FY 1981 Medicaid Quality Control Disallowances, DAB No. 1332 (1992)

Department of Health and Human Services

DEPARTMENTAL APPEALS BOARD

Appellate Division

SUBJECT: FY 1981 Medicaid Quality Control Disallowances

DATE:  May 26, 1992
Docket Nos. 90-86 90-87 90-88 90-89
Decision No. 1332

DECISION

Alaska, Rhode Island, Washington, and West Virginia (States) appealed
Medicaid Quality Control (MQC) disallowances of federal financial
participation (FFP) by the Health Care Financing Administration (HCFA).
HCFA determined that these States had exceeded target "error rates" in
the operation of their Medicaid programs (Title XIX of the Social
Security Act) during the fiscal year ending September 30, 1981 (FY
1981).  Alaska appealed a disallowance of $794,636 FFP, Rhode Island
$302,919, Washington $888,932, and West Virginia $770,380.  A total of
$2,756,867 FFP is at issue in these appeals.

HCFA relied on estimates of the percentage of each State's Medicaid
payments that resulted from errors in determining recipient eligibility.
The error rate estimates were based on statistical samples required by
the MQC system.  The MQC disallowance regulations in effect for FY 1981
provided for the disallowance of erroneous payment amounts above a
target rate.  These regulations were based on the Michel Amendment
provisions for target rates set using FY 1978 MQC error rate estimates
and for fiscal "penalties" for states which exceeded target rates for FY
1981 and subsequent years.

HCFA asserted that since the FY 1981 MQC error rate estimates were above
the target rates for these States, a disallowance was mandated by the
Michel Amendment.  The States asserted that the MQC estimates relied on
were too imprecise to show that the States had exceeded the target
rates..On the basis of the analysis below, we reverse these
disallowances.  The Michel Amendment and HCFA regulations required a
statistical sampling system that provided reliable evidence that a state
had not met target error rates during FY 1981 and that ensured equal
treatment of all states.  The system implemented by HCFA did not provide
reliable evidence that any of these States had in fact exceeded their
target error rates and did not ensure that these States had received
equal treatment.

The point estimates HCFA relied upon in concluding that the States had
exceeded their target rates were so imprecise that HCFA's reliance upon
them for making the fine distinctions required by the Michel Amendment
and MQC regulations was both arbitrary and unreasonable.  In each case
the sampling error for the excess over the target rate was so great that
the excess figure could not be viewed as being reliable.  Indeed in two
instances where HCFA had determined the excess to be merely a fraction
of a point, HCFA could conclude with 95 percent confidence only that the
true value of the excess resided within a range of 10 or 20 points, so
there remained a substantial probability that each State had in fact
reduced its error rate below the target level and the reduction may have
been as large as 5 or 10 points.  Moreover, there were numerous other
factors, including significant problems with the computation of the base
year rate, that seriously undercut the reliability of the estimates.

HCFA's argument that the States themselves were responsible for the
unreliability of the estimates is both unpersuasive and unsubstantiated.
In any event, regardless of who was responsible for the lack of
precision, HCFA clearly could have taken steps retrospectively to
improve the reliability of the estimates or to compensate for the lack
of reliability, but chose not to do so.  The record strongly suggests
that any such steps would have eliminated these disallowances.  There is
simply no support in the record here for HCFA to continue to assert that
it was compelled to proceed with these disallowances in light of the
demonstrated and significant imprecision in the supporting estimates.

Our analysis primarily addresses what the Michel Amendment required and
the adequacy of the supporting statistical data relied on by HCFA in
determining that these States exceeded their target rates for FY 1981.
These are the only issues we need to reach to resolve these appeals.
There were additional arguments, .principally presented by the States,
that we do not address in light of our decision to reverse these
disallowances for the reasons stated below. 1/


Background

These four appeals are only the latest stage in a process that began
seven years ago involving 31 Quality Control disallowances in the Aid to
Families with Dependent Children (AFDC) and Medicaid programs.  Board
proceedings in these cases have involved lengthy discovery, an
evidentiary hearing, the submission of hundreds of pages of briefing,
hundreds of pages of exhibits (including many reports on the public
assistance program QC systems and presentations by expert
statisticians), as well as numerous stays or other delays to permit
states to seek Congressional intervention or to permit negotiations
between the parties.

In 1985 the Family Support Administration and HCFA issued 31 Quality
Control (QC) disallowances to 26 states.  The Board consolidated these
disallowances, 22 in the AFDC program and nine in the Medicaid program,
in order to resolve common issues.

The AFDC disallowances were ultimately permanently waived by statute.
The Board addressed the nine remaining Medicaid disallowances of
$12,280,908 in DAB No. 948. 2/  The Board held that HCFA had incorrectly
valued the excess resource errors identified in the QC samples on which
the disallowances were based; we remanded the appeals to HCFA for
revaluation and any consequent recalculation of state error rates and
disallowances..The excess resource recalculations resulting from DAB No.
948 eliminated the disallowances completely for five of the nine states.
The remaining four states, the appellants in this proceeding, had their
disallowances reduced, but not totally eliminated.

The Michel Amendment

The QC requirements imposed on HCFA and the States by the Michel
Amendment are at the heart of this controversy.  The Michel Amendment
established a series of target error rates for states to reach in their
AFDC and Medicaid programs, with ultimately a national standard of four
percent to be met by September 30, 1982.  The Michel Amendment also
provided for "penalties" should a state fail to make required error rate
reductions for a particular fiscal year.  The regulations at 42 C.F.R.
.431.802 (1981) were based on the Michel Amendment.  45 Fed. Reg. 6,326
(January 25, 1980).

Section 201 (the Michel Amendment) of H.R. 4389 referenced the
conference report for H.R. 4289 (the FY 1979 Supplemental Appropriations
Act, Public Law No. 96-38) and in turn was referenced in Public Law No.
96-123 (the Continuing Resolution for FY 1980). 3/

The conference report on the 1979 Supplemental Appropriations Act stated
in pertinent part as follows:

     The conferees wish to express their very serious concern over the
     unacceptably high rate of errors still occurring in the student.
     assistance, cash assistance (AFDC) and Medicaid programs.

     . . . the conferees direct the Secretary to issue regulations
     requiring that all States must reduce their AFDC and Medicaid
     erroneous excess payment rates to 4 percent by September 30, 1982,
     in equal amounts each year beginning in fiscal year 1980.  If the
     States fail to make the required reductions, penalties shall be
     applied equal to the amount the federal share exceeds the dollar
     error rate for the years in question.  . . .  The conferees also
     direct that the base period to be used in determining the yearly
     reductions for each State shall be the April-September 1978 review
     period for AFDC and the July-December 1978 sampling period for
     Medicaid, and that the Department establish a system of determining
     error rates that insures equal treatment of all States.

H.R. Conf. Rep. No. 331, 96th Cong., 1st Sess. 34 (1979).

As adopted by the House on August 2, 1979, the Department of Labor, and
Health, Education, and Welfare and Related Agencies Act (H.R. 4389)
provided as follows:

     Sec 201.  . . . Provided further, That the requirements pertaining
     to AFDC and Medicaid error rates, as specified in the conference
     report on the fiscal 1979 Supplemental Appropriations Act (P.L.
     96-38), shall be carried out except where the Secretary determines,
     in certain limited cases, that states are unable to reach the
     required reduction in a given year despite a good faith effort.

125 Cong. Rec. H7108, H7131 (August 2, 1979).

The Continuing Resolution for Fiscal Year 1980 provided in pertinent
part for appropriations as follows:

     Such amounts as may be necessary for projects or activities
     provided for in the Departments of Labor, and Health, Education,
     and Welfare and Related Agencies Appropriation Act, 1980 (H.R.
     4389), at a rate of operations, and to the extent and in the
     manner, provided for in.     such Act, as adopted by the House of
     Representatives on August 2, 1979 . . . .

Pub. L. 96-123, .101(g), 93 Stat. 925 (1979) (emphasis added).

The Medicaid Quality Control System and the Michel Amendment
Disallowance Regulations

Prior to the establishment of the MQC system at issue here in April
1978, similar quality control reviews had been operating in both the
AFDC and Medicaid programs, although no program-wide disallowances based
on quality control findings had been imposed. 4/

The MQC system implemented in April of 1978 attempted to arrive at a
measure of Medicaid payments erroneously made on behalf of individuals
who were ineligible for Medicaid by examining a statistical sample of
Medicaid cases.  Each state was required to have a QC system.  The state
had to select and review a statistically valid sample group of cases to
determine whether each sampled case met Medicaid eligibility
requirements in the review month.  In subsequent months, the state QC
staff assembled paid claims for sample cases for services received
during the review month to determine the amount of dollars associated
with each sample case.  Generally, payments on behalf of individuals
found to be ineligible were counted as erroneous Medicaid payments.
HCFA then selected a subsample of each state's sample and reviewed those
cases.  Through the application of a statistical technique called
regression analysis, the results of the federal subsample reviews were
used to modify the states' determinations.  Estimates of error rates
were computed by comparing the total dollars paid in error to the total
dollars paid by Medicaid for a given period.

The MQC system regulation at 42 C.F.R. .431.800(d) (1981) required the
states to operate MQC systems in accord with the "policies, sampling
methodology, review procedures, and reporting forms and requirements
specified" by HCFA in its MQC manuals.  The regulations provided for
statistical samples of cases to identify eligibility errors and for six
month review periods.  (For example, the July-December 1978 base period
used to set the Michel Amendment target rates was the first Medicaid
review period for the 1978 MQC system.).The regulations based on the
Michel Amendment error rate standards required the states to make equal
incremental reductions reaching the four percent standard by September
30, 1982.  The regulations provided that the then current MQC system at
42 C.F.R. .431.800 would be used and that disallowances would be based
on the state payment error rate, i.e., the rate of eligibility payment
errors detected by the MQC system for an annual or other review period.
As provided for by the Michel Amendment, the regulations called for
waivers of disallowances should the Secretary of Health and Human
Services determine that a state's failure to meet a required error rate
target was either beyond the state's control or despite its good faith
effort.  The regulations do not establish the statistical sampling
techniques for the MQC system, but rather rely on HCFA's QC Manuals.
The preamble to the final regulations stated that HCFA had determined to
continue to use the point estimate as "the most accurate estimate of a
State's true error rate [since] it is the estimate generally used in
conventional statistical practice."  The preamble also discussed the
expectation that states would complete valid samples based on HCFA's
directions in QC Manuals.  HCFA also stated that the failure to conduct
a sample in accordance with an approved or approvable plan may result in
invalid or unreliable estimates.  45 Fed. Reg. 6,326, 6,331 (January 25,
1980). 5/

Under the Michel Amendment regulations, target rates for FY 1981 were
calculated by subtracting from each state's July-December 1978 base
period rate one-third of the.difference between the state's base period
error rate and four percent. 6/  Disallowances were calculated using the
number of percentage points by which a state exceeded its 1981 target
rate.


Analysis

The Estimates Used for the Disallowances Were Not Reliable Evidence that
the States in Fact Exceeded the Target Rates.

HCFA's determinations that the States exceeded their target error rates
were derived from two separate estimates using statistical sampling
methodologies authorized by HCFA and implemented by HCFA and the States.
The determinations required an estimate of each State's error rate for
the FY 1978 base period and then an estimate of each State's error rate
for the period at issue, FY 1981.  Whenever a state's estimate of the FY
1978 base period rate was less than 4.00, HCFA gave that state the
benefit of a base period rate of 4.00.  HCFA then subtracted the
estimates of the FY 1978 rate from the estimates of FY 1981 rate in
order to determine the "Excess Over Target (EOT)."  The estimates for
each State involved here and the resulting Excess Over Target were as
follows:

CHART 1:  The Disallowance Estimates

       Target     1981      EOT rate       rate

 Alaska              4.00       7.86      3.86 Rhode Island
 4.03       4.55      0.52 Washington          4.98       5.86
 0.88 West Virginia       6.54       8.64      2.10

For each of the State's estimates, HCFA used what is called the "point
estimate."  (The Excess Over Target was in fact a composite "point
estimate.")  A "point.estimate" is the single most likely estimate of
the true value of what is being measured by a particular statistical
methodology.  (The "true" value of what is being measured by statistical
sampling usually cannot be determined, and most certainly could not be
determined here.)  Although the point estimate is the most likely
estimate of the true value that can be provided by a particular
methodology, the reliability and precision of the point estimate as a
reflection of the true value can vary dramatically depending on the
methodology used.  Indeed, a point estimate derived from a particular
methodology may be so unreliable and imprecise that it is worthless as
an estimate of the true value for the purpose being served.

The "confidence interval" surrounding the point estimate provides an
effective means of measuring (and hence comparing) how reliable and
precise a point estimate is as a reflection of the true value.  A
confidence interval is a statistician's calculation of the range of
values (equidistant on either side of the point estimate) within which
the statistician can say with a specified degree of certainty that the
true value would occur.  For example, a 95 percent confidence interval
is that range of values that is expected to contain the true value with
a 95 percent degree of certainty.

In the instant case, HCFA provided the 95 percent confidence intervals
for the composite point estimate of the Excess Over Target. 7/  The
following chart provides a comparison of the Excess Over Target with the
width of the confidence intervals for the composite estimate represented
by the Excess.  (The "width" is the difference between the lower and
upper bounds of the interval.)  The next chart compares how each State
would have performed under a recomputed 1981 rate if HCFA had used the
lower bound of the confidence interval for the .composite estimate
rather than the composite point estimate.

CHART 2:  The Width of the 95 Percent Confidence Intervals for the
Composite Estimate of the Excess Over Target

   Target    1981     EOT     Width rate      rate

 Alaska          4.00      7.86     3.86    12.44 Rhode Island
 4.03      4.55     0.52    19.56 Washington      4.98      5.86
 0.88    10.28 West Virginia   6.54      8.64     2.10    11.84

CHART 3:  The States' Performance if HCFA Had Used the Lower Bound of
the Confidence Interval for the Composite Estimate of the Excess Over
Target

     Target       1981 rate recomputed

 Alaska            4.00         1.64 Rhode Island      4.03
 -5.23 Washington        4.98         0.72 West Virginia     6.54
 2.72

The above two charts demonstrate that the confidence intervals for the
composite estimate of the Excess were so wide that it is impossible to
conclude with a reasonable certainty that the States failed to meet
their target rates at all, much less that they failed to meet the rates
by specific percentages as required by the Michel Amendment.  For
example, HCFA determined, based on estimates of Rhode Island's
performance for base year FY 1978 and for the audited year FY 1981, that
Rhode Island had failed to meet the target rate of 4.03 in the audited
year by only 0.52.  The composite 95 percent confidence interval for the
two estimates on which the excess determination of 0.52 was based was
19.56!  This means that HCFA had determined with 95 percent certainty
that the true value was somewhere on a scale between -5.23 and 14.33.
Indeed, even the 90 percent confidence interval for the separate
estimate of Rhode Island's FY 1981 performance was 10.65.  See States
Opening Brief -- Part II at 72.  Such wide confidence intervals for such
a small excess mean that the statistical data in question for Rhode
Island is completely unreliable for making the fine distinctions
required by the Michel Amendment.  The confidence intervals indicate
that there is almost as great a chance that Rhode Island met the target
rate as that it did not.  Indeed, based on these confidence.intervals,
HCFA cannot even rule out the possibility that the State's program was
error free! 8/

Moreover, almost identical conclusions can be reached concerning
Washington's performance, where HCFA had determined that Washington
exceeded its target rate by only 0.88 and had a composite 95 percent
confidence interval for the Excess Over Target of 10.28.

Even for the two remaining states that had somewhat larger Excesses,
West Virginia with an Excess of 2.10 and Alaska with an Excess of 3.86,
the composite confidence intervals of 11.84 for West Virginia and 12.44
for Alaska were such that there was a substantial probability for each
State that it met its target rate and indeed that it had a FY 1981
performance that was well below its target rate.  Although HCFA had
determined that Alaska had exceeded its target rate by 3.86, the width
of the 95 percent confidence interval for the composite point estimate
of the Excess was 12.44 points, and the possibility consequently existed
at the lower bound of the confidence interval that Alaska's performance
actually was 1.64, which was 2.36 points below its target.  Likewise,
although West Virginia exceeded its target rate by 2.10, the width of
the 95 percent confidence interval for the composite point estimate of
the Excess was 11.84 and the possibility still existed at the lower
bound of the confidence interval that West Virginia's performance was
2.72, which was 3.82 points below the target of 6.54.

Moreover, there were substantial other factors that would suggest that
the statistical data here were completely unreliable.

o  The FY 1978 base period estimates were unreliable and understated the
true error rates.  The underlying data was so replete with errors that
the United States General Accounting Office (GAO) recommended in several
interrelated reports that disallowances not be based on the 1978 base
period data; GAO found that there was gross understatement of errors by
the 1978 MQC process as well as significant deficiencies in the AFDC QC
data used by.the MQC system. 9/  States Exhibits (Exs.) 52, 53 and 54.
Moreover, HCFA's attempt to correct the base period rates was not
adequate.  HCFA provided a "strictly optional" reexamination of the base
period data for the federal subsample cases only.  There is no reason to
conclude that this limited reexamination would have produced valid base
period rates since AFDC cases previously found eligible and state sample
cases other than those in the federal subsample were not reevaluated.
HCFA provided an extremely short time frame for reopening, which indeed
had already passed before the states knew what the FY 1981 error rates
were.  Contrary to HCFA's assertions, the States are not bound to accept
the target rates as binding simply because there had been this very
limited opportunity to reopen and correct the FY 1978 rates. 10/  There
is no reason to conclude that this opportunity would have fully
corrected for either the extent to which the 1978 rates understated
errors or the fact that the MQC systems in 1978 and 1981 were not
comparable (as the GAO reports noted, HCFA continued to improve and
streamline administration of the system so that the 1981 system was far
more effective at error identification).  In any event, the opportunity
to correct is of no legal significance here since the question presented
is whether there is overall an adequate factual predicate for a
disallowance, an inquiry which also involves MQC statistical data other
than the FY 1978 rate..o  The FY 1981 error rates suffered from
substantial imprecision.  As with the confidence interval for the
composite estimate of the Excess, the confidence intervals for the FY
1981 rates show that the range of possible true values for the FY 1981
rates was so broad that it is not possible to conclude that the point
estimate was a reliable indicator of the true error rates for these
States.  See States Opening Brief -- Part II at 80-83; States Ex. 122
(HCFA letter of June 11, 1991 listing standard errors for original and
revised 1981 rates).  Moreover, the States presented persuasive expert
opinion and testimony on the effect of certain statistical phenomena on
the 1981 rates which provided further evidence of the inadequacy of
these rates as a basis for a disallowance, which HCFA was not able to
rebut.  States Exs. 55 and 86 and Transcript of 1987 Hearing.  The error
rate estimates were skewed upward for FY 1981 by outlier cases (cases of
higher value than the norm) in the sample.  See States Ex. 70. 11/  This
lends credence to the States' assertion that when estimates are
imprecise, as these were, the presence of outlier cases in the sample
required the use of statistical techniques to test and correct the
imprecise rates.  HCFA also failed to take into account the effect on
these rates of penalty bias and extreme value bias in the MQC system.
12/  Here, the imprecision in the rates exacerbates any .penalty bias
and increases the likelihood that over time a state will be sanctioned
when its true error rate was within the applicable tolerance.  The rates
for the States may also have been affected by a statistical phenomenon
called extreme value bias.  In essence, in this type of system where
there is a set of values stated for the 50 states, the values for those
states at the high or low end of the set of statistical values will tend
to be over and under estimates, respectively, of the actual value they
measure.  Accordingly, it is of some significance here that these States
were among the higher range of values for FY 1981 so that statistically
these estimates are likely to be overstatements of the true error rates.
Compared to all other states in the nation, Alaska had the second lowest
error rate estimate in the base period and the second highest error rate
estimate in FY 1981.  (Alaska's base error rate was determined to be 0.7
percent!)  The States' unrebutted evidence indicates that both estimates
are suspect because they are by comparison so unstable and because they
are so extreme in relation to all other states.  Also, beyond the
imprecision inherent in the small federal subsample size, the States
pointed to the effects of measurement error on the accuracy of the
rates.  It was undisputed that particularly at this point in the QC
process some error findings resulted either from the lack of an
objective definition of error (i.e., the absence of a clearcut policy or
requirement) or from different information available to the federal and
state reviewers.

HCFA's Argument that the States Themselves Were Responsible for the
Unreliability of the Estimates Is Both Unpersuasive and Unsubstantiated.

HCFA consistently argued that the States themselves were responsible for
the high level of imprecision in the estimates.  HCFA particularly
relied on two factors in support of this argument.  HCFA asserted that
the States had discretion in determining the size of their samples and
could have increased the precision of their estimates by increasing the
size of their samples.  HCFA also argued that the imprecision was the
result of the poor quality of the state reviews of case errors which in
turn resulted in low correlations between state and federal QC findings
and thereby required greater reliance for the error rate estimate on the
small federal subsample..We conclude that HCFA's arguments are without
any merit for the following reasons:

o  The record demonstrates that, more than any other factor, the size of
the federal subsample determined the level of imprecision in the
estimates.  In fact, Dr. Morris Hansen of Westat, Inc., the chief
architect of the QC program, stated:  "The size of the Federal subsample
is the principal factor in determining their [the error rate estimates']
precision." 13/  Thus, HCFA, which was in complete control of the
subsampling process and which had considerable discretion concerning the
size of the subsample, has to accept primary responsibility for the
ultimate lack of precision of the error rate estimates at issue.

o  It was HCFA's correlation measure that determined whether the error
rate estimate would rely principally on the smaller federal subsample or
the larger state sample.  Where the correlation between state and
federal error findings was low, the state's ultimate error rate estimate
would be based to a greater degree on the smaller federal subsample and
the attendant precision of the resulting error rate estimate would be
diminished.  Yet the findings of low correlation that actually occurred
here, which in fact diminished the precision of the error rate
estimates, did not necessarily reflect any failings or carelessness in
the State reviews, as HCFA argued.  The low correlation resulted only
because the small federal subsamples contained a very small number of
error cases and the effect of any disagreement from among this very
small number could easily result in a finding of low correlation.  In
fact, when all cases in the subsample, not just the error cases, were
compared with the state samples, all four of the States here had overall
agreement rates ranging from 95.5 percent to 98.3 percent.  States
Opening Brief -- Part II at 102.

o  Even the disagreement on a very small number of error cases which
occurred here cannot be said to reflect on the quality of either the
state or the federal review until those disagreements have been
resolved.  In fact, some of the original disagreements were resolved in
favor of the States in our earlier decision. 14/  The
remaining.disagreements have yet to be resolved and therefore cannot be
said to reflect negatively on the quality of the States' reviews.
(Among other issues, the disagreements still include an aspect of the
excess resource issue, which as noted above, we do need not to resolve
here.)  Furthermore, the States alleged, and HCFA did not effectively
refute, that the disagreements often involved cases where there may not
have been clear and unequivocal answers, cases that raised issues of
evolving federal policy.  See States Opening Brief -- Part II at
102-104.  This very possibility in the AFDC context was addressed by
Congress in the QC reform legislation for AFDC.

o  The States persuasively argued that, since the imprecision of their
estimates rested so heavily on the results of the federal subsample
review, they lacked any substantial opportunity to make improvements in
the precision of the estimates, and indeed, were unaware of the wide
confidence intervals associated with the estimates until those
statistics were obtained in the discovery proceedings in the first phase
of this case.  States Opening Brief -- Part II at 104.

The Michel Amendment Does Not Compel These Disallowances in the Absence
of Evidence Adequate to Show That the States Exceeded Their Error Rate
Targets for FY 1981.

HCFA consistently asserted that the Michel Amendment compelled these
disallowances notwithstanding any deficiencies in the underlying MQC
error rate estimates.  We disagree.  Rather, we conclude that the Michel
Amendment and the MQC regulations at 42 C.F.R. .431.802 authorized HCFA
to impose a disallowance only where there was valid and reliable
statistical data supporting the conclusion that a state exceeded its
error rate targets.  Moreover, nothing in the language of the various
statutory or legislative materials which comprise the Michel Amendment
sanction provisions or in the MQC regulations supports HCFA's assertion
here that we must reject the States' challenge to HCFA's use of this
data as a challenge to the MQC system itself.

HCFA acknowledged that the error rate data was imprecise and that
accepted statistical techniques might permit the development of more
precise rates for FY 1981.  Transcript of 1987 Hearing, Day 2 at 95-98.
Nevertheless, HCFA asserted in these appeals that disallowances were
mandated since the Michel Amendment adopted the existing MQC system and
that system had produced these rates.  This argument is not persuasive.
The actual language .used in the conference report manifests an intent
to exact fiscal "penalties" only for states that fail to improve their
program administration to meet the error rate targets.

While the MQC system was a sampling system and thus necessarily involved
an estimation process, Congress cannot reasonably be said to have
intended that the results of the sampling be adopted regardless of its
statistical validity. 15/  To the contrary, Congress can be presumed to
have intended that sound techniques would be employed.  Congress
specifically cautioned that the system for determining rates must ensure
equal treatment of the states. 16/  For us to accept HCFA's assertion
that it is obliged to pursue these disallowances notwithstanding the
significant imprecision in the supporting estimates would substantially
undercut the object of the MQC disallowance process of providing some
fiscal incentive to reduce erroneous payments.  The result HCFA espouses
would undercut the statutory scheme because disallowances would be based
not on a state's failure to meet the target but on the vagaries of
imprecise data.

We see no basis in the MQC system regulation and HCFA pointed to no part
of the MQC manuals that would prevent HCFA from developing more precise
error rate estimates for these States using accepted statistical
techniques.  In this regard, we do not read the Michel Amendment
language as establishing a mandate to disallow Medicaid monies
notwithstanding the adequacy of the supporting .estimates.  Congress
established the Michel Amendment targets for reductions by the states in
the levels of erroneous expenditures as part of the overall efforts to
improve program administration and reduce erroneous expenditures.  While
the system required by the conferees in 1979 specified use of the FY
1978 base period data to set the target rates, the conferees acted
before GAO or HHS had analyzed the 1978 data and learned that there were
deficiencies in that data.  By providing that funds would be disallowed
to the extent the error rates for FY 1981 and subsequent years exceeded
the target rates based on the FY 1978 rates, the Michel Amendment was an
obvious incentive for the states to reduce errors, but not an end in
itself.  The Michel Amendment and the QC disallowance regulations do not
provide for a lesser standard of proof for a MQC system-based
disallowance than would apply to disallowance actions generally.  HCFA
was not authorized or directed to disallow Medicaid monies without
reasonable factual support.  The object of imposing disallowances was
not to generate revenue or blindly apply the results of the QC sampling
process notwithstanding the lack of reliability of any particular rate.
Thus, implicit in the requirements of the Michel Amendment and the QC
regulations is a determination with an adequate factual basis for
concluding that the FY 1981 target error rates were exceeded.

Moreover, HCFA's characterization of this case as dealing only with how
much misspent money states would be permitted to retain ignores the
history and purpose behind the MQC system and disallowances based on
that system.  Accordingly, we are not persuaded by HCFA's assertion that
despite "the less than ideal statistical precision" of the MQC error
rate estimates these disallowances must be upheld since HCFA would have
had the authority to disallow all erroneous payments if the Michel
Amendment had not specifically provided for FFP up to the amount of each
year's target rate.  HCFA cited Maryland v. Mathews, 415 F.Supp. 1206
(D.D.C. 1976), for the proposition that the states were not entitled to
FFP for erroneous payments.

We conclude, however, that HCFA misread Maryland v. Mathews.  As
relevant here, that case stands for the proposition that the Secretary
had the authority by regulation to establish an empirically-based
tolerance level providing for FFP in certain erroneous payments in
recognition of the impossibility of operating completely error-free
programs.  Here, the Secretary exercised that authority in regulations
and established such a .tolerance.  Thus, the question presented is
whether the evidence is sufficient to establish that the tolerance was
exceeded.

In this regard, we note that the history of quality control efforts in
Medicaid and the other public assistance programs reveals a tension
between competing purposes of improving administration, and thus
reducing erroneous payments, and recovering erroneous payments.  This
particular dispute arose in the context of a tolerance level whereby FFP
was permitted in a specified level of erroneous payments.  Assuming that
the Michel Amendment is a legislative enactment, as HCFA argued it was,
it was then the culmination of previous efforts in both AFDC and
Medicaid (and Food Stamps) to reduce errors.  By setting the tolerance
level and the interim targets, Congress implicitly recognized the
principle that FFP was available in some level of erroneous payments.
While the Board has upheld HCFA's general authority to disallow
erroneous expenditures, it does not follow from this authority that
where there is an established tolerance level HCFA may determine that
the tolerance was not met using significantly imprecise or unreliable
data.  Without reliable factual evidence that the States failed to meet
the error targets, no disallowance was warranted.

HCFA Could Have Taken Steps to Improve the Reliability of the Estimates
or to Compensate for the Lack of Reliability But Chose Not to Do So.

Perhaps the most troubling aspect of these appeals was that HCFA did not
set any minimum standards of precision despite the problems with the FY
1978 base period and FY 1981 estimates.  It appears that there was no
point at which HCFA would have required some form of re-sampling or some
alternative methodology or auditing technique to compensate for the
extreme imprecision of the MQC estimates.  Obviously those states that
had less precise estimates, as the ones here, were placed at a
considerable disadvantage when compared to states with more precise
estimates.  The states with less precise estimates were just as much at
jeopardy for a QC disallowance even though their error rate estimates
would be much less likely to be a reflection of their true performance
in administering the Medicaid program.

One of the mandates of the Michel Amendment is that DHHS "establish a
system for determining error rates that insures equal treatment for all
States."  The four States.here were all among the top ten states having
imprecise point estimates.  The lack of a required minimum level of
precision together with the wide disparities in the precision of the
point estimates for all the states appears to be inconsistent with the
mandate of equal treatment in the Michel Amendment.  The goal was to
induce the States to improve their performance, not to subject them to
haphazard disallowances based on imprecise estimates that are not
adequate evidence of their actual performance.

Throughout the lengthy proceedings that have surrounded these
disallowances since originally issued, the States have identified a
variety of ways by which HCFA could have improved the precision of the
estimates or could have compensated for the lack of precision.  The
suggested techniques, some of which are widely used throughout
government and even within this Department, include:  stratified
sampling, reliance on the lower bound of the confidence interval rather
than the point estimate, anti-penalty bias modifications, elimination of
"outlier" cases, using a "process control band" with the point estimate,
"empirical Bayes estimates," and use of the States' samples modified by
federal sub-sample findings without the use of the regression formula.
While we recognize that some of these techniques may be more compatible
with the overall QC system HCFA has adopted, we do not need here to
assess the merits of any particular technique.  The record strongly
suggests that the application of at least some of these techniques would
have eliminated the disallowances.  When HCFA was specifically asked
during the hearing held for the earlier appeals whether it was legally
precluded from improving the precision or compensating for the lack of
precision, HCFA did not so argue.  Transcript of 1987 Hearing, Day 2 at
95-96.  We find that the availability of these techniques and the lack
of a legal barrier to their application here is a further indication
that the use of highly imprecise estimates is arbitrary and
unreasonable.


Conclusion

Based on the foregoing analysis, we reverse the FY 1981 MQC
disallowances for Alaska, Rhode Island, Washington, and West Virginia
based on our conclusion that the statistical MQC data upon which HCFA
relied was so .imprecise that it was inadequate proof that these States
in fact failed to meet MQC targets for error reduction for FY 1981.

 


 _______________________________ Donald F. Garrett

 


 _______________________________ Norval D. (John) Settle

 


 _______________________________ Cecilia Sparks Ford Presiding
 Board Member


1.  Among these other matters is the States' assertion that an excess
resource error should be counted as an error in the review month only if
in prior months an individual had not incurred sufficient medical
expenses to offset the excess resource amount.  While the States
presented a substantial question concerning whether it was necessary to
eliminate the multiple counting effect to properly value excess resource
errors (a different issue than we resolved in FY 1981 Medicaid Quality
Control Disallowances, DAB No. 948 (1988)), we did not need to reach
that question here.

2.  Both parties agreed that the record from DAB No. 948 should be
incorporated into the record for these appeals.

3.  The States argued that the Michel Amendment was never enacted by
Congress and that the FY 1981 regulations were therefore invalid.  At an
early stage in the proceedings of the prior appeal, the Board ruled that
it did not have jurisdiction to address the validity of the regulations,
since the Board is "bound by all applicable laws and regulations."  45
C.F.R. .16.14.  The States appealed this issue to court, but the United
States District Court for the District of Columbia granted the
Department's Motion to Dismiss on grounds of ripeness.  Alabama v.
Bowen, Civ. Act. No. 86-0841 (November 26, 1986).  The States continue
to maintain that the disallowances are unlawful because the Michel
Amendment regulations lack statutory authority, but did not further
pursue this issue before the Board because of the earlier jurisdictional
ruling.  States Opening Brief -- Part II at 6.

4.  For a discussion of the developments in quality control reviews, see
DAB No. 948, at 3-6.

5.  State plan requirements for the MQC system were established by 42
C.F.R. .431.800 (1979).  43 Fed. Reg. 45,188 (September 29, 1978) as
amended by 44 Fed. Reg. 17,935 (March 23, 1979).  Rules and procedures
for the disallowance of FFP in erroneous Medicaid payments detected
through the MQC system were established by 42 C.F.R. .431.801.  44 Fed.
Reg. 12,591 (March 7, 1979).  Effective only for FYs 1981 and 1982,
these initial disallowance regulations were replaced by the regulations
based on the Michel Amendment that are at issue here, 42 C.F.R. .431.802
(1981).  45 Fed. Reg. 6,326 (January 25, 1980).  These Michel Amendment
disallowance provisions are now at 42 C.F.R. .431.862 (1991).  55 Fed.
Reg. 22,142 (May 31, 1990).  The current Medicaid Eligibility Quality
Control system regulations implementing section 1903(u) of the Act are
at 42 C.F.R. ..431.800 -- 431.836 (1991).  55 Fed. Reg. 22,142 (May 31,
1990).  Other regulatory changes over the last decade are not relevant
here and thus are not described.

6.  HCFA furnished the following examples:  If a state's 1978 error rate
were seven percent, its FY 1981 target error rate would have been six
percent (seven percent minus one-third the difference between seven
percent and four percent).  Similarly, if the state's 1978 error rate
were 10 percent, its FY 1981 target rate would be eight percent (10
percent minus one-third the difference between 10 percent and four
percent).  HCFA Brief on Common Issues -- Part II at 9, n. 12.

7.  We note that the composite 95 percent confidence intervals for the
Excess Over Target were calculated during prior Board proceedings based
on the unrevised 1981 error rate estimates.  The parties here agreed to
continue to use the earlier data as an indicia of the precision of the
Excess amounts now before us based on the revised 1981 estimates.  See
States Opening Brief --Part II at 67-74.  In fact, the 95 percent
confidence intervals surrounding solely the revised 1981 estimates were
also very wide, and thus suggest that continued reliance on the
composite intervals by the parties was reasonable.

8.  We do not here address the rather esoteric issue of the possibility
contemplated by the confidence interval that the State's "real"
performance was better than perfect.

9.  The GAO 1981 Report to Congress "Medicaid's Quality Control System
Is Not Realizing Its Full Potential" stated at page 16 --

 Regarding the accuracy of the base period data, HHS agreed that
 there were problems. . . .  HCFA believes that the accuracy of
 MQC data has substantially and steadily improved . . . .  While
 we trust that the accuracy has improved from that of the base
 period and will improve further when our recommendations are
 implemented, the current formula for Medicaid MQC disallowances
 still measures progress in reducing error rates from the base
 period.  To the extent that the base period data are
 inaccurate--and . . . we found significant inaccuracies--any
 disallowances made under the formula would be inaccurate . . . .
 and disallowances calculated using the current formula would not
 be fully supported.

10.  The seven states which elected to reopen their rates had their base
period rates increased.  West Virginia and Rhode Island were among those
seven.

11.  It was likely that the outlier cases substantially increased the
error rate estimates; HCFA did nothing to correct the estimates to
account for this effect.  For example, for Rhode Island, there was an
outlier case with a dollar value of $2,873 in the sample for the
April-September 1981 sample period.  Without this outlier case the error
rate for this sample period would have been .16% not 6.83% and the FY
1981 rate would be 2.06% not 5.51%.  States Opening Brief -- Part II at
75; States Ex. 55 at 18-19.  While this data is for the Rhode Island
rate prior to its recalculation based on the valuation of excess
resource errors to account for DAB No. 948, no adjustment was made to
correct for the effect of such outlier cases.  Therefore, we conclude
that the distorting effect of such cases which this data illustrates was
still a factor for the recalculated rates directly at issue here.

12.  Penalty bias exists because over time states can only be sanctioned
and never credited.  Consequently, to the extent the rates are over or
under estimates this will not balance out.

13.  Letter dated February 8, 1988 from Dr. Morris Hansen, States Ex.
130 at 1.

14.  The States demonstrated that the resolution of just a small number
of error cases had a tremendous impact on the error rate estimates.

15.  We note that it is well established that acceptable evidence can be
produced by valid statistical sampling techniques.  See Rosado v. Wyman,
322 F.Supp. 1173, 1180-1 (1977).  We have specifically stated that valid
statistical sampling produces a result which has a high degree of
probability of being at least as accurate as a case-by-case review.
Tennessee Dept. of Health and Environment, DAB No. 898 (1987).  This
decision in no way affects existing authority upholding the use of
statistical sampling techniques.

16.  While HCFA asserted that by using a federal subsampling process the
MQC system assured equal treatment among the states, this one
characteristic of the MQC sampling process does not preclude HCFA from
taking other reasonable steps to assure that no state is unfairly
determined to have exceeded its target rate for a particular