United States Court of Appeals
FOR THE DISTRICT OF COLUMBIA CIRCUIT
Argued November 3, 2020 Decided August 13, 2021
No. 18-5326
UNITEDHEALTHCARE INSURANCE COMPANY, ET AL.,
APPELLEES
v.
XAVIER BECERRA, IN HIS OFFICIAL CAPACITY AS SECRETARY
OF HEALTH AND HUMAN SERVICES, ET AL.,
APPELLANTS
Appeal from the United States District Court
for the District of Columbia
(No. 1:16-cv-00157)
Weili J. Shaw, Attorney, U.S. Department of Justice,
argued the cause for appellants. With him on the briefs were
Ethan P. Davis, Acting Assistant Attorney General, and Mark
B. Stern, Attorney. Michael S. Raab, Attorney, entered an
appearance.
Daniel Meron argued the cause for appellees. With him
on the brief was Matthew M. Shors.
David W. Ogden, Brian M. Boynton, and Kevin M. Lamb
were on the brief for amicus curiae America’s Health
Insurance Plans in support of appellees.
2
Before: ROGERS, PILLARD and WALKER, Circuit Judges.
Opinion for the Court filed by Circuit Judge PILLARD.
PILLARD, Circuit Judge: UnitedHealthcare Insurance
Company and other Medicare Advantage insurers under the
umbrella of UnitedHealth Group Incorporated (collectively,
UnitedHealth) challenge a rule the Centers for Medicare and
Medicaid Services (CMS) promulgated under the Medicare
statute, 42 U.S.C. §§ 1301-1320d-8, 1395-1395hhh. The
Overpayment Rule is part of the government’s ongoing effort
to trim unnecessary costs from the Medicare Advantage
program. Neither Congress nor CMS has ever treated an
unsupported diagnosis for a beneficiary as valid grounds for
payment to a Medicare Advantage insurer. Consistent with
that approach, the Overpayment Rule requires that, if an
insurer learns a diagnosis it submitted to CMS for payment
lacks support in the beneficiary’s medical record, the insurer
must refund that payment within sixty days. The Rule
couldn’t be simpler. But understanding UnitedHealth’s
challenge requires a bit of context.
As explained in more detail below, people who are
eligible for Medicare may elect to receive their health
insurance through a private insurer under Medicare
Advantage rather than directly through the government under
traditional Medicare, and approximately forty percent of
beneficiaries have chosen Medicare Advantage. CMS pays
private Medicare Advantage insurers, in a prospective lump
sum each month, the amount it expects a month’s care would
otherwise cost CMS in direct payments to healthcare
providers treating the same beneficiaries under traditional
Medicare. For each Medicare Advantage beneficiary, CMS
pays the insurer a per-capita amount that varies according to
3
demographic characteristics and diagnoses that CMS has
determined, based on its past experience in traditional
Medicare, to be predictive of healthcare costs.
Payments to the Medicare Advantage program depend on
participating insurers accurately reporting to CMS their
beneficiaries’ salient demographic information and medically
documented diagnosis codes. To better control erroneous
payments, including those garnered from reported—but
unsupported—diagnoses, Congress in 2010 amended the
Medicare program’s data-integrity provisions. The
amendment specified a sixty-day deadline for reporting and
returning identified overpayments and confirmed that such
payments not promptly returned may trigger liability under
the False Claims Act. See id. § 1320a-7k(d). CMS
promulgated the Overpayment Rule to implement those
controls on Medicare Advantage. See 42 C.F.R. § 422.326.
As relevant here, the Overpayment Rule establishes that, if a
Medicare Advantage insurer has received a payment
increment for a beneficiary’s diagnosis and discovers that
there is no basis for that payment in the underlying medical
records, that is an overpayment that the insurer must correct
by reporting it to CMS within sixty days for refund. See
Medicare Program; Contract Year 2015 Policy and Technical
Changes to the Medicare Advantage and the Medicare
Prescription Drug Benefit Programs, 79 Fed. Reg. 29,844,
29,921 (May 23, 2014) (hereinafter Overpayment Rule), J.A.
64.
UnitedHealth claims that it is unambiguous in the text of
the Medicare statute that the Overpayment Rule is subject to a
principle of “actuarial equivalence,” and that the Rule fails to
comply. See 42 U.S.C. § 1395w-23(a)(1)(C)(i). But actuarial
equivalence does not apply to the Overpayment Rule or the
statutory overpayment-refund obligation under which it was
4
promulgated. Reference to actuarial equivalence appears in a
different statutory subchapter from the requirement to refund
overpayments, and neither provision cross-references the
other. Further, the actuarial-equivalence requirement and the
overpayment-refund obligation serve different ends. The role
of the actuarial-equivalence provision is to require CMS to
model a demographically and medically analogous
beneficiary population in traditional Medicare to determine
the prospective lump-sum payments to Medicare Advantage
insurers. The Overpayment Rule, in contrast, applies after the
fact to require Medicare Advantage insurers to refund any
payment increment they obtained based on a diagnosis they
know lacks support in their beneficiaries’ medical records.
UnitedHealth contends that the actuarial-equivalence
principle reaches beyond its statutory home to impose an
implied—and functionally prohibitive—legal precondition on
the requirement to return known overpayments. As
UnitedHealth would have it, Congress clearly intended
enforcement of the statutory overpayment-refund obligation,
which the Overpayment Rule essentially parrots, to depend on
a prior determination of actuarial equivalence. That principle,
UnitedHealth says, prevents CMS from recovering
overpayments under the Rule unless CMS first shows that the
rate of payment errors to healthcare providers in traditional,
fee-for-service Medicare is lower than the rate of payment
errors to the Medicare Advantage insurer, or that CMS
comprehensively audited the data from traditional Medicare
before using it in the complex regression model—the CMS
Hierarchical Condition Category (CMS-HCC) risk-adjustment
model—that predicts the cost to insure Medicare Advantage
beneficiaries.
There is no legal or factual basis for UnitedHealth’s
claim. Actuarial equivalence is a directive to CMS. It
5
describes the goal of the risk-adjustment model Congress
directed CMS to develop. It does not separately apply to the
requirement that Medicare Advantage insurers avoid known
error in their payment requests. It assuredly does not
unambiguously demand that, before CMS can collect known
overpayments from Medicare Advantage insurers, it must
engage in unprecedented self-auditing to eliminate an
imagined bias in the body of traditional Medicare data CMS
used in its regressions. The implausibility that Congress
would have so intended is underscored by the lack of
parallelism between the context and effects of, on one hand,
unsupported diagnoses in the traditional Medicare data CMS
uses to model generally applicable risk factors and, on the
other, the specific errors the Overpayment Rule targets.
Even if actuarial equivalence applied as UnitedHealth
suggests, it would be UnitedHealth’s burden to show the
systematically skewed inaccuracies on which its theory
depends, which it has not done. Also fatal to UnitedHealth’s
claim is that it never challenged the values CMS assigned to
the risk factors it identified or the level of the capitation
payments resulting from CMS’s risk-adjustment model. It
cannot belatedly do so in the guise of a challenge to the
Overpayment Rule.
UnitedHealth’s next claim relies on the Medicare
statute’s requirement that CMS annually compute and publish
certain traditional Medicare data “using the same
methodology as is expected to be applied in making
payments” to Medicare Advantage insurers. Id. § 1395w-
23(b)(4)(D). That “same methodology” requirement does not
bear on the overpayment-refund obligation. Meant to
facilitate Medicare Advantage insurers’ bidding for contracts
with CMS, that requirement merely clarifies that, in
computing the data it publishes, CMS must use the same risk-
6
adjustment model that it already uses to set monthly payments
to Medicare Advantage insurers; like the actuarial-
equivalence requirement, it says nothing about what
constitutes an “overpayment.”
UnitedHealth’s final claim is that the Overpayment Rule
is arbitrary and capricious in violation of the Administrative
Procedure Act (APA). That claim hinges on what
UnitedHealth sees as an unexplained inconsistency between
the Overpayment Rule and another error-correction
mechanism to which Medicare Advantage insurers are
subject: Risk Adjustment Data Validation (RADV) audits.
With those audits, CMS proposed a systemic adjustment
involving the traditional Medicare data used to model risk
factors to account for any errors in that data set before
requiring any contract-level repayments from insurers.
UnitedHealth sees inconsistency in obligating repayments
under the Overpayment Rule without any such adjustment.
But the system-level adjustment that CMS said it would apply
in the context of contract-level RADV audits came in direct
response to concerns about actuarial equivalence. Because
we hold that the actuarial-equivalence requirement does not
pertain to the statutory overpayment-refund obligation or the
Overpayment Rule challenged here, and the two error-
correction mechanisms are plainly distinguishable in other
ways, CMS’s one-time intention to apply the adjustment in
one context but not the other was reasonable.
In sum, nothing in the Medicare statute’s text, structure,
or logic applies actuarial equivalence to its separate
overpayment-refund obligation, and thus the Overpayment
Rule does not violate actuarial equivalence. For much the
same reasons, we reject UnitedHealth’s claim that the Rule
violates the statute’s “same methodology” requirement, and
we also deny its claim that the Rule is arbitrary and capricious
7
as an unexplained departure from prior policy. We therefore
reverse the district court’s grant of summary judgment to
UnitedHealth and its resulting vacatur of the Overpayment
Rule and remand for the district court to enter judgment in
favor of CMS.
BACKGROUND
Overpayment to Medicare Advantage insurers is a serious
drain on the Medicare program’s finances. In 2016 alone,
audits of the data submitted by Medicare Advantage insurers
to CMS showed that CMS paid out an estimated $16.2 billion
for unsupported diagnoses, equal to “nearly ten cents of every
dollar paid to Medicare Advantage organizations.” United
States ex rel. Silingo v. WellPoint, Inc., 904 F.3d 667, 673
(9th Cir. 2018) (citing James Cosgrove, U.S. Gov’t
Accountability Off., GAO-17-761T, Medicare Advantage
Program Integrity: CMS’s Efforts to Ensure Proper Payments
1 (2017), https://www.gao.gov/assets/690/685934.pdf).
UnitedHealth is the Nation’s largest provider of Medicare
Advantage plans. Meredith Freed et al., A Dozen Facts About
Medicare Advantage in 2020, Kaiser Family Found. (Jan. 13,
2021), https://www.kff.org/medicare/issue-brief/a-dozen-
facts-about-medicare-advantage-in-2020/.
A. Statutory and regulatory background
1.
Since 1965, most older adults and many people with
disabilities in the United States have received their health
insurance through Medicare, administered by CMS. In
Medicare Parts A and B, or “traditional” Medicare, CMS
itself acts as the insurer, paying healthcare providers directly
for beneficiaries’ medical services. Medicare Part A covers
inpatient hospital treatment and other institutional care and is
8
generally provided without charge to Medicare-eligible
individuals. But for outpatient services, like visits to doctors’
offices, the Medicare statute provides Medicare-eligible
individuals a choice of whether and how to receive such
coverage: They can receive that, too, by having the
government pay providers for services, under Medicare Part
B; or they can opt for private insurance paid for at least in part
by the government, under Medicare Part C, also known as
Medicare Advantage (and formerly known as
Medicare+Choice).
Unlike Medicare Part A, coverage under Medicare Part B
and Medicare Advantage generally requires payments from
beneficiaries to the government or, if applicable, private
insurance companies. Medicare Advantage insurers must
provide coverage of at least the same services as Medicare-
eligible individuals would receive through traditional
Medicare, 42 U.S.C. § 1395w-22(a), and those private
insurers often attract subscribers by offering additional
benefits, such as dental and vision coverage, that they are able
to include due to efficiencies and other cost-saving measures.
More than twenty-four million Americans, or nearly forty
percent of all Medicare beneficiaries, choose to receive their
health insurance through Medicare Advantage. See generally
Freed et al., supra.
Medicare Parts A and B and Medicare Advantage pay
healthcare providers in different ways. Under Medicare Part
A, CMS pays a hospital or institutional care provider based on
a beneficiary’s diagnoses at the time of discharge, which
translate to a “Diagnosis-Related Group.” Under Medicare
Part B, CMS pays outpatient providers on a fee-for-service
basis under fee schedules that set the payment for each service
provided, such as an office visit, examination, or
immunization. A beneficiary’s diagnoses do not directly
9
affect the level of payment made to a healthcare provider
under Part B, but because a service is reimbursable only if it
is “reasonable and necessary for the diagnosis or treatment of
illness or injury,” 42 U.S.C. § 1395y(a)(1)(A), providers still
must generally submit diagnosis codes to CMS showing why
a beneficiary received the services that she did.
Private Medicare Advantage insurers likewise pay
healthcare providers based on the services provided to
beneficiaries but, as noted above, under Part C those insurers
themselves receive in advance a monthly lump sum from
CMS for every beneficiary that they enroll, without regard to
the services that the beneficiaries will actually receive. The
prospective, lump-sum payment approach has the potential to
curb costly and unnecessary overtreatment that the fee-for-
service approach tends to encourage, and it favors
preventative care and other health-protective measures,
enabling cost efficiencies that can elude a fee-for-service
system. See Advance Notice of Methodological Changes for
CY 2004 Medicare+Choice Payment Rates, at 5 (Mar. 28,
2003), J.A. 115. The core idea is that a Medicare Advantage
insurer that covers all of a beneficiary’s health care at least as
well as traditional Medicare but does so at lower cost may
pocket the difference as earned revenue, or pass along that
revenue to beneficiaries in the form of extra benefits meant to
entice and retain subscribers.
2.
It is the Medicare statute that requires CMS to pay
Medicare Advantage insurers in advance, on a monthly basis,
for each of the Medicare-eligible beneficiaries that they
insure. 42 U.S.C. § 1395w-23(a)(1)(A). The statute also
requires CMS to adjust those monthly, per-capita payments to
reflect what traditional, fee-for-service Medicare paid in a
10
base year for a beneficiary population modeled—by reference
to demographics, diagnoses, and other factors CMS selects—
to be actuarially equivalent to the Medicare Advantage
insurer’s beneficiary population. Id. § 1395w-23(a)(1)(C)(i).
Specifically, Congress instructed that the Secretary of Health
and Human Services (HHS)
shall adjust the payment amount . . . for such risk
factors as age, disability status, gender, institutional
status, and such other factors as the Secretary
determines to be appropriate, including adjustment for
health status . . . , so as to ensure actuarial
equivalence. The Secretary may add to, modify, or
substitute for such adjustment factors if such changes
will improve the determination of actuarial
equivalence.
Id. The point of the Secretary’s discretion to select, and
obligation to apply, risk factors is “to ensure that [Medicare
Advantage insurers] are paid appropriately for their plan
enrollees (that is, less for healthier enrollees and more for less
healthy enrollees).” Medicare Program; Establishment of the
Medicare Advantage Program, 70 Fed. Reg. 4588, 4657 (Jan.
28, 2005), J.A. 92. Indeed, “the goal of risk adjustment” is
“to pay [Medicare Advantage] plans accurately.” 152 Cong.
Rec. S438-02 (daily ed. Feb. 1, 2006) (statement of Sen.
Grassley).
Specifically, identifying salient risk factors enables CMS
to determine prospectively, based on Medicare Advantage
beneficiaries’ actuarially relevant, known demographic and
health characteristics, the per-capita payment rate that will
fairly compensate that Medicare Advantage insurer. More
broadly, the demographic- and health-adjusted, capitated
payment scheme is designed to blunt the incentives to enroll
11
only the healthiest, and thus least expensive, beneficiaries
while steering clear of the sickest and costliest—thereby
rewarding Medicare Advantage insurers to the extent that they
achieve genuine efficiencies over traditional Medicare in
addressing the same health conditions. See Gregory C. Pope
et al., Risk Adjustment of Medicare Capitation Payments
Using the CMS-HCC Model, Health Care Fin. Rev., Summer
2004, at 119, 119-20, J.A. 487-88; see also H.R. Rep. No. 105-
217, at 585 (1997) (Conf. Rep.); H.R. Rep. No. 108-391, at
524-25 (2003) (Conf. Rep.).
To adjust the monthly payments, CMS uses a model—
called the CMS Hierarchical Condition Category, or CMS-
HCC, risk-adjustment model—that it periodically studies and
improves based on clinical information and cost data. The
model isolates demographic characteristics CMS has
determined to be predictive of differing costs of care,
including the risk factors expressly mentioned in the statute:
age, sex, disability status, and whether the beneficiary lives in
a long-term institutional setting. See 42 U.S.C. § 1395w-
23(a)(1)(C)(i). It adjusts for health status by isolating cost-
predictive diagnoses. CMS uses expert judgment to
determine, for example, “which diagnosis codes should be
included, how they should be grouped, and how the
diagnostic groupings should interact for risk adjustment
purposes.” Gregory C. Pope et al., Evaluation of the CMS-
HCC Risk Adjustment Model: Final Report 8 (Mar. 2011),
J.A. 525. Diagnostic categories must be reasonably specific
and clinically meaningful. And, to fine-tune its predictive
utility, CMS’s model accounts for interactions between
multiple diagnoses where total joint costs are more than
additive. CMS also establishes a hierarchy of diagnoses to
avoid double counting, zeroing out the cost effects of less
severe disease manifestations when a patient also has a more
12
severe diagnosis that fully accounts for treatment costs for
both. Id.
CMS’s risk-adjustment model applies a regression
analysis to the mass of data from traditional Medicare for a
previous year to convert each demographic and health
characteristic into an expected cost of coverage. See id. at 2,
J.A. 519. CMS inputs traditional Medicare beneficiaries’
data, including the diagnosis codes that healthcare providers
are required to report (even though, as noted above, CMS
itself bases Medicare Part B payments on services, not
diagnoses), along with the total cost for covering those
beneficiaries. The model isolates the anticipated cost of care
associated with each demographic and health characteristic by
first determining the average marginal cost of that
characteristic in dollars and then dividing that dollar amount
by traditional Medicare’s average cost per beneficiary. That
process produces a “relative factor” for each demographic and
health characteristic. The model “use[s] data from a large
pool of beneficiaries (full sample sizes over 1 million for the
CMS-HCC models) to estimate predicted costs on average for
each of the component factors (e.g., age-sex, low income
status, individual disease groups).” Id. at 5, J.A. 522. Using
regression analysis on such a vast data sample mutes the
effect of individual errors in traditional Medicare data, so long
as errors are not so widespread or systemically skewed as to
raise or lower the values of particular relative factors. See id.;
see also Amy Gallo, A Refresher on Regression Analysis,
Harv. Bus. Rev. (Nov. 4, 2015), https://hbr.org/2015/11/a-
refresher-on-regression-analysis.
To enable CMS to apply those relative factors to pay
Medicare Advantage insurers at the correct risk-adjusted rate,
the insurers must report to CMS the salient demographic and
health characteristics of each of their Medicare-eligible
13
beneficiaries. 42 C.F.R. § 422.310(b), (d). CMS then
combines the relative factors for a particular beneficiary to
arrive at her individualized overall “risk score.” See Pope et
al., Evaluation of the CMS-HCC Risk Adjustment Model:
Final Report 15, J.A. 532. CMS posits that an “average
beneficiary” in traditional Medicare has a risk score of 1.0. If
a Medicare Advantage beneficiary has a risk score of exactly
1.0, CMS pays the insurer the base payment rate for that
beneficiary’s location. For Medicare Advantage beneficiaries
with risk scores above 1.0, meaning they are of higher-than-
average risk, CMS pays insurers more than the base payment
rate; for beneficiaries with risk scores below 1.0, the
payments are correspondingly lower than the base rate. But
Medicare Advantage beneficiaries are not presumptively
scored as 1.0; the per-capita payments that CMS makes to
insurers instead depend on an aggregation of the
beneficiaries’ cost-predictive demographic and diagnostic
factors.
CMS illustrates the operation of relative factors with an
example:
[U]nder the 2014 model, a 72-year-old woman living
independently (relative factor 0.348), with diabetes
without complications (relative factor 0.118), and
multiple sclerosis (relative factor 0.556) would have a
total risk score of 1.022, which means that she is
expected to cost Medicare slightly more than the
average traditional Medicare beneficiary (who would
by definition have a risk score of 1.0).
Gov’t Br. 7 (citing Announcement of CY 2014 Medicare
Advantage Capitation Rates and Medicare Advantage and
Part D Payment Policies and Final Call Letter, at 67-68 (Apr.
1, 2013), J.A. 276-77). In other words, as a woman near the
14
younger end of the Medicare-eligible population and living
outside any long-term institutional setting, this sample
beneficiary starts with a risk score well below the overall
Medicare average. The fact that she suffers from diabetes
raises her risk score, but not by much, presumably because
she has not experienced complications and ordinary diabetes
care is not as costly as many other conditions common among
older Americans. The larger bump, putting her over the
average predicted cost of care even for the cost-intensive
Medicare population, is that she suffers from multiple
sclerosis. A Medicare Advantage insurer providing coverage
to this woman therefore “would be paid 102.2 percent of the
relevant base rate.” Id. at 8.
This example illustrates the importance of risk-adjusted
payment. Assume a similar woman, but without her
diagnoses. With a risk score of just 0.348, her care would
then be predicted to be far less expensive than that of the
average Medicare beneficiary, whose risk score is, by
definition, 1.0. If Medicare Advantage insurers were paid an
unadjusted base rate for every beneficiary, they could receive
an enormous, and unjustified, net surplus insofar as they
enrolled beneficiaries with such low anticipated costs.
Conversely, an unadjusted, per-capita base payment would
likely fall far short of fairly compensating a Medicare
Advantage insurer for the costs of care for the woman in the
example with both of the posited diagnoses, and the shortfall
would only grow with any added complications or diagnoses
she developed.
There is some evidence that Medicare Advantage insurers
in fact have tended to attract healthier-than-average
beneficiaries—perhaps because of the additional premiums
they may charge, and the well-established correlation between
wealth and health. See Is Medicare Advantage More Efficient
15
than Traditional Medicare?, Nat’l Bureau of Econ. Rsch.
(Mar. 2016), https://www.nber.org/bah/2016no1/medicare-
advantage-more-efficient-traditional-medicare; see also Pope
et al., Risk Adjustment of Medicare Capitation Payments
Using the CMS-HCC Model, at 119-20, J.A. 487-88; Pope et
al., Evaluation of the CMS-HCC Risk Adjustment Model:
Final Report 7, J.A. 524. Without the corrective provided by
risk-adjusting the capitated payment amounts, payment levels
would not be fair, and incentives to attract the healthy and
deflect the sick would be overwhelming.
CMS determines the base payment rate—which, again, is
the amount a Medicare Advantage insurer would receive for
any beneficiary with a risk score of exactly 1.0, and which is
the denominator for calculation of every capitated payment to
Medicare Advantage—by reference to traditional Medicare’s
per-capita expenditures in a particular place and bids
submitted by Medicare Advantage insurers. Each county in
the United States has its own base rate, and every year
Medicare Advantage insurers bid for contracts after CMS
announces each county’s benchmark for the coming year. See
42 U.S.C. § 1395w-23(b)(1)(B). To inform Medicare
Advantage insurers’ bids to participate in the program, the
Medicare statute requires CMS to compute and publish, on an
annual basis, the “average risk factor” for traditional
Medicare beneficiaries in each county. Id. § 1395w-
23(b)(4)(D). The statute specifies that the published average
risk factor must be “based on diagnoses for inpatient and
other sites of service, using the same methodology as is
expected to be applied in making payments under subsection
(a),” i.e., the subsection that includes the actuarial-
equivalence requirement. Id. UnitedHealth separately claims
the “same methodology” criterion supports its challenge to the
Overpayment Rule.
16
3.
CMS’s regulations have long obligated Medicare
Advantage insurers to certify the accuracy of the data that
they report to CMS. Since 2000, those regulations have made
it “a condition for receiving a monthly payment” that a
Medicare Advantage insurer
agrees that its chief executive officer (CEO), chief
financial officer (CFO), or an individual delegated the
authority to sign on behalf of one of these officers,
and who reports directly to such officer, must request
payment under the contract [with CMS] on a
document that certifies (based on best knowledge,
information, and belief) the accuracy, completeness,
and truthfulness of relevant data that CMS requests.
42 C.F.R. § 422.504(l); see also United States ex rel. Swoben
v. UnitedHealthcare Ins. Co., 848 F.3d 1161, 1168 & n.2 (9th
Cir. 2016) (citing 42 C.F.R. § 422.502(l) (2000)). CMS’s
regulations specifically apply that obligation to the data
Medicare Advantage insurers report to CMS to identify their
beneficiaries’ actuarially salient attributes—i.e., demographic
and health characteristics, including diagnosis codes. See 42
C.F.R. § 422.504(l)(2) (referencing data reported under 42
C.F.R. § 422.310).
But, as Congress has recognized, even accurate diagnosis
codes that Medicare Advantage insurers report can lead to
disproportionately high payments to insurers. That is because
Medicare Advantage insurers have a financial incentive to
code intensely—i.e., to make sure that they report to CMS
their beneficiaries’ every diagnosis—given that their monthly,
per-capita payments are higher to the extent that their
beneficiaries have more or graver diagnoses. Meanwhile,
healthcare providers to traditional Medicare beneficiaries lack
17
that same incentive because their payments from CMS depend
on services rendered, not diagnoses. See U.S. Gov’t
Accountability Off., GAO-12-51, Medicare Advantage: CMS
Should Improve the Accuracy of Risk Score Adjustments for
Diagnostic Coding Practices 2 (Jan. 2012), J.A. 546. Thus, if
one were to imagine that traditional Medicare and Medicare
Advantage had identical populations of beneficiaries, the
latter would generally end up reporting more diagnoses (and
therefore appear sicker and receive additional payments) even
though their true health conditions were the same. To account
for that difference in incentives and coding practices,
Congress enacted a Coding Intensity Adjuster that reduces the
risk scores of all Medicare Advantage beneficiaries by a
specified percentage. See Health Care and Education
Reconciliation Act of 2010, Pub. L. No. 111-152,
§ 1102(e)(3)(D), 124 Stat. 1029, 1046. For 2019, Congress
set that reduction at a minimum of 5.9 percent. 42 U.S.C.
§ 1395w-23(a)(1)(C)(ii)(III). The Coding Intensity Adjuster
does not, however, address unsupported or inaccurate codes
reported by Medicare Advantage insurers, but only the
practice, relative to traditional Medicare, of overreporting
codes that are nonetheless accurate.
UnitedHealth’s challenge to the Overpayment Rule
adverts to yet another data-integrity measure providing for
Risk Adjustment Data Validation, or RADV, audits. To
supplement the regulatory obligations on Medicare Advantage
insurers to certify the accuracy of the diagnosis codes and
other data they report to CMS, and because CMS cannot
confirm in real time the data insurers submit for their millions
of beneficiaries, CMS seeks to confirm that its payments to
insurers are correct by retrospectively spot-checking the data
submissions going back several years. See 42 C.F.R.
§ 422.310(e); see also Medicare Program; Policy and
Technical Changes to the Medicare Advantage and the
18
Medicare Prescription Drug Benefit Programs, 74 Fed. Reg.
54,634, 54,674 (Oct. 22, 2009), J.A. 96. For these RADV
audits, CMS selects a subset of Medicare Advantage insurers
and compares a sample of their reported diagnosis codes to
the underlying medical charts and records for the relevant
beneficiaries. See Medicare Program; Policy and Technical
Changes to the Medicare Advantage and the Medicare
Prescription Drug Benefit Programs, 74 Fed. Reg. at 54,674,
J.A. 96. The Medicare Advantage insurers must return to
CMS any payments that an audit reveals were based on
unsupported diagnoses—that is, diagnoses reported to CMS
but that the audit found lack support in the relevant
beneficiaries’ medical record documentation. See id.
CMS has conducted such audits for well over a decade,
and their results show that a significant number of reported
diagnoses are in fact unsupported. See, e.g., U.S. Dep’t of
Health & Human Servs., Off. of Inspector Gen., Risk
Adjustment Data Validation of Payments Made to PacifiCare
of Texas for Calendar Year 2007, A-06-09-00012, at 4 (May
2012), J.A. 471 (stating that the risk scores for forty-three out
of 100 sampled beneficiaries of the audited insurer “were
invalid because the diagnoses were not supported”); U.S.
Dep’t of Health & Human Servs., Off. of Inspector Gen., Risk
Adjustment Data Validation of Payments Made to PacifiCare
of California for Calendar Year 2007, A-09-09-00045, at i
(Nov. 2012), J.A. 476 (stating that the risk scores for forty-
five out of 100 sampled beneficiaries “were invalid because
the diagnoses were not supported by the documentation that
[the Medicare Advantage] insurer provided”).
Medicare Advantage insurers’ obligation to return
mistaken payments pursuant to RADV audits differs from
their obligation under the Overpayment Rule: With the
former, insurers are required to refund payments based on
19
unsupported diagnoses that CMS discovers through its audit,
whereas with the latter, insurers are required to refund
payments based on unsupported diagnoses that they
themselves discover through the course of their business.
CMS also audits traditional Medicare data, although it does so
through different mechanisms that may result in a lower
percentage of traditional Medicare payment claims being
audited than Medicare Advantage ones. See Gov’t Br. 35-38;
Appellees Br. 42-43.
In 2008, CMS announced an expansion of its RADV
audit program for Medicare Advantage: Rather than requiring
repayments only for the unsupported diagnosis codes
identified in the limited sample itself, CMS would take the
payment error in an audited sample, extrapolate that error rate
across CMS’s entire contract with that Medicare Advantage
insurer, and require the insurer to make a repayment based on
the extrapolated, or contract-level, degree of error. See
Medicare Program; Policy and Technical Changes to the
Medicare Advantage and the Medicare Prescription Drug
Benefit Programs, 74 Fed. Reg. at 54,674, J.A. 96; see also
Announcement of Calendar Year (CY) 2009 Medicare
Advantage Capitation Rates and Medicare Advantage and
Part D Payment Policies, at 22 (Apr. 7, 2008). (Because not
all errors are created equal—that is, some are more costly
than others—the extrapolated error rate would account for the
magnitude of the errors by factoring in the difference between
original and corrected payment amounts in an audited
sample.) In late 2010, CMS sought comments on its proposal
for contract-level RADV audits, and in early 2011 various
commenters, including UnitedHealth and the American
Academy of Actuaries, objected.
One criticism the commenters leveled at expanded
RADV audits was that, “[u]nder sound actuarial principles, it
20
is impossible to know whether [Medicare Advantage insurers]
have been paid accurately by conducting a review of the
medical records supporting [Medicare Advantage] coding,
without also considering the medical records supporting
[traditional Medicare] coding.” Aetna Inc.’s Comments on
Proposed Payment Error Calculation Methodology for Part C
Organizations Selected for Contract-Level RADV Audits, at 4
(Jan. 21, 2011), J.A. 298. In other words, “CMS must audit
and validate both [a Medicare Advantage insurer’s data and
the traditional Medicare data that goes into the risk-
adjustment model] before extrapolating any potential RADV
audit results” and requiring the insurer to return amounts
thereby identified as excessive. Humana Inc., Comment on
RADV Sampling and Error Calculation Methodology, at 3
(Jan. 21, 2011), J.A. 334. “If it does not, CMS will
dramatically underpay [Medicare Advantage insurers] for the
benefits they provided to Medicare beneficiaries,” in violation
of the Medicare statute’s actuarial-equivalence requirement.
Id.; see also id. at 5, J.A. 336.
In a move that UnitedHealth describes as important
context for this case, CMS responded to the comments by
announcing in 2012 that it would apply a Fee-for-Service, or
FFS, Adjuster before requiring repayments based on contract-
level RADV audits. With the FFS Adjuster, Medicare
Advantage insurers would be liable for repayments only to the
extent that their extrapolated, contract-level payment errors,
i.e., the dollar amounts that they received in error, exceed any
offsetting payment error in traditional Medicare. CMS said
that it would determine the actual amount of the FFS Adjuster
“based on a RADV-like review of records submitted to
support [traditional Medicare] claims data.” Notice of Final
Payment Error Calculation Methodology for Part C Medicare
Advantage RADV Contract-Level Audits, at 5 (Feb. 24,
2012), J.A. 398.
21
But CMS then conducted an empirical study from which
it discovered that “errors in [traditional Medicare] claims data
do not have any systematic effect on the risk scores calculated
by the CMS-HCC risk adjustment model, and therefore do not
have any systemic effect on the payments made to [Medicare
Advantage insurers].” CMS, Fee for Service Adjuster and
Payment Recovery for Contract Level Risk Adjustment Data
Validation Audits 5 (Oct. 26, 2018) (hereinafter CMS Study),
J.A. 731. That result is unsurprising. Providers paid on a fee-
for-service basis, as is the case in Medicare Part B, would
appear to lack incentives that bear on Medicare Advantage
insurers to overreport costly diagnoses or other factors
predictive of worse-than-average health, and any
underreporting of diagnoses is likely the result of not catching
the least costly beneficiaries with a given diagnosis (perhaps
because they require little or no treatment), which would tend
to reduce the average cost of a particular condition. See Gov’t
Br. 45-46. And individual errors within the mass of data used
to model a relative factor would tend to have little to no effect
on the factor’s value, given the large sample sizes—on the
order of one million beneficiaries, see Pope et al., Evaluation
of the CMS-HCC Risk Adjustment Model: Final Report 5, J.A.
522—together with “the fact that the relative factors are
summed across each enrollee’s [hierarchical condition
categories] and then across a plan’s enrollment, lead[ing] the
inaccuracies to mitigate each other due to offsetting effects,”
CMS Study at 5, J.A. 731. Based on the study results, CMS
announced in October 2018 that it would not, after all, use an
FFS Adjuster for contract-level RADV audits. See CMS
Study at 5-6, J.A. 731-32. That conclusion is preliminary,
and the review and rulemaking are ongoing. See Oral Arg.
Tr. 14:4-22. In the meantime, CMS does not use any FFS
Adjuster in that context.
22
4.
Against the backdrop of concern about costly errors in
the data reported by Medicare Advantage insurers, but before
CMS even solicited comments on the proposed FFS Adjuster
to contract-level RADV audits it ultimately deemed
unnecessary, Congress enacted the provision that undergirds
the Overpayment Rule. The Patient Protection and
Affordable Care Act, Pub. L. No. 111-148, 124 Stat. 119
(2010), obligates Medicare Advantage insurers to report and
return any overpayment that they receive from CMS within
sixty days of identifying it, 42 U.S.C. § 1320a-7k(d)(1), (2).
The Act defines “overpayment” as “any funds that a person
receives or retains under [the Medicare or Medicaid
programs] to which the person, after applicable reconciliation,
is not entitled.” Id. § 1320a-7k(d)(4)(B). In section 1320a-
7k(d)(3), it establishes that failure to report and return a
known overpayment within sixty days of discovering it
violates the False Claims Act, 31 U.S.C. § 3729 et seq., which
carries the potential for treble damages and other serious
penalties, see id. § 3729(a)(1).
In 2014, CMS promulgated the Overpayment Rule to
implement the statutory requirement to report and return
overpayments. The Rule similarly defines “overpayment” as
“any funds that [a Medicare Advantage insurer] has received
or retained under [the Medicare Advantage program] to which
the [Medicare Advantage insurer], after applicable
reconciliation, is not entitled.” Overpayment Rule, 79 Fed.
Reg. at 29,958 (codified at 42 C.F.R. § 422.326(a)), J.A. 85.
In the Rule’s preamble, CMS explained that, among other
things, any “diagnosis that has been submitted [by a Medicare
Advantage insurer] for payment but is found to be invalid
because it does not have supporting medical record
23
documentation would result in an overpayment.” Id. at
29,921, J.A. 64.
One commenter on the proposed Overpayment Rule, a
Medicare Advantage insurer not a party to this case, had
objected that it ran afoul of the Medicare statute’s actuarial-
equivalence requirement because it did not also require an
adjuster akin to the FFS Adjuster that CMS had proposed two
years earlier in the context of contract-level RADV audits.
See id.; see also J.A. 50-51 (comment from Humana on
proposed rule). In the final Rule, which does not provide for
such an adjuster, CMS stated that it “disagree[d] with the
commenter” because the “RADV methodology does not
change [CMS’s] existing contractual requirement that
[Medicare Advantage insurers] must certify (based on best
knowledge, information, and belief) the accuracy,
completeness, and truthfulness of the risk adjustment data
they submit to CMS.” Overpayment Rule, 79 Fed. Reg. at
29,921, J.A. 64. Nor, said CMS, did the statutory
overpayment-refund obligation, as implemented by the Rule,
“change the long-standing risk adjustment data requirement
that a diagnosis submitted to CMS by [a Medicare Advantage
insurer] for payment purposes must be supported by medical
record documentation.” Id. at 29,921-22, J.A. 64-65.
B. Factual and procedural history
UnitedHealth filed this challenge to the Overpayment
Rule in January 2016. Following the district court’s denial of
CMS’s motion to dismiss in March 2017, the parties cross-
moved for summary judgment. On September 7, 2018, the
court granted UnitedHealth’s motion in full and vacated the
Overpayment Rule. See UnitedHealthcare Ins. Co. v. Azar,
330 F. Supp. 3d 173, 192 (D.D.C. 2018).
24
The district court held that the Overpayment Rule
violated the Medicare statute’s requirement of “actuarial
equivalence.” Id. at 187. It concluded that the Rule would
“inevitabl[y]” lead to the loss of actuarial equivalence, id. at
185, because “payments for care under traditional Medicare
and Medicare Advantage are both set annually based on costs
from unaudited traditional Medicare records, but the 2014
Overpayment Rule systematically devalues payments to
Medicare Advantage insurers by measuring ‘overpayments’
based on audited patient records,” id. at 184. The court
emphasized that CMS had actually “recognized and
mitigated” “the same actuarial problem” when, in 2012, it
provisionally committed to using an FFS Adjuster for
contract-level RADV audits to account for the fact that
extrapolating an error rate across a Medicare Advantage
insurer’s entire contract effectively corrected for any
unsupported codes in the insurer’s data. Id. Relying on much
the same reasoning, the court held that the Rule also violated
the Medicare statute’s “same methodology” requirement. Id.
at 187. The court then deemed the Rule arbitrary and
capricious in violation of the APA as an unexplained
departure from CMS’s prior policy, namely, its stated intent
to use an FFS Adjuster in the context of contract-level RADV
audits. Id. at 187-90. The court noted only in passing that
CMS had not yet determined an appropriate amount of any
FFS Adjuster for contract-level RADV audits. See id. at 188.
The district court also rejected the Overpayment Rule’s
imposition of a negligence standard of liability for failure to
identify and report an overpayment. The Rule as promulgated
provided that a Medicare Advantage insurer “has identified an
overpayment when the [insurer] has determined, or should
have determined through the exercise of reasonable diligence,
that the [insurer] has received an overpayment.” 42 C.F.R.
§ 422.326(c) (emphasis added). But section 1320a-7k(d)(3)
25
of the Medicare statute provides that an overpayment that is
not timely reported and returned “is an obligation (as defined
in section 3729(b)(3) of title 31),” i.e., the False Claims Act,
under which liability requires proof of “knowingly”
submitting false claims for payment to the government, 31
U.S.C. § 3729(a). The False Claims Act defines “knowingly”
as having “actual knowledge” or acting “in deliberate
ignorance” or “reckless disregard of the truth or falsity of the
information.” Id. § 3729(b)(1)(A). The district court thus
held the Rule’s negligence-based liability inconsistent with
the False Claims Act’s “knowingly” standard.
UnitedHealthcare, 330 F. Supp. 3d at 190-91. The court held
that the final Rule’s negligence-based definition of
“identified”—which the proposed rule had defined to track
the False Claims Act’s fault standard before CMS adopted the
negligence standard in the final version—also violated the
APA because it was not a logical outgrowth of the proposed
rule. Id. at 191-92. CMS’s appeal does not challenge either
of those two holdings regarding the Rule’s negligence
standard; it contests only the district court’s rulings on
actuarial equivalence, same methodology, and the question
whether the Rule was arbitrary and capricious as an
unexplained departure from the FFS Adjuster CMS had
proposed to adopt in the context of RADV audits. See Gov’t
Br. 20-22.
In November 2018, CMS moved for partial
reconsideration, which the court denied in January 2020.
CMS based that motion on the results of the October 2018
study of the error rate in traditional Medicare, conducted as
groundwork for the anticipated FFS Adjuster for contract-
level RADV audits. As noted above, the results of that study
were made public several weeks after the district court’s
summary judgment ruling in this case. The study revealed
that “errors in [traditional Medicare] claims data do not have
26
any systematic effect on the risk scores calculated by the
CMS-HCC risk adjustment model,” undermining the case for
an adjuster. CMS Study at 5, J.A. 731; see also
UnitedHealthcare Ins. Co. v. Azar, No. 16-cv-157, 2020 WL
417867, at *1, *3 (D.D.C. Jan. 27, 2020), J.A. 801, 805. In
denying the motion, the district court stated that it “need not
linger on the details of the[] arguments” regarding the validity
of the study and CMS’s preliminary conclusion not to apply
any FFS Adjuster to contract-level RADV audits.
UnitedHealthcare, 2020 WL 417867, at *5, J.A. 811. The
court deemed it “sufficient to say that [UnitedHealth’s]
arguments [opposing the study] are fully explained and the
government does not adequately respond.” Id.
CMS timely appealed on November 6, 2018, and we
removed the case from abeyance in February 2020 following
the district court’s denial of reconsideration.
Finally, it bears noting that the issue of actuarial
equivalence has come up in other litigation between the
parties. The federal government and qui tam plaintiffs have
pursued several False Claims Act cases against Medicare
Advantage insurers in the last several years, charging failures
to report and return overpayments that the insurers knew were
based on unsupported diagnoses. At least some such cases
are still pending. See, e.g., United States ex rel. Poehling v.
UnitedHealth Grp., Inc., No. 16-cv-8697 (C.D. Cal.); United
States ex rel. Osinek v. Kaiser Permanente, No. 13-cv-3891
(N.D. Cal.). Medicare Advantage insurers, including
UnitedHealth, have raised actuarial equivalence as a defense
to False Claims Act liability. See Appellees Br. 55. At least
one court has rejected that defense, see United States ex rel.
Ormsby v. Sutter Health, 444 F. Supp. 3d 1010, 1067-71
(N.D. Cal. 2020), while another denied the government’s
request for an early partial summary judgment on that basis,
27
see United States ex rel. Poehling v. UnitedHealth Grp., Inc.,
No. 16-cv-8697, 2019 WL 2353125, at *1, *5-8 (C.D. Cal.
Mar. 28, 2019), but has not finally resolved the issue.
DISCUSSION
We review a district court’s grant of summary judgment
de novo. See, e.g., Clarian Health W., LLC v. Hargan, 878
F.3d 346, 352 (D.C. Cir. 2017). Under the APA, we must
“hold unlawful and set aside agency action” that is “arbitrary,
capricious, an abuse of discretion, or otherwise not in
accordance with law” or “in excess of statutory jurisdiction,
authority, or limitations, or short of statutory right.” 5 U.S.C.
§ 706(2). The party challenging agency action bears the
burden of proof. See, e.g., Abington Crest Nursing & Rehab.
Ctr. v. Sebelius, 575 F.3d 717, 722 (D.C. Cir. 2009) (citing
City of Olmstead Falls v. FAA, 292 F.3d 261, 271 (D.C. Cir.
2002)).
A. The Overpayment Rule does not violate the
Medicare statute’s requirement of “actuarial
equivalence”
UnitedHealth’s central challenge to the Overpayment
Rule is that it violates the Medicare statute’s command to
CMS to adjust payment amounts to a Medicare Advantage
insurer based on risk factors “so as to ensure actuarial
equivalence” between that insurer’s beneficiary population
and the traditional Medicare beneficiaries whose healthcare
cost data CMS uses to calculate capitated, monthly payments
to the insurer. 42 U.S.C. § 1395w-23(a)(1)(C)(i).
UnitedHealth argues that the Rule “results in different
payments for identical beneficiaries because it relies on both
supported and unsupported codes to calculate risk in
[traditional Medicare], but only supported codes in the
[Medicare Advantage] program,” which “necessarily means
28
that [Medicare Advantage] plans are not paid the same as
CMS for identical beneficiaries”—and in fact are “inevitably
underpa[id].” Appellees Br. 22-23; see also id. at 26-27. In
other words, UnitedHealth objects to CMS’s reliance on
minimally audited traditional Medicare data in the risk-
adjustment model that CMS uses to calibrate the monthly
payment rates for Medicare Advantage insurers, while CMS
at the same time obligates insurers to refund each individual
payment that they know is not supported by a beneficiary’s
medical records. Id. at 26. The Overpayment Rule,
UnitedHealth seems to say, disrupts actuarial equivalence
between Medicare Advantage and traditional Medicare
insofar as data from traditional Medicare that is used to model
the expected cost of a given diagnosis is subject to laxer
documentation standards than is a diagnosis a Medicare
Advantage insurer reports in support of payment.
UnitedHealth claims, and the district court agreed, that
before CMS may lawfully apply the Overpayment Rule, it
must implement one of two measures to remedy the claimed
imbalance. First, CMS could devise and apply an adjuster
akin to the FFS Adjuster it had intended to use (but since has
preliminarily decided is unwarranted) in the context of
contract-level RADV audits of Medicare Advantage insurers’
risk-adjustment data. In that scenario, Medicare Advantage
insurers would be liable for overpayments only to the extent
that their payment error rate exceeded that of traditional
Medicare. Alternatively, CMS could comprehensively audit
traditional Medicare data before using it in the risk-
adjustment model that sets Medicare Advantage insurers’
monthly payments. Only then would UnitedHealth be
prepared to accept that the traditional Medicare data used to
arrive at relative factors did not contain the unsupported codes
that, it asserts, should bar CMS from recouping overpayments
pursuant to the Rule for codes that a Medicare Advantage
29
insurer reported to CMS but later discovered were
unsupported by beneficiaries’ medical records.
There are two main problems with UnitedHealth’s
argument. First, nothing in the Medicare statute’s text,
structure, or logic makes the actuarial-equivalence
requirement in section 1395w-23(a)(1)(C)(i) applicable to the
overpayment-refund obligation in section 1320a-7k(d) or to
the Overpayment Rule promulgated under that section.
Second, even if the actuarial-equivalence requirement did
indirectly relate to Medicare Advantage insurers’
overpayment-refund obligation, we could not here invalidate
the Overpayment Rule as violating actuarial equivalence.
UnitedHealth notably does not challenge the risk-adjustment
model itself or the resultant values CMS assigned to any
relative factor. Nor did it provide evidence that the obligation
to refund overpayments, as defined by the Medicare statute
and the Rule, in fact has led or will lead to systematic
underpayment of Medicare Advantage insurers relative to
traditional Medicare.
1.
We have not previously decided any case involving
“actuarial equivalence” as referenced in section 1395w-
23(a)(1)(C)(i) for the Medicare Advantage program. In the
context of the Employee Retirement Income Security Act
(ERISA), we have said that “[t]wo modes of payment are
actuarially equivalent when their present values are equal
under a given set of actuarial assumptions.” Stephens v. U.S.
Airways Grp., Inc., 644 F.3d 437, 440 (D.C. Cir. 2011).
UnitedHealth and CMS agree that “actuarial equivalence” in
this provision of the Medicare statute means that CMS aims to
pay the same amount to Medicare Advantage insurers for
their beneficiaries’ care as CMS would spend on those same
beneficiaries if they were instead enrolled in traditional
30
Medicare. See Gov’t Br. 1; Appellees Br. 26; see also
Defendants’ Memorandum in Support of Their Cross-Motion
for Summary Judgment and in Opposition to Plaintiffs’
Motion for Summary Judgment at 28, UnitedHealthcare Ins.
Co. v. Azar, 330 F. Supp. 3d 173 (D.D.C. 2018) (No. 16-cv-
157), J.A. 688.
The parties disagree about whether the Overpayment
Rule even implicates the actuarial-equivalence requirement.
UnitedHealth assumes the Overpayment Rule creates a
sweeping obligation that effectively requires Medicare
Advantage insurers to self-audit all their data. It thus asserts
that, because of actuarial equivalence, before CMS may
police overpayments in the manner of the Overpayment Rule,
CMS must either audit traditional Medicare data before it
goes into the risk-adjustment model or, alternatively, adopt a
systemic corrective similar to the FFS Adjuster CMS
contemplated in the context of proposed contract-level RADV
audits. In the context of the RADV audit expansion, the
insurers’ objection was that applying a sampled payment error
rate across an entire contract would effectively audit all of an
insurer’s data while leaving unaudited the traditional
Medicare data used to set monthly payments in the first place,
thus requiring the application of an adjuster that would also
effectively audit all of the data on the traditional Medicare
side. Here, UnitedHealth asserts much the same: that the
Overpayment Rule essentially requires insurers to audit all of
the data they submit to CMS (especially given the prospect of
liability under the False Claims Act), leaving that data set
with no unsupported codes, while traditional Medicare data
remains unaudited, leaving that data set with a significant
number of unsupported codes. And, UnitedHealth says, the
presence of unsupported codes in traditional Medicare data
depresses the value of relative factors, so removing
unsupported codes from a Medicare Advantage insurer’s data
31
but not traditional Medicare’s will cause CMS to underpay
insurers.
UnitedHealth’s premise is unsupported. Nothing in the
Overpayment Rule obligates insurers to audit their reported
data. As the district court held, see UnitedHealthcare, 330 F.
Supp. 3d at 190-91, and CMS does not here dispute, see Gov’t
Br. 22, 30, the Rule only requires insurers to refund amounts
they know were overpayments, i.e., payments they are aware
lack support in a beneficiary’s medical records. That limited
scope does not impose a self-auditing mandate.
No part of the Medicare statute or the Overpayment Rule
supports UnitedHealth’s challenge. The statute’s actuarial-
equivalence requirement does not apply to the separate
statutory obligation on insurers to refund overpayments they
erroneously elicit from CMS; nor, by the same token, does
actuarial equivalence apply to the Overpayment Rule that
implements that statutory obligation and, in relevant part,
essentially parrots it. Compare 42 U.S.C. § 1320a-
7k(d)(4)(B) (defining “overpayment” as “any funds that a
person receives or retains under [the Medicare or Medicaid
programs] to which the person, after applicable reconciliation,
is not entitled”), with 42 C.F.R. § 422.326(a) (defining
“overpayment” as “any funds that [a Medicare Advantage
insurer] has received or retained under [the Medicare
Advantage program] to which the [Medicare Advantage
insurer], after applicable reconciliation, is not entitled”).
Nothing in the text of either the actuarial-equivalence
requirement in section 1395w-23(a)(1)(C)(i) or the
overpayment-refund obligation in section 1320a-7k(d) applies
the former to the latter. There is no cross-reference or other
language suggestive of overlap, nor does UnitedHealth so
contend. Indeed, even the district court acknowledged that
the overpayment-refund obligation does not “state how
32
‘overpayments’ and ‘actuarial equivalence’ in payments are
related.” UnitedHealthcare, 330 F. Supp. 3d at 181.
More specifically, nothing in either provision renders
actuarial equivalence a defense against the obligation to
refund any individual, known overpayment. Notably,
Congress through the Affordable Care Act strengthened
Medicare Advantage insurers’ data-reporting obligations by
requiring insurers to report and return overpayments within
sixty days of their discovery, and it made specific provision
for False Claims Act liability for those that do not. In so
doing, Congress made no reference to the Medicare statute’s
longstanding actuarial-equivalence requirement, let alone any
suggestion that it could be interposed as a defense. See 42
U.S.C. § 1320a-7k(d).
If anything, the text of section 1395w-23(a)(1)(C)(i)
limits the scope of the actuarial-equivalence requirement. It
states that CMS “shall adjust the payment amount under
subparagraph (A)(i) and the amount specified under
subparagraph (B)(i), (B)(ii), and (B)(iii)” for demographic
and health characteristics “to ensure actuarial equivalence.”
Those cross-referenced subparagraphs identify the manner in
which CMS “shall make monthly payments under this section
in advance to each [Medicare Advantage] organization.” Id.
§ 1395w-23(a)(1)(A). Section 1395w-23(a)(1)(C)(i)’s
reference to risk-adjusting the amount paid to Medicare
Advantage insurers “under” certain cross-referenced
subparagraphs, and those subparagraphs’ focus on the
predetermined monthly payments made to insurers “under this
section,” indicate that the actuarial-equivalence requirement is
not broadly applicable, but instead limited to the specified
context of CMS’s calculation and disbursement of monthly
payments in the first instance. Cf. Davis v. Pension Benefit
Guar. Corp., 734 F.3d 1161, 1170 (D.C. Cir. 2013)
33
(interpreting ERISA’s actuarial-equivalence requirement as
limited by statutory text and structure).
Stephens v. U.S. Airways Group, Inc., cited by the district
court in support of its holding, see UnitedHealthcare, 330 F.
Supp. 3d at 185-86, actually cuts the other way. There, we
held that an ERISA actuarial-equivalence requirement did not
obligate the airline to pay pensioners interest on requested
lump-sum payments made well after annuity payments would
have begun had the same benefit been disbursed periodically.
Stephens, 644 F.3d at 440. When we held that interest was
required under IRS regulations regarding unreasonable delay
of such payments, id.; see also id. at 442, we were also clear
that the lump-sum payments did not violate actuarial
equivalence where the airline “accurately calculated [the]
lump sums to be the ‘actuarial equivalent’ of the annuity
option as of the annuity start date,” id. at 440. Because the
actuarial equivalence of the annuity and lump-sum payments
had been calculated based on a common initial payment date,
and the statute was silent on whether interest was owed when
an otherwise actuarially equivalent pension was paid later, we
declined to grant the interest claim on that basis. Id.
Here, the Medicare statute is similarly silent, as it speaks
not at all to whether the actuarial-equivalence requirement in
section 1395w-23(a)(1)(C)(i) bears on section 1320a-7k(d)’s
requirement to refund overpayments. That is, the statute
never says that the later refund of individual, known
overpayments implicates the earlier-in-time requirement that
the lump-sum monthly payments to Medicare Advantage
insurers be set as if an insurer’s beneficiary pool were
actuarially equivalent to traditional Medicare’s population. In
the face of such silence, actuarial equivalence is satisfied
consistently with Stephens so long as CMS reasonably
concluded when it set its monthly payments to UnitedHealth
34
that the traditional Medicare data it used was sufficiently
accurate and free of systemic biases that modeling based on
that data would generate relative-factor values enabling CMS
to “adjust the payment amount” to UnitedHealth “so as to
ensure actuarial equivalence.” 42 U.S.C. § 1395w-
23(a)(1)(C)(i). As discussed in the next section, there is no
evidence of any such systemic skew in traditional Medicare
data, and, indeed, UnitedHealth never challenged the values
CMS assigned to the relative factors. CMS permissibly reads
the Medicare statute to authorize it to recover overpayments
for diagnosis codes UnitedHealth submitted but knew or
learned were unsupported—and to do so without first either
remaking its underlying actuarial-equivalence calculation to
prove that traditional Medicare data is completely free of
unsupported diagnoses, or re-defending its calculation as
already accounting for unsupported diagnoses.
As CMS points out, the actuarial-equivalence
requirement is not an “entitle[ment] . . . to a precise payment
amount” for a Medicare Advantage insurer, but only “an
instruction to the Secretary regarding the design of the risk
adjustment model as a whole . . . describ[ing] the type of
‘payment amount[s]’ that the risk adjustment model should
produce”; “[i]t does not directly govern how CMS evaluates
the validity of diagnoses or defines ‘overpayment.’” Reply
Br. 5-6 (third alteration in original); see Gov’t Br. 42-43. To
that end, the Medicare statute grants the agency considerable
discretion in determining how to structure the risk-adjustment
model to achieve actuarial equivalence. See 42 U.S.C.
§ 1395w-23(a)(1)(C)(i).
The actuarial-equivalence requirement and the
overpayment-refund obligation apply to different actors,
target distinct issues arising at different times, and work at
different levels of generality. The actuarial-equivalence
35
provision directs CMS to develop a system of relative factors
to use in adjusting the amount of the monthly payments to
each Medicare Advantage insurer. See id. It calls on CMS to
use its expert judgment to identify cost-predictive risk factors
in the Medicare population and to analyze the data
accumulated in traditional Medicare to determine average
costs associated with those factors.
The point of that exercise is to enable CMS to pay only
as much for coverage of Medicare Advantage beneficiaries as
it would if they were instead enrolled in traditional Medicare,
notwithstanding differences between the actual populations—
for example, that Medicare Advantage populations have
tended to be healthier than traditional Medicare’s population.
See Reply Br. 20-21 (citing Pope et al., Risk Adjustment of
Medicare Capitation Payments Using the CMS-HCC Model,
at 119, J.A. 487). Thus, the actuarial-equivalence requirement
is focused on accounting for the distinct profiles of each
insurer’s beneficiary population, listing “age, disability status,
gender, institutional status, and . . . health status” as
potentially relevant considerations in the risk-adjustment
model. 42 U.S.C. § 1395w-23(a)(1)(C)(i). Significantly,
section 1395w-23(a)(1)(C)(i)’s use of the qualifier “actuarial”
necessarily implies an assessment made at the group or
population level, not the individual level, so as to support
credible statistical inferences. Cf. Pope et al., Evaluation of
the CMS-HCC Risk Adjustment Model: Final Report 5, J.A.
522 (explaining that “risk assessment is designed to
accurately explain the variation at the group level, not at the
individual level, because risk adjustment is applied to large
groups,” and that “the Actuarial Standard Board’s Actuarial
Standard of Practice for risk classification” requires that “risk
classes are large enough to allow credible statistical
inferences”). By contrast, the overpayment-refund obligation
in both the Medicare statute and the Overpayment Rule
36
corrects particular mistaken payments to Medicare Advantage
insurers that exceed what the relevant medical records
support.
Finally, applying actuarial equivalence to the Medicare
statute’s separate obligation to refund particular, known
overpayments would seriously undermine that obligation,
with the potential for absurd consequences. As UnitedHealth
acknowledged at oral argument, under its view of actuarial
equivalence as a defense against its obligation to reimburse
CMS for known overpayments, a Medicare Advantage insurer
could be entitled to retain payments that it knew were
unsupported by medical records so long as CMS had not
established that the insurer’s overall payment error rate was
higher than traditional Medicare’s payment error rate. See
Oral Arg. Tr. 50:12-18. Indeed, under that line of thinking, a
Medicare Advantage insurer could knowingly submit
unsupported diagnosis codes and retain payment for them
unless and until CMS established—based on fully audited
data of both traditional Medicare and the Medicare Advantage
insurer at issue—that the particular overpayment resulted in a
net gain to the insurer relative to traditional Medicare. There
is no basis on which we can conclude that Congress intended
the distinct actuarial-equivalence requirement to so thwart the
overpayment-refund obligation—an obligation that, again,
Congress strengthened through the Affordable Care Act
without any reference to the accuracy or actuarial equivalence
of the prospective monthly payments that CMS calculates and
disburses to Medicare Advantage insurers. Congress gave no
sign that it was limiting the obligation in the way
UnitedHealth now suggests.
UnitedHealth asks us to rewrite the statutory
overpayment-refund obligation, which was the basis for the
Overpayment Rule, by narrowing the capacious “any funds”
37
to which a Medicare Advantage insurer “is not entitled,” 42
U.S.C. § 1320a-7k(d)(4)(B), with an actuarial-equivalence
exception. But in the absence of any textual or structural
connection between the two provisions, we decline to hold
that the actuarial-equivalence requirement in section 1395w-
23(a)(1)(C)(i) applies to the overpayment-refund obligation in
section 1320a-7k(d) or the Overpayment Rule CMS
promulgated to comply with that provision.
2.
Even if the Medicare statute could theoretically support
UnitedHealth’s reading, we lack the necessary grounds here
to invalidate the Overpayment Rule as a violation of actuarial
equivalence. Recall that UnitedHealth’s claim is that CMS
cannot demand that UnitedHealth refund overpayments unless
CMS shows it meets what UnitedHealth posits as a
symmetrical auditing or error-correction obligation regarding
traditional Medicare. But Congress has spelled out distinct
obligations for traditional Medicare and Medicare Advantage,
such as the Coding Intensity Adjuster that applies to the latter
program but not the former, see id. § 1395w-
23(a)(1)(C)(ii)(III); and CMS has long employed different
audit mechanisms for the claims submitted by healthcare
providers for traditional Medicare beneficiaries as compared
to the data submitted by Medicare Advantage insurers to
enable CMS to calculate accurate risk scores for Medicare
Advantage beneficiaries, see Gov’t Br. 16-19, 35-38.
Congress’s and CMS’s use of measures tailored to the
differing structures of and incentives in the two programs
makes sense; indeed, it could be irrational not to use distinct
tools as needed to respond to different problems.
UnitedHealth does not challenge the Coding Intensity
Adjuster imposed by Congress. And UnitedHealth has never
38
taken the opportunity that arises annually to challenge the
accuracy of the risk-adjustment model or pricing when CMS
announces the relative factors and base payment rates that it
will use for the upcoming year. See Oral Arg. Tr. 12:12-
13:16; see also Ormsby, 444 F. Supp. 3d at 1068 n.442. We
accordingly accept the unchallenged validity of the overall
design of the model, the risk factors considered by CMS
pursuant to its discretion under section 1395w-23(a)(1)(C)(i),
and the accuracy of the resultant values of relative factors.
UnitedHealth cannot now use actuarial equivalence to litigate
belated objections to the risk-adjustment model or the level of
its monthly payments through the back door of the
Overpayment Rule.
UnitedHealth has failed to provide any logical or
empirical basis to question the accuracy of traditional
Medicare data. UnitedHealth asserts that the obligation to
refund overpayments, at least as defined by the Overpayment
Rule, leads to systematic underpayment of Medicare
Advantage insurers relative to traditional Medicare. But it is
by no means “inevitable” that Medicare Advantage insurers
will be underpaid without the correctives that UnitedHealth
would require. UnitedHealthcare, 330 F. Supp. 3d at 185,
187. Congress and CMS have long recognized that the uses
of and incentives bearing on data in traditional Medicare and
Medicare Advantage are very different, and accordingly have
designed a range of distinct obligations and error-correction
mechanisms for the two programs. As is by now familiar,
CMS pays healthcare providers for Medicare Part B
beneficiaries on a fee-for-service basis; thus, whereas
providers may have incentives to overtreat those beneficiaries,
they lack incentives to overreport diagnosis codes. By
contrast, Medicare Advantage insurers, which CMS pays
based on their beneficiaries’ demographic and health
characteristics, including diagnoses, have financial incentives
39
to code intensely and overreport diagnoses but not necessarily
to overtreat beneficiaries. See Advance Notice of
Methodological Changes for CY 2004 Medicare+Choice
Payment Rates, at 5, J.A. 115; U.S. Gov’t Accountability
Off., Medicare Advantage: CMS Should Improve the
Accuracy of Risk Score Adjustments for Diagnostic Coding
Practices 2, J.A. 546.
UnitedHealth complains of “a substantial number” of
unsupported diagnosis codes in the minimally audited
traditional Medicare data set. Appellees Br. 26. But
UnitedHealth identifies no reason why the traditional
Medicare data that goes into the risk-adjustment model would
suffer systematically from unsupported codes like those the
Overpayment Rule targets, i.e., codes lacking substantiation
in medical records. If anything, the fact that providers for
traditional Medicare beneficiaries are generally paid based on
services, not diagnoses, would seem to tend toward
underreporting, not overreporting, of diagnoses within
traditional Medicare. The underlying premise of
UnitedHealth’s overall position is that traditional Medicare
data includes a significant rate of unsupported diagnosis codes
that ultimately depresses the payments to Medicare
Advantage insurers. But the different ways the programs’
reimbursement schemes work in practice make that premise
implausible.
Nor has UnitedHealth established another premise of its
position—that the unsupported codes it posits in traditional
Medicare would both be materially analogous to those the
Overpayment Rule targets, and would cause UnitedHealth to
be underpaid. To start, it is not even clear which kind of
payment error in traditional Medicare, relative to Medicare
Advantage, UnitedHealth believes is overlooked to its
detriment. UnitedHealth identifies the problem in traditional
40
Medicare as “a substantial number” of unsupported codes, id.,
though, as discussed more below, it does not specify what, if
any, payment implications it sees as necessarily attending
them. To the extent that unsupported codes in traditional
Medicare would be associated with erroneous payments that
CMS need not have made to healthcare providers—i.e.,
overpayments analogous to any CMS makes to Medicare
Advantage insurers and targets with the Overpayment Rule—
that kind of error would, if anything, tend to raise, not lower,
overall payments to Medicare Advantage insurers. That is,
because CMS’s expenditures on traditional Medicare
contribute to setting the base rate later used to make payments
to Medicare Advantage insurers, the more money CMS
spends on traditional Medicare, the higher the baseline for its
expenditures on Medicare Advantage.
UnitedHealth nonetheless defends its position and the
district court’s ruling as founded “on straightforward math:
Including unsupported codes when allocating costs on the
traditional Medicare side, then excluding those same codes
when determining payment amounts on the [Medicare
Advantage] side, will underpay plans.” Id. at 27.
UnitedHealth’s math does not add up. To illustrate its
assertion of inevitable underpayment, UnitedHealth riffs on
CMS’s example involving a 72-year-old woman living
independently (relative factor 0.348), with diabetes without
complications (relative factor 0.118), and multiple sclerosis
(relative factor 0.556), who would have a total risk score of
1.022. See Gov’t Br. 7. But for UnitedHealth that woman is
a twin: Her sister (Twin A) is a traditional Medicare
beneficiary, and she (Twin B) is “identical in all respects” but
is a Medicare Advantage beneficiary. Appellees Br. 32.
UnitedHealth asks us to imagine that the diabetes code for
both twins (who, again, are identical) is “unsupported.” Id. It
says that, under the Overpayment Rule, the woman’s
41
Medicare Advantage insurer “would need to delete her
unsupported diabetes code after identifying it, and the
resulting risk score for Twin B would be 0.904.” Id. So, if
her sister, Twin A, “cost CMS $10,000 to insure . . . the
[Medicare Advantage] plan would receive only $8,845 to
insure its identical beneficiary (0.904/1.022 x $10,000).” Id.
at 32-33.
UnitedHealth’s twin example ignores that unsupported
codes are likely to occur for different reasons and with
differing effects in the two programs: Unlike an unsupported
diabetes code associated with Twin B in Medicare Advantage,
which leads to an unwarranted increase in payment to the
insurer, the mere existence of an unsupported diabetes code
for Twin A in traditional Medicare does not mean CMS spent
more money on that beneficiary. That is, CMS’s expenditure
for Twin A (at least in fee-for-service Part B) is not likely to
have been higher if she were miscoded as diabetic than it
would be without that error. CMS’s expenditure on the twin
in traditional Medicare would increase only if CMS paid for
treatment corresponding to that unsupported code. But if
Twin A’s unsupported diabetes code is only an administrative
error that does not correspond to treatment actually provided
and paid for, UnitedHealth’s hypothetical uses the wrong
starting point, and so the wrong figures, for Twin A’s side of
the comparison. Her costs in traditional Medicare from the
outset (and even if her unsupported diabetes code is never
caught) would be at the same, lowered level as Twin B’s in
Medicare Advantage once that diabetes code was removed—
in both cases, the payment level appropriate for a non-
diabetic.
Even assuming Twin A’s unsupported diabetes code were
associated with erroneous payment by CMS, one would need
to know more about the nature and scale of such errors to
42
determine whether they could have affected the results of the
regression analysis used to calculate relative factors, and in
what direction. For example, if UnitedHealth is assuming that
Twin A’s unsupported diabetes code triggered payment for
treatment that had no medical purpose, UnitedHealth still has
not made its case of inevitable underpayment. Specifically, if
an unsupported code in traditional Medicare pairs with
diabetes treatment for which CMS paid, UnitedHealth has not
explained how, in coding it as just that—a cost of diabetes
treatment, however unnecessary—CMS would inevitably
depress the value of the relative factor for diabetes. As
UnitedHealth sees it, every unsupported diabetes code in
traditional Medicare lowers the value of the relative factor for
diabetes, as CMS’s expenditure on diabetes is divided among
more and more beneficiaries. But UnitedHealth does not
account for the possibility of an unsupported code associated
with payment by CMS, which would enlarge both the total
costs and the beneficiary pool in traditional Medicare and
thus, if anything, tend to keep constant the value of the
relative factor at issue.
Alternatively, if UnitedHealth’s concern is with a
diabetes code that is unsupported because treatment was
delivered, medically necessary, and paid for, but just
administratively associated with the wrong code—diabetes
rather than celiac disease, for example—it also has not shown
inevitable underpayment. In such a case, a data point that
should have gone into the regression analysis supporting the
relative factor for celiac disease would have instead been part
of the data crunched to arrive at the diabetes relative factor.
But, without any basis to conclude that any such errors occur
at scale or in any particular pattern, the misattribution of some
costs in the data cannot be assumed to distort CMS’s analysis.
43
The implications of any unsupported diabetes code in
traditional Medicare are quite different from those of the same
unsupported code in Medicare Advantage. The former will
not lead to Medicare Advantage insurers’ inevitable
underpayment because, as already noted, any erroneous code
in traditional Medicare is aggregated with millions of others
in the regressions called for under the risk-adjustment model.
Errors that are isolated and random, not systemic, cannot
alone be assumed to affect the value of a relative factor that
bears on how much CMS will pay Medicare Advantage
insurers for beneficiaries with any particular condition. An
unsupported code submitted by a Medicare Advantage
insurer, in contrast, triggers overpayment in every case. That
is because individual codes in that program are used to
determine payments, not as data points in a complex and
rigorous statistical model.
In sum, UnitedHealth has given no reason to think that
miscoding in traditional Medicare necessarily leads to any
inflated or deflated relative factors and, if it did, which ones
are affected in which direction. We cannot assume based on
UnitedHealth’s reasoning alone that Medicare Advantage
insurers are inevitably underpaid under any of the
circumstances possible in its example.
What’s more, the empirical evidence that we do have—
CMS’s October 2018 study concerning an FFS Adjuster in the
context of contract-level RADV audits—suggests that
Medicare Advantage insurers are not underpaid relative to
traditional Medicare, contrary to UnitedHealth’s and the
district court’s belief that underpayment is inevitable.
Through that study, CMS “found that errors in [traditional
Medicare] claims data do not have any systematic effect on
the risk scores calculated by the CMS-HCC risk adjustment
model, and therefore do not have any systematic effect on the
44
payments made to [Medicare Advantage] organizations.”
CMS Study at 5, J.A. 731. In fact, CMS determined that the
impact of errors in traditional Medicare data “is less than one
percent on average and in favor of the [Medicare Advantage]
plans.” Id.
Together with its opposition to CMS’s motion for partial
reconsideration before the district court, UnitedHealth
submitted a declaration from an actuarial expert “reflect[ing]
[the expert’s] professional interpretation” of CMS’s study.
Declaration of Julia Lambert at 2, UnitedHealthcare Ins. Co.
v. Azar, 2020 WL 417867 (D.D.C. Jan. 27, 2020) (No. 16-cv-
157), J.A. 771. UnitedHealth’s expert criticized the study by
asserting that the underlying data in fact showed that, “if you
take [a Medicare Advantage insurer] with risk profiles
identical to those in the [traditional Medicare] data, the
[insurer] would be underpaid if the relative factors generated
using both supported and unsupported data [from traditional
Medicare] were applied only to supported codes in the
[insurer’s] data.” Id. at 19, J.A. 788. But neither CMS’s
study nor UnitedHealth’s expert’s declaration tells us what
happens when a Medicare Advantage insurer removes some,
but not all, unsupported codes from its data, as is the reality
here with the overpayment-refund obligation for only known
overpayments. Indeed, UnitedHealth’s expert’s declaration
unquestioningly presumes that, as a result of the Overpayment
Rule, a Medicare Advantage insurer’s data will consist of
only supported codes. See id. UnitedHealth has not shown,
though, that the overpayment-refund obligation, as defined by
the Overpayment Rule and limited to codes known to lack
support, in fact will result in Medicare Advantage insurers
receiving payment for only supported codes, or that there is a
point at which the removal of some, even if not all,
unsupported codes from an insurer’s data would violate
actuarial equivalence.
45
The burden of proof is UnitedHealth’s to show that the
Overpayment Rule is unlawful. See, e.g., Abington Crest, 575
F.3d at 722 (citing City of Olmstead Falls, 292 F.3d at 271).
In the absence of such proof—or even persuasive logic in
UnitedHealth’s favor—we could not here invalidate the
Overpayment Rule as violating actuarial equivalence even if
we held that such requirement bore on the overpayment-
refund obligation.
B. The Overpayment Rule does not violate the
Medicare statute’s requirement of “same
methodology”
UnitedHealth’s second claim—that the Overpayment
Rule violates the Medicare statute’s “same methodology”
requirement in section 1395w-23(b)(4)(D)—is likewise
without merit. As explained above, each county in the United
States has its own base payment rate, which provides the
starting point for the monthly, per-capita payment to a
Medicare Advantage insurer covering a beneficiary in that
area. Every year, Medicare Advantage insurers bid for
contracts after CMS announces the county-specific
benchmarks for the coming year. See 42 U.S.C. § 1395w-
23(b)(1)(B). The base rate for a given county is then
determined by the benchmark derived from traditional
Medicare’s per-capita expenditures in the county and the
winning bid submitted by a Medicare Advantage insurer. An
insurer covering a beneficiary with a risk score of 1.0 can
expect to receive the base rate for the beneficiary’s home
county, whereas beneficiaries with risk scores higher or lower
than 1.0 will draw prorated payments above or below the base
rate, respectively.
As UnitedHealth acknowledges, the annual computation
and publication requirement in section 1395w-23(b)(4) is
46
meant to facilitate Medicare Advantage insurers’ yearly
submission of viable, competitive bids for contracts with
CMS. See Appellees Br. 33-34. In a section titled “Annual
announcement of payment rates,” the Medicare statute
requires CMS to compute and publish annually the “average
risk factor” for traditional Medicare beneficiaries on a county-
by-county basis, “using the same methodology as is expected
to be applied in making payments under subsection (a).” 42
U.S.C. § 1395w-23(b)(4)(D). Subsection (a) is, at this point,
familiar: It contains the actuarial-equivalence requirement
and governs the design of the risk-adjustment model. See id.
§ 1395w-23(a)(1)(C)(i).
The “same methodology” requirement plays a specific
role in the computation and publication of data to aid the
bidding process. It does not impose a substantive limit on the
operation of the risk-adjustment model, which is governed by
a separate provision. Nor does it have any bearing on whether
a particular payment to a Medicare Advantage insurer
constitutes an “overpayment.” Rather, the requirement to
“us[e] the same methodology” clarifies that CMS, in
computing the traditional Medicare data it publishes, must use
the same risk-adjustment model that it already uses to set
monthly payments to Medicare Advantage insurers, not
devise a new model or method for that purpose. Thus, for the
same reasons that support our holding regarding
UnitedHealth’s actuarial-equivalence claim, we conclude that
the Overpayment Rule simply does not implicate the
Medicare statute’s separate “same methodology” requirement.
C. The Overpayment Rule is not an unexplained
departure from prior policy
UnitedHealth’s third and final claim on appeal is that
CMS’s response to a comment calling for the use of an
47
adjuster under the Overpayment Rule was arbitrary and
capricious in violation of the APA. That comment advocated
“appl[ication of] the principles adopted by CMS in the RADV
audit context” to argue that “the sole instance in which an
‘overpayment’ can be determined” is when CMS first has
shown that the overall payment error for a given Medicare
Advantage insurer is higher than that in traditional Medicare.
Overpayment Rule, 79 Fed. Reg. at 29,921, J.A. 64.
In 2012, CMS proposed to use an FFS Adjuster in the
context of contract-level RADV audits used to review
Medicare Advantage insurers’ risk-adjustment data. It did so
in response to objections by Medicare Advantage insurers and
the American Academy of Actuaries that failure to use an
adjuster would violate the Medicare statute’s requirement of
“actuarial equivalence.” Specifically, those commenters had
argued that the actuarial-equivalence requirement prohibited
CMS from using traditional Medicare data—which is subject
to minimal auditing—to make monthly payments to Medicare
Advantage insurers in the first instance, but then requiring an
insurer to return some portion of those payments once CMS
had effectively audited all the insurer’s data by applying an
extrapolated payment error rate to its entire contract with
CMS. See, e.g., Aetna Inc.’s Comments on Proposed
Payment Error Calculation Methodology for Part C
Organizations Selected for Contract-Level RADV Audits, at 4
& 18-22, J.A. 298 & 312-16; Humana Inc., Comment on
RADV Sampling and Error Calculation Methodology, at 2-5
& 12, J.A. 333-36 & 343. Notably, the Academy did not
object to the proposed Overpayment Rule based on actuarial
equivalence, and CMS has preliminarily decided not to use an
FFS Adjuster for contract-level RADV audits after all because
“errors in [traditional Medicare] claims data do not have any
systematic effect on the risk scores calculated by the CMS-
HCC risk adjustment model.” CMS Study at 5, J.A. 731.
48
Because, as discussed above, the Overpayment Rule does
not violate, or even implicate, actuarial equivalence, CMS had
no obligation to consider an FFS Adjuster or similar
correction in the overpayment-refund context. Contract-level
RADV audits, which would effectively eliminate—and
require repayment for—all unsupported codes in a Medicare
Advantage insurer’s data, are an error-correction mechanism
that is materially distinct from the Overpayment Rule
challenged here, which requires only that an insurer report
and return to CMS known errors in its beneficiaries’
diagnoses that it submitted as grounds for upward adjustment
of its monthly capitation payments. Thus, CMS was not
required to provide further explanation of its decision. See
Motor Vehicles Mfrs. Ass’n v. State Farm Mut. Auto. Ins. Co.,
463 U.S. 29, 43 (1983). CMS’s response to the comment
reiterated Medicare Advantage insurers’ longstanding
obligations, under other of CMS’s regulations not challenged
here, see, e.g., 42 C.F.R. § 422.504(l), to certify the accuracy
of the data that they report to CMS, see Overpayment Rule,
79 Fed. Reg. at 29,921-22, J.A. 64-65. Its response was
therefore reasonable. See id. 1
1
As mentioned above, CMS has since proposed not to use an FFS
Adjuster in the context of contract-level RADV audits. See CMS
Study at 5, J.A. 731. We express no opinion on whether the
actuarial-equivalence requirement in section 1395w-23(a)(1)(C)(i)
of the Medicare statute requires such an adjuster in that context.
For current purposes, it suffices that the contexts of contract-level
RADV audits and overpayment refunds are plainly distinguishable,
such that CMS did not need to further explain, when it issued the
Overpayment Rule in 2014, why it then intended to use an adjuster
in the former context but not the latter.
49
* * *
For the foregoing reasons, we hold that the Overpayment
Rule does not violate the Medicare statute’s “actuarial
equivalence” and “same methodology” requirements and is
not arbitrary and capricious as an unexplained departure from
prior policy. We accordingly reverse the judgment of the
district court and remand this case with orders to enter
judgment in favor of Appellants.
So ordered.