Ariz. Health Care Cost Containment Sys. v. Ctrs. for Medicare & Medicaid Servs.

Full title: Arizona Health Care Cost Containment System, Plaintiff, v. Centers For…

Court: United States District Court, District of Arizona

Date published: Jul 20, 2023

No. CV-21-00952-PHX-DWL (D. Ariz. Jul. 20, 2023)

Opinion

CV-21-00952-PHX-DWL

07-20-2023

Arizona Health Care Cost Containment System, Plaintiff, v. Centers For Medicare and Medicaid Services, Defendant.

Dominic W. Lanza United States District Judge

ORDER

Dominic W. Lanza United States District Judge

This lawsuit arises from a long-running dispute between a state agency, the Arizona Health Care Cost Containment System (“AHCCCS”), and a federal agency, the Centers for Medicare and Medicaid Services (“CMS”), over Medicaid funding. In 2018, after years of negotiations, CMS concluded that a previous award to AHCCCS of about $124 million in such funding should be reduced by about $20 million. CMS calculated this $20 million disallowance figure by using various statistical sampling methods that are discussed in more detail below. AHCCCS appealed the disallowance decision to an administrative agency, the Health and Human Services Departmental Appeals Board (“DAB”), arguing, inter alia, that CMS should have utilized different statistical sampling methods that would have resulted in a disallowance of only about $ 12 million. In December 2019, DAB issued a final decision upholding CMS's decision and approach.

The merits of this lawsuit concern AHCCCS's request for judicial review of DAB's decision. Such review is made available by 42 U.S.C. § 1316(e)(2)(C), which provides that a dissatisfied party may seek judicial review of a DAB decision by filing an action in federal district court within 60 days of the decision's issuance. Such review is deferential and governed by the “arbitrary and capricious” standard that generally applies to challenges to agency action. Here, the parties have presented their merits-based arguments by way of an opening brief (Doc. 39), response brief (Doc. 40), and reply (Doc. 41).

This case also presents a wrinkle that has the potential to eliminate AHCCCS's ability to seek any merits-based review of DAB's decision. Due to an apparent email failure, AHCCCS's counsel did not become aware of DAB's decision until May 2021, which was long after the 60-day statutory deadline for seeking judicial review had expired. During earlier stages of the case, CMS moved to dismiss on this basis, but the Court declined to order outright dismissal on the ground that “AHCCCS may be entitled to equitable tolling . . . which would excuse its failure to comply with the statutory deadline for seeking review.” (Doc. 18 at 2.) Now that discovery has concluded, CMS has moved for summary judgment as to the equitable tolling issue. That motion is also fully briefed. (Docs. 40-42.)

For the following reasons, CMS's motion for summary judgment as to the equitable tolling issue is denied. However, on the merits, AHCCCS has not established that DAB's decision was arbitrary or capricious. Thus, DAB's decision is affirmed and this action is terminated.

BACKGROUND

I. Medicare Reimbursement Framework

The plaintiff in this action, AHCCCS, is responsible for administering Arizona's Medicaid program. Wood v. Betlach, 922 F.Supp.2d 836, 839 (D. Ariz. 2013). The defendant, CMS, oversees state Medicaid programs on behalf of the federal government. Broadly, Medicaid is a cooperative federal-state program under which the federal government provides funding to the states to assist with medical expenses for lower-income populations. Id. at 839 (“Medicaid was enacted, in part, to enable states ‘to furnish . . . medical assistance on behalf of families with dependent children and of aged, blind, or disabled individuals, whose income and resources are insufficient to meet the costs of necessary medical services.'”) (citations omitted). “States that wish to receive federal funds through Medicaid must submit a state plan for approval by the Secretary of DHHS [Department of Health and Human Services].” Id. Upon approval, such a plan entitles the state to federal financial participation (“FFP”) in certain amounts as determined by statute. 42 U.S.C. § 1396b.

Once a state has received FFP, CMS has the authority to audit the state's funding allocation. See generally 42 C.F.R. §§ 430.32(a), 430.33(a)(2). If CMS determines that Medicaid has overpaid the state, CMS issues a notice of disallowance. 42 C.F.R. § 430.42(a). The state then has a variety of ways to challenge the disallowance. First, the state can request reconsideration. 42 U.S.C. §§ 1316(e)(1). Second, the state can appeal the disallowance decision (or the denial of reconsideration as to the disallowance decision) to DAB. Id. § 1316(e)(2)(A). Finally, the state may obtain judicial review of DAB's decision by filing an action in a United States District Court. Id. § 1316(e)(2)(C).

II. Relevant Facts

The facts set forth below are derived from the parties' summary judgment submissions, other documents in the record, and the administrative record.

A. The Disallowance Decision By CMS

Between January 1, 2004, and June 30, 2006, AHCCCS “claimed approximately $184 million ($124 million Federal share) for Medicaid school-based health services.” (Doc. 27-2 at 45.) Between 2007 and 2009, the Office of Inspector General (“OIG”) of the Department of Health and Human Services (“HHS”) conducted an audit of the claims submitted by AHCCCS for the 2004-2006 timeframe. (Id. at 52.)

At some point in 2009, CMS issued a draft report concluding that AHCCCS owed a “refund to the Federal Government” in the amount of $21,288,312 “for unallowable school-based health service[s].” (Id. at 68.) On October 8, 2009, AHCCCS sent a letter to the auditor disputing the $21,288,312 figure contained in the draft report. (Id. at 68-70.)

On March 22, 2010, OIG issued its final report of the audit, concluding that AHCCCS “was improperly reimbursed at least $21,288,312 in Federal Medicaid funds for school-based health services.” (Id. at 40-70.) To compute this total, OIG “extracted claims data from its Prepaid Medical Management Information System for 9,542,514 Medicaid school-based health services claimed for the period January 1, 2004, through June 30, 2006” with some exclusions,¹ resulting in “9,542,367 services.” (Id. at 62.) The remaining services were then grouped into “530,029 student-months” with some exclusions,²resulting in a total “sampling frame” of “528,543 student-months for which [AHCCCS] claimed a total of $182,790,631 ($123,614,883 Federal share).” (Id.)

1 The exclusion was for claims “funded by Title XXI (Children's Health Insurance Program).” (Doc. 27-2 at 62.)

2 The exclusions were for months with “a net claimed amount of zero” or a “net negative claimed amount” and various months that had been previously reviewed by either CMS or Arizona's Office of Program Integrity. (Doc. 27-2 at 62.)

Next, OIG used a “simple random sample” of “100 student-months.” (Id. at 62-63.) “The source of [the] random numbers for selecting sample units was the Office of Audit Services (OAS) statistical software.” (Id. at 63.) Based on this 100 student-month sample, OIG audited the individual services and found 46 student months with “deficiencies” that had a value (per federal share) of $6,764. (Id. at 64.) OIG then calculated an estimate of the unallowable amounts using a “90-Percent Confidence Interval.” (Id.) The point estimate was $35,750,384, with an upper limit of $50,212,457 and a lower limit of $21,288,312. (Id.) OIG also included an appendix that detailed the sample months and a legend indicating which types of services were deficient. (Id. at 65-67.)³

3 In layman's terms, “[t]he total dollar error (i.e., overpayment) is $35.8 million (this is the ‘point estimate'). The upper/lower bound, or level of precision, is +/- $14.5 million (at a 90% confidence level). Therefore [under that confidence level], the total dollar error is between $21.3 million and $50.2 million.” (Doc. 27-3 at 193, emphasis omitted.)

Over the next several years, AHCCCS and CMS engaged in discussions concerning OIG's recommendation, which included AHCCCS providing additional documentation to CMS. (See, e.g., Doc. 27-2 at 4-5; Doc. 27-3 at 179-82.)

On June 26, 2018, after “CMS reviewed additional documentation provided by the state,” “CMS issued a formal notice of disallowance in the amount of $19,923,489.” (Doc. 27-2 at 32-35.)⁴

4 As explained by the DAB, the “unallowable expenditures in the sample” was reduced from “$6,764 to $6,502.55.” (Doc. 27-2 at 4.)

On October 23, 2018, AHCCCS's request for reconsideration was denied. (Id. at 37-38.)

B. The Appeal To DAB

On December 15, 2018, AHCCCS submitted a notice of appeal to DAB. (Id. at 2729.) AHCCCS's primary objection was that “OIG's sampling methodology produced results that are unreliable, and the resulting extrapolation of the disallowance amount is incorrect and invalid.” (Id. at 28. See also id. at 133 [briefing].) Included it in the notice were expert reports from Milliman, an actuarial accounting firm (id. at 72-85), and EconLit, an economic litigation firm (id. at 87-97).

The Milliman report stated that “[t]he sample size of 100 is too small to obtain an extrapolation of dollar error that fits into the norms of statistical inference and in fact is too small to even measure the overall percentage of student-months paid in error.” (Id. at 73.) The Milliman report also asserted that the random sample of 100 student-months was “not representative of the population from which it is drawn.” (Id.) To support these conclusions, the Milliman report asserted that the calculations used by OIG averaged an overpayment per month of $6,764, meaning the average overpayment per student month in the 100-month sample was $67.64. (Id. at 74-75.) Because OIG's report stated that the confidence interval was 90%, the Milliman report stated that the standard deviation could be calculated as $166.34 per student-month. (Id. at 75.) The Milliman report concluded that the sample size should have been at least 1,632 student-months (as opposed to 100 student-months). (Id. at 75.) The Milliman report also questioned whether the 100 studentmonth sample exhibited “similar characteristics to the overall population.” (Id. at 76.) It elaborated: “The mean student-months size in the sample was $233.88 . . . per studentmonth and the mean size of the student-months dollars selected in the sample of size 100 was $322.12.” (Id. at 76.) Based on this difference, the Milliman report concluded that “the sampled population differs from the overall population and much higher studentmonths were included in the sample than exist in the overall population.” (Id. at 78.) Finally, as for the disallowance categories (nursing care; attendant care; bus transportation; occupational, speech, hearing and physical therapy; and psychologist, family therapist, professional counselor, guidance counselor, social worker), the Milliman report concluded that two categories were over-represented (transportation and nursing care) and two were underrepresented (counseling/evaluation and attendant care). (Id. at 79.)

The EconLit report raised similar concerns. (Id. at 88.) The EconLit report was offered in rebuttal to OIG's report by Dr. Alan Kvanli.⁵ (Id.) Based on the numbers in the OIG audit, the EconLit report concluded that the precision percentage in OIG's calculations was 40%. (Id. at 89.) The EconLit report asserted that OIG's “level of precision . . . is unacceptably low, rendering the asserted values statistically meaningless and unreliable” and that the selected sample resulted in “a significant overstatement of both the point estimate and lower limit of the confidence interval.” (Id.) The EconLit report further concluded that a cluster sample calculation was required, which would put the lower disallowance limit somewhere in the ballpark of $12.8 million (as opposed to $21 million). (Id. at 96; see also id. at 109 [cluster calculation].)

5 EconLit spent a considerable portion of its report criticizing Dr. Kvanli's biases as a “specialist in defending statistical audits” as opposed to being an “independent and unbiased expert in the field of statistics.” (Doc. 27-2 at 90 nn. 7-10; id. at 90-92 [asserting that Dr. Kvanli's published materials conflict with his opinions].)

On January 8, 2019, DAB indicated its receipt of AHCCCS's notice of appeal and made several evidentiary rulings. (Id. at 116-23.)⁶ The parties' briefing concluded in September 2019. (Id. at 241-62.)

6 Before issuing its decision, the parties fought bitterly over expert reports through surreplies, objections, and motions to strike. (See, e.g., Doc. 27-2 at 204-62; see also Id. at 5-6 [summarizing the disputes and outcomes].)

On December 27, 2019, DAB issued a final decision. (Id. at 1-26.) Broadly speaking, DAB construed AHCCCS's appeal as “rais[ing] objections to the OIG's statistical sampling and estimation methods and to the OIG's finding that expenditures for certain speech therapy services were ineligible for FFP” and concluded: “We overrule these objections and therefore affirm the disallowance.” (Id. at 1.) More specifically, as for the confidence interval, DAB cited several decisions that explain that a 90% confidence interval is appropriate so long as it is “properly derived using valid methods,” “is reliable evidence of the amount of unallowable costs charged to federal funds,” and “protects the State with a 95% degree of confidence from having to refund more than the true but unknown amount of the FFP overpayment.” (Id. at 6-7, cleaned up.) DAB concluded that CMS met its initial “burden in this proceeding to show [its sampling and estimation methods] are ‘scientifically valid' and ‘yielded ‘reliable evidence' of the amount of FFP improperly claimed for school-based health services.” (Id. at 7.) This burden was met through the “declarations of [CMS's] expert, Dr. Smith, who has 10 years of experience working as a statistician and completed 60 quarter hours of graduate-level coursework relating to statistics.” (Id.) As for AHCCCS's challenge to sample size, DAB found that AHCCCS had “conceded that the point estimate's precision has ‘nothing to do with whether OIG's lower bound is correct.'” (Id. at 10 [“[T]he State did not (in its reply or sur-surreply briefs) press its initial claim that the sample size was too small.”].) DAB also concluded that “the size of the sample in this case did not render the population estimate supporting the disallowance invalid, and that any imprecision in the point estimate likely benefitted the State because it widened the confidence interval whose lower bound supports the disallowance.” (Id. at 11.)

Turning to AHCCCS's challenge to the representativeness of the sample, DAB found “Dr. Smith's analysis persuasive.” (Id. at 12.) More specifically, DAB credited Dr. Smith's explanation that “random sampling, if executed properly with respect to a properly defined population, provides substantial assurance that the resulting sample's relevant attributes will sufficiently represent those in the population and lead to unbiased estimates.” (Id.)⁷ DAB further concluded that AHCCCS's arguments that “the OIG ‘did nothing to account for the unrepresentativeness of the sample,' and that this failure ‘produc[ed] a lower bound that is incorrect, statistically invalid, and far too high'” were “unconvincing for at least three reasons.” (Id. at 13.) “First, the State sidesteps Dr. Smith's main point, which is that the confidence interval's upper and lower bounds account for samplepopulation differences because they are calculated using the expected error, or variability, in the point estimate due to those differences.” (Id.) “Second, the State does not contend that the disparity between average FFP-paid amounts in the sample and population is so large or unusual that the sample cannot be relied upon to produce valid interval estimates.” (Id. at 14.) “Third, the State's suggestion that the sample's alleged non-representativeness actually caused the OIG's lower-bound estimate to be higher than it would otherwise have been is unfounded.” (Id.)

7 DAB also noted that “[tlhe parties agree with the basic maxim that a random draw does not guarantee a sample that is a perfectly representative subset of the population.” (Doc. 27-2 at 13.)

As for whether OIG was required to use a cluster sample analysis, DAB concluded that it was “unpersuaded that the OIG used an incorrect or inappropriate estimator to project its sample findings to the population.” (Id. at 18.) DAB explained that the difference between the two calculation methods was that “OIG used a mean difference estimator” whereas AHCCCS used a “ratio estimation approach.” (Id. at 15.) DAB noted that the mean difference estimator took the amount of overpayment ($6,503) and divided it by the sample number (100 student-months) to get the sample mean of $65.03. (Id.)

Next, DAB explained, the $65.03 figure was multiplied by the total student months in the population (528,543) to get a point estimate that, using the 90% confidence interval, OIG calculated the lower bound at $19,923,489. (Id. at 15-16, 26.) In contrast, DAB explained, AHCCCS's ratio estimation approach took the amount of overpayment ($6,503) and divided it by the FFP paid in the sample ($32,212), resulting in a quotient of 0.20188. (Id. at 15.) Using that quotient, AHCCCS then multiplied 0.20188 by the total amount of FFP paid during the audited years ($123.6 million) to create the point estimate of $24,953,704. (Id.) Using the 90% confidence interval, the lower bound was calculated to be $11,945,011. (Id.)

DAB's opinion contains a detailed summary of the reasons that Dr. Smith provided for not following AHCCCS's preferred methodology. (Id. at 16-18.) As the opinion explains, Dr. Smith stated that the “ratio estimator allows a statistician to account for known ‘auxiliary' information (here, that information is the amount of FFP paid for a student-month) when estimating the value of the population characteristic of interest (FFP paid in error).” (Id. at 16.) Dr. Smith concluded that the FFP paid per student-month and the FFP paid in error are “likely related” and therefore the ratio estimator can create a more precise point estimate. (Id.) However, Dr. Smith explained that “it is ‘not necessary' that the point estimate be as accurate and efficient as possible given CMS's reliance on the lower limit of the confidence interval.” (Id.) Dr. Smith continued that any imprecision in the calculations (when using the mean-difference estimator) benefits AHCCCS because “[w]hen the precision is worse, the amount subtracted is greater.” (Id. at 16-17.) Dr. Smith explained that “OIG does not use a ratio estimator because it can be ‘biased,' and that ‘the lower limit of the ratio estimator tends to be less reliable than the lower limit of the difference estimator.” (Id. at 18.) Regarding the unit of measure, Dr. Smith asserted that “defining the sampling unit as a student-month (rather than as a cluster of ‘dollars paid') and selecting a random sample of student-months were valid sampling methods and sufficed to produce an ‘unbiased' estimate of the relevant population parameter.” (Id. at 17.) Dr. Smith also asserted that the availability of alternative units of measure (like service categories or dollar values) “does not undermine the unbiased nature of the sample as long as the total [dollar-amount] for each student-month is finite.” (Id.) Finally, Dr. Smith opined that that “the State failed to show that the ratio estimator was actually ‘unbiased or more precise' than the mean difference estimator.” (Id. at 18.)

DAB chose to credit Dr. Smith's assertions on these issues, concluding that “[k]ey assertions by EconLit lack foundation, allowing Dr. Smith's opinions on the subject, which we find facially plausible, to stand unrebutted.” (Id.)⁸ Those key assertions included that “it was necessary for the OIG to use a ratio estimator merely because the sampling units could be viewed, or should have been originally defined, as ‘clusters' of ‘dollars paid'” (id.); that “a ratio estimator must be used to correct for a non-representative sample” (id.); that the “ratio estimator is unbiased for the population of student-months defined by the OIG” (id. at 19); and that AHCCCS's “lower-bound estimate is ‘more conservative' than the OIG's” (id. at 20).

8 “For example, EconLit cited no literature, statistical theory, or evidence of accepted norms of audit sampling to support its suggestion that it was necessary for the OIG to use a ratio estimator merely because the sampling units could be viewed, or should have been originally defined, as ‘clusters' of ‘dollars paid.'” (Doc. 27-2 at 18.)

In conclusion, DAB explained: “Given that the OIG's lower-bound estimate was derived using valid sampling and estimation techniques and is unbiased, the State cannot prevail merely by touting a lower interval estimate derived from a sample statistic (the ratio estimator) whose suitability has not been adequately demonstrated.” (Id. at 20-21.) As for AHCCCS's argument that “precision matters” as it relates to the point-estimate, DAB concluded that “the state failed to provide any statistical calculations or analysis to support its claim that the population point estimate generated by the ratio estimator was actually more precise than the point estimate yielded by the mean difference estimator.” (Id. at 21.) Therefore, DAB upheld the $19,923,489 disallowance. (Id. at 26.)

C. Transmission Of DAB's Decision

As a general matter, DAB does not send paper copies of its decisions and instead uses email. (Doc. 40-2 at 2 ¶ 8 [“[T]he Board's general practice is to serve all decisions or other Board issuances to the parties or their representatives only via DAB E-File.”]; Doc. 16-1 ¶¶ 4-6 [same].)⁹ Here, although DAB issued its decision on December 27, 2019, AHCCCS's counsel in this matter (“Counsel”)¹⁰ did not learn of the decision until May 8, 2021, “when [he] went to the DAB website . . . to see if there was someone [he] was supposed to contact when a decision seemed overdue.” (Doc. 16-1 ¶ 13.)

9 “The DAB system alerts the parties by e-mail when any document is filed in a case. The message comes from a source identified as notification@dab.efile.hhs.gov and is captioned ‘Message from the Departmental Appeals Board E-Filing System: New Document Filed.' The message describes the new document and says, ‘Click here to log into the e-filing system and view this submission under the docket number.'” (Doc. 16-1 ¶ 6.)

10 Counsel was the “sole counsel” litigating the disallowance. (Doc. 16-1 ¶ 2.)

Counsel contends that he had “no notification from the DAB either in the in-box, deleted messages, or archives of my e-mail messages.”¹¹ (Doc. 16-1 ¶ 14.) Instead, the last notice he received from DAB in relation to this case was on September 10, 2019, after filing AHCCCS's final memorandum.¹² (Doc. 16-1 ¶¶ 7-8; accord Doc. 40-2 at 3 ¶ 15.) Counsel asserts that, after the briefing sequence was complete, he “checked [his] e-mail daily for any ruling by the DAB either on the case as a whole or on ancillary evidentiary disputes that had been briefed.”¹³ (Doc. 16-1 ¶ 7.) He also emailed AHCCCS's in-house counsel monthly between October 2019 and April 2021, explaining that “[w]e are awaiting a decision.” (Id. ¶ 9.)¹⁴

11 CMS objects to this statement because “Plaintiff has not shown that its counsel performed a search of the messages he sent or received on December 27, 2019 that were deleted by its counsel.” (Doc. 40 at 13.)

12 On January 16, 2020, CMS emailed Jeffrey Tegen, the “Assistant Director, Business & Finance at AHCCCS,” requesting a payment of “$19,923,489 with interest of $647,290, citing a December 27, 2019 decision by the DAB upholding the disallowance.” (Doc. 163 ¶¶ 2, 5.) Tegen did not forward the email to the AHCCCS Office of Legal Assistance or to Counsel. (Doc. 16-3 ¶ 7. See also Doc. 16-2 ¶ 6 [confirming that Counsel did not receive actual notice of the decision until May 2021].) At the time, Tegen knew “the attorneys who were handling the dispute were aware AHCCCS had not retained the funds” “so there was no reason for [him] to contact the attorneys about this notice.” (Doc. 16-3 ¶¶ 7-8.)

13 During his deposition in this case, Counsel stressed that he keeps emails “forever” and “[a]s long as the machine will hold them.” (Doc. 40-5 at 5.) However, Counsel did not contact his email provider to “see if their system had any outages on December 27th, 2019.” (Id. at 6.)

14 CMS asserts that its “Appellate Division's records indicated that on March 3, 2020, [the DAB decision] was posted to the public website” (Doc. 40-2 at 4 ¶ 20), but it is not clear from the website when the decision was posted. Arizona Health Care Cost Containment System, DAB No. 2981 (2019), https://www.hhs.gov/about/agencies/dab/decisions/board-decisions/2019/board-dab-2981/index.html [https://perma.cc/N9ES-S3F2].

On May 11, 2021, three days after learning of DAB's decision, Counsel inquired with DAB via email about why he was not notified of the decision via DAB's usual practice of sending notification emails. (Doc. 40-2 at 13.) Counsel also reached out to opposing counsel for a copy of the notification email, which, initially, opposing counsel could not locate. (Doc. 16-1 ¶ 16.) Later, opposing counsel stated in a declaration that the email had been sent to the “Region 9” mailbox, which was the “registered user for HHS in this appeal.” (Doc. 40-4 at 2 ¶ 7; Doc. 40-4 at 8.)

On May 12, 2021, Counsel was informed that the “DAB E-File notification records . . . reflect that a notification email of the Board's decision was sent” to Counsel's email address on “December 27, 2019, at 11:09 am.” (Doc. 16-1 ¶ 15; Doc. 40-2 at 10, 15.) Counsel was further informed that DAB's records “do not reflect that the notification email was returned or bounced back as undeliverable.” (Doc. 16-1 ¶ 15; Doc. 40-2 at 4 ¶ 23; Doc. 40-3 at 2-3 ¶¶ 5-8.)

III. Procedural History

On June 1, 2021, AHCCCS initiated this action by filing a complaint. (Doc. 1.)

On June 2, 2021, AHCCCS filed a first amended complaint (“FAC”). (Doc. 5.)

On August 6, 2021, CMS moved to dismiss the FAC for lack of subject-matter jurisdiction. (Doc. 11.) CMS argued that “[t]here is no dispute that this case was brought more than 60-days from December 27, 2019, and therefore cannot satisfy the requirements of 42 U.S.C. § 1316(e)(2)(C)” and that “Plaintiff has not pled or demonstrated that equitable tolling should be applied to overcome the jurisdictional requirements.” (Id. at 78.)

On January 28, 2022, after full briefing (Docs. 16, 17), the Court granted CMS's motion to dismiss but allowed AHCCCS to file a second amended complaint (“SAC”). (Doc. 18.) As relevant here, the Court concluded that AHCCCS “can plead facts showing that the deadline to file the action is subject to equitable tolling.” (Id. at 15-16.) For example, viewing the facts as briefed, the Court found that “[t]he most reasonable conclusion is that the DAB E-File system malfunctioned or some other technological failure beyond AHCCCS's control prevented the email from being successfully transmitted” and “[a]lthough Counsel could have regularly checked the DAB website to see if a decision had been posted, this would have been an unjustifiable waste of time- there was no reason to check the website when Counsel had every reason to believe he would receive the decision by email.” (Id. at 12.)

On February 2, 2022, AHCCCS filed the SAC. (Doc. 19.)

On September 16, 2022, AHCCCS filed its opening brief. (Doc. 39.)

On October 17, 2022, CMS filed a combined response brief and motion for summary judgment as to equitable tolling. (Doc. 40.)

On November 16, 2022, AHCCCS filed a combined reply brief and response to CMS's summary judgment motion. (Doc. 41.)

On December 1, 2022, CMS filed a reply in support of its summary judgment motion. (Doc. 42.) Neither side requested oral argument.

DISCUSSION

I. Equitable Tolling

A. Standard of Review

When, as here, the material facts are undisputed, the equitable tolling inquiry is a question of law. Valenzuela v. Kraft, Inc., 801 F.2d 1170, 1172 (9th Cir. 1986) (“Because the facts which Valenzuela relies on to establish tolling are not disputed, this is also a question of law which we review de novo.”); Long v. Paulson, 349 Fed.Appx. 145, 146 (9th Cir. 2009) (same).

B. Background Law

There is a “rebuttable presumption in favor of equitable tolling for statutes of limitations.” Smith v. Davis, 953 F.3d 582, 592 (9th Cir. 2020). See also id. (“Equitable tolling operates apart from any statutory provision.”). Here, CMS does not dispute that equitable tolling is potentially available. (Doc. 40 at 17-22; Doc. 42 at 3 n.2.)

A litigant “is entitled to equitable tolling only if he shows (1) that he has been pursuing his rights diligently, and (2) that some extraordinary circumstance stood in his way and prevented timely filing.” Hollandv. Florida, 560 U.S. 631, 649 (2010) (internal quotation marks omitted). As for the first element, a litigant must demonstrate “that he has been reasonably diligent in pursuing his rights not only while an impediment to filing caused by an extraordinary circumstance existed, but before and after as well, up to the time of filing his claim in federal court.” Smith, 953 F.3d at 598-99. “The diligence required for equitable tolling purposes is ‘reasonable diligence,' not ‘maximum feasible diligence.'” Holland, 560 U.S. at 653 (citations omitted). The second element is “met only where the circumstances that caused a litigant's delay are both extraordinary and beyond its control.” Menominee Indian Tribe of Wis. v. United States, 577 U.S. 250, 257 (2016). “[Principles of equitable tolling . . . do not extend to . . . a garden variety claim of excusable neglect.” Irwin v. Dep't of Veterans Affairs, 498 U.S. 89, 96 (1990). The “extraordinary circumstance” must have “prevented a petitioner acting with reasonable diligence from making a timely filing.” Smith, 953 F.3d at 600. This is not “a rigid ‘impossibility' standard,” and the court “is not bound by ‘mechanical rules' and must decide the issue based on all the circumstances of the case before it.” Id.

C. The Parties' Arguments

In its opening brief, AHCCCS argues equitable tolling should apply because “Counsel did not receive a transmission” from the DAB E-file system, there “is nothing to indicate Counsel was the cause of this problem,” and “Counsel watched for the DAB decision daily.” (Doc. 39 at 10.)

CMS, in turn, argues that AHCCCS “cannot satisfy the heavy burden to demonstrate that equitable tolling should apply here” because the “DAB E-File system was functioning properly on December 27, 2019” and therefore there was no “‘extraordinary circumstance' preventing Plaintiff from timely filing the Complaint.” (Doc. 40 at 19-20.) CMS further contends that Counsel “had a duty to regularly monitor the docket of this case” and “admitted to not checking the docket . . . for over 1 year and 7 months.” (Id. at 20.) CMS thus argues that “Plaintiff's counsel's neglect in monitoring his case docket . . . is not an ‘extraordinary circumstance' to justify equitable tolling.” (Id. at 21.)

In response, AHCCCS argues that “the evidence the Court previously considered regarding this issue has now been confirmed by depositions and discovery. CMS fails to demonstrate any reason for the Court to change its prior analysis, much less to enter summary judgment against AHCCCS.” (Doc. 41 at 1, citing Doc. 18.) AHCCCS further argues that CMS has not established that DAB's “decision was received, since the DAB system apparently does not track receipt or reading of its notifications.” (Id. at 2, emphasis omitted.) Finally, as for the cases cited by CMS to condemn Counsel for not checking the docket, AHCCCS argues that most of those cases “do not discuss equitable tolling but focus instead on application of different standards under FRCP Rule 60 for relief from judgment, FRCAP Rule 4 for appeals as of right from district court decisions, and Bankruptcy Rule 9022” and the rest of the cases are factually distinguishable. (Id. at 4-7.) In reply, CMS reiterates its position that “[g]iven that the undisputed record shows that the DAB's e-file system was working properly, confirms that it sent the email to the State's counsel with no issues or bounce backs, and no other party receiving notice through the DAB's e-file had this same issue-in fact, the regional counsel's email account received the notice-the State's counsel's denial of receipt is not sufficient to establish equitable tolling.” (Doc. 42 at 4.) CMS also reiterates its position that Counsel's “failure to monitor the status of the case further demonstrates that equitable tolling is not warranted.” (Id. at 5.) According to CMS, “[t]he obligation to monitor the case docket does not, in the words of the State ‘turn practice on its head,' but is a rather minimal burden and would be in this context.” (Id. at 6.)

D. Analysis

As noted, 42 U.S.C. § 1316(e)(2)(C) requires a state seeking review of an adverse DAB decision to file an action for judicial review within 60 days of when the decision was issued. Here, the decision was issued on December 27, 2019, but AHCCCS did not initiate this action until June 1, 2021, well outside the 60-day window. In determining whether equitable tolling is warranted here, the Court considers the following undisputed facts to be particularly salient: Counsel consistently received notifications from the DAB E-File system via email before the decision was issued (Doc. 16-1 ¶¶ 5, 18); nothing suggested that it was risky or insufficient to rely on such email notifications (id. ¶¶ 18-19); Counsel consistently checked his email for a decision after the briefing process was completed (id. ¶ 18); a paper notice of the decision was never issued (Doc. 40-2 at 2 ¶ 8); Counsel lacked actual notice of the decision until May 2021 (Doc. 16-3 ¶ 7; Doc. 16-2 ¶ 6); and there is no obvious explanation as to why the email did not appear in Counsel's email inbox when the DAB's system indicates that it was delivered (Doc. 16-1 ¶¶ 15, 21).

Given these facts, the Court agrees with AHCCCS that the reasoning from the order denying CMS's motion to dismiss (Doc. 18) remains intact. Counsel's reliance on the DAB E-File notification was reasonably diligent under the circumstances. The DAB had sent such notifications in the past. Although Counsel could have regularly checked the DAB website to see if a decision had been posted, this would have been an unjustifiable waste of time-there was no reason to check the website when Counsel had every reason to believe he would receive the decision by email. Thus, under the circumstances, checking “daily” for a decision to arrive via email was reasonably diligent. It was also reasonably diligent for Counsel to wait to check the docket index until May 8, 2021, at a time when “the delay had begun to seem suspicious” (Doc. 16-1 ¶ 20), and it was reasonably diligent for AHCCCS to file this action 24 days later, on June 1, 2021. Smith, 953 F.3d at 599 (“In determining whether reasonable diligence was exercised courts shall ‘consider the petitioner's overall level of care and caution in light of his or her particular circumstances.'”).

CMS argues that Counsel had “an obligation to monitor [his] cases' status by periodically checking the Court's docket” and that his failure to do so establishes an absence of reasonable diligence. (Doc. 40 at 14, 18 [collecting cases].) The Court disagrees for the reasons set forth in the motion-to-dismiss order-considering that e-filing systems and email transmissions generally function properly, and courts and attorneys justifiably rely on them to do so, the apparent failure of the DAB E-Filing system to successfully email the decision to Counsel (the only form of notification the DAB uses) was an extraordinary circumstance beyond AHCCCS's control. Diaz v. Kelly, 515 F.3d 149, 155 (2d Cir. 2008) (“[A] court's failure to send notice within a reasonable time after entry of an order . . . can provide a basis for equitable tolling.”); cf. Hansen v. Astrue, 2012 WL 1551887, *4 (W.D. Pa. 2012) (“[I]t would be inequitable if it turned out that Hansen had to suffer the consequences of an error made by . . . an electronic malfunction in the Court's ECF system.”). The most reasonable conclusion based on the undisputed facts is that the DAB E-File system malfunctioned or some other technological failure beyond AHCCCS's control prevented the email from being successfully transmitted.

CMS's cited cases do not persuade the Court otherwise. For example, in Yeschick v. Mineta, 675 F.3d 622 (6th Cir. 2012), the plaintiff moved for relief under Rule 60(b)'s excusable neglect standard after his counsel did not receive email notices for discovery or the defendant's motion for summary judgment. Id. at 627, 629. Counsel in that case “failed to update his email address on file with the district court” and, even after he was informed that his email had not been receiving notifications, “did not check the docket in [plaintiff's] case until more than a month” had passed. Id. at 630. Such conduct is distinguishable from the facts here, where Counsel maintained a current email address, went in search of the source of the error within three days of learning of the decision, and filed this action less than a month later. Similarly distinguishable is Fox v. American Airlines, Inc., 389 F.3d 1291 (D.C. Cir. 2004), in which counsel failed to realize American Airlines had filed a motion to dismiss, thus resulting in the automatic granting of the motion. Id. at 1293. There, counsel had filed other documents referencing the outstanding amended complaint (over a month after the motion to dismiss was filed) and failed to notice he had not yet “receive[d]a timely answer to the amended complaint.” Id. at 1294. Here, Counsel had no concrete expectation of a timeline for DAB to render a decision, so there was no obligation for him to be on notice that a decision was overdue and check the docket in search of the overdue decision.

Also unavailing is CMS's argument that, because a judge of this Court issued a decision in one of Counsel's other “DAB appeal” cases within a four-month timeframe, Counsel should have been on notice of the likely issuance of DAB's decision in this case within a similar timeframe. (Doc. 40 at 13-14.) The obvious problem with this argument is that federal district courts are dissimilar to DAB-they are composed of different personnel, governed by different laws and regulations, and operate on different timeframes. Indeed, AHCCCS has proffered evidence establishing why it was reasonable for Counsel to assume that DAB was taking an unusually long time to issue a decision in this case- namely, “it had taken eight years for CMS to finalize its disallowance after the audit in this case” and he suspected “DAB was wrestling with the novel issue [AHCCCS] had raised regarding whether CMS must use ‘cluster sampling' for the kind of audit it did in this case (as opposed to the ‘simple random' methodology CMS had used).” (Doc. 16-1 ¶¶ 10-12.)

Nor does Snyder v. Barry Realty, Inc., 60 Fed. App'x 613 (7th Cir. 2003), support CMS's position. There, the Seventh Circuit affirmed the dismissal of the Snyders' lawsuit under Rule 60(b) because they failed to appear for trial. Id. at 614-15. Admittedly, neither the attorney nor the Snyders checked the docket, so they did not see the “scheduling and dismissal orders.” Id. However, the Seventh Circuit only reviewed the district court's dismissal order under an abuse of discretion standard. Id. The court remarked that “[a]lthough the district court could have excused the Snyder's extended carelessness, the facts did not compel it to do so. Thus, because the district court adequately considered the Snyders' reasons for their neglect, it did not abuse its discretion when it denied their Rule 60(b) motion.” Id. Thus, Snyder does not stand for the proposition that a district court must order dismissal under such circumstances. More important, this case involves dissimilar facts (as discussed, Counsel had a variety of case-specific reasons to believe his methodology for staying abreast of case developments was appropriate and to believe that no DAB decision had been issued) and a somewhat different legal inquiry.

The other cases cited by CMS are distinguishable for similar reasons, as the attorneys in those cases engaged in more-culpable conduct than Counsel and/or lacked Counsel's case-specific reasons for believing that DAB's decision had not yet issued. See, e.g., Kuhn v. Sulzer Orthopedics, Inc., 498 F.3d 365, 371-72 (6th Cir. 2007) (denying reopening of an appeal under FRCP 4(a) when, inter alia, an attorney had not registered for CM/ECF email notifications and had particular reason to check the docket because he had “take[n] the unusual step of asking a court to promptly issue a written order”); United States ex rel. McAllan v. City of New York, 248 F.3d 48, 53-54 (2d Cir. 2001) (denying an appeal under FRCP 4(a) where, inter alia, counsel waited too long to file another notice of appeal (i.e., 30 days rather than 14 days) after the district court reissued the underlying decision in an attempt to correct earlier docketing irregularities); Henken v. IW Tr. Funds, 568 F.Supp.3d 870, 876-77 (S.D. Ohio 2021) (denying relief under FRCP 60(b) where plaintiff's counsel knew he “could not use his CM/ECF username or password to access the docket” yet failed to attempt to fix these problems and also admitted “that the COVID-19 pandemic made him ‘not concerned' about the case's progress”); Konarski v. Rankin, 2015 WL 10793428, *1-*2 (D. Ariz. 2015) (denying relief under FRCP 60(b) where counsel had been aware of his “e-mail system encompassing significant glitches” and was subjectively aware of the Ninth Circuit's decision remanding the case).

For similar reasons, the Court is unpersuaded by CMS's contention that Counsel's failure to anticipate the decision and check the docket is a “garden variety of excusable neglect.” (Doc. 40 at 22.) In response, AHCCCS argues that “for anyone registered as a DAB E-filing recipient, monitoring the docket serves no purpose unless one has reason to suspect the docket will disclose something the email system failed to deliver.” (Doc. 41 at 3-4.) The Court agrees with AHCCCS. The equitable tolling doctrine does not mandate “maximum feasible diligence.” Holland, 560 U.S. at 653 (citations omitted). Here, unlike in the cases cited by CMS, Counsel had no reason to suspect that reliance on email notification was misplaced and various reasons to suspect that DAB's issuance of the final decision would be delayed (which, in turn, meant that the absence of email notifications wasn't suspicious). Although it may have been prudent to check the docket, the Court concludes under the somewhat unusual circumstances of this case that Counsel's daily checks of his email constituted reasonable diligence.¹⁵ This conclusion is not undermined by the fact that CMS's counsel happened to check the docket on January 30, 2020. (Doc. 40-4 ¶ 4.) Because CMS's counsel knew he was not receiving email notifications (id. ¶ 7), checking the docket was a reasonably necessary way for him to stay apprised of case developments.

15 CMS notes that “as a DAB registered user, [Counsel] was informed that ‘an electronic filer assumes the risk of all errors not solely attributable to a DAB E-File Malfunction' and was required to ensure that DAB e-mail notifications are not blocked by spam or other filters.'” (Doc. 40 at 20.) But the fact that DAB's E-File system contains a disclaimer related to email notifications doesn't undermine the reasonableness of Counsel's approach. This disclaimer doesn't change the fact that, before this incident, Counsel consistently received DAB E-File notifications.

CMS also criticizes AHCCCS for “not tak[ing] any additional steps” to get to the bottom of the non-receipt issue, including “contacting its counsel's email service provider to see if there were any outages on the day the email was sent or to about [sic] obtain records regarding any emails sent on that day. So, Plaintiff cannot even point to its email server as being the cause for the non-receipt.” (Doc. 40 at 19-20.) In a related vein, CMS contends that AHCCCS cannot overcome the “presumption of receipt of the e-mail” to establish extraordinary circumstances. (Id., citing Am. Boat Co. v. Unknown Sunken Barge, 567 F.3d 248 (8th Cir. 2009).) In response, AHCCCS disputes CMS's assertion that “Plaintiff has not shown that its counsel performed a search of the messages he sent or received on December 27, 2019, that were deleted by its counsel.” (Doc 41 at 5 [“The deleted messages from that day were retained, searched, and produced to CMS.”].) Moreover, AHCCCS argues that “no outage was experienced, and even if an outage had occurred this would still have been an extraordinary event beyond AHCCCS's control.” (Id. at 6.) In reply, CMS reiterates: “Given that the undisputed record shows that the DAB's e-file system was working properly, confirms that it sent the email to the State's counsel with no issues or bounce backs, and no other party receiving notice through the DAB's e-file had this same issue-in fact, the regional counsel's email account received the notice-the State's counsel's denial of receipt is not sufficient to establish equitable tolling.” (Doc. 42 at 4.)

CMS's arguments miss the mark. CMS has not identified any Ninth Circuit caselaw suggesting that a presumption of delivery applies to email messages, and even if there were such a presumption, it has been overcome here-AHCCCS has proffered an array of evidence that suggests Counsel never received the email and CMS has simply identified theoretical additional steps AHCCCS could have pursued in its quest to prove this negative. As the Court has already explained, it was extraordinary, under the particular circumstances at issue here (i.e., updated contact information, no notice of any prior or subsequent issue with email notifications, consistent reliance on those email notifications, no actual notice of the decision, no evidence of carelessness, and case-specific reasons to expect that the issuance of the decision would be delayed), for Counsel to lack timely notice of DAB's decision. Counsel's failure to pinpoint the exact mechanism that explains why the message was not received does not undermine the reasonableness of his actions or the extraordinary nature of the circumstances. Smith, 953 F.3d at 590 (“Because equity requires a court to deal with the case before it, complete with its unique circumstances and characteristics, courts must take a flexible approach in applying equitable principles.”).

Finally, CMS argues that “Plaintiff's actual notice of the DAB's decision on January 16, 2020 indicates that equitable tolling is not warranted in this case.” (Doc. 40 at 21.) This argument lacks merit. The uncontroverted evidence shows that although Tegen, a non-attorney, received notice of CMS's attempt to collect money from AHCCCS (which included notice of the DAB's decision), he did not convey that notice to AHCCCS's inside counsel or to Counsel. (Doc. 16-2 ¶ 6; Doc. 16-3 ¶¶ 2, 5, 7.) As explained at the motion-to-dismiss stage, AHCCCS is a large state agency. It cannot be the case that sending notice of DAB's decision to anyone who works there, regardless of their role, is sufficient to provide notice. Although it would have been serendipitous if the finance department had, for some reason, forwarded a copy of the decision to the people at AHCCCS who were involved in the litigation (or to Counsel), the absence of such serendipity does not alter the conclusion that an extraordinary circumstance prevented AHCCCS, despite its diligence, from pursuing review within the 60-day statutory period.

In sum, AHCCCS was reasonably diligent in pursuing its rights and extraordinary circumstances stood in its way and prevented timely filing. Accordingly, CMS's motion for summary judgment on the basis of equitable tolling is denied.

II. Judicial Review Of DAB's Decision

A. Standard Of Review

Judicial review of an administrative decision is generally limited to the record that was before the agency when it rendered its decision. Fed. Power Comm 'n v. Transcon. Gas Pipe Line Corp., 423 U.S. 326, 331 (1976).

Under the Administrative Procedure Act (“APA”), a reviewing court shall “hold unlawful and set aside agency action, findings, and conclusions found to be . . . arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with the law.” 5 U.S.C. § 706(2)(A). “The Supreme Court has held that the ultimate standard of review under 5 U.S.C. § 706(2)(A) is a narrow one, noting that a court is not empowered by section 706(2)(A) to substitute its judgment for that of the agency.” Nw. Motorcycle Ass 'n v. U.S. Dep't of Agric., 18 F.3d 1468, 1472 (9th Cir. 1994) (citations omitted).

“[W]here, as here, the district court is reviewing the decision of an administrative agency which is itself the finder of fact . . . summary judgment is an appropriate mechanism for deciding the legal question of whether the agency could reasonably have found the facts as it did.” Occidental Eng'g Co. v. I.N.S., 753 F.2d 766, 769-70 (9th Cir. 1985). “In reviewing an agency's decision under section 706(2)(A), a court must consider whether the decision was based on a consideration of the relevant factors and whether there has been a clear error of judgment. After considering the relevant data, the court must articulate a satisfactory explanation for its action including a rational connection between the facts found and the choice made. In order for an agency decision to be upheld under the arbitrary and capricious standard, a court must find that evidence before the agency provided a rational and ample basis for its decision.” Nw. Motorcycle, 18 F.3d at 1471 (citations and quotation marks omitted). See also Audubon Society of Portland v. Haaland, 40 F.4th 917, 924 (9th Cir. 2022) (“Arbitrary and capricious review is deferential, but an agency must articulate a satisfactory explanation for its action including a rational connection between the facts found and the choice made.”) (cleaned up).

B. The Parties' Arguments

AHCCCS's overarching position is that it identified various errors in how CMS performed the statistical sampling that resulted in the $20 million disallowance decision, the correction of which would have reduced the disallowance to approximately $12 million, and that DAB acted arbitrarily and capriciously by overlooking those errors and affirming CMS's methodology. (Doc. 39 at 12-29.) According to AHCCCS, “[t]he sampling issue presented by this case is straightforward, though sometimes complex in its mathematical underpinnings.” (Id. at 12.) Although AHCCCS's briefing on the sampling issue is voluminous, AHCCCS essentially raises two primary (if related) objections to CMS's approach. First, AHCCCS objects to the choice of the sample, which it variously criticizes as too small, not random, and unrepresentative. (See, e.g., id. at 16-20.) Second, AHCCCS objects to CMS's decision to use a particular methodology (the “mean difference estimator”) rather than AHCCCS's preferred methodology (“ratio estimation approach” or “cluster sampling”) to correct for the problems with the sample selection. (Id. at 20-25.) Along the way, AHCCCS advances various reasons why DAB's rejection of its arguments on those issues was not just incorrect, but arbitrary and capricious. Among other things, AHCCCS contends that “OIG has [n]ever used cluster sampling (or any other means) to correct for an unrepresentative sample” (id. at 15); that “[n]either CMS nor its expert Dr. Smith denies that cluster sampling is an accepted statistical methodology or that it can be used to correct for an unrepresentative sample, or that it can be applied to this sample” (id.); that “CMS offers no evidence that OIG or CMS considered whether this sample [of student months] was unrepresentative, much less whether it could and should be adjusted” (id. at 16); and that CMS's ultimate defense of its methodology was not that the methodology was more accurate than AHCCCS's, but simply that its methodology was “acceptable” and “reasonable” (id. at 25-26). As relief, AHCCCS asks the Court to “require CMS to recompute the disallowance using the cluster sample analysis set forth by EconLit.” (Id. at 29.)

In response, CMS contends that “AHCCCS has not met its burden to establish that the DAB's decision was arbitrary, capricious, or unsupported by substantial evidence” and therefore “the Court should affirm the DAB's decision.” (Doc. 40 at 22.) More specifically, CMS argues that DAB correctly concluded that it met its burden of scientific validity and reliable evidence through Dr. Smith's methodology, which included “(1) identifying an appropriate sampling unit, the student-month; (2) defining a finite target population consisting of non-overlapping sampling units that had an equal chance of being selected; (3) drawing a simple random sample from the target population using widely accepted statistical software (RAT-STATS) developed by the federal government; (4) using a mean-per-unit estimator to derive an unbiased point estimate of unallowable FFP in the population; and (5) accounting for the uncertainty of the unbiased point estimate by calculating a two-sided 90 percent confidence interval around the unbiased point estimate.” (Id. at 23.) In response to the argument that CMS was required to use the cluster-sampling methodology because it is more precise, CMS argues “the State fails to cite any legal authority to back up this manufactured burden. Indeed, by this logic it would be the agency's burden in any disallowance to explain why it declined to use one of multiple alternative sampling methodologies, which would be an onerous and illogical burden to place on the agency without supporting legal authority.” (Id. at 24.) CMS contends that it was not required “to produce the most precise, efficient, or best method, but rather a valid and reliable method. . . . The existence of a different sampling methodology does not negate this finding.” (Id.) CMS further contends that AHCCCS's “sample size and precision critiques are unpersuasive” because “the level of assurance in the lower bound of the 90 percent confidence interval is not impacted by the point estimate's precision,” which AHCCCS did not dispute below, and “the point estimate's imprecision will tend to benefit the grantee when the disallowance is based on the lower bound of the relevant confidence interval.” (Id. at 24-25.)

In its reply, AHCCCS contends that “CMS's response avoids the issue.” (Doc. 41 at 9.) It synthesizes its position as follows:

The parties' differing sampling analyses use the same data. Both are mathematically “reliable,” but only AHCCCS's corrects for OIG's unrepresentative sample that used 100 student-months with a mean value of $322.12 even though the other 528,443 student-months had a mean value of $233.88. The DAB admitted the sample was unrepresentative but dismissed the unrepresentativeness as “slight.” OIG could have used a ratio estimator analysis to correct for this disparity but did not. . . .

The questions the Response avoids are whether applying the lower bound of a range based on, and skewed by, an unrepresentative sample is reasonable under the circumstances when the result can be readily corrected and whether failure to make such a correction results in an overstated disallowance that is arbitrary and capricious.

(Id. at 9-10.)

C. Analysis

1. The Big Picture

It is important to begin by identifying several things that AHCCCS does not dispute. First, AHCCCS does not take issue with CMS's decision to engage in statistical sampling as part of the auditing and disallowance process. To the contrary, AHCCCS agrees that “[s]ampling is used as a necessary convenience to fairly approximate the result that would be obtained in a claim-by-claim analysis.” (Doc. 41 at 10.) Second, AHCCCS also agrees that, at least in the abstract, “OIG used a generally reliable methodology.” (Id. at 16.) Distilled to its essence, then, AHCCCS's argument is that DAB acted arbitrarily and capriciously by upholding CMS's sampling methodology, which was supported and explained by CMS's experts, instead of the even-better sampling methodology devised by AHCCCS's dueling experts.

Although AHCCCS's briefing does an admirable job of diving into the intricacies of some of the disputed statistical concepts at issue in this case, that briefing does not (at least as far as the Court can tell) identify a single case in which a federal court, applying the deferential arbitrary-and-capricious standard, has ever overturned an agency decision based on such a challenge. This is unsurprising. Although the Court's independent research indicates that a handful of such cases do exist, the overwhelming consensus is that federal courts should not (and, under the APA, may not) second-guess an agency's reasoned decision to follow a particular statistical sampling methodology, at least when (as here) the chosen methodology is rational and supported by expert testimony. For example, Louisiana ex rel. Guste v. Verity, 853 F.2d 322 (5th Cir. 1988), involved a challenge to an agency rule regarding the use of shrimp trawlers to reduce the incidental death of sea turtles. Id. at 328-29. The challengers argued, as AHCCCS argues here, that the agency improperly relied on certain statistical “extrapolations” that were based on “unscientifically small” samples. Id. The Fifth Circuit affirmed, explaining that “[a]lthough we believe appellants' challenge is not totally without merit, we are mindful that under the arbitrary-and-capricious standard, our deference to the agency is greatest when reviewing technical matters within its area of expertise, particularly its choice of scientific data and statistical methodology. In reviewing such technical choices, we must look at the decision not as the chemist, biologist or statistician that we are qualified neither by training nor experience to be, but as a reviewing court exercising our narrowly defined duty of holding agencies to certain minimal standards of rationality. Accordingly, where, as here, the agency presents scientifically respectable conclusions which appellants are able to dispute with rival evidence of presumably equal dignity, we will not displace the administrative choice.” Id. at 329 (cleaned up).

Similarly, Colorado Wild, Heartwood v. U.S. Forest Service, 435 F.3d 1204 (10th Cir. 2006), involved a challenge to an agency's forest salvaging rule on the ground that “the methodology employed by the [agency] in promulgating” the rule was arbitrary and capricious because the agency did “not use either (1) a mean analysis which excluded the largest salvage projects (i.e., removing them from the sample subset as statistical outliers) or (2) a median analysis.” Id. at 1214-16. The Tenth Circuit affirmed, explaining that “[f]rom our admittedly lay perspective and this record, [the agency's] rejection of the median because of the over-representation of small projects (and its focus on the acres of activity) does not seem irrational. We cannot substitute our views on statistics (including skewed data and outlier analysis) for those of the [agency] and insist that one measure or another be used.” Id.

And again, Superior Home Health Services, L.L.C. v. Azar, 2018 WL 3717121 (W.D. Tex. 2018), which involved a challenge to a Medicare disallowance decision, the plaintiff argued that “the Council erred in concluding as a matter of law that substantial evidence in the record supported the statistical sampling and extrapolation methodologies” and “made clear in its submissions to this Court that its expert witnesses would have employed a different methodology, perhaps even one that would have yielded a significantly more accurate result.” Id. at *4, *8. Despite those arguments, the district court affirmed, explaining that “it is not the function of this Court on judicial review to dictate which sampling and extrapolation methodologies must be used in administrative proceedings. What matters for the purpose of this appeal is the substantial evidence in the record supporting the Secretary's finding that the methodologies used . . . satisfied all relevant legal and administrative requirements. His decision pursuant to those findings was neither arbitrary nor capricious.” Id. at *8.¹⁶

16 See also Rio Home Care, LLC v. Azar, 2019 WL 1411805, *1 (S.D. Tex. 2019) (affirming Medicare disallowance decision, where the plaintiff argued “that the selection of the sample deviated from applicable statistical standards [and] an impermissible extrapolation method was used,” because “[w]hen considered in light of the deferential standard of review that applies, Rio has failed to show that the Council's decision upholding the overpayment determination lacks substantial evidence, that it is arbitrary and capricious, or that it is otherwise contrary to law. Although Rio relies on opinion evidence from its well-qualified expert statistician in challenging the sampling and extrapolation methodology, the Council's decision is supported by substantial evidence, including evidence from several qualified statisticians.”); Pruchniewski v. Leavitt, 2006 WL 2331071, *10-11 (M.D. Fla. 2006) (affirming Medicare disallowance decision, even though the plaintiff presented evidence that “the sampling process employed in his review was fundamentally unfair because it was based on an inadequate sample size and was unrepresentative” and the district court acknowledged that “if the standard of review was different and a better showing had been made, perhaps I would recommend a different result,” because “courts have approved a variety of sampling sizes against due process challenges . . . [and] despite the contrary views, the ALT s conclusion that the approach used by the Carrier was statistically valid is also supported by substantial evidence”); Zirkle Fruit Co. v. U.S. Dep't of Labor, 442 F.Supp.3d 1366, 1383 (E.D. Wash. 2020) (upholding agency's determination of a prevailing wage rate, where the agency relied on certain statistical sampling methods to compute the rate and the plaintiff argued that those methods were arbitrary and capricious, because the agency “made difficult decisions” and “applied reliable-if limited and imperfect-statistical models” and the challenger had “failed to show that [the agency] made sufficiently egregious statistical or mathematical errors to warrant setting that decision aside”).

By citing these cases, the Court does not mean to suggest that affirmance is required whenever a plaintiff seeks to challenge an agency's choice to utilize a particular statistical sampling methodology. Review under the APA is not a rubber stamp, and there may very well be circumstances where an agency's use of a particular methodology is arbitrary and capricious. But this is not such a case. DAB put much thought and effort into the underlying decision and provided an extensive discussion of why it agreed with CMS's expert's decision to use the particular sampling methodologies at issue (and to not use the competing sampling methodologies being proposed by AHCCCS's experts). This, alone, undermines AHCCCS claim for relief under the APA.

2. AHCCCS's Discrete Challenges

As explained below, the Court also concludes that DAB's conclusions as to some of the discrete, in-the-weeds methodological disputes raised by AHCCCS were not arbitrary or capricious. As noted, it not clear that this level of detail is even required in this context. Louisiana ex rel. Guste, 853 F.2d at 328-29 (“In reviewing such technical choices, we must look at the decision not as the chemist, biologist or statistician that we are qualified neither by training nor experience to be, but as a reviewing court exercising our narrowly defined duty of holding agencies to certain minimal standards of rationality. Accordingly, where, as here, the agency presents scientifically respectable conclusions which appellants are able to dispute with rival evidence of presumably equal dignity, we will not displace the administrative choice.”) (cleaned up). Nevertheless, in an abundance of caution, the Court provides it.

a. CMS's Burden

During the underlying proceeding, CMS had the initial burden to justify its disallowance. (Doc. 39 at 14; Doc. 40 at 23.) Puerto Rico Dep t of Health, DAB No. 2385, at 5 (2011) [https://perma.cc/35TR-6YJ9] (“Where a grantee challenges the use of statistical sampling, the Board looks to whether the agency has shown that sampling is appropriate in the context of the particular disallowance and whether the sampling and extrapolation methodology used to calculate that disallowance is scientifically valid.”); Mid-Kansas Cmty. Action Program, Inc., DAB No. 2257, at 4 (2009) [https://perma.cc/YB2F-NURV] (“[T]he agency has the burden of showing that sampling is appropriate to use in the context of the particular disallowance and that the sampling and extrapolation methodology used to calculate that disallowance is scientifically valid.”).

On that issue, DAB concluded:

CMS carried that burden based on the declarations of its expert, Dr. Smith, who has 10 years of experience working as a statistician and completed 60 quarter hours of graduate-level coursework relating to statistics. Citing authoritative literature in the field of statistical sampling and estimation, Dr. Smith credibly asserted that the OIG used valid procedures and methods to estimate the amount of FFP claimed for unallowable school-based health services during the 30-month audit period. Those procedures and methods included:

• identifying an appropriate sampling unit - the student-month;

• defining a finite target population (the sampling frame), consisting of nonoverlapping sampling units (student-months) that had an equal chance ofbeing selected;

• drawing a simple random sample from the target population using widely accepted statistical software (RAT-STATS) developed by the federal government;

• using a mean-per-unit estimator or a mean difference estimator, a statistic calculated based on the sample findings, to derive an “unbiased” point estimate of unallowable expenditures for schoolbased health services in the population; and

• “accounting] for the uncertainty of the unbiased point estimate” by calculating a two-sided 90 percent confidence interval around the unbiased point estimate.

(Doc. 27-2 at 7-8, record citations omitted.)

DAB's conclusion that CMS met its initial burden of validity was not arbitrary or capricious and is supported by substantial evidence. Dr. Smith's report explained that the 100-month sample was appropriate for “any population with finite values . . . . The proof does not require that each item in the population be indivisible.” (Doc. 27-4 at 4 ¶ 10.) He went on to explain that a simple random sample is composed of “sample units” that are “unique and are selected independently with equal probability.” (Id. at 9 ¶ 27.) To reinforce his opinion, Dr. Smith referenced Theorem 2.1 from a statistics textbook. William G. Cochran, Sampling Techniques 18 (3d. Ed. 1977) [https://perma.cc/TSQ9-KTED] (“Simple random sampling is a method of selecting n units out of the N such that every one of the . . . distinct samples has an equal chance of being drawn. The units in the population are numbered from 1 to N. A series of random numbers between 1 and N is then drawn, either by means of a table of random numbers or by means of a computer program that produces such a table. At any draw the process used must give an equal chance of selection to any number in the population not already drawn.”).

Additionally, the 100-student month sample was generated by RAT-STATS, as required by OIG's Office of Audit Services.¹⁷ (Doc. 27-4 at 3 ¶ 7.) To calculate the “unbiased”¹⁸ point estimate, again using the Sampling Techniques approach, OIG computed the “sample average” and then “multiplied [it] by the number of items in the population,” creating the unbiased point estimate of $34,368,773. (Id. at 4-5 ¶¶ 9-10, 14.) Then, to account for any “uncertainty of the unbiased point estimate,” OIG calculated the confidence limits on either side of the point estimate. (Id. ¶¶ 12-14.) See also Cochran, supra, at 27 (explaining how to calculate confidence limits using the mean of the sample). Finally, the lower bound of the confidence limit-here, $19,926,220-was used as the imposed disallowance. (Id. ¶¶ 13-14.)

17 HHS OIG, Office of Audit Services, RAT-STATS 2010 Companion Manual (Version 1), available at https://oig.hhs.gov/documents/rat-stats/833/CompManual201004js.pdf [https://perma.cc/GN3W-NCZF] (“This program generates unduplicated quantity of random numbers.”).

18 Unbiased means “on average across potential samples, it will not benefit CMS or a state when compared to a review of all items.” (Doc. 27-4 at 4 ¶ 9; Doc. 27-2 at 166.)

As explained, Dr. Smith's approach is well-supported by statistical literature. Moreover, the 90% confidence interval appears to be the standard in disallowance cases. See, e.g., P.R. Dep't of Health, DAB No. 2385, at 9 (“A 90% confidence interval means that there is a 10% probability that the true value of the error rate falls outside the confidence interval; or a 5% probability that the true value is greater than the upper limit or bound of the confidence interval, and a 5% probability that it is below the lower limit. Thus, since the disallowance was based on the lower limit of the confidence interval, and not the point estimate, there was a 95% probability that the true value was above the lower limit. [In other words,] the state was protected with a 95% degree of confidence from having to pay an amount greater than the true value of erroneous payments.”).

b. The Sample

AHCCCS also challenges the 100-month sample as unrepresentative because the average FFP paid per month was $322.12 in the sample, while the average for the entire population was $233.88. (Doc. 39 at 16-20.) DAB discussed, at length, why it found this challenge unavailing. (Doc. 27-2 at 11-14.) The Court concludes that this reasoning was not arbitrary or capricious and was supported by substantial evidence.

First, DAB found Dr. Smith's explanation of random sampling and statistics persuasive. (Id. at 11.) More specifically, Dr. Smith explained that a random sample is not always a perfectly representative slice of the population, but instead, other statistical mechanisms (like confidence intervals) can account for the differences in the sample. (Id.) Dr. Smith, acknowledging the discrepancy in the average sample calculations, explained:

This argument must fail if it is possible to use the sample to reliably estimate the very quantity [FFP-paid amounts] that [the State] claims is not represented. In fact, when OIG used the same [estimation] method it used for the refund amount [the FFP overpayment] to calculate paid amounts it obtained a 90 percent confidence interval ($122,949,342 to $217,560,575) that contained the actual paid amount in the population ($123,614,883). As expected, even though the average paid amounts are higher in the sample than in the population, the confidence interval calculated from the sample still captures the correct population total.

(Id. at 12.) DAB also credited Dr. Smith's explanation that the confidence interval can then “show the range of values that can be said to include the true population value with a pre-specified level of confidence (absent any bias).” (Id. at 13.) As Dr. Smith explained (and DAB accepted), not only does “the State [not] allege that the OIG's methods introduced bias into the sample results,” but it also failed to demonstrate that the differences between the “sample and the population” were due to anything other than “random selection.” (Id.) Therefore, Dr. Smith explained (and DAB accepted), the confidence interval provided a sufficient correction for the variation. (Id.)

Once again, regardless of whether this Court would have reached the same conclusions if asked the resolve this question in the first instance, DAB's resolution of the question was not arbitrary and capricious. This conclusion is not undermined by AHCCCS's contention that, as a general matter, an “unbiased estimate should deviate as little as possible from the ‘true' value one is trying to estimate,” and, to that end, “[a] random sample cannot ensure representativeness with respect to any characteristic” especially when one can “randomly select payments that are unusually high.” (Doc. 39 at 16.) DAB rejected these arguments on the ground that AHCCCS's assertions were based on unsupported portions of the EconLit report. (Doc. 27-2 at 18 [“Key assertions by EconLit lack foundation . . . .”].) This conclusion was not arbitrary and capricious and was supported by substantial evidence. Under the subheading “Is the Sample Representative?,” EconLit cited no authority for its conclusions, while simultaneously making broad pronouncements as to the significance of the non-representativeness of the sample. (Doc. 27-3 at 308-09.)

Likewise, AHCCCS relies heavily on EconLit's conclusion that OIG's simple random sample “overestimates the average student-month payment by 37.7%.” (Doc. 273 at 309; Doc. 39 at 17 [referring to the discrepancy as “too high, and prejudicial to Arizona”].) Although the Court can understand this argument as an intuitive matter, the report itself cites no authority that a 37.7% discrepancy is an unacceptable margin in the statistical field and, more important, does not explain why the confidence interval could not correct for it. The Court cannot say that DAB acted arbitrarily or capriciously in disagreeing with these unsupported assertions, particularly where the counterarguments it accepted were supported by the testimony of a qualified expert. The report of Dr. Al Kvanli (Doc. 27-3 at 184-89) explained that “the average FFP-paid amount in the sample was 1.64 deviations above the population mean for that parameter,” which was “slightly on the high side” but not “unusually so.” (Id. at 188.) DAB found that this conclusion was worthy of credence according to the Federal Judicial Center's Reference Manual on Scientific Evidence.¹⁹ Federal Judicial Center, Reference Manual on Scientific Evidence 244, https://www.fjc.gov/sites/default/files/2015/SciMan3D01 .pdf [http s://perma.cc/WF5K-UAG9] (“Random errors larger in magnitude than two or three times the standard error are unusual. Confidence intervals make these ideas more precise. . . . Those who say ‘within 2 standard errors' will be correct about 95% of the time.”). This reasoning was not arbitrary or capricious and was supported by substantial evidence.

19 “The Reference Manual on Scientific Evidence, Third Edition assists judges in managing cases involving complex scientific and technical evidence by describing the basic tenets of key scientific fields from which legal evidence is typically derived and by providing examples of cases in which that evidence has been used.” Federal Judicial Center, Reference Manual on Scientific Evidence, Third Edition, Research: Special Topics https://www.fjc.gov/content/reference-manual-scientific-evidence-third-edition-1 [https://perma.cc/3L5U-U8R3].

c. Precision

The parties agree that, under OIG's sampling structure, the precision percentage is 40%. AHCCCS argues that this rate is “so imprecise” as to be “statistically meaningless and therefore unreliable.” (Doc. 39 at 18.) CMS replies that the rate is less important here because the broad confidence interval (which is in turn a product of the imprecise point estimate) protects AHCCCS against any imprecision in the point estimate; in other words, the point estimate itself is not what is being imposed as the disallowance amount. (Doc. 40 at 6-7.) In reply, AHCCCS seems to argue that although the math is technically correct, it was unreasonable to use under the circumstances here. (Doc. 41 at 10-11 [“The Response argues CMS did not have to ‘produce the most precise, efficient, or best method, but rather a valid and reliable method.' This is only half-true. The DAB has held, ‘[A]s the proponent of the [sampling] method, [the government] has the burden of showing not only that its sampling methodology ‘produces reliable evidence to support the amount of the disallowance' but also that it is ‘reasonable under the particular circumstances.'”].)

DAB credited CMS's explanation:

“Precision” is an attribute of the point estimate. It is the degree to which that estimate varies across potential samples; in other words, precision captures the uncertainty inherent in the point estimate. . . .

In his declaration, CMS's expert, Dr. Smith, acknowledged that “[b]etter precision can be achieved by pulling larger samples” (as well as by using more complex sampling procedures or different methods of measuring the population characteristic of interest). Dr. Smith also acknowledged that ‘[p]recision and the general match between the sample and the population is an important consideration when relying on a point estimate.” However, Dr. Smith emphasized that precision is less salient in this case because the estimate supporting the disallowance is not the unbiased point estimate but, rather, the lower limit of the confidence interval around that estimate. Dr. Smith explained that any imprecision in the point estimate arising from “sample design and choice of estimation method” is “account[ed] for” in calculating the confidence interval's lower bound, and that “[b]y design, the level of assurance provided by the confidence interval is not related to the precision of the point estimate.”

In response to these statements, the State did not (in its reply or sur-surreply briefs) press its initial claim that the sample size was too small.²⁰ Nor did

the State dispute Dr. Smith's opinion that imprecision in the point estimate is accounted for in calculating the confidence interval's lower limit and “does not impact the confidence associated with the lower limit that CMS is using for the overpayment calculation in this case.” Indeed, Petitioner conceded that the point estimate's precision has “nothing to do with whether OIG's lower bound is correct.” Furthermore, the Board has recognized in other cases that the point estimate's imprecision tends to benefit the grantee when the disallowance is based on the lower bound of the relevant confidence interval. Accordingly, we find that the size of the sample in this case did not render the population estimate supporting the disallowance invalid, and that any imprecision in the point estimate likely benefitted the State because it widened the confidence interval whose lower bound supports the disallowance.

(Doc. 27-2 at 9-10, internal citations and record citations omitted.)

20 In its opening brief, AHCCCS clarifies that “[t]here was no ability to draw another, larger sample for comparison” so it chose to criticize “OIG's analysis and conclusions drawn from the sample.” (Doc. 39 at 18 n.3.) The Court interprets this to mean that AHCCCS is not specifically contesting the sample size. In any event, had that issue been squarely raised here, AHCCCS has failed to show that DAB's decision was arbitrary and capricious. P.R. Dep't of Health, DAB No. 2385, at 10 (“While a larger sample may have resulted in a higher confidence interval, PRDH provides no reason to believe that the benefit to PRDH (or HRSA) would have justified the additional costs of conducting a 600 sample prescription audit.”). See also Reference Manual on Scientific Evidence, supra, at 246 (“There is no easy answer to this sensible question. Much depends on the level of error that is tolerable and the nature of the material being sampled.”).

Once again, this analysis was not arbitrary and capricious and is supported by substantial evidence. EconLit's report, which cited a textbook written by Dr. Kvanli, asserted that as a general principle, “[t]he narrower your confidence interval, the better, for the same level of confidence.” (Doc. 27-3 at 193.) To illustrate this principle, Dr. Kvanli's textbook explains that a range between 2 and 50 minutes is “practically worthless” compared to a range between 25 and 27 minutes. (Id. at 193-94.) Dr. Kvanli's textbook does not, however, provide support for the proposition that any larger range is “statistically meaningless” (as opposed to simply less precise). More important, as DAB explained, it is generally accepted that lower precisions, which typically result from low sample sizes due to efficiency concerns, tend to benefit the audited party because the wider the gap confidence level, the more generous its upper/lower bounds. (Doc. 27-2 at 6.) The Court- which, it cannot be overstated, is not sitting as a factfinder choosing de novo between two experts but is instead reviewing DAB's decision under the APA's deferential standard of review-cannot say this reasoning was arbitrary or capricious.

d. Cluster Sampling Versus Ratio Sampling

AHCCCS also argues that “academic literature requires a non-representative sample to be evaluated as a cluster sample using a ratio estimator and states that it is especially important to analyze the sample as a cluster sample rather than a simple random sample when the values of the numerator and the value of the denominator in the ratio (here, improper payments over claimed payments) are correlated.” (Doc. 39 at 20, emphasis added, citing Doc. 27-3 at 235.)

In response, CMS argues that “the State fails to cite any legal authority to back up this manufactured burden. Indeed, by this logic it would be the agency's burden in any disallowance to explain why it declined to use one of multiple alternative sampling methodologies, which would be an onerous and illogical burden to place on the agency without supporting legal authority.” (Doc. 40 at 24.) CMS further argues that “the State fails to provide the academic literature that supports its argument and fails to prove the mathematical assertions upon which its argument relies. The DAB reviewed the pages of academic literature cited by the state and found that none of them mentions nonrepresentative samples or indicates that a ratio estimator must be used to address potential differences between the sample and the population.” (Id. at 28, citing Doc. 27-2 at 19.) As a final matter, CMS argues that AHCCCS failed to produce mathematical data showing either “that the ratio estimate is unbiased [or] that the ratio estimate is more precise than the difference estimate.” (Id. at 28-29.)

In reply, AHCCCS argues that its math was substantiated in the EconLit report. (Doc. 41 at 13-14.) As for the academic literature, AHCCCS explains that “one does not need a text to know the analysis must be corrected if possible. The texts make clear that a ratio estimator is not affected by an unrepresentative sample. Thus, using the ratio estimator was required, if OIG wanted to use this sample and not skew the resulting disallowance.” (Id. at 14.) AHCCCS concludes that “the DAB simply rubber-stamped the disallowance because OIG used a generally reliable methodology . . . though the uncorrected, skewed sample increased the disallowance by $8,000,000.” (Id. at 16.)

In addressing this issue, DAB noted that “[t]he State's third main criticism assumes that the second is well-founded.” (Doc. 27-2 at 15.) It then explained:

Also unfounded is the State's principal claim that a ratio estimator must be used to correct for a non-representative sample. EconLit cited the following literature to support that claim: Cochran, William G., Sampling Techniques (3rd ed. 1977) at 249; Schaeffer, Richard L., et al., Elementary Survey Sampling (3rd ed.) at 205-06; and Kish, L., Survey Sampling at 204. CMS provided copies of these cited pages as attachments to Dr. Smith's second declaration. As Dr. Smith observed, none of the cited pages mentions nonrepresentative samples or indicates that a ratio estimator must be used to address potential differences between the sample and the population. Those pages, Dr. Smith stated, merely confirm that “precision improvements [in the point estimate] are possible when there is a close relationship between the available auxiliary information [FFP paid for a student-month] and the quantity of interest [FFP paid in error].'

In addition, the State failed to substantiate its assertion that the ratio estimator is unbiased for the population of student-months defined by the OIG. The parties apparently agree that a ratio estimate is unbiased if two conditions (specified in Cochran, Sampling Techniques (3rd ed. 1977)) are met: (1) the “relationship between the paid amounts and error amounts is a straight line through zero”; and (2) “the variance of the paid amounts is proportional to the variance of the error amounts.” EconLit asserted that “[b]oth of these conditions are true for the population in the instant matter,” but we see nothing in the record to back up that assertion. EconLit did not, in particular, provide statistical or other quantitative analysis showing that the relationship between FFP payments and FFP overpayments is a “straight line” - that is, a relationship of direct proportionality or perfect correlation - nor did EconLit demonstrate that the variance of paid amounts is proportional to the variance of payment-error amounts. EconLit asserts in its Rebuttal Report that it provided evidence of “high correlation” - a strong linear relationship -between “improper payments and claimed payments.” But merely “high” correlation does not appear to satisfy the condition set out in the Cochran textbook, which is that the relationship be one of perfect correlation (a straight line). Furthermore, EconLit did not, so far as we can determine, measure the correlation between a student-month's improper (disallowed) and claimed payments. Instead, EconLit measured the “correlation between the percent of reimbursed dollars . . . disallowed in a student-month and the total number of dollars reimbursed” in the month; found that these two variables were “effectively not correlated” (with a correlation coefficient of negative 0.068); and then concluded - without supporting or reasoning - that this relationship confirms that, “on average, disallowed payments are a fixed percent of claimed payments.”

(Id. at 18-20, record citations omitted.)

This reasoning was not arbitrary or capricious and is supported by substantial evidence. For example, AHCCCS argues that “Cochran's ‘Sampling Techniques points out that when the number of clusters in the population is known (which is true in the instant matter), it is preferable to analyze the sample as a cluster sample rather than a simple random sample.'” (Doc. 39 at 24, citing Doc. 27-3 at 235.) The basis for that conclusion is EconLit's report, which cites Cochran at page 249. (Doc. 27-3 at 235.) But it was not arbitrary or capricious for DAB to conclude that this particular page of Cochran does not support EconLit's contention. Indeed, the page explains there are “numerous methods of sample selection and estimation that have been produced for cluster units of unequal sizes” and that in some cases using the “population mean per cluster unit” has been “found to be of poor precision.” (Doc. 27-4 at 102.) Perhaps more important, DAB did not simply rely on its own independent review of the dense concepts, statements, and mathematical formulas set out in the statistical textbooks cited by AHCCCS but instead relied in part on the interpretation of those textbooks provided by CMS's expert, Dr. Smith, who avowed in one of his reports that Cochran's textbook and “[o]ther finite sampling textbooks . . . do not require or recommend the type of representative testing performed by AHCCCS.” (Doc. 27-4 at 12 ¶ 36.) DAB's decision to credit Dr. Smith's expert viewpoint on this complicated topic is precisely the sort of conclusion that cannot be second-guessed by courts applying the APA's deferential standard of review.²¹

21 The same analysis applies to Schaeffer and Kish, also cited in the EconLit report. For example, nothing in the Schaeffer text suggests that a cluster evaluation is required when the population is known. (Doc. 27-4 at 104 [“Often the number of elements in the population is not known in problems for which cluster sampling is appropriate.”].) Kish provides general background on the best applications of a ratio estimator but does not explain how it is superior to other methods, much less required in the scenario here. (Id. at 103 [“A few examples can illustrate some uses of ratio estimates. From Surveys of Consumer Finances [1950], mean liquid assets per household were obtained; these can be expanded with the Census Bureau's count of households to provide estimates of aggregate liquid assets. From the same surveys, estimates of assets per income were also obtained, and these can also be expanded with income data from outside sources.”].)

Accordingly, IT IS ORDERED that:

1. CMS's motion for summary judgment as to equitable tolling (Doc. 40) is denied.

2. AHCCCS's merits challenge to DAB's decision (Doc. 39) is denied.

3. DAB's decision (Doc. 27-2 at 1-26) is affirmed.

4. The Clerk shall enter judgment accordingly and terminate this action.

Dated this 20th day of July, 2023.