneficiaries                  1991 costs of all FFS             None
                                                                         beneficiaries
                                      People who died within the sample year (1992)
                                      HMO enrollees                      Costs of people who died          None
                                                                         in FFS in 1991
                                      FFS beneficiaries                  Costs of people who died          None
                                                                         in FFS in 1991



Joiners’ Prior-Use Costs              In estimating SACHMO, we used the prior-use costs of joiners as a baseline
Used to Estimate All HMO              in estimating the (unobserved) expected FFS costs of all HMO enrollees.
Enrollees’ Costs                      Adjusting these baseline costs for regression toward the mean and death
                                      costs translates the joiners’ costs into enrollees’ costs.

                                      Our analysis of HMO enrollees from several years suggested that new HMO
                                      enrollees (joiners) in a given year tend to be similar—in terms of cost
                                      histories prior to joining an HMO—to longer-term HMO enrollees. Therefore,
                                      we assumed that enrollees’ costs could be estimated by adjusting joiners’
                                      costs for expected cost changes after enrollment. This assumption enabled
                                      us to estimate costs for all HMO enrollees on the basis of a subset who had
                                      FFS costs in the prior year. (If the data had not supported this assumption,
                                      we would have had to collect FFS costs on all HMO enrollees prior to their
                                      enrollment. Because some enrollees had been HMO enrollees for several
                                      years while Medicare eligible, this more comprehensive task would have
                                      required complex adjustments to account for changes in price levels,
                                      medical practice patterns, and technology across years. In fact, such an


                                      38
                                       Because the demographic characteristics of each group of beneficiaries may be different, and
                                      because health care costs vary by those characteristics, it would be inappropriate to compare average
                                      costs between groups without controlling for such demographic differences. Therefore, the average
                                      cost estimates of all groups were made comparable, or normalized, by dividing each group average
                                      cost by the group’s risk-adjustment factor—as determined by HCFA. In effect, each cost estimate
                                      corresponds to a representative individual within the group who has a risk adjustment factor of 1.0.



                                      Page 25                                       GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix I
Methodology




approach would not have been possible for beneficiaries who enrolled in
an HMO upon becoming Medicare eligible.)

We tested our assumption that joiners’ costs—with some
adjustments—are representative of enrollees’ costs by examining joiners’
costs over several years. Noting that most enrollees were joiners in earlier
years,39 we examined whether the relationship of joiners’ costs in the base
year to average costs of those remaining in the FFS system was similar to
the relationship of joiners’ costs in earlier years, relative to FFS
beneficiaries’ costs. We found that the ratio of joiners’ to FFS beneficiaries’
costs remained relatively stable over time. Therefore, we concluded that
joiners’ costs (in the base year) are representative of the
just-prior-to-enrollment costs of enrollees from many years before the
base year.40

The ratio of joiners’ costs to FFS beneficiaries’ costs showed no trend and
did not differ greatly from year to year. In fact, in all the years we
examined, the ratio varied by less than 10 percent of its 3-year average.41
This suggests that, relative to FFS beneficiaries, soon-to-be HMO enrollees in
1992 and 1993 (who constituted about 25 percent of all HMO enrollees in
1994) were very similar to soon-to-be HMO enrollees in 1994. Ratios for
each of three California counties for the years 1992 through 1994 are
shown in table I.2.42




39
 Beneficiaries who enrolled in a risk contract HMO immediately upon becoming eligible for Medicare
were excluded from our joiner group because their costs were not observable until or unless they
disenrolled. These “age-ins” composed about 24 percent of all new HMO enrollees in California during
1992-94. These age-ins may be included as enrollees in the following year when they meet the enrollee
criteria. For the purposes of our analysis, we assumed that the costs of age-ins, when they became
enrollees, were like those of all other HMO enrollees. (That is, they resembled joiners from earlier
years.) We based this assumption on the fact that death rates for 65-year-old FFS beneficiaries are
about 25 percent higher than for 65-year-old risk contract program age-ins. This finding is consistent
with the differences (and age-related trend) in death rates we observed between joiners and FFS
beneficiaries (see table II.5).
40
  If the empirical relationship between joiners’ costs and FFS beneficiaries’ costs is not stable across
years, the prior-use costs of enrollees (from multiple prior years) could provide an alternate baseline
for enrollee costs. Moreover, this option should be considered when the number of joiners in any given
year is insufficient to obtain a reasonable estimate of baseline enrollee costs. This option would
minimize the influence of outlier observations on the baseline estimate. As noted in app. III, we found
that a minimum of 500 joiners per county appeared to provide reasonably stable baseline average cost
measures. Furthermore, counties below that threshold did not display significant excess payments.
41
  The variation in cost ratios was greatest for Sacramento, the smallest county in terms of HMO
enrollment. This suggests that our method to estimate excess payments may be less precise for
low-enrollment counties than for high-enrollment counties.
42
  App. III describes our data set.



Page 26                                        GAO/HEHS-97-16 Medicare HMO Excess Payments
                                       Appendix I
                                       Methodology




Table I.2: Ratios of Monthly Average
Costs of New Risk HMO Enrollees to                                                                  Year
FFS Beneficiaries’ Costs for Three                                         1992                 1993                  1994                 1995
California Counties, 1992-94
                                       Los Angeles
                                         Joiners                           $161                 $184                  $189                 $178
                                         FFS
                                         beneficiaries                       333                  362                  399                   365
                                         Ratio                               .48                   .51                  .47                  .49
                                       San Diego
                                         Joiners                             162                  195                  191                   183
                                         FFS
                                         beneficiaries                       285                  315                  342                   314
                                         Ratio                               .57                   .62                  .56                  .58
                                       Sacramento
                                         Joiners                             159                  204                  198                   187
                                         FFS
                                         beneficiaries                       268                  298                  318                   295
                                         Ratio                               .59                   .68                  .62                  .63
                                       Note: To reduce the computational burden for the purposes of this example, we did not normalize
                                       these cost measures to reflect the costs of an average beneficiary, and we excluded the costs of
                                       the disabled and of the FFS and joiner beneficiaries who died in the year of reference.
                                       Normalizing these cost measures would bring the FFS costs closer to the joiners’ costs. On the
                                       basis of our other analyses, however, we believe that normalization would only increase the ratio
                                       levels by about .1, which would not significantly alter the cost relationships of FFS to joiners either
                                       across years or counties.




Prior-Use Costs of Joiners             After a beneficiary joins an HMO, it is hypothesized that the beneficiary’s
Adjusted for                           cost is likely to increase relative to his or her FFS costs in the year prior to
Regression-Toward-the-                 enrolling. Such cost increases seem likely for two reasons. First,
                                       beneficiaries may postpone discretionary care in the months prior to
Mean Effect                            joining an HMO so that they can take advantage of HMOs’ typically lower
                                       copayments. Second, beneficiaries may be more likely to join HMOs during
                                       a spell of unusually good health. This expectation that costs increase is
                                       known as “regression toward the mean” (RTM). To the extent that RTM
                                       occurs, unadjusted prior-use costs of joiners understate the initial average
                                       health care costs of new HMO enrollees, as well as the costs of all HMO
                                       enrollees.

                                       HCFA’s method for determining HMO capitation rates implicitly assumes that
                                       RTM is full (100 percent) and immediate. That is, HCFA assumes that, upon
                                       enrolling in an HMO, joiners’ costs immediately increase to equal the
                                       average cost of FFS beneficiaries. Although it is reasonable to expect some



                                       Page 27                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                            Appendix I
                            Methodology




                            RTM,no evidence supports a 100-percent effect that occurs so soon after
                            enrollment.

                            We estimated the degree of RTM likely to occur and used this estimate to
                            adjust joiners’ prior-use costs so they more accurately represented all
                            enrollees’ costs. We derived our estimate of the regression effect, which
                            we term the “regression-toward-the-mean adjustment factor” (RTMF), from
                            actual FFS cost data for beneficiaries whose cost and demographic
                            characteristics resembled those of joiners and from the actual distribution
                            of enrollees’ HMO tenure. Our analysis of 1995 data suggested that the RTMF
                            was about half of the maximum potential effect—50 percent, as opposed
                            to the 100-percent RTMF that HCFA’s methodology implicitly assumes. (For
                            further discussion of the RTMF, see app. II.)


Prior-Use Costs Adjusted    Because new HMO enrollees, by definition, do not die during the period just
for Death Costs             prior to their enrollment, prior-use cost data understate the costs of HMO
                            enrollees who die during the year. The costs associated with the final
                            months of life—“death-related costs”—are typically substantial.
                            Consequently, we accounted for them to avoid underestimating SACHMO.
                            We assumed that the costs of an HMO enrollee who died equal the costs of
                            an FFS beneficiary who died. To find the average cost estimate for the
                            deceased, we divided the calendar year total costs of all FFS beneficiaries
                            deceased in 1991 in each county by the number of months those
                            beneficiaries were alive during the year.

                            Our adjustment was equivalent to imposing a 100-percent RTM effect on the
                            costs of HMO enrollees who died during the base year. Because favorable
                            selection can result in HMOs’ having lower mortality rates than FFS, we
                            imputed death costs only for HMO enrollees who died during the year. This
                            approach accounted for excess payments to HMOs in counties where
                            mortality rates were lower in HMOs than in FFS.


Calculating County-Rate     After estimating the average expected costs of serving all of a county’s
Excess Payments That Are    beneficiaries in FFS (SACALL), we could estimate the excess capitation
Due to Using Only FFS       payments that resulted from HCFA’s method of calculating SAC and the
                            county rate. The formula for determining capitation rates can be
Beneficiaries’ Experience   expressed as the following:
to Set Rates

Equation 3




                            Page 28                            GAO/HEHS-97-16 Medicare HMO Excess Payments
                        Appendix I
                        Methodology




                        However, HCFA estimates average costs using only beneficiaries actually in
                        FFS, so that HCFA’s formula is actually this:



Equation 4




                        Consequently, the excess capitation rate can be estimated by the
                        following:


Equation 5




                        The risk factor term is specific to individual beneficiaries. On the basis of
                        their demographic characteristics, it can take on values greater or less
                        than 1.0. The total of county rate excess payments for a given county is
                        obtained by summing the individual level excess payment amounts,
                        expressed by equation 5. We applied this methodology to California’s 58
                        counties to estimate county-rate excess payments for 1995, 1996, and 1997.
                        Our estimates are presented in appendix III.


                        This section describes the steps we followed to estimate aggregate excess
Method for Estimating   payments to HMOs, that is, total excess payments caused by the full effect
Medicare’s Aggregate    of favorable selection on the rate-setting formula. Our method compares
Excess Payments         what Medicare paid for risk contract HMO enrollees to what Medicare
                        would have paid for the same enrollees had they not joined HMOs. Although
                        this method establishes a benchmark for excess payments against which
                        HMO payment reforms can be measured, we do not suggest that HCFA use
                        the methodology described below to adjust capitation rates because it was
                        not designed or tested as a rate-setting methodology.43




                        43
                          In order to use this method to adjust rates, HCFA would need data that only become available after
                        the contract year; hence, the method would have to be applied retroactively. Because the current
                        payment method is prospective, such a change in approach could have consequences for the operation
                        of the program that are not yet well understood.



                        Page 29                                      GAO/HEHS-97-16 Medicare HMO Excess Payments
             Appendix I
             Methodology




                  Step 1

             We estimated the average cost of HMO enrollees (ACHMO) using the same
             prior-use approach described above. After our adjustments for RTM and
             death-related costs were applied, ACHMO was representative of the costs of
             a group of HMO enrollees with the demographic characteristics of new HMO
             enrollees (joiners).44

                  Step 2

             We used HCFA’s method to calculate a county average capitation rate.
             Because ACHMO reflected the demographic characteristics of only
             joiners, we calculated the average capitation rate for the joiner population
             (CAP_RATEJAVG) so that it, too, reflected the demographic characteristics
             of only joiners. Specifically, we adjusted the 1995 county rate up or down
             according to the average risk factor of that county’s joiners.

                  Step 3

             We calculated the percent aggregate excess payment (PAEP) to risk
             contract HMOs in each county using the following formula:

Equation 6




             CAP_RATEJAVG and ACHMO reflect the demographic characteristics only of
             joiners, but the cost characteristics of all HMO enrollees. Because these
             terms affect both the numerator and denominator, PAEP is demographically
             neutral—that is, demographic characteristics are canceled out in the
             expression.

             To find aggregate excess payments that corresponded to actual HMO
             enrollees, we multiplied PAEP by total payments to risk HMOs by county.

             We applied this methodology to estimate aggregate excess payments to
             HMOs in California’s 58 counties in 1995. (See app. III.)




             44
               We used 1994-95 data to define joiners, enrollees, and FFS beneficiaries for this analysis.



             Page 30                                          GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix II

Adjustments for Regression Toward the
Mean and Death-Related Costs in Estimating
Excess Payments to Medicare HMOs
              As explained in appendix I, establishing the Medicare capitation rate for
              HMOs on the basis of the cost of serving beneficiaries hinges on estimating
              the expected FFS costs of HMO enrollees (SACHMO). In turn, adequately
              estimating SACHMO requires adjusting HMO enrollees’ observed prior-use
              costs for the increases expected to occur after they enroll. This increase
              has been labeled regression toward the mean because enrollees’ average
              health costs, which are relatively low before joining the HMO, begin to rise
              over time and approach (“regress” toward) the average cost of similar
              beneficiaries who remain in FFS. This appendix describes our methodology
              to account for the RTM effect, including the high health care costs typically
              incurred during the last months of life. Although we drew on previous
              studies, available data required that we develop a new method of adjusting
              prior-use estimates of enrollees’ costs for RTM.

              HCFA  implicitly assumes than HMO enrollees’ costs fully regress (increase)
              to the mean of FFS immediately upon enrollment. Studies have generally
              found that, after a beneficiary enrolls in an HMO, his or her service use and
              costs rise. Nonetheless, HCFA’s assumption that RTM is full and immediate
              receives no empirical support in the literature.45 For example, Beebe found
              significant increases in the first year after enrollment and moderate
              increases thereafter. After 3 years, estimated costs of HMO enrollees were
              94 percent of those of comparable FFS beneficiaries; by year 6, enrollees’
              estimated costs had risen modestly to 96 percent of FFS beneficiaries’
              costs.46 A more recent study by Hill and others found that RTM closed half
              the gap in costs between HMO joiners and FFS beneficiaries.47




              45
                Studies do differ, however, in their estimates of how fully and rapidly the costs of HMO enrollees
              regress toward the mean. While some have found that differences in cost between enrollees and the
              FFS population rapidly shrink after enrollment, others have found that initial cost differences are quite
              persistent. (See James Beebe, “Medicare Reimbursement and Regression to the Mean,” Health Care
              Financing Review, 9 (3) (spring 1988), p. 9.)
              46
                J. Beebe, “Medicare Reimbursement and Regression to the Mean,” pp. 9-22. This study estimates RTM
              by tracking over time the costs of a “proxy joiner cohort”—that is, a group of beneficiaries who
              resemble new HMO enrollees but remain in FFS.
              47
               J. Hill, R. Brown, D. Chu, and J. Bergeron, The Impact of the Medicare Risk Program on the Use of
              Services and Costs to Medicare, report to HCFA (Washington, D.C.: Mathematica Policy Research, Inc.,
              Dec. 3, 1992). This study derives an estimate of RTM by comparing the estimated cost ratio of all
              enrollees with that of joiners. Joiners’ costs were estimated by prior use, and enrollees’ costs, by a
              survey of service use.



              Page 31                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                     Appendix II
                     Adjustments for Regression Toward the
                     Mean and Death-Related Costs in Estimating
                     Excess Payments to Medicare HMOs




                     We allow our estimate of RTMF to differ between groups of beneficiaries,
Methodology Allows   depending on whether they survived or died during the 4-year period that
RTM Factor to Vary   we analyzed. The association between mortality and average costs is well
by Beneficiary       documented by previous studies. For example, Lubitz and others found
                     that people in their last 12 months of life have costs that are significantly
Survival Status      higher than those of other Medicare beneficiaries and account for a
                     disproportionate share (about 28 percent) of health care expenditures.
                     Similarly, average costs during the final 2 and 3 years of life, while not as
                     large, are also considerably higher than the average for all beneficiaries.48
                     This pattern is illustrated in figure II.1.




                     48
                      See J. Lubitz, J. Beebe, and C. Baker, “Longevity and Medicare Expenditures,” New England Journal
                     of Medicine, 332 (15) (1995), pp. 999-1003; J. Lubitz and R. Prihoda, “Medicare Services in the Last 2
                     Years of Life,” Health Care Financing Review, 5(3) (1984), pp. 117-31; J. Lubitz and G. Riley, “Trends in
                     Medicare Payments in the Last Year of Life,” New England Journal of Medicine, 328(15) (1993), pp.
                     1092-96.



                     Page 32                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                                          Appendix II
                                          Adjustments for Regression Toward the
                                          Mean and Death-Related Costs in Estimating
                                          Excess Payments to Medicare HMOs




Figure II.1: Annual Medicare Payments in the Years Preceding Death

1990 Dollars (in Thousands)

16

15

14

13

12

11

10

 9

 8

 7

 6

 5

 4

 3

 2

 1

 0
     9       8            7     6         5           4          3           2          1           0
     Years Before Death


                                          Note: Figure shows costs for people who died at age 75.

                                          Source: J. Lubitz, J. Beebe, and C. Baker, “Longevity and Medicare Expenditures.”




                                          Page 33                                     GAO/HEHS-97-16 Medicare HMO Excess Payments
                                    Appendix II
                                    Adjustments for Regression Toward the
                                    Mean and Death-Related Costs in Estimating
                                    Excess Payments to Medicare HMOs




                                    The relationship between the degree of RTM experienced by HMO enrollees
                                    and their proximity to death has not been addressed by previous studies.
                                    Nonetheless, it is possible that enrollees surviving different lengths of time
                                    after joining an HMO would experience different degrees of RTM. For
                                    example, it is plausible that HMO enrollees in their last year of life might
                                    experience complete RTM, while those many years from death might
                                    experience little.

                                    In our analysis, we allowed for the possibility that the appropriate RTM
                                    adjustment for a group of beneficiaries may depend on their proximity to
                                    death. Table II.1 presents the definitions of the beneficiary categories and
                                    the percentage of HMO enrollees (for California in sample year 1992) in
                                    each category.

Table II.1: Classification of HMO
Enrollees by Survival Status                                                                                  Percentage of all HMO
                                    Category of enrollee               Status                                            enrolleesa
                                    I                                  Survived 4 or more years                                     83.8
                                    II                                 Survived at least 1 year but
                                                                       less than 4 years                                            12.9
                                    III                                Survived less than 1 year                                         3.3
                                    a
                                    Percentages are based on 1992 Medicare risk HMO enrollees in California and include those
                                    who disenrolled in subsequent years.

                                    Source: GAO analysis of HCFA data on Medicare beneficiaries.




                                    To estimate RTMF for enrollees who survive for 4 or more years (category I
Method Used to                      enrollees), we developed an approach that generally follows Beebe’s 1988
Estimate the RTM                    methodology. That is, we used 4 years of longitudinal data on a sample of
Factor for Category I               the FFS Medicare population to track the cost experience over time of two
                                    proxy cohorts—one representing HMO joiners and one representing FFS
Enrollees                           beneficiaries. Our method involved four steps.

                                    1. We randomly drew two samples—one reflecting the distribution of age,
                                    sex, and costs of new HMO enrollees (joiners)49 and the second reflecting
                                    the distribution of age, sex, and costs of beneficiaries who remained in FFS.

                                    2. We then computed, for each of 4 years, the ratio of the average annual
                                    cost of the proxy HMO joiners to the cost of the proxy FFS beneficiaries.

                                    49
                                     In app. I, we defined new HMO enrollees (joiners) as beneficiaries with 6 or more months of FFS
                                    experience in the prior year and 7 or more months of HMO experience in the year that they join the
                                    HMO.



                                    Page 34                                       GAO/HEHS-97-16 Medicare HMO Excess Payments
                         Appendix II
                         Adjustments for Regression Toward the
                         Mean and Death-Related Costs in Estimating
                         Excess Payments to Medicare HMOs




                         3. Next, we used these cost ratios to estimate how rapidly and fully the
                         costs of HMO joiners converged toward those of FFS beneficiaries.

                         4. Finally, we combined the cost ratios with data on HMO enrollees’ tenure
                         within each county to produce a county-specific RTMF.


Description of FFS       We assembled a longitudinal data set that contained the claims for
Beneficiary Data Set     approximately 1.4 million California beneficiaries who were continuously
                         enrolled in FFS Medicare between 1991 and 1994. Only beneficiaries who
                         were eligible for part A and part B and who remained in the FFS sector for
                         the entire 4-year period were included.50 People under age 65 who were
                         eligible for Medicare because of a disability and people with end-stage
                         renal disease were excluded.


Methodology for          We constructed two proxy cohorts, one with the same demographic mix
Constructing the Proxy   and 1991 service cost distribution as the Medicare HMO joiners, and the
HMO Joiner and Proxy     other with the demographics and cost distribution of continuing FFS
                         beneficiaries. To do this, we divided the FFS data set into 10 age and sex
FFS Cohorts              subgroups51 and further divided each subgroup into 25 smaller strata
                         according to the cost of services they received in 1991. We then selected
                         two stratified random samples—one for each proxy cohort—from each
                         demographic subgroup. We limited each sample to 20 percent of the size
                         of its corresponding demographic subgroup within the FFS data set. The
                         sample sizes within each cost stratum were determined by the actual cost
                         distribution of HMO joiners and continuing FFS beneficiaries.

                         Table II.2 lists the cost strata for one demographic subgroup: females aged
                         65 to 69. Columns 2 and 3 show the percent distribution of the actual FFS
                         and joiner populations across 25 cost categories. For example, among
                         females aged 65 to 69, 19.2 percent of the FFS population and 39.9 percent
                         of the joiner population had no Medicare charges in 1991.




                         50
                          We excluded those who died during the 1991 through 1994 period from our analysis. Our treatment of
                         people who die within 4 years of enrollment is discussed in the following sections pertaining to
                         category II and III enrollees.
                         51
                          These groups are (1) male, aged 65-69; (2) female, 65-69; (3) male, 70-74; (4) female, 70-74; (5) male,
                         75-79; (6) female, 75-79; (7) male, 80-84; (8) female, 80-84; (9) male, 85+; and (10) female, 85+.



                         Page 35                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                                          Appendix II
                                          Adjustments for Regression Toward the
                                          Mean and Death-Related Costs in Estimating
                                          Excess Payments to Medicare HMOs




Table II.2: 1991 Distribution Across Cost Categories of HMO Joiners and FFS Beneficiaries, 65- to 69-Year-Old Females
                                                                                    Number of beneficiaries
                                   Percentage distribution of              Longitudinal
                                         beneficiaries                    population of           Proxy FFS       Proxy joiner
Cost                                     FFS                   Joiner            beneficiariesa          cohort sample            cohort sample
$0                                       19.2                    39.9                     37,595                    8,362                   17,392
1-99                                      9.4                      9.8                    20,264                    4,104                    4,267
100-199                                   8.5                      7.9                    18,329                    3,687                    3,439
200-299                                   7.4                      6.1                    15,981                    3,213                    2,646
300-399                                   6.2                      4.6                    13,584                    2,690                    2,021
400-599                                   9.1                      6.9                    20,228                    3,963                    3,004
600-799                                   6.2                      4.6                    13,832                    2,692                    2,011
800-999                                   4.5                      2.9                    10,066                    1,956                    1,252
1,000-1,499                               7.1                      4.6                    16,059                    3,083                    2,004
1,500-1,999                               3.9                      2.4                     8,869                    1,699                    1,035
2,000-2,499                               2.6                      1.4                     5,807                    1,112                         590
2,500-2,999                               1.9                      1.2                     4,411                      840                         535
3,000-3,499                               1.5                      0.9                     3,356                      638                         404
3,500-3,999                               1.2                      0.7                     2,678                      507                         297
4,000-4,499                               1.0                      0.6                     2,263                      432                         262
4,500-4,999                               0.9                      0.5                     2,091                      395                         207
5,000-5,999                               1.6                      0.8                     3,608                      677                         362
6,000-6,999                               1.2                      0.7                     2,787                      522                         304
7,000-7,999                               0.9                      0.6                     2,049                      391                         255
8,000-9,999                               1.3                      0.6                     2,944                      558                         279
10,000-14,999                             1.9                      1.1                     4,312                      811                         476
15,000-24,999                             1.7                      0.7                     4,011                      742                         321
25,000-49,999                             1.0                      0.5                     2,377                      440                         214
50,000-74,999                             0.2                      0.0                        379                       70                        21
75,000-99,999b                            0.0                      0.0                         83                       15                         7
                                                                                                                           c
Total                                     100                     100                    217,963                   43,599                   43,605c

                                          a
                                              Composed of people in Medicare FFS for 48 consecutive months, from 1991 through 1994.
                                          b
                                           Because of insufficient representation in the population, beneficiaries with costs in the first year
                                          of $100,000 or more were excluded from the analysis.
                                          c
                                            The totals in columns 5 and 6 each represent 20 percent of the total in column 4, which is the
                                          entire category of 65- to 69-year-old female beneficiaries. The slight difference between column 5
                                          and column 6 totals is due to rounding error associated with sampling the 25 cost strata.




                                          Page 36                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                           Appendix II
                           Adjustments for Regression Toward the
                           Mean and Death-Related Costs in Estimating
                           Excess Payments to Medicare HMOs




Ratio of Proxy HMO         Within each demographic group, we calculated the ratio of the proxy HMO
Joiners’ Costs to Proxy    joiner cost average to the proxy FFS cost average for each of 4 years (1991
FFS Beneficiaries’ Costs   through 1994). The results are presented in figure II.2, which shows that
                           the pattern of changes in the cost ratios over time displays a high degree of
                           consistency across demographic groups.




                           Page 37                                GAO/HEHS-97-16 Medicare HMO Excess Payments
                                         Appendix II
                                         Adjustments for Regression Toward the
                                         Mean and Death-Related Costs in Estimating
                                         Excess Payments to Medicare HMOs




Figure II.2: Regression-Toward-the-Mean Patterns for 10 Demographic Groups of Proxy HMO Enrollees
Proportion of FFS Average Cost




 1




                                                                     J                        I
                                                                                              G
                                                                                 H
0.9                                                                                           E

                                                                 F                            C
                                                                                              D
                                                                                              A


0.8
                                                                                              B




0.7
                                                                         A    Male, 65-69
                                                                         B    Female, 65-69
                                                                         C    Male, 70-74
                                                                         D    Female, 70-74
                                                                         E    Male, 75-79
                                                                         F    Female, 75-79
0.6                                                                      G    Male, 80-84
                                                                         H    Female, 80-84
                                                                         I    Male, 85+
                                                                         J    Female, 85+



0.5
 1991                       1992                          1993                          1994
 Year




                                         Page 38                                GAO/HEHS-97-16 Medicare HMO Excess Payments
                                         Appendix II
                                         Adjustments for Regression Toward the
                                         Mean and Death-Related Costs in Estimating
                                         Excess Payments to Medicare HMOs




                                         The weighted average (across demographic groups) of these cost ratios is
                                         shown in table II.3.52 These ratios show how rapidly and fully the costs of
                                         the overall proxy HMO joiner cohort are likely to converge toward the costs
                                         of the proxy cohort in FFS.

Table II.3: Costs of Proxy HMO Joiners
Relative to Those of Proxy FFS                                                                      Tenure in HMO (in years)
Beneficiaries, 1991-94                                                               Year prior
                                                                                             to
                                                                                    enrollment           Year 1          Year 2         Year 3
                                         Cost ratio                                    (1991)a           (1992)          (1993)         (1994)
                                         Proxy HMO/proxy FFS                                  .64            .85             .88            .90
                                         a
                                          As in our modified prior-use methodology for estimating excess payments, the year prior to
                                         enrollment is the benchmark for estimating HMO enrollee costs.



                                         These cost ratios show that HMO enrollee costs (represented by proxy HMO
                                         joiners’ costs) are about two-thirds of comparable FFS beneficiary costs in
                                         the year before enrollment, suggesting significant favorable selection.
                                         However, once beneficiaries enroll, their costs are expected to increase
                                         significantly relative to FFS costs in the first year; the proxy HMO cohorts’
                                         costs rose from 64 percent to 85 percent of FFS cost. In the second year of
                                         HMO enrollment, enrollees’ relative costs are expected to rise moderately,
                                         and they did—from 85 percent to 88 percent. In the third year, enrollees’
                                         relative costs are expected to show a further, slight increase. By the end of
                                         the third year, enrollees’ expected costs—as represented by their proxy
                                         cohort’s costs—had regressed about 71 percent; the difference between
                                         enrollees’ costs and those of FFS beneficiaries had declined from
                                         36 percent to 10 percent. The slight increases in the proxy enrollees’ costs
                                         (relative to the FFS beneficiaries’ costs) after the first year suggest that
                                         complete regression either will not occur or will take many years.53


Calculating the RTMF                     We used the information on the joiners’ estimated cost increases over time
From the Estimated Cost                  (presented in table II.3) to construct an RTMF for each county. Table II.4
Ratios                                   illustrates the calculations for a hypothetical county (based on California
                                         data). First, we used our estimates to calculate the increase in expected

                                         52
                                          The weights are assigned according to the proportion of the actual HMO joiner group that is
                                         accounted for by each demographic group.
                                         53
                                          Several peer reviewers commented that, because proxy HMO enrollees are drawn from the FFS
                                         population, our method is conservative and may somewhat overestimate the degree of RTM. Our
                                         proxy HMO enrollees are, after all, FFS beneficiaries who chose not to join an HMO. If their reason for
                                         not joining an HMO was health-related, one could expect their costs (within each 1991 cost stratum) to
                                         exhibit greater increases over time than those of actual HMO enrollees.



                                         Page 39                                        GAO/HEHS-97-16 Medicare HMO Excess Payments
                                       Appendix II
                                       Adjustments for Regression Toward the
                                       Mean and Death-Related Costs in Estimating
                                       Excess Payments to Medicare HMOs




                                       FFScosts of people who had been enrolled in an HMO for 1, 2, or 3 or more
                                       years—relative to their prior-use costs. (See table II.4, row 1.) Computing
                                       a weighted average of these increases—where the weights reflect the
                                       tenure distribution of HMO enrollees in a given county—yielded a county’s
                                       RTMF. (A tenure distribution representative of all California counties is
                                       presented in table II.4, row 2.) The RTMF of 1.40 combines information
                                       about how quickly and fully RTM occurs (row 1) with these data on the
                                       tenure of HMO enrollees.

Table II.4: Example of Derivation of
Regression-Toward-the-Mean                                                                                  Number of years in HMO
Adjustment Factor From Cost Ratios     Measure                                                                   1               2     3 or more
                                       Benchmark cost proportion: the cost ratio for
                                       each year divided by the cost ratio for the year
                                       prior to enrollmenta                                                  1.33             1.38            1.41
                                       Tenure distribution: proportion of HMO enrollees
                                       for the county (from actual enrollment data)b                           .11             .18             .71
                                       RTMF: a weighted average of benchmark cost
                                       proportions, using the tenure distribution as
                                       weightsc                                                                               1.40
                                       a
                                           For example, 1.38=.88/.64.
                                       b
                                        The values shown here are for illustration. They represent the tenure distribution of enrollees for
                                       all California counties in 1993.
                                       c
                                         This number is for a hypothetical county: RTMF = (.11 • 1.33) + (.18 • 1.38) + (.71 • 1.41) =
                                       1.40. We constructed actual RTMF values for each county in each year on the basis of tenure in
                                       that county in the year.

                                       Source: GAO calculations based on HCFA Medicare claims and enrollment data for 1992.




                                       We could not estimate an RTMF for category II enrollees with the method
Method Used to                         that we used for category I enrollees. That method requires constructing
Estimate the RTM                       proxy cohorts of HMO joiners and FFS beneficiaries, but the number of
Factor for Category II                 category II enrollees—those who survive between 1 year and 4 years after
                                       enrollment—was insufficient to do so.
Enrollees
                                       We chose to assume full RTM for the year a joiner died and to apply our
                                       estimate of RTMF for category I enrollees to category II enrollees prior to
                                       the year they died. Research indicates that individuals’ costs tend to rise
                                       most sharply in the months before death,54 so we assumed the costs of


                                       54
                                        The average cost of FFS beneficiaries who will live for 3 or more years (alive in 1995) is about
                                       one-fifth the average of those FFS beneficiaries in their final (calendar) year of life (that is, those who
                                       died in 1991). This finding is consistent with the work of Lubitz and others. See footnote 47.



                                       Page 40                                          GAO/HEHS-97-16 Medicare HMO Excess Payments
                          Appendix II
                          Adjustments for Regression Toward the
                          Mean and Death-Related Costs in Estimating
                          Excess Payments to Medicare HMOs




                          category II enrollees in their year of death regressed fully to the mean of
                          FFS beneficiaries’ costs. With respect to the year or years before this last
                          year of life, when individuals’ costs generally rise less sharply, we applied
                          the category I RTMF estimate to category II enrollees, which represented a
                          significant increase in prior-use costs. If these assumptions over- or
                          underestimate the RTMF for category II enrollees, the effect on the estimate
                          of the county adjusted average per capita cost (AAPCC) rate will be quite
                          small, given the limited number of category II enrollees.55


                          The average costs of HMO joiners in the year of their death (in this case
The RTM Factor for        1991) cannot be estimated. After all, joiners must live beyond the prior-use
Category III Enrollees    year (1991) to become HMO enrollees. This means that we lacked data to
                          estimate the extent to which category III enrollees’ average costs (in the
                          year of their death) might remain below the costs of comparable FFS
                          beneficiaries. Consequently, to account for enrollees’ death-related costs
                          that prior-use estimates cannot capture, we assigned to HMO enrollees who
                          died in 1992 the costs of FFS beneficiaries with comparable demographic
                          characteristics who died in 1991. Similarly, we used the costs of FFS
                          beneficiaries who died in the prior-use year to approximate the costs of
                          FFS beneficiaries who died in the sample year (1992). By setting the
                          death-related costs of HMO enrollees equal to those of FFS beneficiaries, we
                          assumed that, among category III enrollees, RTM in costs was complete.


Favorable Selection       Although our method for estimating excess payments to HMOs assumed
Indicated by Relatively   that no difference existed in death-related costs between HMO and FFS
Low HMO Death Rates       enrollees, it did not assume that the respective death rates were equal. As
                          table II.5 shows, the death rates (per 100) of beneficiaries enrolled in HMOs
                          are significantly lower than those of beneficiaries in FFS. This finding is
                          consistent over time and across demographic groups. The lower death
                          rates among HMO enrollees are a measure of favorable selection.
                          Consequently, these lower death rates are partly responsible for the
                          findings of excess payments to HMOs reported in appendix III.




                          55
                           HCFA may have sufficient national data on category II enrollees to empirically estimate the RTM
                          effect on these enrollees.



                          Page 41                                       GAO/HEHS-97-16 Medicare HMO Excess Payments
                                            Appendix II
                                            Adjustments for Regression Toward the
                                            Mean and Death-Related Costs in Estimating
                                            Excess Payments to Medicare HMOs




Table II.5: Death Rates, per 100, of Aged Medicare Beneficiaries by Demographic Group and Year, 1992-94
                                          1992                            1993                                         1994
Demographic                          FFS                HMO                  FFS                HMO                  FFS                 HMO
Male
65-69                                 2.8                  2.1                2.8                 2.1                 2.8                 2.1
70-74                                 3.9                  3.1                3.9                 2.9                 3.8                 2.9
75-79                                 6.2                  4.6                6.1                 4.6                 6.2                 4.6
80-85                                 9.6                  7.0                9.7                 7.1                 9.8                 7.1
85+                                  16.9                12.3                16.9                12.3                17.8                12.7
Female
65-69                                 1.7                  1.2                1.7                 1.1                 1.8                 1.2
70-74                                 2.5                  1.7                2.5                 1.7                 2.6                 1.7
75-79                                 4.0                  2.7                4.0                 2.7                 4.3                 2.7
80-85                                 6.2                  4.2                6.5                 4.2                 6.7                 4.2
85+                                  13.3                  8.7               13.9                 8.7                14.6                 9.1
Weighted meana                        5.2                  3.7                5.2                 3.6                 5.2                 3.5
                                            a
                                             To control for differences in the demographic composition of the FFS and HMO populations,
                                            population group means are weighted by the proportion of the FFS population in each
                                            demographic group.




                                            We summarize below the source of empirical evidence we used to
Summary of                                  estimate the RTM experience for each category of enrollee, and how this
Adjustments for RTM                         evidence was used to arrive at a corresponding RTM adjustment factor.


Category I Enrollees                        We used FFS data on cohorts of beneficiaries whose costs and
                                            demographic characteristics were comparable with those of HMO enrollees
                                            to simulate their RTM experience. On the basis of this simulation, we
                                            estimated an RTMF (a numerical factor) to adjust the average cost of
                                            category I enrollees upward.


Category II Enrollees                       Because of insufficient sample size of cost strata, we could not conduct a
                                            simulation of proxy HMO enrollees’ costs to estimate an RTMF. However,
                                            research indicates that individuals’ costs tend to rise most sharply in the
                                            months before death. Consequently, we assumed these enrollees’ costs
                                            regressed fully to the mean of FFS beneficiaries’ costs. With respect to the
                                            year or years before the last year of life (when costs generally rise less
                                            sharply), we applied the category I RTMF estimate to category II enrollees.



                                            Page 42                                     GAO/HEHS-97-16 Medicare HMO Excess Payments
                         Appendix II
                         Adjustments for Regression Toward the
                         Mean and Death-Related Costs in Estimating
                         Excess Payments to Medicare HMOs




Category III Enrollees   We could not conduct a category I-type simulation. Prior-use data
                         provided only limited insight on the RTM experience for these enrollees.
                         Consequently, we assumed that the costs of category III enrollees
                         displayed complete RTM, that is, that their costs in the sample year were no
                         different on average than costs for comparable FFS beneficiaries.

                         By making these RTM-related adjustments to our prior-use-based estimates
                         of HMO enrollees’ costs, we significantly lowered our estimates of HMO
                         excess payments from what they would have been otherwise. Appendix III
                         presents estimates of excess payments affected by the RTM adjustments
                         described above.




                         Page 43                                GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix III

Estimates of Medicare Excess Payments to
HMOs in California

               This appendix discusses our estimates of the amount of excess payments
               Medicare has made to California HMOs that participate in its risk contract
               program, in order to indicate the size and significance of this problem in
               Medicare’s method of setting capitated rates. The appendix details the
               savings that could be realized by adopting our method to improve the
               county rate. These savings are implied by our estimates of county-rate
               excess payments for the years 1995, 1996, and 1997. The appendix also
               addresses aggregate excess payments to Medicare HMOs—the sum of
               county-rate and risk-adjuster-related excess payments—for 1995.

               To reduce the computational burden, we limited our efforts to the 58
               counties of California. Because risk contract program enrollees are
               concentrated in relatively few states,56 demonstrating the magnitude of
               excess payments did not require us to produce estimates for every county
               nationwide. We selected the counties of California because (1) about
               36 percent of all risk contract enrollees reside there, (2) rates of
               beneficiary enrollment in risk HMOs vary substantially across the 58
               counties, and (3) in recent years, California has experienced rapid growth
               in HMO enrollment. Although our estimates pertain to a large portion of the
               risk contract program, we cannot project our estimates nationwide or to
               other states with demographically similar counties.

               We constructed all our estimates from individual-level claims data,57 using
               data from two HCFA sources: (1) the Enrollment Database File (EDB)58 and
               (2) the HCFA claim files, which contain Medicare claims submitted by FFS
               providers.59 We combined individual expenditure information with EDB
               data to produce a single enrollment/expenditure file containing
               information on approximately 4.3 million California residents.


               56
                See Medicare HMOs: Growing Enrollment Adds Urgency to Fixing HMO Payment Problem
               (GAO/HEHS-96-21, Nov. 8, 1995). Two states (California and Florida) account for more than half of
               Medicare risk HMO enrollees.
               57
                 Compared with HCFA’s rate-setting method, our improvement involves greater disaggregation of the
               claims data. We needed individual-level data for a key step in estimating excess payments: isolating the
               FFS costs of beneficiaries remaining in FFS from the costs of those about to join an HMO.
               58
                The claim files contain detailed enrollment and entitlement data for all individuals who are or have
               ever been Medicare beneficiaries. Data items include age, sex, Medicare entitlement status, state and
               county of residence, and date of HMO enrollment.
               59
                We extracted claims information from seven separate files for 1991-94: inpatient hospital, outpatient,
               home health agency, skilled nursing facility, hospice, physician/supplier, and durable medical
               equipment. We obtained expenditure information from the “payment amount” portion of the claim.
               Also, following HCFA’s methodology, we added pass-through and per-diem expenses to the payment
               amount for inpatient claims. From the claim files, we computed annual expenditures for individual
               beneficiaries enrolled in the FFS program and produced separate part A and part B subtotals for the
               years 1991-94.



               Page 44                                        GAO/HEHS-97-16 Medicare HMO Excess Payments
                                            Appendix III
                                            Estimates of Medicare Excess Payments to
                                            HMOs in California




                                            Table III.1 presents estimates of county-rate excess payments in dollar
Estimates of                                amounts and as a percentage of risk contract program expenditures for
County-Rate Excess                          each county. (The estimates are weighted averages of the excess payments
Payments                                    in the rates for aged (parts A and B) and disabled (parts A and B).) The
                                            counties are ranked by excess payment amounts for 1997. We have
                                            included in table III.1 only those counties for which the number of new
                                            risk HMO enrollees exceeded 500 in the base year.60,61 With respect to the
                                            excluded counties, the county-rate excess payments (in each year) total
                                            less than 3 percent of total county-rate excess payments in the state.


Table III.1: Medicare County-Rate Excess Payments for 20 California Counties in Dollars and as a Percentage of Program
Payments, 1995-97
                                                                               County-rate excess payment as percentage of
                           County-rate excess payment amount (in millions)            risk contract program payments
County                              1995                  1996                 1997                 1995                  1996                  1997
Los Angeles                       $135.3               $119.4                $182.7                  6.56                 5.32                  7.62
San Diego                            37.3                 20.2                  57.5                 5.12                 2.43                  6.37
Orange                               38.5                 28.4                  46.5                 6.37                 4.17                  6.31
San Bernardino                       23.4                 21.1                  29.5                 5.79                 4.61                  5.99
Riverside                            17.5                 25.4                  21.3                 3.70                 4.86                  3.78
Alameda                                 •                   5.7                 12.5                     •                1.75                  2.96
Sacramento                            3.2                   4.1                 10.2                 1.62                 1.40                  2.77
Contra Costa                            •                   4.9                  9.8                     •                1.94                  2.92
Ventura                               6.6                   4.7                  8.8                 4.80                 2.91                  4.97
Santa Clara                           2.3                   4.4                  8.4                 1.18                 1.48                  2.19
Kern                                  4.4                   5.3                  4.6                 3.74                 3.67                  2.87
Sonoma                                  •                     •                  3.9                     •                    •                 2.68
Stanislaus                              •                     •                  3.8                     •                    •                 3.08
San Mateo                             2.9                   2.7                  3.7                 2.25                 1.53                  1.70
San Luis Obispo                         •                     •                  2.9                     •                    •                 4.54
San Francisco                         4.0                   1.4                  2.9                 2.44                 0.66                  1.12
Santa Barbara                           •                   2.1                  2.4                     •                2.67                  2.70
Butte                                 0.2                   0.3                  1.1                 0.79                 0.81                  2.51
Fresno                                  •                     •                  0.6                     •                    •                 0.79
                                                                                                                                        (continued)
                                            60
                                             The base year is 3 years prior to the contract year. We use base-year data to be consistent with
                                            HCFA’s practice of calculating county rates from base-year enrollment and cost data.
                                            61
                                             Joiner cost estimates are the starting point for estimates of all risk HMO enrollees’ costs, so accuracy
                                            of joiner cost estimates is important. Given this, we sought to minimize the undue influence of outlier
                                            observations on our estimates. After examining our estimates for a wide range of joiner sample sizes,
                                            we concluded that a sample size of 500 would dampen outliers’ influence and yield reasonable
                                            estimates.



                                            Page 45                                         GAO/HEHS-97-16 Medicare HMO Excess Payments
                                     Appendix III
                                     Estimates of Medicare Excess Payments to
                                     HMOs in California




                                                                                    County-rate excess payment as percentage of
                    County-rate excess payment amount (in millions)                       risk contract program payments
County                        1995                 1996                  1997                   1995                  1996                  1997
San Joaquin                      •                      •                   0.1                     •                      •                 0.14
Total                       $275.7               $249.9                $413.2
Weighted averagea                                                                                5.26                  3.72                  5.13

                                     Notes: Excess payment amounts are based on projections of risk contract program payments.
                                     (By contrast, percentage rates of excess payment depend only on HCFA’s county AAPCC and
                                     risk adjuster and our estimate of the baseline county cost.) We projected 1995 payments by
                                     annualizing HCFA risk contract program payments for October through November 1995. We
                                     projected the 1996 and 1997 payments by updating the 1995 projection to account for
                                     (1) changes in the HMO payment rates (AAPCC) from 1995 to 1996 and (2) changes in
                                     enrollment since 1995 that were assumed equal to the 1994-95 rate of enrollment growth.

                                     Bullets indicate that the estimate was not sufficiently precise to be reported, because the county
                                     had fewer than 500 joiners during the base year.
                                     a
                                       These weighted average percentages are the ratios of total excess payments to risk contract
                                     program expenditures. Each weighted average pertains only to the counties listed. The weighted
                                     averages are not comparable across years because the number of counties differs from year to
                                     year. However, the percentages for a given county can be compared across years.



                                     Table III.1 shows that, for California in 1996, the estimated excess
                                     payments solely attributable to the county rate are substantial.
                                     Consequently, elimination of this component of excess payments—in one
                                     state—would save Medicare several hundred million dollars annually. This
                                     potential saving equals about 5 percent of risk contract program
                                     expenditures in California.

                                     As rates of risk HMO enrollment increase in future years, county-rate
                                     excess payments may increase as well. (As a result, the longer-term
                                     savings from eliminating county-rate excess payment could well exceed
                                     the immediate savings.) This conclusion follows from three premises:

                                     1. Across counties in each year, the higher the HMO enrollment rate, the
                                     higher the county-rate excess payment as a share of risk contract outlays.
                                     (More technically, the relationship between the county-rate excess
                                     payment—as a share of risk contract outlays—and the share of Medicare
                                     beneficiaries in the county enrolled in a risk HMO is positive and
                                     statistically significant.)62 This premise implies that the degree of favorable
                                     selection in a county does not decline as enrollment rates rise—at least
                                     over their observed range of variation.


                                     62
                                       The correlation coefficients between the excess payment and enrollment percentages for each of the
                                     3 years are .84, .82, and .74. All are significant at the 1-percent level. These correlations pertain only to
                                     the counties listed in table III.1.



                                     Page 46                                           GAO/HEHS-97-16 Medicare HMO Excess Payments
                   Appendix III
                   Estimates of Medicare Excess Payments to
                   HMOs in California




                   2. The enrollment rate for risk HMOs will increase nationwide and in
                   California.

                   3. As the national and state enrollment rates increase, the number of
                   counties with substantial risk HMO enrollment will increase.

                   In sum, in California, growing enrollment is likely to have two effects on
                   excess payments. The more straightforward effect will be to raise excess
                   payments because a given excess payment per enrollee will be multiplied
                   by a larger number of enrollees. Less obvious, however, will be higher
                   enrollment’s tendency to raise the excess payment per enrollee. That is, if
                   favorable selection continues to occur while HMO enrollment increases, the
                   average cost of beneficiaries remaining in FFS can also increase, leading to
                   higher excess payments per HMO enrollee. As a result of these two effects,
                   the statewide total estimate of county-rate excess payments will increase
                   with HMO enrollment, between 1995 and 1997, from about $276 million to
                   about $413 million.63


                   Table III.2 presents our estimates of aggregate excess payment by county.64
Estimates of        Only those counties for which the number of new HMO enrollees (joiners)
Aggregate Excess   exceeded 500 in 1995 are presented in the table.65 The counties are ranked
Payments           by excess payment amounts. We estimated that aggregate excess
                   payments totaled about $1 billion in 1995. This amount represents about
                   16 percent of Medicare’s payments to California HMOs under the risk
                   contract program in 1995. Like county-rate excess payments, aggregate
                   excess payments are concentrated in the five counties ranking highest in
                   risk contract program enrollment. Together, these counties account for
                   more than 75 percent of our estimate of statewide aggregate excess
                   payments.



                   63
                     Contrary to expectation, excess payments fell between 1995 and 1996, because of the introduction of
                   the Medicare Fee Schedule in 1992. (Recall that we used 1992 cost data to estimate the 1996
                   county-rate excess payment.) This new fee schedule coincided with an unusually large decline in
                   Medicare physician service volume growth—from an average of almost 9 percent in 1990-91 to about
                   2 percent in 1992. As a result, average part B costs for FFS beneficiaries declined in 1992. The lower
                   FFS costs caused a narrowing of the cost disparity between HMO enrollees and FFS beneficiaries.
                   64
                     HCFA actually determines four sets of HMO base-payment rates for each county: (1) aged part A,
                   (2) aged part B, (3) disabled part A, and (4) disabled part B. The estimates in table III.1 are a weighted
                   average of the biases in the rates for aged (parts A and B) and disabled (parts A and B). (HCFA also
                   determines separate statewide rates for beneficiaries with end-stage renal-disease. We excluded these
                   beneficiaries from our estimates.)
                   65
                    The counties excluded from the table account for less than 1 percent of the sum of aggregate excess
                   payments of all California counties.



                   Page 47                                          GAO/HEHS-97-16 Medicare HMO Excess Payments
                                 Appendix III
                                 Estimates of Medicare Excess Payments to
                                 HMOs in California




Table III.2: Aggregate Excess
Payments by County for 1995 in                                                                              Aggregate excess
Millions of 1995 Dollars                                                    Aggregate excess        payment as a percentage
                                                                             payment amount          of risk contract program
                                 County                                          (in millions)                      payments
                                 Los Angeles                                             $429.0                                20.8
                                 Orange                                                    121.3                               20.0
                                 San Diego                                                 113.2                               15.5
                                 San Bernardino                                             71.9                               17.8
                                 Riverside                                                  66.7                               14.1
                                 Alameda                                                    30.5                               14.8
                                 Ventura                                                    29.4                               21.3
                                 Contra Costa                                               25.2                               15.6
                                 San Francisco                                              17.4                               10.7
                                 Santa Clara                                                16.2                                8.2
                                 Kern                                                       16.0                               13.6
                                 San Mateo                                                    9.2                               7.0
                                 Fresno                                                       8.7                              19.7
                                 Santa Barbara                                                7.9                              12.5
                                 Sonoma                                                       6.7                               9.5
                                 San Joaquin                                                  6.4                              15.8
                                 Solano                                                       5.2                              15.9
                                 Placer                                                       5.1                              21.2
                                 Sacramento                                                   4.4                               2.2
                                 Santa Cruz                                                   4.2                              30.7
                                 Marin                                                        3.4                               9.7
                                 Stanislaus                                                   2.9                               4.2
                                 Yolo                                                         1.7                              10.6
                                 San Luis Obispo                                              1.5                               3.6
                                 Monterey                                                     1.1                               9.6
                                 Butte                                                         .5                               2.4
                                 Total                                                 $1,005.6
                                 Weighted average                                                                              16.4

                                 Note: Excess payment amounts (but not percentages) are based on county-level projections of
                                 risk contract program payments for 1995. We projected 1995 payments by annualizing actual
                                 HCFA risk contract program payments for October through November 1995.



                                 A comparison of the percentages shown in tables III.1 and III.2 indicates
                                 that county-rate excess payments account for roughly one-quarter of




                                 Page 48                                     GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix III
Estimates of Medicare Excess Payments to
HMOs in California




aggregate excess payments.66 This result suggests that, even if the
imprecision in the estimates of excess payment due to the county rate
were substantial, correction of the county rate on the basis of those
estimates would not lead Medicare to underpay HMOs as a group. In effect,
the component of aggregate excess payment due to inadequate risk
adjustment acts as a cushion for the county-rate correction.




66
 Alternatively, about three-quarters of aggregate excess payments result directly from inadequate risk
adjustment.



Page 49                                        GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix IV

Comments From the Department of Health
and Human Services and Our Evaluation




              Page 50     GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix IV
Comments From the Department of Health
and Human Services and Our Evaluation




Page 51                              GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix IV
Comments From the Department of Health
and Human Services and Our Evaluation




Page 52                              GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix IV
Comments From the Department of Health
and Human Services and Our Evaluation




Page 53                              GAO/HEHS-97-16 Medicare HMO Excess Payments
                           Appendix IV
                           Comments From the Department of Health
                           and Human Services and Our Evaluation




                           The following is GAO’s comment on the Department of Health and Human
                           Services’ letter dated March 26, 1997.


                           In commenting on a draft of this report, HHS agreed that, because of
GAO Comment                favorable selection, the current payment method results in substantial
                           overpayments to Medicare managed care plans. Moreover, HHS did not
                           dispute that our recommended rate-setting revision would save money.
                           However, HHS cited our proposed revision as potentially “inequitable,”
                           possibly burdensome to implement, and “only an interim measure” until
                           HCFA develops better health status adjusters. As discussed below, we
                           believe that certain features make our recommended revision evenhanded,
                           easy to implement, and important to adopt, regardless of the likely
                           improvements to risk adjustment now under consideration. The details of
                           our reasoning follow.


Recommended Revision       HHS  stated that our proposed revision is not equitable because it would
Would Improve Payment      differentially affect HMO payments based on the managed care penetration
Rate Accuracy and Target   rate within each county. This is not accurate. Nothing in our proposed
                           refinement to the Medicare payment method would tie HMO payments to
Excess Payments            HMO penetration rates. Our recommendation is to include an estimated FFS
Reductions                 cost for HMO enrollees in the formula used to calculate the county rate. By
                           making the estimate of a county’s average Medicare costs more accurate,
                           this revision would reduce payments most in counties where cost
                           disparities between the FFS and HMO beneficiaries are greatest. Our
                           recommended approach would leave the county payment rate unchanged
                           despite high managed care enrollment—if HMO and FFS beneficiaries in a
                           county have the same average cost.

                           HHS  also expressed concern that, with the adoption of our revision,
                           counties with relatively low AAPCC rates but high Medicare managed care
                           penetration rates could be “very adversely affected.” Our approach is
                           targeted and would not reduce Medicare rates in counties with no cost
                           disparities between the FFS and HMO beneficiaries. Under our approach, a
                           county with a low AAPCC rate but no cost disparities would see no change
                           in its county payment rate—even if the HMO penetration rate in that county
                           was high. In contrast, an across-the-board payment rate cut—which, as
                           HHS notes, is part of the administration’s fiscal year 1998 budget
                           proposal—would affect high AAPCC and low AAPCC counties equally,
                           regardless of how costly a county’s beneficiaries might be. Our proposed
                           revision would reduce but not eliminate excess HMO payments.



                           Page 54                              GAO/HEHS-97-16 Medicare HMO Excess Payments
    Appendix IV
    Comments From the Department of Health
    and Human Services and Our Evaluation




    Consequently, substantial excess payments would probably remain to
    cushion HMOs from any resulting reduction in the county rate. (See p. 49.)

    To illustrate what HHS believes is the potential for our modified payment
    method to produce inequitable results, HHS constructed an example
    involving two hypothetical counties. HHS contends that the example shows
    a paradoxical result: under our modified method, HHS asserts, HMOs in
    county A would receive higher capitation payments than HMOs in county B
    even though HMO enrollees in county A are healthier than those in county
    B. As explained below, this conclusion is incorrect.

•   Our recommendation would yield HMO payment rates in line with Medicare
    law, because they would be set on the basis of the estimated average FFS
    cost of all beneficiaries in a county. HHS did not acknowledge that under
    the current method both counties’ HMOs receive the same rate even though
    county A HMOs serve healthier beneficiaries than county B HMOs. Our
    method would reduce excess payments to HMOs in both counties, although
    HMOs would still receive payments exceeding their enrollees’ expected per
    capita costs. Moreover, our method would increase payments to HMOs in
    counties experiencing adverse selection—that is, in instances where a
    county’s HMOs have enrollees whose expected costs exceed those of FFS
    users.
•   HHS’ example also runs counter to the experience of the counties we
    examined. Our data show that counties with low HMO penetration rates
    tend to have low excess payments relative to counties with high
    penetration rates. For example, excess HMO payments are lower in
    Sacramento, which had 5.6 percent of its Medicare beneficiaries enrolled
    in HMOs, than in Los Angeles, which had 25.5 percent enrolled in HMOs.
    Nonetheless, HHS’ example assumes excess payments and HMO penetration
    are inversely related (higher penetration rate, lower excess payments).
    Though some counties may display this pattern, the counties we examined
    do not.

    In discussing its example, HHS seemingly endorses the current method of
    paying Medicare HMOs as an interim strategy and, consequently, considers
    it appropriate to ignore the problem of large excess payments in counties
    like A, at least for several years. In contrast, our recommended
    modification of the current method would reduce excess payments
    significantly and promptly. While it is true that HMOs in B would be paid
    less than in A, correcting such discrepancies is the role of improved health
    status adjusters.




    Page 55                              GAO/HEHS-97-16 Medicare HMO Excess Payments
                          Appendix IV
                          Comments From the Department of Health
                          and Human Services and Our Evaluation




Recommended Revision      HHS commented that our modification to the current payment method may
Could Be Readily          be difficult to implement, citing both conceptual issues and resource
Implemented               requirements. For example, HHS suggested that “the issue of when to begin
                          counting for the regression (toward the mean) effect is problematic”
                          because many beneficiaries switch plans or switch between managed care
                          and FFS. To overcome this potential difficulty, HCFA could consider time
                          spent in various HMOs with brief spells in FFS as continuous enrollment in
                          managed care. If the beneficiary spent a significant length of time in FFS,
                          HCFA could reset the regression effect for that beneficiary to zero. This
                          approach would be conservative in that it would tend to increase the
                          estimated FFS costs of HMO enrollees and thus yield rates favorable to HMOs.

                          In addition, HHS expressed concern that “if separate [RTM factor] estimates
                          are required for each county the [computational] burden could be very
                          great.” Separate estimates of RTM factors for each county are not needed.
                          We estimated the RTM factor using statewide data, although we used HMO
                          tenure levels at the county level in conjunction with the RTM factor to
                          adjust county costs.

                          HHS believes that implementing our refinement to the current method
                          would require a significant amount of resources. Given the modest
                          resources (two analysts) that we used in conducting our analysis, and that
                          our proposed change would not entail collecting new data, we believe that
                          the additional resources needed to implement our refinement would be
                          small. Moreover, the likely benefits greatly outweigh such costs. As our
                          report indicates, the payoff from this effort would probably be hundreds of
                          millions of dollars in Medicare savings each year.


Recommended Revision Is   HHS states that our payment method revision is an interim solution to the
Fundamental to Fixing     HMO overpayment problem. HHS also notes that HCFA is working to develop
Excess Payment Problem    a new payment methodology incorporating health status adjusters that
                          might be phased in starting in calendar year 2001. Together, these
                          assertions could imply that our approach is unnecessary.

                          Our revision, however, is not an interim solution. It is an important first
                          step toward—and most likely will be a component of—a comprehensive
                          solution. By addressing the effect of favorable selection in the county rate,
                          our revision makes an essential adjustment to the rate on which the rest of
                          an HMO’s capitation payment is based. The revision could be implemented
                          as early as calendar year 1998. This would allow the government, at the
                          very least, 3 years to make partial reductions in excess HMO



                          Page 56                              GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix IV
Comments From the Department of Health
and Human Services and Our Evaluation




payments—amounting to saving hundreds of millions of taxpayer dollars
in each of those years. Moreover, our recommended correction of the
county rate would complement improved health status adjusters to
provide the foundation for a more efficient, accurate, and equitable
redesign of Medicare’s method of HMO payment.




Page 57                              GAO/HEHS-97-16 Medicare HMO Excess Payments
Appendix V

GAO Contacts and Staff Acknowledgments


                  Jonathan Ratner, Associate Director, (202) 512-7107
GAO Contacts      Scott L. Smith, Project Director, (202) 512-5713
                  Richard M. Lipinski, Project Manager, (202) 512-3597


                  The following team members also made important contributions to this
Staff             report: James Cosgrove, Assistant Director; Thomas Dowdal, Assistant
Acknowledgments   Director; Craig Winslow, Senior Attorney; George M. Duncan, Senior
                  Evaluator; and Hannah F. Fein, Senior Evaluator.




(101369)          Page 58                          GAO/HEHS-97-16 Medicare HMO Excess Payments
Ordering Information

The first copy of each GAO report and testimony is free.
Additional copies are $2 each. Orders should be sent to the
following address, accompanied by a check or money order
made out to the Superintendent of Documents, when
necessary. VISA and MasterCard credit cards are accepted, also.
Orders for 100 or more copies to be mailed to a single address
are discounted 25 percent.

Orders by mail:

U.S. General Accounting Office
P.O. Box 6015
Gaithersburg, MD 20884-6015

or visit:

Room 1100
700 4th St. NW (corner of 4th and G Sts. NW)
U.S. General Accounting Office
Washington, DC

Orders may also be placed by calling (202) 512-6000
or by using fax number (301) 258-4066, or TDD (301) 413-0006.

Each day, GAO issues a list of newly available reports and
testimony. To receive facsimile copies of the daily list or any
list from the past 30 days, please call (202) 512-6000 using a
touchtone phone. A recorded menu will provide information on
how to obtain these lists.

For information on how to access GAO reports on the INTERNET,
send an e-mail message with "info" in the body to:

info@www.gao.gov

or visit GAO’s World Wide Web Home Page at:

http://www.gao.gov




PRINTED ON    RECYCLED PAPER
United States                       Bulk Rate
General Accounting Office      Postage & Fees Paid
Washington, D.C. 20548-0001           GAO
                                 Permit No. G100
Official Business
Penalty for Private Use $300

Address Correction Requested