Baystate Medical Center v. Leavitt

545 F. Supp. 2d 20, 2008 U.S. Dist. LEXIS 25861, 2008 WL 836012

• Court: District Court, District of Columbia
• Decided: March 31, 2008
• Docket: Civil Action 06-1263 (JDB)
• Status: Published

Opinion

MEMORANDUM OPINION

JOHN D. BATES, District Judge.

The Secretary of the Department of Health and Human Services, through the Centers for Medicare and Medicaid Services (“CMS”), is responsible for providing payments known as “disproportionate share hospital” adjustments to hospitals that serve a significantly disproportionate share of low income patients, as set forth under the Medicare statute, Title XVIII of the Social Security Act, 42 U.S.C. § 1395 et seq. Plaintiff Baystate Medical Center (“Baystate”) seeks judicial review of the Secretary’s final decision concerning the calculation of its adjustments for fiscal years 1993-1996 (Count One) or, in the alternative, seeks production of certain supplemental security income (“SSI”) entitlement records from the Social Security Administration that would allegedly affect the amount of its adjustments (Counts Two and Three).

The parties have agreed that Count One against the Secretary should be resolved first because, if resolved in Baystate’s favor, it will be dispositive of the case. The parties have filed cross-motions for summary judgment, and a motions hearing was held on February 14, 2008. 1 Plaintiffs in Auburn Regional Med. Ctr. v. Leavitt, Civil Action No. 07-2075 (D.D.C. filed Nov. 15, 2007) (collectively, the “Auburn plaintiffs”) were allowed to participate in the briefing as amici because they have raised many of the same issues as Baystate. 2 For the reasons explained below, the Court will grant in part and deny in part both Baystate’s and the Secretary’s motions for summary judgment and remand the case to the Secretary for further proceedings.

BACKGROUND

I. Statutory and Regulatory Background

Through a complex statutory and regulatory regime, the Medicare program reimburses qualifying hospitals for the services they provide to eligible elderly and disabled patients. See generally County of Los Angeles v. Shalala, 192 F.3d 1005, 1008 (D.C.Cir.1999). The “operating costs of inpatient hospital services” are reimbursed under a prospective payment system (“PPS”) — that is, based on prospectively determined standardized rates — but subject to hospital-specific adjustments. 42 U.S.C. § 1395ww(d); see generally In re Medicare Reimbursement Litig., 309 F.Supp.2d 89, 92 (D.D.C.2004), aff'd, 414 F.3d 7, 8-9 (D.C.Cir.2005). One such adjustment is the “disproportionate share hospital” (“DSH”) adjustment which requires the Secretary to provide an additional payment to each hospital that “serves a significantly disproportionate number of low-income patients.” 42 U.S.C. § 1395ww(d)(5)(F)(i)(I). Congress concluded that the additional DSH payment was necessary for hospitals serving a disproportionate share of low-income patients because they have higher costs per case. See H.R.Rep. No. 99-241, at 16-17 (1985), reprinted in 1986 U.S.C.C.A.N. 579, 594-95. Whether a hospital qualifies for the DSH adjustment, and the amount of the adjustment it receives, depends on the “disproportionate patient percentage” determined by the Secretary under a statutory formula. 42 U.S.C. § 1395ww(d)(5)(F)(v)-(vii). This percentage is a “proxy measure for low income.” See H.R.Rep. No. 99-241, at 16, reprinted in 1986 U.S.C.C.A.N. at 594.

The disproportionate patient percentage is the sum of two fractions, commonly referred to as the Medicaid fraction (often called the Medicaid Low Income Proxy) and the Medicare fraction (the Medicare Low Income Proxy). 42 U.S.C. § 1395ww(d)(5)(F)(vi); Jewish Hospital, Inc. v. Secretary of Health and Human Servs., 19 F.3d 270, 272 (6th Cir.1994). The Medicaid fraction reflects the number of hospital inpatient days attributable to patients who were eligible for medical assistance under a State Medicaid Plan, but not Medicare Part A. See 42 U.S.C. § 1395ww(d)(5)(F)(vi)(II); In re Medicare Reimbursement Litig., 309 F.Supp.2d at 93. In contrast, the Medicare fraction reflects the number of hospital inpatient days attributable to Medicare Part A patients who are also entitled to Supplemental Security Income (“SSI”) benefits at the time of their hospital stays, and thus is often referred to as the SSI fraction. See 42 U.S.C. § 1395ww(d)(5)(F)(vi)(I).

The SSI fraction' — and issues pertaining to identifying SSI patients — lies at the heart of this case. The DSH provision defines the SSI fraction as:

the fraction (expressed as a percentage), the numerator of which is the number of such hospital’s patient days for such period which were made up of patients who (for such days) were entitled to benefits under part A of this subchapter and were entitled to supplemental security income benefits (excluding any State supplementation) under subchap-ter XVI of this chapter, and the denominator of which is the number of such hospital’s patient days for such fiscal year which were made up of patients who (for such days) were entitled to benefits under part A of this subchapter.

Id. § 1395ww(d)(5)(F)(vi)(I) (hereinafter, “DSH provision” or “subparagraph (d)(5)(F)(vi)”). In other words, calculation of the numerator and denominator each require tallying the number of patient days for patients entitled to benefits under Part A of the Medicare Act; the numerator differs from the denominator in that it captures an additional restriction — the Medicare patients in the numerator must also be entitled to SSI benefits at the time of their hospital stays. See 51 Fed.Reg. 16772, 16777 (May 6, 1986) (DSH interim final rule). 3 Thus, the greater the number of patient days involving SSI beneficiaries, the greater the hospital’s DSH adjustment.

The SSI program is a federal assistance program for low-income individuals who are aged, blind, or disabled. 42 U.S.C. § 1382. The SSI program is administered by the Social Security Administration (“SSA”). Id. § 1381a. Hence, calculation of the numerator of the SSI fraction requires use of SSA data, in addition to the Medicare inpatient data compiled by CMS.

The Health Care Financing Administration — the predecessor to CMS — announced soon after enactment of the DSH provision that the Secretary, rather than the hospitals, would be solely responsible for computation of the SSI fraction in light of the voluminous data and involvement of another agency. 51 Fed.Reg. 31454, 31459 (Sept. 3, 1986) (DSH final rule). HCFA noted that, as of 1986, the data sources for the computation of the SSI fraction included approximately 11 million billing records from the Medicare inpatient discharge file, and over 5 million records from the SSI file compiled by the SSA. Id. Computation of the SSI fraction would require matching individual Medicare billing records to individual SSI records. Id. Considering the administrative burdens and complexity of the match process, HCFA concluded that it would be solely responsible for the match process, which it would conduct retrospectively for every eligible Medicare hospital on a “fiscal year” basis — that is, based on discharges occurring in the federal fiscal year ending each September 30. Id. at 31459-60; 42 C.F.R. § 412.106(b). 4 HCFA recognized that the statute provid ed for the match process to be based on a hospital’s “cost reporting period” (rather than the fiscal year); thus, HCFA also provided each hospital the option of having its SSI fraction computed based on its own cost reporting period notwithstanding the belief that the two different periods would yield “reasonably close” percentages. 5 51 Fed.Reg. at 16777; 42 C.F.R. § 412.106(b).

The data underlying the match process is drawn from (1) the Medicare Provider Analysis and Review File (“MEDPAR”), which is also referred to as the Medicare Part A Tape Bill (“PATBILL”), and (2) an SSI eligibility file that the SSA transmits annually to CMS. 6 52 Fed.Reg. 33143, 33144 (Sept. 1, 1987); 51 Fed.Reg. at 16777. The SSI fraction is thus determined by computing the number of patient days of those patients entitled to both Medicare Part A and SSI as indicated by matching CMS’s MEDPAR file and the SSA’s SSI eligibility file. 51 Fed.Reg. at 16777.

Medicare payments are initially determined by a “fiscal intermediary” — typically an insurance company that acts as the Secretary’s agent for purposes of reimbursing health care providers. See 42 C.F.R. §§ 421.1, 421.3, 421.100-.128. A fiscal intermediary is required by regulation to apply the SSI fraction computed by CMS. See id. § 412.106(b)(2) and (b)(5). In contrast, the intermediary (rather than CMS) calculates the Medicaid fraction based on annual cost reporting data submitted by the provider. Id. §§ 412.106(b)(4), 413.20. The fiscal intermediary then determines the total amount of DSH reimbursement due the provider, which it sets forth in a Notice of Program Reimbursement (“NPR”). Id. § 405.1803.

A provider dissatisfied with the amount of the award is entitled to request a hearing before the Provider Reimbursement Review Board (“PRRB” or “Board”), an administrative body composed of five members appointed by the Secretary who must be “knowledgeable in the field of payment of providers of services.” 42 U.S.C. § 1395oo(a), (h). The Board has the authority to affirm, modify, or reverse the final determination of the intermediary “even though such matters were not considered by the intermediary in making such final determination,” and the Secretary may then reverse, affirm, or modify the Board’s decision within 60 days thereafter. Id. § 1395oo(d) and (f). The Secretary has assigned the function of reviewing Board decisions to the Administrator of CMS, whose decisions are considered the final decision of the Secretary. 42 C.F.R. § 405.1875.

II. Procedural and Factual Background

Baystate seeks judicial review of the Administrator’s decision on its appeal from four separate notices of program reimbursement issued by the fiscal intermediary for fiscal years 1993, 1994, 1995 and 1996. Baystate alleges that the SSI fractions for those years are understated because the SSA records that CMS relied upon contained inaccurate or incomplete information on the eligibility of patients for SSI and because the match process itself was flawed. 7 Baystate also alleges that CMS applied an incorrect legal standard in determining the number of patients entitled to Medicare Part A. Baystate’s current SSI fractions, in combination with its Medicaid fraction, yielded for those years DSH payments in the range of'$12 million to $17 million each year. 8 AR 3820.

Baystate’s claims were initially heard by the Provider Reimbursement Review Board, which issued a decision on March 17, 2006, granting in part and denying in part the appeal, and remanding the case to the intermediary for recalculation of Bays-tate’s SSI fractions. See Board Decision at 41-42. The Administrator subsequently reversed those parts of the Board’s decision granting Baystate relief, and affirmed CMS’s calculation of the SSI fractions as proper. See Final Decision at 55.

The administrative proceedings before the Board consumed several years. Bays-tate, the intermediary, 9 and CMS engaged in extensive discovery over SSI eligibility records, including a three-day evidentiary hearing on disputed discovery matters in 2003. See Board Decision at 4. Baystate eventually obtained disclosure of a sample of 627 individual-specific SSI eligibility records, but not the complete set of eligibility records it originally sought. Id.; see also Final Decision at 36-37. A subsequent evidentiary hearing took place in September 2004, at which each side presented witnesses consisting largely of current or former CMS and SSA employees, as well as Baystate’s consultants. See Board Decision at 4. The witnesses included, among others, Anthony Dean, the lead CMS computer programmer responsible for the SSI/MEDPAR match process; Daryl Rosenberg, a CMS computer system analyst; Janet O’Leary, CMS’s National Claims History Branch Chief; Cliff Walsh, an SSA branch chief whose branch maintains the SSI master file and associated database; Alan Shafer and Patricia Cribbs, retired SSA employees involved in managing SSI data, now retained by Bays-tate as consultants; and David Pfeil and Gerry Smith, Baystate’s expert consultants. Id. at 10,16,18, 22, 25, 36.

Based on the evidence submitted, the Board made extensive factual findings concerning how CMS conducted the match process, focusing on certain categories of SSI eligibility data and Medicare patient data available to CMS and employed in the process. Most of the factual findings concerning the mechanics of the match process were left in place by the Administrator’s decision (see generally Final Decision at 15-16), and remain undisputed. The Social Security Administration sends CMS a tape each year containing SSI eligibility data covering a 42-month period, a tape that is prepared solely for CMS’s use. Board Decision at 10. This tape, typically sent every March, covers the '36 months of the three prior calendar years and the first six months of the calendar year in which the tape is prepared. 10 Id. Once per quarter, CMS conducts a “MEDPAR run” — that is, matching of the MEDPAR inpatient hospital stay records against the most recent annual SSA tape. Id. at 12. Each quarterly run covers multiple calendar and fiscal years, resulting in 10 MED-PAR runs for each fiscal year. Id. For example, the tenth and final MEDPAR run for the federal fiscal year ending September 30, 1994, occurred in December 1996. Id. Because the cost report settlement period underlying the DSH calculation process takes two to three years to complete, 11 multiple MEDPAR runs are available for any given cost report settlement period. Id. at 32. CMS uses one of the earliest runs to compute the SSI fractions at issue — that is, the MEDPAR run from the June following the end of the federal fiscal year (“the June update”)— rather than up-to-date runs that are available prior to the final cost report settlement for such year. Id. at 10, 32-33.

Baystate alleged that there were five systemic omissions from the SSA annual tapes that caused the SSI fractions for the years at issue to be understated. Id. at 23. Finding no procedural bar to reviewing the SSI fractions and concluding that section 1395ww(d)(5)(F)(vi) required an accurate determination, the Board addressed each category of omissions on its merits. See id. at 6-9. The first two categories of SSI eligibility records omitted are retroactive eligibility determinations by SSA— that is, where SSI benefits are initially denied but subsequently granted on appeal, and where SSI benefits are temporarily on “hold” or “suspense” until additional information is obtained. 12 Id. at 27-29. The Board observed that these ongoing processes take, on average, a year to resolve, and thus the timing of the calculation of the SSI fraction makes a difference in the extent to which retroactive eligibility determinations are incorporated in the SSI fraction. Id. at 28-29, 32-33. The record was undisputed that, due to the timing of CMS’s calculation of the SSI fraction early in the DSH calculation process — in the June following the end of each fiscal year based on the first SSA tape for such year — inpatient hospital stay records from MEDPAR are not matched with SSI records showing the granting or restoration of SSI benefits after each April 1 retroactively to a period within the prior fiscal year. Id. at 32. The Board found this timing improper because CMS has in its possession updated MEDPAR runs based on a subsequent updated SSA tape before the end of the DSH cost report settlement process. Id. The Board thus concluded that CMS did not use the best available data to conduct the matches. Id. at 33.

The next category of omitted SSI eligibility records consists of records considered inactive by the Social Security Administration because of the beneficiary’s death, and which it thus excluded from the tapes sent to CMS until about February 1996. 13 Id. at 23. At that time, CMS directed SSA to cease its practice of omitting inactive data beginning with the tape for fiscal year 1995. Id. at 24. The dispute over this systemic omission — referred to in the record as the “stale records” problem — has remained a live issue, however, because CMS did not take corrective action for past years — that is, for fiscal years 1993 and 1994 — based on a determination that the impact of the omission was not material. Id.

The Board observed that CMS had attempted to assess whether the omission of the “stale records” for past years had a material impact on prior SSI fractions by creating special MEDPAR files that matched the corrected new SSI data against the existing MEDPAR files and original count of SSI days for inpatient hospital stays, thus providing a perspective on how many SSI days had been missed. Id. at 24. The Board observed that Bays-tate’s consultant, Mr. Pfeil, testified that CMS’s summary of the special MEDPAR files reflects that “the ratio of new SSI additions to old SSI deletions was greater than 17:1 for most years,” and showed 400 more SSI days for one unidentified year (apparently 1994). Id. at 24-25; see also AR 5267-68 (Smith testimony indicating that 400 days at issue were from 1994). The Board found that there was no way to know the source of the additional days— that is, whether it reflected a previously omitted “stale record,” unedited data from stays in non-PPS units of a hospital, or data updates reflecting retroactive SSI eligibility determinations. Board Decision at 25. The Board concluded that, although it could not determine the amount of the financial harm, the discrepancy with respect to the 400 additional SSI days was sufficient to demonstrate that the original data was inaccurate and, furthermore, that the “systemic” nature of the error indicated relief was warranted. Id. at 25, 26.

The Board also considered whether CMS knew of the stale records omission prior to February 1996 as part of its inquiry into whether CMS had nonetheless used the “best available data,” which, if so, might warrant denial of relief. Id. at 33-35. Ms. Rosenberg had testified that CMS knew there was “something wrong with the numbers” at least by 1993, because SSI records were inexplicably dropped when DSH percentages were recalculated. Id. at 33. Mr. Pfeil also had been in contact with three CMS employees about this problem in 1993 when CMS could not reconstruct previously published SSI calculations. Id.; AR 5419-29. Thus, the Board concluded that CMS knew or should have known at least by 1993 that there was a problem with the SSI data received from SSA, and thus CMS could not claim that the incomplete tapes were the “best available data.” Board Decision at 25, 35. The Board noted that CMS had lost or destroyed the special MEDPAR files and unidentified SSA tapes some time after 1997, but did not make any findings of bad faith on that basis. Id. at 34-35.

The next category of omitted SSI eligibility records involved persons receiving a forced or manual payment on a temporary basis in lieu of the automated payments that are typically used for SSI payments. In those instances, to prevent a duplicate automated payment, SSA’s practice is to terminate the recipient’s existing record, start a new record that retains a “forced pay” code (COI or M01) but indicates no payment is due, and later create a third record to resume payment when the manual payment action is resolved. Id. at 26. As a result, the SSA annual tapes received by CMS would indicate that “forced pay” recipients have no SSI entitlement. Id. at 26-27. Baystate identified one such stay that was not counted in the sample of SSI records obtained through discovery. Id. at 27. CMS did not dispute that forced pay recipients should be reflected in the count of SSI days in the numerator of the SSI fraction, but submitted that the impact was too small to warrant retroactive correction. Id. However, the Board concluded that, in light of the systemic nature of the problem, and testimony from a former SSA employee, Mr. Shafer, indicating the problem was “common,” the omissions would be likely to deflate Baystate’s SSI fraction. Id. It further concluded that the omission could be corrected by modifying a computer program to check a person’s earlier SSI records when a forced pay code appears on a tape. Id.

The last category of SSI data at issue concerns patients who are not eligible to receive SSI payments, but who have a special status under Section 1619(b) of the Social Security Act, 42 U.S.C. § 1382h(b), that enables them to receive Medicaid assistance based on a past entitlement to SSI payments. Such patients are intentionally excluded from the SSA annual tapes because, in CMS’s view, they are not “entitled to supplemental security income benefits” as required by subparagraph (d)(5)(F)(vi) of the DSH statutory provision. Id. at 30. The Board disagreed, finding that the special status under Section 1619(b) is a “supplemental security income benefit” within the meaning of the DSH provision. Id.

The Board also considered whether, independent of these systemic omissions from the SSI eligibility data, Baystate’s SSI fractions were understated due to flaws in the match process. Baystate contended that certain types of omissions inevitably occurred because CMS failed to use individuals’ own social security numbers to match records, and instead used other allegedly inadequate “patient identifiers” in the MEDPAR database and SSA tapes. 14 Id. at 15.

The Board reviewed extensive testimony from CMS and SSA employees to ascertain which patient identifiers CMS relies upon and the role that social security numbers play. From the MEDPAR database, the record is clear and undisputed. CMS uses a patient’s Health Insurance Claim Account Number (“HICAN”), which consists of a social security or railroad retirement number followed by an alpha-numeric beneficiary identification code indicating the relationship between the patient and the account holder. 15 Id. at 14. However, the HICAN may or may not be the patient’s own social security number. Id. For example a married woman receiving Medicare Part A benefits on the account of her husband, based on his social security status, would have a HICAN composed of the husband’s nine-digit social security number followed by an alpha-numeric beneficiary identification code (“BIC”) indicating her spousal relationship to him. Id.

CMS runs a computer program that matches each HICAN against numbers known as “Title II numbers” derived from the SSA annual tapes. Id. at 14, 16. The “Title II” number is a number used to track an individual’s benefits under the Social Security Retirement program under Title II of the Social Security Act, 16 and consists of the social security number of the Title II record holder (the person whose work history qualifies for benefits) and an alpha-numeric suffix, or BIC, that denotes the relationship between the beneficiary and the Title II record holder. 17 Id. at 13. The Board observed that “[o]f-ten an individual’s HIC[AN] number is the same as his or her Title II number,” thus enabling matches to occur. Id. at 14. But there are undisputed problems with using the Title II number as the identifier: (1) an individual may have more than one Title II number at the same time because he or she may have received benefits under more than one Title II account, and (2) a person’s Title II number may change over time due to marriage, divorce, death of a spouse, or changes in work status. Id. at 13-14. The Board found this variation in Title II numbers significant because SSA’s annual tape includes a field for only one Title II number, ordinarily the oldest Title II number, thus preventing some matches from occurring. 18 Id. at 14.

Another problem identified by the Board was that many individuals receiving SSI are not assigned Title II numbers on the SSA annual tapes. Id. at 16-17, 20. This would, however, result in missed matches only if CMS were using Title II numbers exclusively, without regard to the social security numbers listed on the SSA annual tapes. After considering inconsistent testimony from CMS’s principal MEDPAR programmer, Mr. Dean, at two different hearings, the Board concluded that social security numbers on the SSA tapes were not utilized, and thus individuals receiving SSI payments outside of the Title II program would have been omitted from the match process. Id. at 16-21, 42. Mr. Dean initially testified at a 2003 hearing that he runs a program called SSISORT that reformats the SSI data received from SSA, drops all SSI records that do not include a Title II number, and puts the remaining SSI records in a new output file. Id. at 16. He repeated in other testimony that, if “there is no Title II number, which is all we are ... on our end, that is all we are really concerned about ... this [SSI-SORT] process would not write that record out.... I wouldn’t keep that record.” Id. He also called the other data elements in SSA’s annual tapes, including social security numbers, “useless.” Id.

But at the evidentiary hearing the next year, Dean insisted that social security numbers from the SSA annual tapes are, in fact, used in the matching process. Id. at 17. He testified that CMS runs matches against two numbers (instead of one) — the Title II number, plus a second number that CMS generates from the social security number with the addition of “A” in the BIC position, which he also calls a HICAN. Id. He denied that the SSI-SORT program discards records from the SSA tapes if they do not contain a Title II number, since, under his revised testimony, “we create a record from every record Social Security sends us by generating a HIC[AN] off of the Social Security number” and “we’ll match against numbers that are provided in the Title 2 field and we’ll match against numbers provided in the Social Security field because we generate a [HICAN] from that.” Id. (emphasis added).

The Board discounted the later testimony in light of the strength of Mr. Dean’s original testimony and corroborating evidence. In particular, it noted that CMS’s discovery responses during the administrative proceedings denied that CMS matched based on a second number derived from a social security number: “ ‘[CMS] has no recollection, and no documentation concerning, whether CMS ever saw a need ... to employ secondary or alternative matching criteria’ in addition to or in lieu of, the HICAN.” Id. at 19; see also AR 3650-51. The Board also observed that a long-time CMS employee, Ms. O’Leary, believed that social security number information would not have been incorporated. Board Decision at 19.

Based on its review of the testimony, the Board ultimately concluded that there is a “great likelihood” that matches will be missed because of the “failure to match on a unique identifier and/or multiple criteria,” and further that beneficiaries’ social security numbers and alternative identifiers, such as name and gender, should be utilized, as occurs in a comparable matching process that is in place to monitor SSI payments to patients in nursing homes. Id. at 13, 21. Based on its findings, the Board granted Baystate substantially all of the relief requested on issues concerning the completeness of SSI data and its management during the match process. Id. at 41-42. The Board found that granting this relief would impose no significant administrative burden, since it involved primarily redesigning the computer program to capture the information at issue. Id. at 42.

The Board, however, rejected Bays-tate’s claims that the universe of Medicare Part A patient days was counted based on an incorrect legal standard — a standard that affects the count of days in both the numerator and denominator of the SSI fraction. 19 Id. at 35-40. On this front, Baystate alleged that CMS had been over-inclusive in its count of patient days by counting all days that were “covered” by Medicare Part A, including Medicare HMO days, even if the provider had not received any payment. Id. at 37-40. The Board concluded that the statutory DSH provision and its implementing regulation defined the denominator solely in terms of a patient’s “entitle[ment] to benefits under part A” of Medicare, and thus the count of Medicare Part A days properly included those days for which Medicare Part A provided coverage to the patient, including HMO days, even though the hospital may not have received payment. Id.

On April 7, 2006, CMS notified the parties of the Administrator’s intent to review the Board’s decision, based upon comments submitted by CMS’s Center for Medicare Management and the Intermediary. AR 97-98. The Administrator subsequently affirmed the Board’s holding that CMS had counted the universe of Medicare Part A patient days, including HMO days, using the correct legal standard, but reversed in all other respects. Final Decision at 31-55. He determined that, as a threshold matter, Baystate was precluded from challenging the SSI fraction at all because, in his view, the regulation governing calculation of the SSI fraction, 42 C.F.R. § 412.106(b), authorizes only two calculations — the computation based on the federal fiscal year and, upon request, recalculation of the fraction based on a hospital’s cost reporting period. Id. at 17-21. He thus concluded that further recalculations were unauthorized. 20 Id. at 17-18.

The Administrator then proceeded to address the merits of Baystate’s claims, leaving largely undisturbed the Board’s findings that the types of SSI eligibility data at issue would not be reflected in the SSI fractions, but finding that those omissions were not significant. He focused first on the level of precision that is required in calculating the SSI fraction under section 1395ww. 21 Id. at 23. The Administrator held that the Board erred in holding CMS to a standard of “absolute precision,” which improperly allowed the SSI fraction to be continuously corrected with later updated data and hence precluded the finality of payments. Id. at 23-26. He reasoned that the prospective payment system requires instead only that payment rates be “based on the best available data at the time” and that the “best available data is ‘accurate’ data in fact for purposes of payments under [the inpatient prospective payment system].” Id. at 23 (emphasis added). He viewed this standard as reflected in the agency’s prior statements that it would obtain “reasonably accurate but not perfect calculations.” Id. The efficient administration of a complex regulatory benefits program involving millions of records, he reasoned, necessarily allows room for “acceptable rates of error,” draw ing guidance from other precedents involving allocation of benefits. Id. at 26 (quoting Califano v. Boles, 443 U.S. 282, 285, 99 S.Ct. 2767, 61 L.Ed.2d 541 (1979)). He thus concluded that the law authorizes an “acceptable rate of error,” which was satisfied in this case. Id. at 26-27.

The Administrator observed that, out of the sample of 627 individual-specific SSI eligibility records Baystate had compiled, it had shown only 12 stays that were omitted from its SSI fractions, an error rate of 1.59 percent. Id. at 36-37. Moreover, the sample was likely to exaggerate the number of errors because it consisted of records “for individuals for whom the MED-PAR data ... indicated no SSI days were associated with their stay” and thus would only “increase [the provider’s] Medicare DSH payments if later information was different from the original record.” Id. at 37. The sample was further likely to exaggerate errors because Baystate “had only requested records for those individuals that [it], based on State Medicaid data believed were, or likely were, eligible for SSI.” Id. But assuming the 1.59 percent rate of missed matches was correct, the Administrator concluded that it nonetheless failed to support Baystate’s claim that the best available data was not used or that its reimbursement was affected by the errors. Id.

He reached similar conclusions concerning an “acceptable rate of error” when looking at each individual category of omitted data. He concluded that retroactive SSI eligibility determinations — that is, resolutions in a beneficiary’s favor following a suspension or hold of SSI benefits or upon appeal within SSA- — -would have only a “minimal impact” on a provider’s DSH fraction, which “combined with the need for finality in the process,” makes use of later data (i.e., a subsequent SSA tape) impractical. Id. Drawing from a Federal Register notice addressing the issue of updated SSI data in more detail, he explained:

We understand that many hospitals are concerned that later data matches may produce a different Medicare fraction. However, we believe there needs to be administrative finality to the calculation of a hospital’s Medicare fraction. CMS has previously stated that its goal is to obtain reasonably accurate but not perfect calculations (51 Fed.Reg. 16777). Additionally, our data have shown that 98 to 99 percent of SSI eligibility determinations are made and remain unchanged 6 months after the end of the Federal fiscal year.... Given the time between the end of [a] hospital’s cost reporting periods and when we are furnished with SSI eligibility information for that period, we believe it is highly unlikely that a subsequent data run will produce data that is significantly different than one completed 6 months after the end of the Federal fiscal year.

Therefore, we will use the SSI eligibility information provided to CMS by SSA 6 months after the end of the Federal fiscal year (or April 1) to calculate the Medicare fraction. We will match these data to the MedPAR system once and conduct no further matches after that time.

Id. at 38 (quoting 70 Fed.Reg. 47278, 47439-40 (Aug. 12, 2005)). The Administrator believed the minimal impact was demonstrated by Baystate’s sample, which showed only one patient stay that was omitted from the numerator because of a suspension of SSI benefits that was later reinstated — an error rate of 1/6 of one percent. Id. at 39. The error rate was likely to be even less, he concluded, because retroactive disallowances of benefits were likely to counterbalance any retroac tive reinstatement of benefits. Id. at 39-40.

The Administrator found the error rate for omission of patients receiving a manual or forced payment from SSA to be similarly low — again, only one patient stay from the non-random sample, a 0.15 percent error rate. Id. at 41. He found no basis for concluding the error rate might be higher, citing conflicting testimony on the frequency of forced pay cases: one witness testified she had never seen a forced pay situation, while another testified — without corroborating evidence — that it occurred frequently. Id.

With respect to the omission of “stale records” — that is, records that SSA had terminated as “inactive” due to a beneficiary’s death — the Administrator acknowledged the omissions for fiscal years 1993 and 1994, but concluded, as he did for the other data gap issues, that the omissions were not “significant” for providers. Id. at 43-44. He found that “the average size hospital with the average size SSI population would have had about four stale records omitted from its calculation” in fiscal year 1993 and “about three stale records omitted” in fiscal year 1994, based upon his review of the “special MEDPAR project” that CMS had conducted in 1996, Id. at 44-46. He considered omissions on this level to be “not significant,” which he believed was further supported by historical trends in Baystate’s SSI fractions from 1993 to 2000, as well as trends at the national and state level. Id. at 46. That is, SSI fractions at each of these levels were on a “continuous steady rise from one year to the next” without a “noticeable” increase when the stale records omission was corrected by SSA for the 1995 fiscal year. Id. He thus concluded that CMS’s decision not to retroactively recompute the fractions for 1993 and 1994 was reasonable. Id. at 47.

With respect to the exclusion of “Section 1619(b)” patients from the SSI fraction, the Administrator did not rely on any quantitative findings, instead holding that the exclusion was required under section 1395ww(d)(5)(F)(vi) because such persons are not “entitled to supplementary security income benefits under [subchapter] XVI” as required under the DSH provision. Final Decision at 48-49. He reasoned that Section 1619(b) and the relevant SSA regulation (20 C.F.R. § 416.264) provide that the special status applies only “for purposes of’ Medicaid eligibility, in contrast to SSI payments, in order to allow individuals who are disqualified from receiving SSI payments nonetheless to continue on Medicaid if certain conditions are met. Id. at 49. Because individuals covered by section 1619(b) are not “entitled to” any SSI payment, he concluded that they were properly excluded from calculation of the SSI fraction. Id.

On the subject of whether the match process utilized sufficient patient identifiers from the MEDPAR and SSA tapes to enable matches to occur, the Administrator again reversed the Board, this time based on contrary factual findings as well as an acceptable rate of error. Id. at 42-43. As a factual matter, he rejected Baystate’s allegation that CMS eliminated SSI records where individuals lacked Title II numbers. Id. at 42. He described the testimony of the CMS programmer — Mr. Dean — as “confirming] that CMS generates a Title II number from each social security number on the [SSA] tape,” and thus, if an individual’s record lacks a Title II number, CMS matches based on the number it has created from the social security number. Id.

The Administrator also rejected the contention that the match process should be based on a beneficiary’s own social security number, relying on an earlier determi nation by the Secretary that “they [social security numbers] are not specific to an individual Medicare beneficiary” — apparently referring to the situation where one family member claims benefits under another’s social security number. Id. at 42 (citing 70 Fed.Reg. at 47440-41). As to other patient identifiers, such as name, gender, and date of birth, he noted that the Secretary had previously determined that providers should not be allowed to use alternative criteria in the match process because it would produce “ ‘an inconsistent matching methodology and inconsistent DSH Medicare fraction calculations among providers.’ ” Id. at 43 (quoting 70 Fed. Reg. at 47440^41). The Administrator further found that the decision not to use alternative patient identifiers was supported by the low error rate in Baystate’s sample, which showed only one uncredited SSI stay — a 0.15 percent error rate — resulting from the limitations of HICAN and Title II identifiers. Id. at 43. He thus concluded that the Secretary’s “policy determination” not to use alternative patient identifiers was supported by the record. Id.

The Administrator lastly rejected the Board’s conclusion that there would be no administrative burden in modifying the computer program to correct the alleged errors. 22 He found the Board’s reasoning was inappropriately limited to the single provider before it and failed to consider “the enormity of the Medicare Program,” which he considered necessary because of the “retrospective and wide reaching implications” of the errors at issue. Id. at 53-54. The Administrator thus concluded on multiple grounds that Baystate was not entitled to a recalculation of its SSI fractions for the years at issue, and affirmed CMS’s determination of the fractions as proper. Id. at 54. His decision constituted the final decision of the Secretary of Health and Human Services. Id. at 55.

STANDARD OF REVIEW

Under Fed.R.Civ.P. 56(c), summary judgment is appropriate when the pleadings and the evidence demonstrate that “there is no genuine issue as to any material fact and that the moving party is entitled to judgment as a matter of law.” In a case involving review of a final agency action under the Administrative Procedure Act, 5 U.S.C. § 706, however, the standard set forth in Rule 56(c) does not apply because of the limited role of a court in reviewing the administrative record. See North Carolina Fisheries Ass’n v. Gutierrez, 518 F.Supp.2d 62, 79 (D.D.C.2007); Fund for Animals v. Babbitt, 903 F.Supp. 96, 105 (D.D.C.1995), amended on other grounds, 967 F.Supp. 6 (D.D.C.1997). Under the APA, it is the role of the agency to resolve factual issues to arrive at a decision that is supported by the administrative record, whereas “the function of the district court is to determine whether or not as a matter of law the evidence in the administrative record permitted the agency to make the decision it did.” See Occidental Eng’g Co. v. INS, 753 F.2d 766, 769-70 (9th Cir.1985); see also Northwest Motorcycle Ass’n v. U.S. Dep’t of Agriculture, 18 F.3d 1468, 1472 (9th Cir.1994) (“[T]his case involves review of a final agency determination under the [APA]; therefore, resolution of th[e] matter does not require fact finding on behalf of this court. Rather, the court’s review is limited to the administrative record.”). Summary judgment thus serves as the mechanism for deciding, as a matter of law, whether the agency action is supported by the administrative record and otherwise consistent with the APA standard of review. See Richards v. INS, 554 F.2d 1173, 1177 & n. 28 (D.C.Cir.1977), cited in Bloch v. Powell, 227 F.Supp.2d 25, 31 (D.D.C. 2002), aff'd, 348 F.3d 1060 (D.C.Cir.2003).

Under the APA, the Administrator’s decision can be set aside only if it is “unsupported by substantial evidence,” or “arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law.” 23 5 U.S.C. § 706(2)(A) and (E); see 42 U.S.C. § 1395oo(f)(1) (providing that judicial review of Medicare reimbursement decisions shall be made under APA standards); St. Elizabeth’s Med. Ctr. of Boston v. Thompson, 396 F.3d 1228, 1233 (D.C.Cir.2005). The “scope of review under the ‘arbitrary and capricious’ standard is narrow and a court is not to substitute its judgment for that of the agency.” Motor Vehicle Mfrs. Ass’n of U.S., Inc. v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983). The court must be satisfied that the agency has “examine[d] the relevant data and articulate[d] a satisfactory explanation for its action including a rational connection between the facts found and the choice made.” Id.; Alpharma, Inc. v. Leavitt, 460 F.3d 1, 6 (D.C.Cir.2006). The agency’s decisions are entitled to a “presumption of regularity,” Citizens to Preserve Overton Park, Inc. v. Volpe, 401 U.S. 402, 415, 91 S.Ct. 814, 28 L.Ed.2d 136 (1971), and although “inquiry into the facts is to be searching and careful, the ultimate standard of review is a narrow one.” Id. at 416, 91 S.Ct. 814. The Court’s review is confined to the administrative record, subject to limited exceptions not applicable here. See Camp v. Pitts, 411 U.S. 138, 142, 93 S.Ct. 1241, 36 L.Ed.2d 106 (1973) (“[T]he focal point for judicial review should be the administrative record already in existence, not some new record made initially in the reviewing court.”).

Where an administrative tribunal — here, the Board — issues a decision based on an evidentiary hearing, and a final agency decision follows, “[t]he reviewing court must take the [tribunal’s] findings into ac count as part of the record,” and “the significance to be ascribed to them depends largely on the importance of credibility in the particular case.” Morall v. DEA, 412 F.3d 165, 179 (D.C.Cir.2005) (citations and internal quotation marks omitted). The final decision must “consider relevant contradictory evidence, including evidence that led the [tribunal] to contrary findings of fact and credibility,” and failure to do so may result in reversal. Id. at 180.

A court should review an agency’s interpretation of a statute under the familiar two-step analysis outlined in Chevron, U.S.A., Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 104 S.Ct. 2778, 81 L.Ed.2d 694 (1984). The first step is determining whether Congress has spoken directly to the “precise question at issue,” for if it has, “the court, as well as the agency, must give effect to the unambiguously expressed intent of Congress.” Id. at 842-43, 104 S.Ct. 2778; State of New Jersey v. EPA, 517 F.3d 574, 581 (D.C.Cir. 2008). If, however, the statute is silent or ambiguous on the specific issue, “the question for the court is whether the agency’s answer is based on a permissible construction of the statute.” Chevron, 467 U.S. at 843, 104 S.Ct. 2778. When the agency’s construction of a statute is challenged, its “interpretation need not be the best or most natural one by grammatical or other standards.... Rather [it] need be only reasonable to warrant deference.” Pauley v. BethEnergy Mines, Inc., 501 U.S. 680, 702, 111 S.Ct. 2524, 115 L.Ed.2d 604 (1991) (citations omitted).

DISCUSSION

I. Individuals with Special Status under Section 1619(b)

Baystate contends that CMS improperly omitted several types of patient days from the numerator of the SSI fraction. The Court turns first to the category of patients with Section 1619(b) status. They were omitted from the SSI fraction based on the agency’s interpretation of the DSH statute and the position of Section 1619(b) patients relative to it.

Patients covered by Section 1619(b) are persons whose income rendered them ineligible for SSI payments but who are conferred a special status under Section 1619(b) of the Social Security Act, 42 U.S.C. § 1382h(b), that enables them to remain eligible for Medicaid. The resolution of this issue turns, however, on whether the phrase “patients ... entitled to supplemental security income benefits ... under subchapter XVI” in section 1395ww(d)(5)(F)(vi) includes patients with Section 1619(b) status. That question is resolved under the familiar Chevron framework. As indicated earlier, the first step under Chevron requires the Court to determine whether Congress “has directly spoken to the precise question at issue.” Chevron, 467 U.S. at 842, 104 S.Ct. 2778. At this stage of the analysis, the Court employs the “traditional tools of statutory construction, including examination of the statute’s text, legislative history, and structure, as well as its purpose.” Shays v. Fed. Election Comm’n, 414 F.3d 76, 105 (D.C.Cir.2005) (citation and internal quotation marks omitted). “When Congress has spoken, we are bound by that pronouncement and that ends this Court’s inquiry.” Nat’l Treasury Employees Union v. Federal Labor Relations Auth., 392 F.3d 498, 500 (D.C.Cir.2004). Only when “Congress’s intent is ambiguous” does the Court proceed to the second step of the inquiry and consider “whether the agency’s interpretation is based on a permissible construction of the statute.” New York v. EPA, 413 F.3d 3, 18 (D.C.Cir.2005) (internal quotation marks omitted).

A careful review of the text, the structure, the legislative history and the purpose of both subparagraph (d)(5)(F)(vi) and Section 1619(b) leads the Court to the inescapable conclusion that Congress did not intend that patients ineligible for SSI payments would be counted in the numerator described by subparagraph (d)(5)(F)(vi). Congress created the DSH payment to compensate hospitals that serve a disproportionately large share of low-income patients because they generally have higher costs than are accounted for by standardized Medicare payments. H.R.Rep. 99-241, at 16 (1985), reprinted in 1986 U.S.C.C.A.N. 579, 594. But Congress did not measure those costs directly, and instead set forth a “proxy measure for low-income.” Id. The proxy measure for low income in the SSI fraction is determined by the number of patient days for patients “entitled to supplemental security income benefits ... under subchapter XVI.” § 1395ww(d)(5)(F)(vi). This proxy makes sense because subchapter XVI provides that certain categories of individuals — the aged, blind, or disabled — shall be “paid benefits” only if their income and resources fall below certain thresholds. See 42 U.S.C. §§ 1381a, 1382.

Section 1619(b) status, however, is for those who do not meet those low income and resource thresholds, but who may nonetheless be continued on Medicaid assistance based on their past SSI eligibility in order to avoid creating a disincentive to return to work. 24 Section 1619(b) states:

[F]or purposes of subchapter XIX of this chapter [42 U.S.C. §§ 1396-1396v, Medicaid], any individual who was determined to be a blind or disabled individual eligible to receive a benefit under section 1382 of this title or any federally administered State supplementary payment for a month and who in a subsequent month is ineligible for benefits under this subchapter (and for any federally administered State supplementary payments) because of his or her income shall, nevertheless, be considered to be receiving supplementary security income benefits for such subsequent month provided that-Ncertain enumerated requirements are met] ...

42 U.S.C. § 1382h(b) (emphasis added). In simpler terms, Section 1619(b) provides that, “for purposes of’ Medicaid, those eligible for SSI payments who subsequently become “ineligible for benefits” under the subchapter governing supplemental security income will nevertheless be “considered” to be receiving such benefits upon certain conditions.

The operative language of Section 1619(b) for determining the reach of its special status is the limiting opening phrase: “for purposes of subchapter XIX of this chapter.” Subchapter XIX, 42 U.S.C. §§ 1396-1396v, deals exclusively with Medicaid, while Medicare is covered by subchapter XVIII. See In re Medicare Reimbursement Litig., 309 F.Supp.2d at 93 (contrasting eligibility for Medicaid under subchapter XIX with Medicare under Part A of subchapter XVIII). Thus, although the statute goes on to state that an individual “ineligible for benefits under this subchapter” will be “considered to be receiving supplemental security income benefits,” he or she will be so considered only “for purposes of subchapter XIX”— that is, Medicaid. Indeed, by describing the individual as “ineligible for benefits under this subchapter [XVI],” Section 1619(b) makes the very point that Section 1619(b) status is not a “benefit under sub- chapter XVI” — the criteria for counting SSI days in the statutory DSH provision, § 1395ww(d)(5)(F)(vi).

The legislative history confirms that Congress intended Section 1619(b) status to serve a limited function with respect to the continuation of Medicaid benefits. Congress explained that it intended to address the “work disincentive problem” that was created when the loss of Medicaid benefits, along with loss of SSI payments, “significantly outweigh the potential gain from [employment] earnings.” See S.Rep. No. 96-408, at 44-45 (1979), reprinted, in 1980 U.S.C.C.A.N. 1280, 1322-23; see also H.R. Conf. Rep. 96-944, at 49-50 (1980), reprinted in 1980 U.S.C.C.A.N. 1392, 1397-98 (discussing need to continue Medicaid for individuals engaging in employment activity). Section 1619(b) thus establishes a work incentive for an individual eligible for SSI payments by allowing him or her to receive Medicaid under subchap-ter XIX even when “ineligible for benefits under this subchapter [XVI] ... because of his or her income”—that is, the SSI payments established under 42 U.S.C. § 1382. See S.Rep. No. 96-408, at 44-45; H.R. Conf. Rep. 96-944, at 49.

The limited nature of the special status is reiterated in the SSA’s implementing regulations, which state: “The special SSI eligibility status applies for purposes of establishing or maintaining your eligibility for Medicaid.” 20 C.F.R. § 416.264 (1995) (emphasis added). The regulation further provides that: “For these purposes, we continue to consider you to be a blind or disabled individual receiving benefits even though you are in fact no longer receiving regular SSI benefits or special SSI cash benefits.” Id. (emphasis added).

Baystate contends, however, that the quoted language from the regulation actually establishes that the SSA considers Section 1619(b) status a “benefit” under subchapter XVI because the regulation states “we continue to consider you to be a blind or disabled individual receiving benefits.” See Pl.’s Mem. at 33-34 (quoting 20 C.F.R. § 416.264). The problem with Baystate’s spin on this part of the regulation is that it ignores the opening qualification “[f]or these purposes” — that is, eligibility for Medicaid — and also ignores the rest of the sentence' — “even though you are in fact no longer receiving regular SSI benefits or special SSI cash benefits.” The import of this language is unmistakable — the SSA recognizes that Section 1619(b) individuals are “in fact” not receiving “benefits” but will “consider” them so solely for the purpose of maintaining eligibility for Medicaid. The legislative history recognizes that such individuals are not actually entitled to SSI benefits. See H.R. Conf. Rep. 96-944, at 49 (“When a disabled SSI recipient’s earnings rise to the point that he no longer qualifies for federal SSI benefits ... he would nevertheless continue to retain eligibility for Medicaid ... as though he were an SSI recipient.”) (emphasis added).

Baystate then highlights a handful of Federal Register preambles published by the SSA in an attempt to demonstrate that the SSA, the agency charged with administering Section 1619(b), construes that section to confer a “supplemental security income benefit under subchapter XVI.” See Pl.’s Reply Mem. at 33-34 (citing 70 Fed.Reg. 57132, 57133 (Sept. 30, 2005), 61 Fed.Reg. 31022, 31024 (June 19, 1996), and 59 Fed.Reg. 41400, 41402 (Aug. 12, 1994)). To be sure, the word “benefit” appears in conjunction with Section 1619(b) on some occasions. But those preambles do not establish that the SSA views Section 1619(b) status as anything more than a status that maintains an individual’s entitlement to Medicaid. 25 Indeed, the SSA did not purport to construe Congress’ intent in defining the SSI fraction in these isolated Federal Register excerpts, nor was SSA suited to do so since it does not administer the Medicare program. See Ragsdale v. Wolverine World Wide, Inc., 535 U.S. 81, 91, 122 S.Ct. 1155, 152 L.Ed.2d 167 (2002) (an agency “may not exercise its authority in a manner that is inconsistent with the administrative structure that Congress enacted into law.”) (citations and internal quotation marks omitted). More significantly, under Chevron, this type of isolated preamble language is of no moment, for “the court, as well as the agency, must give effect to the unambiguously expressed intent of Congress.” Chevron, 467 U.S. at 842-43, 104 S.Ct. 2778. Those preambles are, for that additional reason, uninstructive.

The opening phrase of Section 1619(b) is clear in limiting the reach of its special status only to Medicaid eligibility. Congress, then, has spoken directly to the question at issue — patients ineligible for payment of supplemental security income benefits because of their incomes should not be counted in the SSI fraction notwithstanding their continued eligibility for Medicaid under Section 1619(b). But even if the phrase “patients ... entitled to supplemental security income benefits ... under subchapter XVI” in subparagraph (d)(5)(F)(vi) were ambiguous, which it is not, the Court would defer to the Administrator’s interpretation under the second step of the Chevron inquiry. See Am. Bar Ass’n v. FTC, 430 F.3d 457, 468 (D.C.Cir. 2005) (a court “will uphold the [agency’s] interpretation of [an] ambiguous statute if that interpretation is ‘permissible,’ that is, if it is ‘reasonable’ ”). Exclusion of patients with incomes exceeding the SSI eligibility thresholds is in harmony with the principle that the DSH adjustment is a proxy for measuring the number of low income patients served by a Medicare provider. The Court thus rejects Baystate’s challenge to the Secretary’s interpretation of the statutory definition of the SSI fraction as excluding individuals with Section 1619(b) status.

II. Calculation of the SSI Fraction Based on the Best Available Data

The Court next turns to the largely record-based issues concerning whether the numerator of the SSI fraction was calculated based on an adequate set of SSI eligibility records. Resolution of this aspect of the case requires disentangling several overlapping and intertwined issues, including whether retroactive relief is available and warranted on this record, the standard of accuracy that applies to CMS’s calculation of the SSI fraction, whether the best available data was used, and whether acknowledged errors in the data can be tolerated under the law. The Court believes it makes sense to first consider the issues concerning the standard of accuracy and data errors, and then address whether retroactive correction is required, inasmuch as the latter issue turns on the nature of the errors.

A. The Best Available Data Standard

Baystate contends that the Administrator’s decision is fatally flawed because the definition of the SSI fraction in section 1395ww(d)(5)(F)(vi) does not authorize the kinds of estimates and approximations he accepted, but instead requires an accurate calculation of the number of patient days comprising the fraction. See Pl.’s Mem. at 22-23, PL’s Reply at 6-9. Under Bays-tate’s standard, any “known errors” in the process are prohibited. PL’s Reply at 6. The Secretary responds that he is entitled to Chevron deference with respect to his conclusion that subparagraph (d)(5)(F)(vi) allows for an “acceptable rate of error,” and urges the Court to reject the proposition that complete accuracy is required in a system involving millions of SSI and Medicare records. See Def.’s Mem. at 29-37. Framing the issue solely in terms of the level of accuracy required leads the Court down an errant path, however, as neither Baystate nor the Secretary posits that the outcome of this case depends on whether one specific error rate or another exists. 26 Instead, the parties disagree over whether the data and processes utilized were reasonably designed to yield accurate results, and the Court agrees that CMS’s reliance on the best available data, or absence thereof, ultimately provides the basis for resolving this issue.

The starting point is subparagraph (d)(5)(F)(vi), which describes the SSI fraction as composed of a numerator consisting of “the number of such hospital’s patient days” attributable to patients who were entitled to benefits under Medicare Part A and supplemental security income, and a denominator consisting of “the number of such hospital’s patient days” attributable to patients who were entitled to benefits under Medicare Part A. Baystate contends that the repeated use of the term “the number” requires the agency to calculate the SSI fraction with accuracy, rather than making an estimate. But the Administrator did not disagree with this point. Instead, he stated that the SSI fraction, like other rates under the Medicare prospective payment system, must be based on “the best available data,” and that such data is “accurate” for the purpose of payments under PPS. Final Decision at 23. He further recognized that he must make “reasonably accurate” calculations, but admitted that such calculations would not necessarily be “perfect,” implying that the degree of accuracy would be limited by any flaws in the “best available data.” Id.

This articulation of the agency’s duty is consistent with the repeated recognition in the case law that the agency must use “the most reliable data available” to produce figures that can be considered sufficiently “accurate.” See Methodist Hosp. of Sacramento v. Shalala, 38 F.3d 1225, 1230 (D.C.Cir.1994) (holding that, where the agency had used “the most reliable data available” in determining a regional wage index, the agency was not required to recalculate Medicare payments for past years based on subsequently corrected data); Mt. Diablo Hosp. v. Shalala, 3 F.3d 1226, 1233 (9th Cir.1993) (holding that, where the agency had used “the most reliable data available at the time,” the agency was not required to recalculate a different wage index although data failed to account for part-time workers); see also County of Los Angeles v. Shalala, 192 F.3d 1005, 1020-23 (D.C.Cir.1999) (holding that agency must explain why a more recent database it had considered “reliable” for certain purposes was not used in calculating other Medicare payments where “accurately forecasting” payments depended on use of updated hospital stay data); Alvarado Community Hosp. v. Shalala, 155 F.3d 1115, 1125 (9th Cir.1998) (approving standard requiring “the most reliable data available,” and noting that the most recent data available was “highly significant to an accurate determination”). These cases teach that the accuracy of any particular index, payment or, in this case, the SSI fraction, cannot be weighed in a vacuum, but instead must be evaluated by reference to the data that was available to the agency at the relevant time. Baystate expressed its agreement with this very principle at the motions hearing. See Mot. Hr’g Tr. at 29 (“[If] the Secretary had engaged a process reasonably calculated to achieve an accurate number, and ... did in fact use at all relevant junctures the best available data, and ended up somehow short of this sort of universal theoretical number that would be 100 percent perfect, I would submit in that case there may not be an error in a sense.”)

The issue, then, is not whether the statute’s reference to “the number” in subpar-agraph (d)(5)(F)(vi) allows any room for error, but rather whether the Administrator reasonably concluded that the SSI fractions for the years at issue were based on the “best available data” at the time the DSH payment was finally determined by the fiscal intermediary. 27 Whether the “best available data” has been used is an inquiry based on the evidence in the record, rather than, as the Secretary suggests (Mem. at 29-30, 37), an issue involving statutory interpretation under Chevron. See County of Los Angeles, 192 F.3d at 1020-22 (evaluating whether the agency used the most reliable data available in computing Medicare payments based on “the evidence before the agency” and sufficiency of agency explanations, without reference to Chevron). Hence, the Court will conduct a record-based inquiry applying the traditional arbitrary and capricious and substantial evidence standards of review.

B. Specific Record Issues Pertaining to Best Available Data

Baystate contends on three primary grounds that the data comprising the SSI fraction was not the “best available data”: first, CMS possessed updated SSI eligibility data that it excluded from the SSI fraction only due to the running of the match process early in the DSH cost report settlement period; second, CMS categorically omitted “stale records” for two fiscal years and manual payment records for all four fiscal years; and third, CMS allegedly failed to use patient identifiers that maximize the number of matches during the match process. The Court addresses each of these issues in turn.

1. SSA Tapes Incorporating Updated SSI Eligibility Data

Turning first to the SSI eligibility data, the Court will consider Baystate’s contention that use of the most recent SSI eligibility data within the cost report settlement period is a significant factor in producing an accurate numerator because SSI eligibility status is updated on later SSA tapes. As discussed earlier, CMS receives a tape from SSA every year (by early April) containing SSI eligibility information covering the three pri- or calendar years, and matches the Medicare inpatient hospital stay records from MEDPAR to that SSA tape the following June to determine the SSI fraction for the prior fiscal year. See Final Decision at 15; Board Decision at 10, 12. One additional updated SSA tape is received — for which an additional match process is run — before the cost report settlement period concludes and the DSH payment is finally calculated, but the information from that subsequent tape and match process goes unused. See Board Decision at 32-33. Baystate contends that use of the match from the first SSA tape after the fiscal year cannot be considered the “best available data” in light of CMS’s possession of an updated SSA tape for that year well before the cost report settlement period closes. The Court agrees.

There is no dispute that there are two types of patients at risk of being omitted from the first SSA tape but who may be captured on the second tape: (1) persons whose SSI benefits are on “hold and suspense” pending SSA’s receipt of additional information; and (2) persons who successfully appeal a denial of SSI benefits and, hence, receive a retroactive award. Indeed, CMS stated in its response to interrogatories that, after its SSI calculation using the first SSA tape, “SSA may update its SSI eligibility records for one or more of the hospital’s patients with respect to the month of discharge of such patient or patients. SSA’s latest SSI eligibility data may show SSI eligibility for the month of discharge for individuals for whom SSI eligibility was not shown in the SSI eligibility data that CMS used to calculate a hospital’s Medicare fraction.” See AR 3634. The Administrator thus recognized that some of these patients may be included in the later SSA tape, but nonetheless rejected use of the later data on two grounds: the “minimal impact” on the provider’s SSI fraction and the “need for finality in the process.” See Final Decision at 38.

Those two justifications for rejecting use of an updated SSA tape are fatally flawed. First, the Administrator acknowledges that there may be a change of up to one to two percent in the SSI eligibility determinations, but then simply by administrative fiat deems this percentage change “minimal.” See id. at 38 (“our data have shown that 98 to 99 percent of SSI eligibility determinations ... remain unchanged 6 months after the end of the Federal fiscal year”). This approach turns the “best available data” standard on its head; it puts the issue of accuracy — here, the impact on the SSI fraction — before data quality. This makes no sense if the SSI fraction is, as the Administrator held, considered “accurate” only by virtue of its reliance on the “best available data.” See Final Decision at 23 (“The Administrator finds that ... [the] fraction must be based on the ‘best available data.’ ... The best available data is ‘accurate’ data in fact for purposes of payments under IPPS [inpatient prospective payment system].”).

Furthermore, the term “minimal,” as used by the Administrator, ultimately is just a subjective and conclusory label applied without reasoned analysis. Like the labels “minor” and “not major” used by the agency in Sierra Club v. Mainella, 459 F.Supp.2d 76 (D.D.C.2006), such an “unbounded term cannot suffice to support an agency’s decision because it provides no objective standard for determining what kind of differential makes one impact more or less significant than another.” 459 F.Supp.2d at 101; see also Tripoli Rocketry Ass’n, Inc. v. Bureau of Alcohol, Tobacco, Firearms, and Explosives, 437 F.3d 75, 81 (D.C.Cir.2006) (rejecting an agency’s characterization of a chemical reaction as “much faster” than another as an insufficient explanation “because it says nothing about what kind of differential makes one ... ‘much faster’ than another. Ten millimeters per second? A hundred? A thousand?”). Moreover, rejection of data based on a finding that the percentage change is “small” is particularly problematic in light of this Circuit’s recognition that even a modest percentage difference can be “substantial” given the enormity of the Medicare program. See County of Los Angeles, 192 F.3d at 1010.

Nor does the Administrator’s interest in “administrative finality” demonstrate that the second SSA tape following a fiscal year could not be used. The Board concluded that the cost report settlement process takes about two to three years after the end of a hospital’s cost reporting period — a factual finding that was left intact by the Administrator’s decision. Board Decision at 32. This finding is borne out for each of the provider years at issue in this case. See AR 497A (NPR dated September 29, 1995, for fiscal year ending September 30, 1993); AR 500 (NPR dated September 13, 1996 for fiscal year ending September 30, 1994); AR 502 (NPR dated September 25, 1997 for fiscal year ending September 30, 1995); AR 6472 (NPR dated September 29, 1998 for fiscal year ending September 30, 1996). Given this timing, then, for each of these fiscal years, a more recent SSA tape (along with an updated quarterly MEDPAR run showing matches of Medicare patients and SSI recipients) was available many months before the close of the cost report settlement period. The Administrator’s interest in “administrative finality” could just as easily have been served by setting the cut-off point for consideration of updated SSA data at the end of the quarter following the second SSA tape.

The Secretary contends that the Administrator has the discretion to choose the timing of the SSI fraction calculation, and that it would be unreasonable to require the agency to hold open indefinitely the determination of the SSI fraction simply to acquire further SSA data. See Def.’s Mem. at 37. But the prospect of waiting “indefinitely” is a strawman, neither raised by Baystate nor adopted by the Board, which instead rested its decision on the undisputed finding that updated SSA data were in fact available to CMS before the end of the DSH cost report settlement period for each fiscal year. In short, the Administrator’s rejection of the use of updated SSA tapes available within the DSH cost report settlement period is inconsistent with the “best available data” standard and, hence, arbitrary and capricious.

2. The Omission of Inactive SSI Records and Forced Pay Records

Two other types of records have categorically been omitted from SSA annual tapes and thus are not dependent on CMS’s use of the first rather than the second SSA tape: (1) stale records — that is, records that SSA terminated as “inactive” upon a beneficiary’s death — which were excluded from Baystate’s SSI fraction for 1993 and 1994; and (2) forced pay records — that is, records for recipients who receive manual SSI payments instead of automated ones. These omissions were either conceded by the Administrator or acknowledged by the Secretary before this Court. See Final Decision at 43 (“Up until approximately February 1996, SSA’s annual tapes did not include SSI records that had been ‘terminated’ and were inactive prior to ... transmission of the tapes to CMS”); Def.’s Stmt, of Material Facts As To Which There Is No Genuine Issue ¶ 38 (“individuals who received a forced payment during the period of their inpatient hospital stays were not shown as having been entitled to SSI for such periods on SSA annual tapes.”).

Baystate contends that use of SSA data excluding both inactive SSI records and forced pay records necessarily means that CMS did not use the best available data, and the Board so found. See Pl.’s Mem. at 14-15; Pl.’s Reply at 20-21 & n.22; Board Decision at 25-26, 35, 42 (finding that CMS did not use the best available data due to the omission of these SSI records). The Secretary responds that the Administrator reasonably concluded that the omission of these two types of records had only a “relatively small effect” on the number of matches and he thus requests affirmance of this aspect of the decision. See Def.’s Mem. at 32-33.

The Secretary once again skirts the question whether SSA tapes that categorically excluded these two types of data constituted the best available data. He does not contend that the tapes omitting stale records constitute the best available data — nor did the Administrator — but does suggest that a relevant consideration is CMS’s correction of the stale records problem when it became fully aware of the situation in February 1996. See Def.’s Mem. at 32. Indeed, a plausible argument could be made that, despite the exclusion of these kinds of SSI eligibility data from the SSA tapes, it was nonetheless reasonable for CMS to use the tapes because, although imperfect, it arguably was the best data available to CMS at the time the SSI fractions were calculated. See Mt. Diablo Hosp., 3 F.3d at 1233 (holding that agency’s use of Bureau of Labor and Statistics data to set wage index used in Medicare cost reimbursement was permissible because it was “the most reliable data available at the time” even though it did not reflect regional variations in part-time employment data).

Such a conclusion cannot be sustained on this record. CMS and SSA are two divisions of a single agency, the Department of Health and Human Services, with a close working relationship involving the transmission of SSI data between the two offices. This close working relationship defeats any suggestion that data held by one arm of the agency is not available to another part of the same agency. See In re Sealed Case No. 99-3096, 185 F.3d 887, 896 (D.C.Cir.1999) (considering whether there is a “close working relationship” between agencies to determine whether information is available to an agency in context of assessing its disclosure obligations in a criminal case); In re Agent Orange Product Liability Litig., 597 F.Supp. 740, 796 (E.D.N.Y.1984) (holding that information will be considered known to a government agency “if there is some relationship between [the two] agencies—either some reason for the agency without knowledge to seek the information or a reason for the knowledgeable agency to transmit the information”), aff'd, 818 F.2d 145 (2d Cir.1987). Here, the administrative record amply demonstrates a close working relationship involving the exchange of SSI eligibility data between CMS and SSA; indeed, SSA supplies SSI eligibility information to CMS on a regular basis, with modifications to the SSA tapes at CMS’s request. See Final Decision at 15; Board Decision at 10, 24. Thus, for example, when CMS asked SSA to include the “missing stale records” in the SSA annual tapes, SSA did so. See Final Decision at 44; Board Decision at 24; see also AR 2108 (Mem. from Anne Rudolph, HCFA, to SSA dated Feb. 29, 1996) (“please do not purge the stale records from any future files that you forward to us concerning the SSI data”).

Furthermore, substantial evidence in the record, including the testimony of CMS employee Daryl Rosenberg, supports the Board’s finding that, as early as 1993, CMS was aware that SSI days were dropping out of the numerator of the SSI fraction with each successive recalculation and that there was thus “something wrong with the numbers.” Board Decision at 24 n.120; see also AR 5582-83 (Rosenberg testimony stating that “we knew there was something wrong with the numbers” even before Baystate consultant David Pfeil called in 1993, and further that “the people that were in the hospitals [with numerators] that were going down were dying,” although CMS did not then know that SSA data was the problem). The Administrator left this finding undisturbed, and it is thus appropriate for this Court to consider it. See Morall, 412 F.3d at 179 (“the reviewing court must take the [tribunal’s] findings into account as part of the record”). In light of the intra-agency relationship between the two divisions, and the evidence in the record, there can be no doubt that the inactive and forced pay records were readily available to CMS. Moreover, an SSA tape incorporating the inactive and forced pay records would clearly be superior because it would yield a fuller set of SSI recipients. Thus, SSI eligibility data excluding the stale records and forced pay records cannot reasonably be characterized as the best available data.

The question then becomes whether the Administrator can forego the use of the best available data by finding that the number of patients omitted is too small to be of concern. This was the exclusive focus of the Administrator’s discussion of the stale records and forced pay records problems, rather than any articulation of a best available data theory. See Final Decision at 43-47. With respect to stale records, the Administrator estimated that “the average size hospital with the average size SSI population would have had about four stale records omitted from its calculation” in fiscal year 1993 and “about three stale records omitted” in fiscal year 1994, based on an examination of the results of the “special MEDPAR” project that CMS had conducted in 1996. See id. at 46. He then rejected this quantity as “not signifi cant.” Id. With respect to forced pay-cases, the Administrator again focused solely on the number of forced pay records omitted from the SSI fraction. Id. at 41. He reasoned that, out of the non-random sample of 627 records disproportionately likely to show this kind of error, the provider had shown only one omitted hospital stay involving a manual payment, which he equated to a 0.15 percent error rate. Id. This degree of error, too, was considered to have an insignificant impact that did not warrant further consideration. Id.

The Administrator’s approach is inherently flawed because the standard for accuracy, as discussed above, is intertwined with whether the best available data has been used. Although the Administrator acknowledged this in his threshold discussion of whether an absolute standard of accuracy applied under subparagraph (d)(5)(F)(vi), his consideration of the stale records problem entirely loses sight of that fundamental requirement. Instead of an approach that tolerates errors based on a “best available data” standard, the methodology becomes one of simply tolerating errors because they are deemed “not significant.” This is at odds with the Administrator’s stated requirement that the best available data must be used in order to justify payments under Medicare that are “not perfect” but nonetheless “reasonably accurate.” Id. at 23.

Nor does the ostensibly small size of the error rate, without more, warrant denial of retrospective relief. As the Court noted earlier, the unbounded standards used by the Administrator — whether phrased as “acceptable rate of error,” “minimal,” or, as here, “not significant” — “provide[] no objective standard for determining what kind of differential makes one impact more or less significant than another.” See Sierra Club, 459 F.Supp.2d at 101. In short, the record provides no basis for this Court to condone the Administrator’s silent acceptance of the incomplete SSA tapes as the best available data, nor can the Court discern a reasoned basis grounded in the record for concluding that the projected error rate is “not significant.”

3. Patient Identifiers Used in the Match Process

Baystate also contends that the data utilized in the SSI fraction cannot be considered the best available data because important “patient identifier” data was discarded or ignored. Instead of using each individual’s own social security number, CMS matched Medicare patient records using HICANs on the MEDPAR side and Title II numbers on the SSI side. Bays-tate argues that the Administrator improperly reversed the Board findings concerning three flaws inherent in a match process relying only on HICANs and Title II numbers — the variability that occasionally exists between HICANs and Title II numbers, the problem arising from multiple Title II numbers for one individual but only one Title II “field”, and CMS’s exclusion of SSI records without Title II numbers.

The Court begins with the most serious of the alleged flaws — whether SSI records were systematically excluded from the match process by the SSISORT program described by Mr. Dean whenever they lacked Title II numbers. This issue is of special significance because roughly half of the records on SSA’s annual tapes do not have Title II numbers. Board Decision at 20. The Secretary has mounted virtually no defense of the Administrator’s finding that records lacking Title II numbers are nonetheless included in the match process (Final Decision at 42), instead acknowledging that a remand on this issue may be necessary. See Def.’s Mem. at 35 n.20. Indeed, the Administrator’s finding cannot be sustained because he assumed, without any analysis, that all SSI records were included in the match process, without weighing any of the contrary testimony given by the CMS programmer, Mr. Dean, at an earlier evidentiary hearing. 28 See Final Decision at 42. Where a final agency decision fails to consider “relevant contradictory evidence, including evidence that led the [administrative tribunal hearing the evidence] to contrary findings of fact and credibility,” the decision cannot be sustained. Morall, 412 F.3d at 180.

It is not surprising, then, that the Secretary concedes that the record is “unclear” on the “extent to which CMS attempted to match the numeric portion of the HICANs with the SSNs sent by SSA” where no Title II number exists, and that a “remand would be required” if the Court were to reach this issue. See Def.’s Mem. at 35 n.20. But the Secretary suggests that the flaw is not only a deficiency in the Administrator’s explanation but a gap in the record itself, citing the absence of the computer program. Id. Indeed, the Board also expressed concern about the absence of evidence bearing on the issue of the completeness of the data produced by Dean’s SSISORT process. See Board Decision at 18 (noting that “a list of [Dean’s] available files” from the SSISORT process did not come to light until a few days before the evidentiary hearing and making no findings regarding the files’ contents). In light of these evidentiary gaps, and the Secretary’s concession, the Court agrees that a remand must include further eviden-tiary proceedings before the Board on whether SSI records without Title II numbers were excluded from the match process for the years at issue.

The Court next turns to Baystate’s allegation that the match process reliance on HICANs and Title II numbers, instead of social security numbers, will likely result in missed matches (1) where Medicare or SSI benefits are received pursuant to a current or former spouse’s number, or (2) multiple Title II numbers exist for one person. The Secretary acknowledges that such matches may, in theory, be missed. See Def.’s Mem. at 35 (“if an individual’s title II eligibility status or HICAN changed from one earnings record to another and the admission in question was under the later HICAN, some records could have been missed”); see also Board Decision at 20-21 (describing examples where changes in HICAN and Title II numbers may result in missed matches). Nonetheless, the Secretary insists that both the data and the process are adequate because the actual quantity of such misses is “uncertain,” and is shown in this case by a non-random sample to have occurred only once. See Def.’s Mem. at 35.

The flaw in this defense is that, like the earlier defenses of incomplete data, it bypasses the question of whether CMS is using the best available data, and goes straight to the anticipated small size of the error rate to justify ignoring available data. If social security numbers of individual patients are to be ignored, however, the Administrator must provide a reasoned explanation for rejecting that data in the face of the acknowledged errors. See PPL Wallingford Energy LLC v. FERC, 419 F.3d 1194, 1198 (D.C.Cir.2005) (“To survive review under the arbitrary and capricious standard, an agency must ‘examine the relevant data and articulate a satisfactory explanation for its action including a rational connection between the facts found and the choice made.’ ”) (quoting Motor Vehicle Mfrs. Ass’n, 463 U.S. at 43, 103 S.Ct. 2856). But that was not done. Instead, the Administrator only cross-referenced a “policy” decision made on an earlier occasion not to use social security numbers or other patient identifiers in the match process. Final Decision at 42. That policy decision, however, only states in conclusory terms that a social security number is not specific to an individual Medicare beneficiary or hospital patient. See id. at 42 (“social security numbers are used on a ‘wage earner’ basis that is not necessarily specific to an individual Medicare beneficiary (or hospital patient)”) (quoting 70 Fed.Reg. at 47440-41 (Aug. 12, 2005)). But the Court is unable to discern from that reference why a patient’s social security number would not be considered unique to the patient.

The Administrator also expressed concern that the MEDPAR database does not contain Medicare beneficiaries’ social security numbers, referring again to the prior “policy” decision. Id. at 43 (quoting 70 Fed.Reg. at 47440-31 as stating, “we do not have social security numbers for every Medicare beneficiary in the MEDPAR data”). If that were true, that would certainly bear on whether the SSNs could be considered “available data.” The Board, however, specifically found that CMS does possess such social security numbers in its Medicare Enrollment Database, based on a declaration of CMS’s own employee. See Board Decision at 22 (“Robyn Thomas is the Division Director for CMS’ Division of Information Distribution. She confirmed that CMS’ Medicare Enrollment Data Base (like SSA’s annual tape) contains Medicare beneficiaries’ own Social Security numbers, their names, and dates of birth.”); AR 598 (Thomas Declaration describing Enrollment Data Base). Thus, substantial evidence in the record demonstrates that beneficiary social security numbers are available to CMS.

Indeed, the weight of the evidence indicates that individual patient social security numbers are reliable data for the match process. The Administrator left undisturbed the Board’s finding that individual patient social security numbers are used to coordinate a similar match process with SSA for another purpose — identifying Medicare beneficiaries in nursing homes who receive SSI payments, in order to implement legally required reductions of SSI. Board Decision at 13, 22. The Board further found that alternative identifiers— name and gender — were used in that match process, and that such alternative identifiers were important to ensuring an accurate match. Id. This Circuit has held that, where an agency considers data reliable for one purpose but not another, the agency must explain its reasoning. See County of Los Angeles, 192 F.3d at 1022 (declining to defer to agency’s exercise of discretion where agency had “inadequately explained why the ... data were suitable for one significant calculation but unreliable for another”). That was not done by the Administrator here.

To summarize on the subject of patient identifiers, a remand is necessary to determine whether individuals without Title II numbers were excluded from the match process and also to allow the Administrator to articulate a reasoned explanation, if there is one, for why social security numbers and other patient identifiers (such as the patient’s name) are not the “best available data” that should be incorporated into the match process, considering that the Board has found that they are available to CMS and used in another matching program. 29

4. Other Considerations

It is worth underscoring the difference between the “best available data” standard and the unrealistic standard of perfection that the Secretary fears will be imposed if the use of a better set of data is required. The case law amply supports the proposition that the best available data standard leaves room for error, so long as more reliable data did not exist at the time of the agency decision. This is demonstrated in Methodist Hospital v. Shalala, where this Circuit held that HHS reasonably relied on a regional wage index despite its incorporation of erroneous understated wage data from a large hospital that then caused other hospitals to lose over $730,000 in reimbursements. 38 F.3d at 1228. The reimbursement amounts were upheld because they were based on “the most reliable data available to the Secretary at the time of publication” and thus were considered to “reflect the Secretary’s best approximation" of the regional wage index. Id. at 1230 (emphasis added). Similarly, in another wage index case, the Ninth Circuit held that the agency’s reli-anee on data that failed to account for part-time workers was permissible, despite the resulting underestimation of labor costs, because it was “the most reliable data available.” Mt. Diablo Hosp., 3 F.3d at 1229, 1233.

Recognizing that complex reimbursement determinations in other programs may involve imperfect or incomplete data, this Circuit has observed that an agency could “reasonably have accepted lacunae in data and methodology and even have forgiven ancillary analytic weak points,” so long as the agency employs “reasoned de-cisionmaking” that requires, among other things, an “ ‘adequate factual predicate’ ” for an agency decision and avoidance of “ ‘means that undercut [the agency’s] ends.’ ” 30 See City of Brookings Mun. Tel. Co. v. FCC, 822 F.2d 1153, 1168 (D.C.Cir. 1987) (quoting Aeron Marine Shipping Co. v. United States, 695 F.2d 567, 577-80 (D.C.Cir.1982), and Office of Communication of the United Church of Christ v. FCC, 779 F.2d 702, 707 (D.C.Cir.1985)). Here, ignoring the existence of more reliable data that is available before the DSH adjustment is finally determined — that is, within the cost report settlement period— simply cannot be reconciled with the standard of reasoned decisionmaking. 31 To be sure, the Secretary correctly notes that City of Brookings observed that “[w]e dare not overlook the realities and exigencies of regulatory life in an imperfect world and demand from our lofty perch that which could not reasonably be delivered.” 822 F.2d at 1168. But use of the best available data is firmly recognized by the case law and, indeed, by the Administrator to be essential to the standard of reasoned deci-sionmaking in Medicare reimbursement decisions. It is, as a practical matter, well within that which can “reasonably be delivered” by the agency.

C. Retrospective Relief

In a closely related argument, the Secretary contends that the Administrator acted within his discretion in deciding that Bays-tate is not entitled to retrospective relief, notwithstanding CMS’s failure to use the best available data, because the administrative burden of performing retroactive calculations for Baystate and all other Medicare providers is not warranted in light of the insignificant size of the errors. See Def.’s Mem. at 29-37. There are actually two levels at which the availability of retrospective relief must be assessed. The first is whether either the statute or the DSH regulation categorically prohibits recalculation of the SSI fraction after the NPR has been issued, as the Administrator determined. See Final Decision at 17 (“the regulation does not provide for a recalculation of the SSI ratio based upon updated or later data once it is completed by CMS”). The Court sees no basis for interpreting the statute or regulation to impose a per se rule prohibiting retrospective relief — a position that the Secretary has not defended in this litigation. 32 The Court will therefore consider whether, on this record, the Administrator acted within his discretion in rejecting retrospective relief.

As both parties aptly observe, retrospective relief is not available as a matter of right, but instead is within the agency’s discretion. In Methodist Hospital, the D.C. Circuit held that, where Congress is silent on the question of retrospective revisions to payments in the PPS system, the agency may deny retrospective relief— and instead limit the correction of errors prospectively — so long as it supplies a “reasonable rationale” for that choice. 38 F.3d at 1232. In assessing the reasonableness of that decision, the court weighed several factors: the agency’s articulated interest in determining with “finality” the factors affecting PPS payments; the “administrative burden” of retroactive corrections; and the “incentives for aceura[cy]” yielded by a “prospective only” policy. Id. at 1232-33. In this case, the primary considerations in support of the denial of retroactive relief fall into three categories: (1) the absence of a demonstrated impact on the provider’s SSI fraction, as well as the estimated “insignificant” size of any impact; (2) CMS’s interest in the finality of DSH payments; and (3) the administra- tive burden on CMS of performing recalculations for all providers generally. See Final Decision at 21-27, 36-47, 53-54.

Turning first to whether Baystate demonstrated an impact on its SSI fraction, the Court concludes that substantial evidence in the record shows that Baystate met its burden of producing such evidence. The Administrator acknowledged that Baystate had demonstrated omissions from the numerator for each category of errors at issue, but simply believed these numbers were too small to be considered significant. The Administrator observed that the provider had demonstrated, based on 627 individual-specific SSI eligibility records provided by CMS through the administrative discovery process, that 10 individuals (representing 12 hospital stays) were not credited with SSI days during the match process but were in fact eligible for SSI at the time of discharge. Final Decision at 36-37. Out of these ten cases, one represented a patient who received a retroactive SSI eligibility determination made after a “suspense” case was resolved; one represented a patient who received a manual payment; and one represented a patient missed because of the problem of using HICANs and Title II numbers as the patient identifiers. Id. at 39, 41, 43. The Administrator also estimated that three to four “stale records” would have been omitted each year for an average size hospital, but whether any of the records in Baystate’s sample actually represented omissions due to the exclusion of “stale records” could not be ascertained due to the absence of the original uncorrected SSI tapes for 1993 and 1994. See id. at 44 n.84, 46.

To be sure, the Administrator expressed reservations about the reliability of the numbers produced by the sample because the sample was flawed in some respects. Id. at 37. However, where an agency is in sole possession of the records necessary to prove a party’s claim, the agency may not reject the aggrieved party’s allegations as insufficiently proven unless the agency comes forward with “countervailing evidence or a reason, not based on the insufficiency of the [movant’s] showing, that explains why the .... allegations have not been accepted.” See Atlanta College of Med. and Dental Careers, Inc. v. Riley, 987 F.2d 821, 830-31 (D.C.Cir.1993). This is because “the burden of bringing forward evidence generally shifts when the defendant has greater access to information on a particular issue.” Id. at 831 (citing Geddes v. Benefits Review Bd., 735 F.2d 1412, 1418 (D.C.Cir. 1984)). The D.C. Circuit more recently confirmed this principle in Canadian Commercial Corp. v. Dep’t of the Air Force, 514 F.3d 37, 42 (D.C.Cir.2008), where it held in a reverse-FOIA case that the plaintiff could not be faulted for failing to produce certain evidence during administrative proceedings when only the agency had access to the relevant information, explaining that “the burden of production properly falls upon the party with access to the information to be produced.” Id.

Here, CMS and SSA were in sole possession of the SSI data necessary to determine the scope of the impact with any greater precision. Indeed, Baystate mounted diligent efforts to obtain those records from CMS and SSA, submitting requests for SSI records of all Medicare patients discharged from Baystate from 1993 to 1996 through the administrative discovery process and separately under the Freedom of Information Act. AR 1102-10; AR 1948-53; AR 1984-1987. Those requests were denied because of the privacy interests of the individuals, and Bays-tate ultimately was provided only the 627 SSI records that are presently in the record. See AR 2715-16. Baystate, like the aggrieved parties in Atlanta College, presented “specific allegations supported by all the information available to [it],” which, if undisputed, would prove an impact on its SSI fraction; the burden thus shifts to the agency to provide a justification “not based on the insufficiency of the [plaintiffs] showing” that explains why the allegations were not accepted. Atlanta College, 987 F.2d at 831.

The Administrator did not meet that burden here. Instead, his criticisms of Baystate’s data were centered on “the insufficiency of the [movant’s] showing”— that is, the reliability of the sample itself. See Final Decision at 37. He speculated that the omissions of SSI patients shown by the sample might not reflect actual errors, but instead might indicate SSI eligibility data updated after the DSH determinations were made. Id. He also found that the sample of patients used was likely to capture a disproportionate number of errors in Baystate’s favor because, in creating the sample, Baystate had only requested SSI records for individuals likely to be eligible for SSI and for whom the MEDPAR records indicated no SSI days associated with their stays. Id. at 37, 41, 43. But an agency cannot meet its burden of production under Atlanta College based on flaws in the data the aggrieved party relied upon, where the agency has sole possession of the records necessary to make a fuller demonstration. 987 F.2d at 831. Hence, any failure by Baystate to demonstrate more precisely the impact of the errors is not a permissible basis for denying retrospective relief.

Nor does the ostensibly insignificant size of the error rate, without more, warrant denial of retrospective relief. As the Court explained earlier, see swpra at 38-39, the unbounded standards used by the Administrator — conclusory labels such as “acceptable rate of error,” “minimal,” and “not significant” — provide no objective standard for assessing the financial impacts, particularly in light of the enormity of the Medicare program. Furthermore, the systemic and recurring nature of these errors casts serious doubt on whether the kinds of errors alleged can reasonably be considered “insignificant.”

The Administrator also relied on the importance of administrative finality and certainty in the prospective payment system to support his denial of retrospective relief. Final Decision at 21, 24-26. The “finality” interest purportedly provides “predictability” in the system by preventing unexpected shifts in payments based on later data, and precludes the Secretary from doing a later recalculation even when circumstances might favor a reduced payment to a provider. Id. at 21. However, the interest in administrative finality carries far less weight for DSH payments than other PPS payments. The cases in which courts have denied retrospective relief involved components of the PPS system that were truly prospective — that is, using historical data to project a standardized rate for a future period, thus providing an incentive to control costs. See, e.g., Methodist Hosp., 38 F.3d at 1227, 1232 (discussing the “link between prospectivity and efficiency,” and observing that “[t]o the extent that the Secretary’s prospectivity policy permits hospitals to rely with certainty on one additional element in the PPS calculation rate — the wage index value — the Secretary could reasonably conclude that it will promote efficient and realistic cost-saving targets”). In contrast, the DSH adjustment is entirely retrospective — that is, it is calculated for the past fiscal year, based on hospital-specific data from that earlier period, and paid as an add-on to the amount due to the hospital under the preset PPS rates. See In re Medicare Reimbursement Litig., 414 F.3d at 9 (contrasting Medicare’s “preset na tionally applicable rates” with the “hospital-specific” DSH adjustment). 33 Indeed, this Circuit previously has rejected the Secretary’s contention that interests in “finality and repose” would be greatly undermined by retrospective recalculations of DSH adjustments. Id. at 12-13.

The Court finally considers whether the administrative burden of retrospective correction supports denial of retrospective relief. There is, however, nothing in the administrative record to support the Administrator’s conclusion that the administrative burden is great, other than his conclusory statements. 34 Indeed, the entirety of the Administrator’s analysis on administrative burden is only that “the Board’s decision ignores the enormity of the Medicare Program in finding that the implementation of its order would be no administrative burden, despite its retrospective and wide reaching implications,” referring generally to other “cases pending before [the Board] on this issue.” Final Decision at 53-54. Stated more pointedly, the Administrator considered the relevant administrative burden to be the systemic burden of granting relief to all similarly situated providers — all Medicare providers receiving DSH payments— in contrast to the burden associated with granting retrospective relief to one litigant. See Def.’s Mem. at 31. However, the prospect of systemic recalculations, standing alone, does not present an undue burden warranting denial of retrospective relief. See In re Medicare Reimbursement Litig., 414 F.3d at 12. In that case, the Secretary invoked administrative burden to oppose petitions to reopen past DSH determinations for over 250 Medicare providers where errors in the other DSH fraction— the Medicaid fraction' — had occurred, claiming that doing the recalculations would be “very difficult” and “extraordinarily time-consuming to audit and verify.” Id. The D.C. Circuit rejected that argument, reasoning that the “burden [would] not outweigh the public’s substantial interest in the Secretary’s following the law,” id. — a rationale equally applicable here. Furthermore, the court rejected the proposition that payment of extraordinary sums of money — more than $1 billion — in hundreds of cases constituted an impermissible burden, for “[h]aving to pay a sum one owes can hardly amount to an equitable reason for not requiring payment.” Id. at 13. Thus, the fact that the relief would be systemic, rather than for only one provider, is not by itself a valid ground for denying relief. 35

Moreover, the denial of relief here would have the perverse effect of rewarding CMS for ignoring the best available data and also allowing CMS to continue that practice indefinitely. 36 This is inconsistent with the case law that has generally upheld denial of retrospective relief only where the best available data was used, and compelled retrospective relief where the agency ignored such data. See Methodist Hosp., 38 F.3d at 1230, 1234-35 (affirming denial of retrospective relief where agency had used “the most reliable data available to the Secretary at the time of publication”); County of Los Angeles, 192 F.3d at 1023 (remanding Medicare reimbursement determination to Secretary to retrospectively “recalculate outlier thresholds for [past] fiscal years ... or to offer a reasonable explanation for refusing to use” more current available data); Alvarado Community Hosp., 155 F.3d at 1125-26 (remanding Medicare reimbursement determination to Secretary to “recompute” outlier thresholds for past fiscal years where the “most reliable data” had not been used). Methodist Hospital makes clear that “incentives for aceurac[y]” matter when considering whether the decision to grant or deny retrospective relief is reasonable, and the use of the best available data is obviously fundamental to accuracy. See 38 F.3d at 1233 (observing that denial of retrospective relief encouraged providers to submit carefully prepared data in the first instance, knowing that a subsequent correction would have only prospective effect). Denial of retrospective relief in this case would provide CMS an incentive to continue the status quo because the best available data would be unnecessary whenever the percentage changes to an SSI fraction were, in its view, not significant. For this and the foregoing reasons, the Court concludes that the Administrator’s decision to deny retrospective relief was arbitrary and capricious.

III. The Standard Governing the Calculation of Medicare Part A Patient Days

Although most of the issues raised by Baystate concern the SSI data (i.e., the “numerator” of the SSI fraction) and the integrity of the match process, Baystate also raises a handful of issues that focus on the “denominator” — that is, the standard for counting the number of patient days for patients “entitled to benefits under part A.” See 42 U.S.C. § 1395ww(d)(5)(F)(vi)(I). The Administrator held, like the Board, that in determining which patients are “entitled to benefits under part A of [Medicare],” the calculation includes all “Medicare covered/utilized days” in contrast to “paid days only.” Final Decision at 28, 31-35. Baystate contends that the Administrator should have excluded days for which a provider does not receive payment — primarily, Medicare HMO days. 37 Although the parties refer to this issue as the “denominator” issue, that label is a misnomer because it ignores the inclusion of Medicare Part A entitlement as one of the factors in calculating the numerator of the fraction:

Patient days of those patients entitled to benefits under both Medicare Part A and SSI Patient days of those patients entitled to benefits under Medicare Part A

See 42 U.S.C. § 1395ww(d)(5)(F)(vi)(I); 51 Fed.Reg. at 16777; Pl.’s Demonstrative Exhibit 1. That interplay is significant because Baystate acknowledged at the motions hearing that, even if it prevails on the standard governing the count of Medicare Part A patient days, the overall SSI fraction is unlikely to change. Based in part on that acknowledgment, the Court concludes that Baystate is not entitled to raise this issue on review before this Court.

Application of Baystate’s less inclusive definition of entitlement to Medicare seems, at first glance, to be against a provider’s interest, since a less inclusive definition would also potentially decrease the size of the numerator. However, Baystate’s counsel explained at the motions hearing why the provider actually benefits from the less inclusive definition. The “operative effect” of applying the Administrator’s definition of Medicare days is not upon the SSI fraction, but instead upon the Medicaid fraction because, in that other fraction, any patient “entitled to benefits under part A” is excluded from the numerator, while being retained in the Medicaid fraction denominator. Mot. Hr’g Tr. at 95-96. In contrast, “no change” would be expected in the SSI fraction, when one posits hypothetical numbers that are representative of Baystate’s patient days during the years at issue. Id. at 96. Counsel explained this point using the HMO days as an example:

If we assumed 100 Medicare HMO patient days for any given year, eight of those are days for which the patient was entitled to SSI .... [and] 30, if I have a hundred, are likely to be eligible for Medicaid.... And those numbers are fairly consistent with the reality for these years. Baystate’s Medicaid fraction for these four years [1993-1996] was between 29 and 30 percent of total days actually. And ... its SSI percentage as calculated by the Secretary was between 7 and 9 percent.

Look at the math here. If the Secretary prevails on the argument being made here, that these individuals who are enrolled in an HMO and with respect to whom Medicare Part A payment cannot be made under any scenario, if those days are put into the SSI fraction, ... it’s no change.

Id. at 95-96 (emphasis added). Counsel then reiterated that point: “If I’m starting with an 8 percent ratio of Medicare patients who are SSI and if I add a hundred more HMO [days] and eight of them are [SSI] — no change.” 38 Id. The government agreed that one should reasonably expect the numerator and denominator to be affected proportionately, so the application of a “paid days” standard would not be expected to make a difference in the SSI fraction. Id. at 53-54. 39

Baystate’s counsel also made a persuasive case that, in contrast, the Medicaid fraction would be significantly affected. He explained that

the key is if the Secretary prevails on this argument ... that these HMO patients are entitled to benefits under Medicare Part A, I have all these days in the denominator of the Medicaid fraction, but zero of those 30 in the numerator. And that’s really the operative effect.

Id. at 96. (emphasis added). The impact, then, under the Secretary’s position, would be to decrease the Medicaid fraction; conversely, Baystate’s position would increase the fraction.

The hurdle for Baystate is that the Medicaid fraction is not the subject of this case — only the SSI fraction is at issue. 40 But to obtain review regarding the SSI fraction, Baystate must allege that the error at issue had an impact on the “amount of payment,” pursuant to the administrative and judicial review provisions of the Medicare Act. See § 1395oo(f)(1). This provision states that a provider may obtain Board review of payments computed under section 1395ww(d)—including DSH payments — if the provider “is dissatisfied with a final determination of the Secretary as to the amount of the payment....” 42 U.S.C. § 1395oo(a)(1)(A)(ii) (emphasis added). This limitation on the Board’s review carries over, of necessity, into any subsequent review of a Board decision by the Adminis trator and then by a district court. See § 1395oo(f)(1).

Baystate, however, has failed to allege in this Court—in either its complaint or its briefs — that the Secretary’s “covered days” standard deflates the SSI fraction and hence the “amount of the payment” that is subject to challenge here. See Compl. ¶¶ 358(a)-(Z) and 359(a)-(g) (alleging flaws in the “covered days” standard without reference to impact on total SSI fraction); Pl.’s Mem. at 34-45 (same); Pl.’s Statement of Material Facts Not In Dispute ¶¶ 191-217 (same); Pl.’s Reply at 35-43 (same). 41 Most significantly, Baystate made the opposite point at the motions hearing — “no change” is expected in the SSI fraction because the inclusion of allegedly improper Medicare HMO days would be expected to have a proportionate impact on the numerator and denominator.

In short, the Court finds no basis in section 1395oo to resolve the issue of who is “entitled to benefits under part A” under section 1395ww(d)(5)(F)(vi) because Baystate’s appeal is limited to the SSI fraction, and Baystate acknowledges that resolution of this issue is not expected to affect the SSI fraction. This approach is consistent with the general principle, developed in administrative law cases concerning ripeness and exhaustion, that Article III courts should avoid “unnecessary adjudication” and limit the decision of issues to a “concrete setting.” See John Doe, Inc. v. Drug Enforcement Admin., 484 F.3d 561, 567 (D.C.Cir.2007); Friends of Keeseville, Inc. v. Federal Energy Regulatory Comm’n, 859 F.2d 230, 235-36 (D.C.Cir.1988). To decide the issue of who is “entitled to benefits under part A” because of its impact on Baystate’s Medicaid fraction — which is not at issue in this case — would be to decide the issue wholly out of context, in the absence of an appropriate factual and legal setting. 42 Such an approach would be at odds with Congress’s limitation on review of DSH payments in section 1395oo and with sound principles governing administrative law and justicia-bility.

IV. Summary and Remedy

A brief summary of the Court’s holdings may be helpful. The Administrator acted in accordance with law in holding that CMS properly excluded from the SSI fraction patients who had lost their entitlement to SSI payments notwithstanding their special status under Section 1619(b) of the Social Security Act for other purposes. In contrast, the Administrator’s finding that CMS relied on the best available data in determining the patients entitled to SSI benefits is arbitrary and eapri- cious because CMS failed to use superior data readily available to it, including (1) an updated SSA tape available before the end of the DSH cost report settlement period that would have reflected retroactive SSI eligibility determinations; (2) forced pay SSI records; and (3) inactive SSI records — the “stale records” — omitted from the SSI fractions for 1993 and 1994. With respect to the patient identifiers used in the matching process, a remand is necessary to determine whether individuals without Title II numbers were excluded from the match process and also to allow the Administrator to provide any further explanation of why social security numbers and other patient identifiers (such as the patient’s name) are not considered the “best available data” for use in the match process. Moreover, the Administrator abused his discretion in concluding that retrospective relief should not be granted in this case in those respects. Finally, the issue raised by Baystate concerning the standard for counting days attributable to patients “entitled to benefits under part A” of the Medicare program (Compl. ¶¶ 357, 358(a)-(1), 359(a)-(g), 360(b)(iv)) will be dismissed based on the limitations on judicial review set by section 1395oo.

What remains, then, is the matter of the appropriate remedy in light of these decisions. Baystate has submitted a broad and open-ended 11-paragraph proposed order that would direct the agency to take numerous actions on remand, for example, a requirement to “work with the [SSA] to develop, test, validate, and document appropriate data specifications and systems requirements” for all of the types of data at issue, to “develop and maintain accurate records,” and to “work with the plaintiff hospital to resolve potential errors identified by plaintiff.” See Proposed Order at 1-3. The Court is not empowered, however, to enter this type of relief in an APA case. As this Circuit has explained, in such a case the district court “sits as an appellate tribunal.” County of Los Angeles, 192 F.3d at 1011. “ ‘[U]nder settled principles of administrative law, when a court reviewing agency action determines that an agency made an error of law, the court’s inquiry is at an end: the case must be remanded to the agency for further action consistent with the corrected legal standards.’ ” Id. (quoting PPG Indus., Inc. v. United States, 52 F.3d 363, 365 (D.C.Cir.1995)). A district court may not “retain jurisdiction to devise a specific remedy for the Secretary to follow.” Id. Accordingly, the Court will enter an order remanding this case to the Secretary for further action consistent with this Memorandum Opinion.

CONCLUSION

For the foregoing reasons, the Court will grant in part and deny in part the cross-motions for summary judgment, and remand this matter to the Secretary for further proceedings consistent with this opinion. In light of this disposition, the Court anticipates that the parties will not pursue further litigation on Counts Two and Three, pertaining to disclosure of SSI eligibility records from the Social Security Administration, but Baystate will be given until April 14, 2008, to file a status report concerning those counts. A separate order will be issued herewith.

1 . For ease of reference, the Court will refer to Baystate's memorandum in support of its motion for summary judgment as "Pl.’s Mem.,'' and defendant’s memorandum in support of its cross-motion for summary judgment as "Def.’s Mem.” The Secretary’s decision, issued by the Administrator of CMS, will be cited as the "Final Decision,” and the Provider Reimbursement Review Board decision will be cited as the "Board Decision.” All other citations to the administrative record will be indicated by “AR,” followed by the record page number.

2 . The Court understands that this is, in effect, the "lead” case, and may impact the pending reimbursement claims of hundreds of hospitals for perhaps hundreds of millions of dollars.

3 . Thus, CMS has expressed this fraction as:

Patient days of those patients entitled to both Medicare Part A and SSI (excluding those patients receiving State supplementation only)

Patient days of those patients entitled to Medicare Part A

51 Fed.Reg. at 16777.

4 . All citations to the Code of Federal Regulations are to the version in effect from 1993 to 1996, the fiscal years at issue.

5 . A "cost reporting period” methodology was disfavored because it requires a monthly, rather than a yearly, match. 51 Fed.Reg. at 16777.

6 . The MEDPAR file and the PATBILL file are technically two separate files, with the MED-PAR file using the same data as the PATBILL file "but in a simplified, reformatted record layout.” 52 Fed.Reg. at 33144. CMS uses the two names interchangeably, but technically it actually uses the MEDPAR file. Id.

7 .The SSI fractions for those years is based on the following count of patient days in the numerator and the denominator:

Medicare _SSI Days Covered Days SSI Ratios

1993 4746_65,780_0.07215

1994 5274_64,251_0.08208

1995 5569_63,024_0.08836

1996 5324_57,439_0.09269

AR 5023 (Parties' Stipulations before the Board). The parties have used the term “SSI days” as shorthand for the days attributable to patients entitled to benefits under both Medicare Part A and SSI, and "Medicare covered days” and “Medicare days” as shorthand for the days attributable to patients entitled to benefits under Medicare Part A.

8 . Baystate’s DSH payments for 1993, 1994, 1995, and 1996 were, respectively, $12,098,733; $12,723,254; $14,741,383; and $17,069,231. AR 4820.

9 . The intermediary was represented by the Office of General Counsel at the Department of Health and Human Services. Board Decision at 4.

10 . The tape thus includes a projection of SSI eligibility data for the three months after the March transmission, instead of actual data. Board Decision at 10. That projected data is not relevant to this appeal.

11 . During the cost report settlement period, the provider collects data from the State for the other component of the DSH calculation, the Medicaid fraction. See Board Decision at 32.

12 .Hold and suspense cases occur, for example, when the SSA is looking for a representative payee able and willing to accept checks on behalf of an SSI recipient; when "presumptively disabled” individuals receiving benefits during an initial period are awaiting additional state determinations; or when a state eligibility determination is pending. Board Decision at 27-28.

13 . The record indicates that persons who received manual payments, instead of automated ones, also may have been excluded as “inactive." Board Decision at 23. However, the Board also found that the system of recordkeeping for manual payments independently excludes such persons from the match process, and thus considered the issue of manual payments as a separate issue. Id. at 26-27.

14 . As one might expect, the fundamental requirement in identifying Medicare patients who also are entitled to SSI benefits is that the patient identifier in the SSA file must match the patient identifier in the MEDPAR file. Board Decision at 12.

15 . The parties stipulated that MEDPAR also contains the hospital's Medicare provider number, the dates of admission to and discharge from the hospital; the total length of the inpatient hospital stay; the number of days in the stay that were covered under Medicare Part A; and the number of days in the stay for which the patient was determined, through the match process, to be entitled to federal SSI benefits. Board Decision at 11-12. CMS also has social security numbers for Medicare beneficiaries m its Medicare Enrollment Database. Id. at 22.

16 . An unspoken assumption in the record is that the match process relies on Title II numbers because eligibility for benefits under Title II is the main trigger for Medicare eligibility. See 42 U.S.C. §§ 1395c, 1395i-2, 1395i-2a.

17 . The parties stipulated that the SSA tapes contain the following additional fields for each SSI recipient on the tape: truncated last name and first initial; social security number; date of birth; gender; and 42 monthly indicators (ones and zeros) denoting the payment or non-payment of federal SSI cash benefits during the period covered by the SSA tapes. Board Decision at 10-11.

18 . The Board used the following example to illustrate: " ‘Jane,’ whose Social Security Number is 000-00-0000, initially receives Title II and Medicare under a deceased spouse's account using the decedent’s Social Security number followed by a 'B6' [BIC] (i.e., 111-11-1111B6). Jane subsequently remarries and receives Title II and Medicare on her second spouse’s account, using the second spouse’s Social Security number and a ‘Bl’ [BIC] (i.e., 222-22-2222B1). If Jane is hospitalized after the change in her HICAN number, then the HICAN for the period of the hospitalization would be the second number (222-22-2222B1), which would not match either the first, oldest, Title II number that SSA’s program would pick up [from the SSI master record] and include on its annual tape (111—11—1111B6) or her own social security number with an 'A' ... tagged on at the end (000-00-000A).” Board Decision at 21.

19 . This issue arose in the context of Bays-tate’s discovery of discrepancies in the count of Medicare patient days in the MEDPAR database and a separate database known as the PS & R — the Part A Provider Statistical & Reimbursement Report — which, as its title suggests, is limited to days that the provider actually receives reimbursement from Medicare. Board Decision at 35-36. Sixty-one stays that were indicated in the PS & R were not on MEDPAR, and forty-one stays that were on MEDPAR were not on the PS & R. Id.. To the extent that a pattern or explanation could be discerned, the discrepancy primarily reflected the fact that MEDPAR includes all days "covered” by Medicare, whereas PS & R is limited to "paid days.” See id. at 36-37.

20 . The Administrator also found that Baystate was precluded from challenging the use of the "June updates” to MEDPAR (i.e., using the first SSA tape following a fiscal year) because CMS had provided notice of this practice in various Federal Register preambles. Final Decision at 27. However, the Secretary has not argued waiver in response to Baystate's action in this Court, and based on a review of the Federal Register notices, the Court concludes that the issue does not merit raising sua sponte.

21 . Section 1395ww(d)(5)(F)(vi) provides that the SSI fraction shall be based on "the number” of days attributable to certain types of patients, a term that Baystate has relied upon as setting the level of precision.

22 . The Administrator also criticized the Board for failing to weigh the credibility of Baystate’s consultants, Mr. Pfeil and Mr. Smith, in light of their financial interest in the outcome of the proceedings, namely, a 35 percent contingency fee arrangement. Final Decision at 50-52. He also questioned the credibility of a former SSA employee, Mr. Shafer, who had been retained by Baystate but whose compensation agreement was not available in the record. Id. at 51. The decision does not identify which aspects of the testimony the Administrator believes should be discredited, although suggesting that Pfeil and Smith had an incentive to exaggerate the financial impact of any data omissions. Id. at 50. However, the exact size of the impact has not been germane to the Court’s resolution of this matter, as discussed in more detail below. Furthermore, as indicated by the Administrator’s decision, the existence of most of the types of data omissions at issue has gone largely undisputed. Thus, the Court finds it unnecessary to delve into this aspect of the record further.

23 . Baystate acknowledges that the familiar APA standard of review applies, but also suggests that a heightened standard of review should be applied based on the Secretary's alleged role as a "fiduciary" because the agency alone calculates the SSI fraction without the participation of providers. See Pl.'s Mem. at 3, 9, 29; Pl.’s Reply at 11-12. As defendant points out, however, a fiduciary responsibility can only be created by Congress. See Cobell v. Kempthorne, 455 F.3d 301, 307 (D.C.Cir.2006) (“the trust relationship arises out of statutes,” although the "precise contours” of the obligations are fleshed out in common law); Cobell v. Norton, 240 F.3d 1081, 1098 (D.C.Cir.2001) (“the government’s fiduciary responsibilities necessarily depend on the substantive laws creating those obligations”) (citation and internal quotation marks omitted); North Slope Borough v. Andrus, 642 F.2d 589, 611 (D.C.Cir.1980) ("A trust responsibility can only arise from a statute, treaty, or executive order.”) (citation omitted). Baystate has provided no statutory basis or case law supporting the existence of a trust relationship in the Medicare context between the Secretary and hospitals, nor can the Court discern one.

24 . Ordinarily, an individual receiving SSI payments also thereby qualifies for Medicaid assistance. 42 U.S.C. § 1396a(a)(10)(A)(i)(II). But when monthly earnings exceed the designated amount, the individual loses his or her eligibility to be paid SSI benefits, and hence also loses Medicaid assistance. Id.', see also § 1383(j).

25 . For example, the 1994 preamble cited by Baystate explains that Congress intended "to provide for easy transition ... between the various categories of benefits,” and then refers to SSA revisions to regulations that facilitate easier movement between different “benefit statuses,” including Section 1619(b). See 59 Fed.Reg. at 41402. But one cannot discern that the SSA, through that reference, deemed Section 1619(b) a benefit for any purpose other than Medicaid, especially when by the very same preamble it promulgated § 416.264, which makes clear that "special SSI eligibility status applies for purposes of establishing or maintaining your eligibility for Medicaid.” 59 Fed.Reg. at 41404 (emphasis added).

The reference to Section 1619(b) in the more recent 2005 preamble is even less helpful to Baystate’s position. See 70 Fed.Reg. at 57133. The SSA there discussed new rules governing expedited reinstatement of SSI monetary benefits that were terminated because of work activity, and explained when SSI benefits would be payable to a "reinstated” beneficiary — "when, using normal calculation procedures in subpart D of part 416, we determine you are due a monthly payment, or you are considered to be receiving SSI benefits in a month under section 1619(b).” Id. The phrase "considered to be receiving SSI benefits” simply reflects the language of Section 1619(b), and by inference, its opening limiting clause — "for purposes of subchapter XIX [Medicaid].”

Nor does the Federal Register notice concerning the regulations governing payment of vocational rehabilitation ("VR”) services support Baystate’s position. See 61 Fed.Reg. at 31024. The SSA there created a regulatory term, "disability or blindness benefits,” that included those with Section 1619(b) status. Id. However, the SSA emphasized that this term was "defined for the SSI VR payment regulations only.” Id. (emphasis added). That kind of narrow limitation discounts the notion that Section 1619(b) status should more broadly be recognized as an "[SSI] benefit ... under subchapter XVI” that would impact the Medicare program.

26 . The Auburn plaintiffs participating as ami-ci contend that Congress has set a "materiality threshold”—that is, an "acceptable rate of error” — for requiring retroactive corrections at $10,000, by providing for administrative and judicial review when the amount in controversy is $10,000 or greater. See Auburn's Mem. at 5-6 (citing 42 U.S.C. § 1395oo(a)(1)(A)(ii) and (a)(2)). As the Secretary correctly points out, the amount-in-controversy provision is nothing more than a jurisdictional provision, comparable to the $75,000 amount-in-controversy provision applicable to diversity cases under 28 U.S.C. § 1332. It provides no instruction on when retroactive relief must be granted, and as discussed below, the precedent in this and other circuits reveals that the matter of retroactive relief does not depend simply on the amount of money at stake.

27 . To the extent the Secretary contends that only a "reasonable approximation” or estimate of the patient days count used to compute the SSI fraction is required — without regard to whether the "best available data” was used — the Court rejects that proposition. See Def.’s Mem. at 31, 37. Section 1395ww uses the term "estimate” more than 40 times, which shows that Congress "knew how to enshrine estimates into the rate calculations when it so desired.” Georgetown Univ. Hosp. v. Bowen, 862 F.2d 323, 327 (D.C.Cir.1988). The absence of the term "estimate” from sub-paragraph (d)(5)(F)(vi) necessarily means, then, that errors in calculations cannot be excused solely on the ground that only an “approximation” or "estimate” is required.

28 . For example, the Board observed that Mr. Dean testified if "there is no Title II number, which is all we are, you know, on our end, that is all we are really concerned about ... this process [SSISORT] would not write that record out. So I wouldn't keep that record.” Board Decision at 16 (quoting Dean testimony at AR 5737) (emphasis added). It also noted his testimony that all other data ele-merits on SSA's annual tapes, including the social security numbers, are "useless.” Id. at 16 (quoting Dean testimony at AR 5739). The Board then reviewed Mr. Dean's subsequent contradictory testimony and an array of other evidence before finally concluding that the match process "fails to match SSI eligible beneficiaries who do not receive Title II numbers.” Id. at 17-18, 42.

29 .The Administrator also expressed a concern about potential arbitrary variations in the SSI fraction if other patient identifiers were used, stating that "we do not agree that individual hospitals should be given the choice to run the SSI/MEDPAR data match by alternative criteria. Such variations between providers would result in an inconsistent matching methodology and inconsistent DSH Medicare [SSI] fraction calculations, among providers.” Final Decision at 43. This concern about variability among providers makes no sense, however, for the record is clear that CMS is solely responsible for determining the SSI fraction, rather than providers calculating their own SSI fractions. See 51 Fed.Reg. at 31459; 42 C.F.R. § 412.106(b).

30 . Baystate also has challenged CMS’s overall management of information as "unsound” and suffering from "lack of quality control,” based on CMS's alleged failure to follow federal standards on information management. See Pl.'s Mem. at 26-28. The Administrator made no findings on this issue, nor did the Board. Indeed, the Board stated that CMS’s compliance with the Federal Information Resources Management Regulations and an OMB Circular on validating and testing computer software was "beyond its ability.” Board Decision at 30-31. Thus, there is no administrative decision for the Court to review on this point.

31 . Of course, a court cannot require an agency to develop data that does not presently exist, but that is not the case here. See Southwest Ctr. for Biological Diversity v. Babbitt, 215 F.3d 58, 61 (D.C.Cir.2000) (holding that, where agency was under a duty to make statutory determinations “on the basis of the best ... data available,” the district court erred in imposing "an obligation upon the [agency] to find better data”).

32 . The regulation cited as precluding retroactive corrections, 42 C.F.R. § 412.106(b), only seis forth the method for computing the disproportionate patient percentage, including the SSI fraction. The reference to two different time periods that may be covered — the federal fiscal year or, upon the provider’s request, the provider's cost reporting period— in no manner addresses what limitations, if any, exist on retrospective relief once a problem with the data is revealed. Id.

33 . Nor does the interest in finality offer the "predictability” to providers that the Secretary claims. The Court takes judicial notice of a separate decision of the Administrator claiming the authority to recalculate a DSH adjustment where CMS inadvertently uses the wrong file to calculate the disproportionate patient percentage. See St. Mary's Mercy Med. Ctr. v. Blue Cross/Blue Shield Ass’n/United States Gov’t Servs., LLC, Decision of the Administrator (Aug. 24, 2007) (attachment to Auburn's Mem.) (acknowledging that "CMS recalculated the Provider’s DPP [disproportionate patient percentage] because it inadvertently used the wrong file to calculate the initial DPP,” but asserting such recalculation was not inconsistent with its “policy” of barring recalculating SSI fractions), on appeal sub nom. Trinity Health-Michigan v. Leavitt, Civil Action No. 07-2318 (D.D.C. filed 12/26/07).

34 . The only factual assessment of administrative burden was done by the Board, and the Administrator's decision sets forth no grounds for discrediting that assessment. The Board’s findings include the following: (1) the necessary recalculations could be performed using "current data files” — that is, SSA tapes and MEDPAR files within CMS's possession, and social security numbers from CMS’s Medicare Enrollment Data Base; (2) changes to the computer program to capture the data points at issue would be "routine, simple, and not time consuming”; and (3) the feasibility of conducting a match process using additional patient identifiers, including social security numbers, is demonstrated by a comparable match process done for nursing home patients. Board Decision at 22-23 (citing testimony of Shafer, Walsh, and Dean at AR 5207, 5603, and 5759-61, respectively). To be sure, the task of having the computer programs redo the match process systemwide — that is, for potentially hundreds of Medicare providers — is undoubtedly no small task. But the record indicates that running the computer matching program for all Medicare providers on the relevant databases is already a routine process that is simple enough to do on a quarterly basis. See Def.’s Response to Pl.'s Statement of Material Facts ¶ 9 (acknowledging that "[t]he Secretary ... performs a quarterly match of inpatient hospital stay records with SSI data.”).

35 . The Administrator’s comparison of the systemic burden of granting relief — in his view, “wide reaching”—to the potential harm suffered by only one provider (Baystate) defies the common-sense proposition pointed out by Baystate that systemic burdens should be compared to the systemic impact of the errors. See Pl.’s Reply Mem. at 31. Baystate estimates that the "stale records” problem alone resulted in a systemic impact of $81.3 million for 1993 and 1994, based on the Administrator's finding that three to four hospital stays were omitted per hospital for each of those years. Id. Baystate further estimates that even an error rate as low as 0.15 percent—the rate estimated for missed manual/forced pay SSI days—represents a nationwide impact of $80 to 140 million per year. Id. at 28 n.27. Whether or not these particular dollar figures are correct (see Def.'s Response ¶ 146), they indicate that the systemic monetary impacts of the agency's errors are potentially enormous.

36 . This is not a case where the Administrator has even granted prospective relief for errors (other than the stale records problem). Instead, the Secretary’s broad pronouncements on the insignificance of the omitted SSI records and social security numbers indicate that, except for the stale records, the omissions will continue prospectively. This is 180 degrees opposite the situation in Methodist Hospital, where the Secretary had corrected the error at issue for future reimbursements. 38 F.3d at 1234-35

37 . Baystate also raised the issue of whether "exhausted benefit days” and "Medicare secondary payor” days should be counted as days for which a patient is “entitled to benefits under part A.” Pl.'s Mem. at 35-40; see also 42 U.S.C. § 1395d(b) (addressing scope of Medicare coverage for exhausted benefit days); id. § 1395y(b)(2) (addressing scope of Medicare coverage for Medicare secondary payer days). The Secretary agrees that exhausted benefit days and Medicare secondary payor days not paid under Medicare are properly excluded for the years in question (Def.’s Mem. at 24), and thus there is no controversy over these two categories. See also PL’s Reply Mem. at 35.

38 . This point can be seen mathematically by looking at just one of Baystate's years under review. In 1994, CMS determined that Bays-tate had 5,274 dual-eligible days in its numerator, and 64,251 Medicare Part A "covered days” in the denominator, which resulted in an SSI fraction of 8.2084 percent. AR 5023. If one adds 8 days to the numerator and 100 days to the denominator, the SSI fraction remains virtually unchanged, at 8.2081 percent.

39 . Addressing this issue, the Secretary’s counsel stated: "Covered days versus paid days ... it should not matter in a big picture kind of way which one you choose. The reason is these days go in both the numerator and the denominator or they don't go in both the numerator and the denominator.... [A]s long as you always count covered days or always count paid days it’s hard to know how that would make a difference.” Mot. Hr’g Tr. at 53-54.

40 .In the proceedings before the Board, Bays-tate stipulated that only the SSI fraction was being challenged (AR 194, 5022), and the complaint filed in this action challenging the Administrator’s decision also is limited to judicial review of the SSI fraction. See Compl. ¶¶ 346-60 (Count One).

41 . Baystate argued to the Board that specific discrepancies between the MEDPAR database and the PS & R database (preferred by Bays-tate) may have deflated the SSI fraction (AR 272-74; Board Decision at 41), but Baystate has not pressed that argument here. Moreover, that is different from explaining how categorical application of a “covered days” standard' — the issue it does press — would deflate its SSI fraction.

42 . The Court observes that the Secretary’s interpretation of who is "entitled to benefits under Part A” draws strong support from the plain language of the statute and the current record. The preamble to the 1986 final rule governing DSH payments and also a 1990 preamble on Medicare HMO days indicate that the Secretary has consistently applied this standard based on a patient's entitlement under Part A to Medicare benefits, without regard to whether the provider actually receives a payment. See 51 Fed.Reg. at 31460-61; 55 Fed.Reg. 35990, 35994 (Sept. 4, 1990). But the Court finds it inappropriate to draw any firm conclusions in the absence of more information about what the agency has done in the context of determining providers’ Medicaid fractions inasmuch as that is where the impact of the Secretary’s interpretation will be felt.