Medicare HMOs: HCFA Can Promptly Eliminate Hundreds of Millions in Excess
Payments (Letter Report, 04/25/97, GAO/HEHS-97-16).

Pursuant to a congressional request, GAO provided information on
Medicare's rate-setting method for paying risk contract health
maintenance organizations (HMO), focusing on: (1) the conditions under
which Medicare's method can yield payment rates that are too high; and
(2) a practical improvement to Medicare's method directed at the
problems fostering excess payments.

GAO noted that: (1) contrary to the expectations built into Medicare law
for paying risk contract HMOs, these HMOs have not produced savings for
Medicare; (2) 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 fee-for-service (FFS); (3)
researchers attribute this outcome to "favorable selection", or the
tendency for healthier-than-average individuals to be enrolled in HMOs;
(4) GAO has identified a modification to Medicare's current HMO
rate-setting method that could help reduce excess HMO payments; (5)
central to the current method is an estimate of the average cost, county
by county, of serving Medicare beneficiaries in the FFS sector; (6) 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; (7) although considerable attention
has focused on problems with this process, GAO's 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); (8) HCFA's method of determining the county
rate excludes HMO enrollees' costs in estimating per-beneficiary average
cost; (9) the result is that in counties experiencing favorable
selection, HCFA's method overstates the average costs of all Medicare
beneficiaries and leads to overpayments; (10) GAO's proposed
modification estimates HMO enrollees' expected FFS costs using
information available to HCFA; (11) GAO's approach produces a county
rate that more accurately represents the costs of all Medicare
beneficiaries; (12) in examining the rates HCFA determined for
California's 58 counties in 1995, GAO found that applying its approach
would have reduced excess payments by about 25 percent, or $276 million;
(13) substantially better risk adjustment, which appears to be years
away from implementation, would have targeted the remaining 75 percent;
(14) GAO 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; and (15) 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*

--------------------------- Indexing Terms -----------------------------

 REPORTNUM:  HEHS-97-16
     TITLE:  Medicare HMOs: HCFA Can Promptly Eliminate Hundreds of 
             Millions in Excess Payments
      DATE:  04/25/97
   SUBJECT:  Health maintenance organizations
             Overpayments
             Health care programs
             Health insurance cost control
             Managed health care
             Medical services rates
             Projections
             Health care costs
IDENTIFIER:  Medicare Program
             Medicaid Program
             California
             Medicare Risk Contract Program
             
******************************************************************
** This file contains an ASCII representation of the text of a  **
** GAO report.  Delineations within the text indicating chapter **
** titles, headings, and bullets are preserved.  Major          **
** divisions and subdivisions of the text, such as Chapters,    **
** Sections, and Appendixes, are identified by double and       **
** single lines.  The numbers on the right end of these lines   **
** indicate the position of each of the subsections in the      **
** document outline.  These numbers do NOT correspond with the  **
** page numbers of the printed product.                         **
**                                                              **
** No attempt has been made to display graphic images, although **
** figure captions are reproduced.  Tables are included, but    **
** may not resemble those in the printed version.               **
**                                                              **
** Please see the PDF (Portable Document Format) file, when     **
** available, for a complete electronic file of the printed     **
** document's contents.                                         **
**                                                              **
** A printed copy of this report may be obtained from the GAO   **
** Document Distribution Center.  For further details, please   **
** send an e-mail message to:                                   **
**                                                              **
**                                            **
**                                                              **
** with the message 'info' in the body.                         **
******************************************************************


Cover
================================================================ COVER


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

Medicare HMO Excess Payments

(101369)


Abbreviations
=============================================================== ABBREV

  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

Letter
=============================================================== LETTER


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 option\1 --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 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. 


--------------------
\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. 


   RESULTS IN BRIEF
------------------------------------------------------------ Letter :1

Contrary to the expectations built into Medicare law for paying risk
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 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. 


--------------------
\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. 


   BACKGROUND
------------------------------------------------------------ Letter :2

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



   (See figure in printed
   edition.)

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 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. 


--------------------
\4 Section 1876(a)(4) of the Social Security Act (42 U.S.C. 
1395mm(a)(4) (1994)). 

\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. 


   HOW MEDICARE'S HMO RATE-SETTING
   METHOD CAN LEAD TO EXCESS
   PAYMENTS
------------------------------------------------------------ Letter :3

For HCFA's rate-setting method to produce appropriate rates, the risk
adjusters must reliably differentiate among beneficiaries with
different health status.  Much has been written about the inadequacy
of Medicare's risk adjuster to account for the tendency of HMOs to
experience favorable selection.  More than a decade of research has
concluded that beneficiaries 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
costs of Medicare HMO enrollees ranging from 12 percent\8 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


--------------------
\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). 

\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). 


   HCFA COULD IMPROVE ITS
   RATE-SETTING METHOD BY
   INCLUDING HMO ENROLLEES IN ITS
   CALCULATIONS OF COUNTY AVERAGE
   COST
------------------------------------------------------------ Letter :4

Independent of risk adjustment, modifying the method for calculating
county rate would help reduce Medicare's excess HMO payments.  HCFA
currently estimates the average Medicare costs of a county's
beneficiaries using the costs of only those beneficiaries in
Medicare's FFS sector.  This method would be appropriate if the
average health cost of FFS beneficiaries were the same as that of
demographically comparable HMO enrollees.\12 However, in counties
where there are cost disparities between Medicare's FFS and HMO
enrollee populations, this method can either 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. 

  -- 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. 


--------------------
\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. 

\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. 


      CURRENT COUNTY RATES PRODUCE
      SUBSTANTIAL EXCESS PAYMENTS
---------------------------------------------------------- Letter :4.1

To illustrate the effect of our approach, we analyzed data for
counties with different shares of beneficiaries enrolled in HMOs.\15
We found that our 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. 



                          Table 1
          
            Estimates of Potential Reduction in
           Excess Payments to California HMOs in
               1995, Based on Our Method for
                Calculating the County Rate

                                        County-rate excess
                           County-rate       payments as a
                          estimates of  percentage of risk
                       excess payments    contract program
County                   (in 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
==========================================================
Total (12                       $275.7
 counties)\a
----------------------------------------------------------
\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. 

   Figure 1:  Excess Payments Rise
   With HMO Enrollment

   (See figure in printed
   edition.)

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


--------------------
\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. 

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


      DATA ARE AVAILABLE TO ENABLE
      HCFA TO PROMPTLY ADJUST
      COUNTY RATES
---------------------------------------------------------- Letter :4.2

Because the data we used to estimate HMO enrollees' costs come from
data that HCFA compiles to update HMO rates each year, our method has
two 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. 


   CONCLUSIONS
------------------------------------------------------------ Letter :5

Medicare's HMO rate-setting problems have prevented it from realizing
the 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. 


   RECOMMENDATION TO THE SECRETARY
   OF HEALTH AND HUMAN SERVICES
------------------------------------------------------------ Letter :6

We recommend that the Secretary of Health and Human Services direct
the HCFA Administrator to incorporate the expected FFS costs of HMO
enrollees into the methodology for establishing county rates using
the method we explain in this report and adjust Medicare payment
rates to risk contract HMOs accordingly. 


   AGENCY COMMENTS
------------------------------------------------------------ Letter :7

In commenting on a draft of this report, HHS agreed that, because
Medicare 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.)


---------------------------------------------------------- Letter :7.1

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. 

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


METHODOLOGY
=========================================================== Appendix I

Despite evidence from a number of studies\18 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. 


--------------------
\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. 


   METHOD FOR REDUCING EXCESS HMO
   PAYMENTS BY CORRECTING
   MEDICARE'S COUNTY RATE
--------------------------------------------------------- Appendix I:1

The basic steps HCFA takes to determine capitation payments can be
described as follows. 

 Step 1

HCFA calculates the per capita costs in Medicare FFS, or standard
average 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 ï¿½ 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 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

   (See figure in printed
   edition.)


--------------------
\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. 

\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. 


      SOURCES OF EXCESS PAYMENTS
      TO HMOS
------------------------------------------------------- Appendix I:1.1

Excess payments can occur if HMOs enroll a group of beneficiaries
that is 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
         ROUGH PROXIES FOR
         EXPECTED HEALTH COSTS AND
         DO NOT FULLY ADJUST
         PAYMENTS FOR FAVORABLE
         SELECTION
----------------------------------------------------- Appendix I:1.1.1

HCFA's risk factors adjust for favorable selection using five
characteristics (age, sex, Medicaid eligibility status, institutional
status, and working status) that are relatively poor predictors of
beneficiaries' health care needs.\22 Specifically, the risk factors
are a set of weights--intended to 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. 


--------------------
\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). 


         HCFA'S CAPITATION RATE IS
         INFLATED BY FAVORABLE
         SELECTION
----------------------------------------------------- Appendix I:1.1.2

If HMOs' enrollees tend to be healthier than the average beneficiary
in FFS, then HCFA's method will overestimate the expected cost of
serving 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

   (See figure in printed
   edition.)

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 ï¿½ .95--reflecting just
     the average costs of beneficiaries in the FFS sector--instead of
     $400 ï¿½ .95.  Thus, Medicare would pay HMOs $100 ï¿½ .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 ï¿½ .95 instead of the appropriate
$400 ï¿½ .95. 


--------------------
\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.





      ESTIMATING EXPECTED FFS
      COSTS FOR HMO ENROLLEES
------------------------------------------------------- Appendix I:1.2

We developed a method to estimate the potential FFS costs for HMO
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 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. 


--------------------
\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. 

\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. 


         BENEFICIARIES CLASSIFIED
         ACCORDING TO ENROLLMENT
         STATUS
----------------------------------------------------- Appendix I:1.2.1

Because Medicare allows beneficiaries to switch among specific HMOs
or 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 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. 


--------------------
\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. 


         PRIOR-YEAR FFS SPENDING
         USED TO ESTIMATE
         BASE-YEAR COSTS FOR EACH
         BENEFICIARY CATEGORY
----------------------------------------------------- Appendix I:1.2.2

We adjusted prior-year cost data of joiners to approximate average
costs in the base year for enrollees\36 because their costs (on an
FFS basis) are 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 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, Sample Year 1992

                                Cost estimate
                    --------------------------------------
                                        Adjustment to cost
Beneficiary group   Cost measure        measure
------------------  ------------------  ------------------
HMO enrollees       1991 costs of       Costs increased to
                    people who joined   account for RTM
                    an HMO in 1992      effect
                    (joiners)

FFS beneficiaries   1991 costs of all   None
                    FFS beneficiaries


People who died within the sample year (1992)
----------------------------------------------------------
HMO enrollees       Costs of people     None
                    who died
                    in FFS in 1991

FFS beneficiaries   Costs of people     None
                    who died
                    in FFS in 1991
----------------------------------------------------------

--------------------
\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.



\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. 


      JOINERS' PRIOR-USE COSTS
      USED TO ESTIMATE ALL HMO
      ENROLLEES' COSTS
------------------------------------------------------- Appendix I:1.3

In estimating SACHMO, we used the prior-use costs of joiners as a
baseline in estimating the (unobserved) expected FFS costs of all HMO
enrollees.  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 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



                         Table I.2
          
           Ratios of Monthly Average Costs of New
          Risk HMO Enrollees to FFS Beneficiaries'
            Costs for Three California Counties,
                          1992-94

                                 Year
            ----------------------------------------------
                  1992        1993        1994        1995
----------  ----------  ----------  ----------  ----------
Los Angeles
----------------------------------------------------------
Joiners           $161        $184        $189        $178
FFS                333         362         399         365
 beneficia
 ries
Ratio              .48         .51         .47         .49

San Diego
----------------------------------------------------------
Joiners            162         195         191         183
FFS                285         315         342         314
 beneficia
 ries
Ratio              .57         .62         .56         .58

Sacramento
----------------------------------------------------------
Joiners            159         204         198         187
FFS                268         298         318         295
 beneficia
 ries
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. 


--------------------
\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. 


      PRIOR-USE COSTS OF JOINERS
      ADJUSTED FOR
      REGRESSION-TOWARD-THE-MEAN
      EFFECT
------------------------------------------------------- Appendix I:1.4

After a beneficiary joins an HMO, it is hypothesized that the
beneficiary's cost is likely to increase relative to his or her FFS
costs in the year prior to enrolling.  Such cost increases seem
likely for two reasons.  First, beneficiaries may postpone
discretionary care in the months prior to 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 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 FOR
      DEATH COSTS
------------------------------------------------------- Appendix I:1.5

Because new HMO enrollees, by definition, do not die during the
period just 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
      EXCESS PAYMENTS THAT ARE DUE
      TO USING ONLY FFS
      BENEFICIARIES' EXPERIENCE TO
      SET RATES
------------------------------------------------------- Appendix I:1.6

After estimating the average expected costs of serving all of a
county's beneficiaries in FFS (SACALL), we could estimate the excess
capitation payments that resulted from HCFA's method of calculating
SAC and the county rate.  The formula for determining capitation
rates can be expressed as the following: 

   Equation 3

   (See figure in printed
   edition.)

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

   Equation 4

   (See figure in printed
   edition.)

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

   Equation 5

   (See figure in printed
   edition.)

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. 


   METHOD FOR ESTIMATING
   MEDICARE'S AGGREGATE EXCESS
   PAYMENTS
--------------------------------------------------------- Appendix I:2

This section describes the steps we followed to estimate aggregate
excess payments to HMOs, that is, total excess payments caused by the
full effect of favorable selection on the rate-setting formula.  Our
method compares 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

 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

   (See figure in printed
   edition.)

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.)


--------------------
\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. 

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


ADJUSTMENTS FOR REGRESSION TOWARD
THE MEAN AND DEATH-RELATED COSTS
IN ESTIMATING EXCESS PAYMENTS TO
MEDICARE HMOS
========================================================== Appendix II

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. 


   METHODOLOGY ALLOWS RTM FACTOR
   TO VARY BY BENEFICIARY SURVIVAL
   STATUS
-------------------------------------------------------- Appendix II:1

We allow our estimate of RTMF to differ between groups of
beneficiaries, depending on whether they survived or died during the
4-year period that we analyzed.  The association between mortality
and average costs is well documented by previous studies.  For
example, Lubitz and others found that people in their last 12 months
of life have costs that are significantly 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. 

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

   (See figure in printed
   edition.)

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

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

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

Category of                              Percentage of all
enrollee            Status                 HMO enrollees\a
------------------  ------------------  ------------------
I                   Survived 4 or more                83.8
                     years
II                  Survived at least                 12.9
                     1 year but less
                     than 4 years
III                 Survived less than                 3.3
                     1 year
----------------------------------------------------------
\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. 


--------------------
\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. 


   METHOD USED TO ESTIMATE THE RTM
   FACTOR FOR CATEGORY I ENROLLEES
-------------------------------------------------------- Appendix II:2

To estimate RTMF for enrollees who survive for 4 or more years
(category I enrollees), we developed an approach that generally
follows Beebe's 1988 methodology.  That is, we used 4 years of
longitudinal data on a sample of the FFS Medicare population to track
the cost experience over time of two proxy cohorts--one representing
HMO joiners and one representing FFS 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. 

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. 


--------------------
\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. 


      DESCRIPTION OF FFS
      BENEFICIARY DATA SET
------------------------------------------------------ Appendix II:2.1

We assembled a longitudinal data set that contained the claims for
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. 


--------------------
\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. 


      METHODOLOGY FOR CONSTRUCTING
      THE PROXY HMO JOINER AND
      PROXY FFS COHORTS
------------------------------------------------------ Appendix II:2.2

We constructed two proxy cohorts, one with the same demographic mix
and 1991 service cost distribution as the Medicare HMO joiners, and
the 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 subgroups\51 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. 



                                        Table II.2
                         
                         1991 Distribution Across Cost Categories
                          of HMO Joiners and FFS Beneficiaries,
                                65-to 69-Year-Old Females

                   Percentage distribution of
                         beneficiaries                  Number of beneficiaries
                   --------------------------  ------------------------------------------
                                                 Longitudinal
                                                population of     Proxy FFS  Proxy joiner
                                               beneficiaries\        cohort        cohort
Cost                        FFS        Joiner               a        sample        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
75,000-99,999\b             0.0           0.0              83            15             7
=========================================================================================
Total                       100           100         217,963      43,599\c      43,605\c
-----------------------------------------------------------------------------------------
\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. 


--------------------
\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+. 


      RATIO OF PROXY HMO JOINERS'
      COSTS TO PROXY FFS
      BENEFICIARIES' COSTS
------------------------------------------------------ Appendix II:2.3

Within each demographic group, we calculated the ratio of the proxy
HMO joiner cost average to the proxy FFS cost average for each of 4
years (1991 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. 

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

   (See figure in printed
   edition.)

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 Beneficiaries, 1991-
                                   94

                                           Tenure in HMO (in years)
                                        ------------------------------
                                          Year
                                         prior
                                            to
                                        enroll
                                          ment    Year    Year    Year
                                        (1991)       1       2       3
Cost ratio                                  \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


--------------------
\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. 


      CALCULATING THE RTMF FROM
      THE ESTIMATED COST RATIOS
------------------------------------------------------ Appendix II:2.4

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

                                                Number of years in HMO
                                                ----------------------
                                                                  3 or
Measure                                              1       2    more
----------------------------------------------  ------  ------  ------
Benchmark cost proportion: the cost ratio for     1.33    1.38    1.41
 each year divided by the cost ratio for the
 year prior to enrollment\a
Tenure distribution: proportion of HMO             .11     .18     .71
 enrollees for the county (from actual
 enrollment data)\b
RTMF: a weighted average of benchmark cost                1.40
 proportions, using the tenure distribution as
 weights\c
----------------------------------------------------------------------
\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 ï¿½ 1.33) +
(.18 ï¿½ 1.38) + (.71 ï¿½ 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. 


   METHOD USED TO ESTIMATE THE RTM
   FACTOR FOR CATEGORY II
   ENROLLEES
-------------------------------------------------------- Appendix II:3

We could not estimate an RTMF for category II enrollees with the
method that we used for category I enrollees.  That method requires
constructing proxy cohorts of HMO joiners and FFS beneficiaries, but
the number of category II enrollees--those who survive between 1 year
and 4 years after enrollment--was insufficient to do so. 

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 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


--------------------
\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. 

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


   THE RTM FACTOR FOR CATEGORY III
   ENROLLEES
-------------------------------------------------------- Appendix II:4

The average costs of HMO joiners in the year of their death (in this
case 1991) cannot be estimated.  After all, joiners must live beyond
the prior-use 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
      INDICATED BY RELATIVELY LOW
      HMO DEATH RATES
------------------------------------------------------ Appendix II:4.1

Although our method for estimating excess payments to HMOs assumed
that no difference existed in death-related costs between HMO and FFS
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. 



                                               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 mean\a               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. 


   SUMMARY OF ADJUSTMENTS FOR RTM
-------------------------------------------------------- Appendix II:5

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


      CATEGORY I ENROLLEES
------------------------------------------------------ Appendix II:5.1

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
------------------------------------------------------ Appendix II:5.2

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. 


      CATEGORY III ENROLLEES
------------------------------------------------------ Appendix II:5.3

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. 


ESTIMATES OF MEDICARE EXCESS
PAYMENTS TO HMOS IN CALIFORNIA
========================================================= Appendix III

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. 


   ESTIMATES OF COUNTY-RATE EXCESS
   PAYMENTS
------------------------------------------------------- Appendix III:1

Table III.1 presents estimates of county-rate excess payments in
dollar amounts and as a percentage of risk contract program
expenditures for each county.  (The estimates are weighted averages
of the excess 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 amount (in    County-rate excess payment as percentage
                                    millions)                     of 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
San Joaquin                    --            --           0.1            --            --          0.14
Total                      $275.7        $249.9        $413.2
Weighted average\a                                                     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. 

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


--------------------
\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. 

\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. 

\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. 


   ESTIMATES OF AGGREGATE EXCESS
   PAYMENTS
------------------------------------------------------- Appendix III:2

Table III.2 presents our estimates of aggregate excess payment by
county.\64 Only those counties for which the number of new HMO
enrollees (joiners) exceeded 500 in 1995 are presented in the
table.\65 The counties are ranked 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. 



                        Table III.2
          
          Aggregate Excess Payments by County for
              1995 in Millions of 1995 Dollars

                                          Aggregate excess
                                              payment as a
                      Aggregate excess  percentage of risk
                        payment amount    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 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. 



(See figure in printed edition.)Appendix IV

--------------------
\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. 

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


COMMENTS FROM THE DEPARTMENT OF
HEALTH AND HUMAN SERVICES AND OUR
EVALUATION
========================================================= Appendix III



(See figure in printed edition.)



(See figure in printed edition.)



(See figure in printed edition.)


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


   GAO COMMENT
------------------------------------------------------- Appendix III:3

In commenting on a draft of this report, HHS agreed that, because of
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 WOULD
      IMPROVE PAYMENT RATE
      ACCURACY AND TARGET EXCESS
      PAYMENTS REDUCTIONS
----------------------------------------------------- Appendix III:3.1

HHS stated that our proposed revision is not equitable because it
would differentially affect HMO payments based on the managed care
penetration rate within each county.  This is not accurate.  Nothing
in our proposed refinement to the Medicare payment method would tie
HMO payments to HMO penetration rates.  Our recommendation is to
include an estimated FFS 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.  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. 


      RECOMMENDED REVISION COULD
      BE READILY IMPLEMENTED
----------------------------------------------------- Appendix III:3.2

HHS commented that our modification to the current payment method may
be difficult to implement, citing both conceptual issues and resource
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
      FUNDAMENTAL TO FIXING EXCESS
      PAYMENT PROBLEM
----------------------------------------------------- Appendix III:3.3

HHS states that our payment method revision is an interim solution to
the HMO overpayment problem.  HHS also notes that HCFA is working to
develop 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
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. 


GAO CONTACTS AND STAFF
ACKNOWLEDGMENTS
=========================================================== Appendix V

GAO CONTACTS

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

STAFF ACKNOWLEDGMENTS

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


*** End of document. ***