The remand instructions in St. Mary's I were as follows:

[W]e remand this case for the limited purpose of the PRRB's taking evidence on the question whether the number of Medicare patients found nationally in other ancillary areas at the census hour is sufficient to offset the dilution of Medicare reimbursement created by counting labor/maternity patients in the routine inpatient count.

718 F.2d at 474. The court restated the scope of this remand a few paragraphs later:

We remand the case to the District Court with instructions to remand to the PRRB for the limited purpose of taking evidence on the issue of whether the use of other ancillary services by Medicare beneficiaries at the census-taking hour suffices to compensate for the dilution of Medicare reimbursement caused by including labor/delivery area patients in the calculation of average general routine costs per diem.

Id.

On remand, HHS has stated repeatedly before the District Court that it can produce no evidence on the ancillary -service utilization discussed in the remand instructions. St. Mary of Nazareth Hospital v. Heckler, 587 F. Supp. at 938, J.A. at 8. The government has, however, requested the District Court to remand the case to the PRRB for the taking of evidence on the costs of routine care. HHS asserts that it can demonstrate that its treatment of labor/delivery patients results in a reimbursement system consonant with Congress's mandate in 42 U.S.C. Sec. 1395x(v)(1)(A) (1982).

To understand why the District Court properly decided the case in favor of the hospitals' position, we must review at some length both the Medicare reimbursement scheme and this panel's opinion in St. Mary's I.

Hospitals are reimbursed for their Medicare expenses in three separate categories: general routine care, special routine care, and ancillary care. Routine care is for those services for which a separate fee is not generally charged, such as the room, food, and routine nursing services. 42 C.F.R. Sec. 405.452(b), at 149 (1984). General routine care is for routine services in the ordinary beds of the hospital. Special routine care is for routine services in areas such as intensive care or coronary care, where the costs of routine care are likely to be much greater than in general routine-care areas and where Medicare agrees that the area provides "special care." See id.; 42 C.F.R. Sec. 405.452(d)(10) (1980); see also Villa View Community Hospital, Inc. v. Heckler, 728 F.2d 539 (D.C.Cir.1984) (per curiam) (upholding Secretary's determination that unit for cardiac care not requiring bedside monitor was not a special care unit). Ancillary care is for services, such as surgery or x-rays, for which a separate fee is charged. 42 C.F.R. 405.452(b), at 148. Labor/delivery services are an ancillary charge.

Medicare reimburses hospitals for general routine care according to the following formula:

                   Total cost of routine services       Medicare
Reimbursement  =  -------------------------------  x
                  Total number of inpatient days      inpatient days

Under the challenged policy, the number of inpatient days included patients in ancillary-care areas at midnight (the "census-taking hour") of the counted day as well as patients in routine-care areas. Included among such inpatients were those in the labor/delivery area at midnight, just as all other patients in ancillary areas at midnight were counted. Excluded from routine costs were the costs of providing labor/delivery services, just as all ancillary costs were excluded from the calculation of routine costs. Ancillary costs were reimbursed separately.

As long as inpatients counted in the denominator generate the services counted in the numerator, Medicare's formula is a sensible one. The fraction--total cost of services divided by days of service--represents the average cost per inpatient day of routine services. When this average daily cost is multiplied by the number of Medicare inpatient days, the result would be the cost of Medicare patients' routine services.

Counting patients in ancillary-care areas as inpatients for the purposes of calculating routine costs does not ordinarily distort the reimbursement. The fact that a patient happens to be in an ancillary-care area at midnight is usually nothing more than an artifact of the particular time when the patient receives ancillary services. Such a patient was receiving routine services before the receipt of ancillary care and will receive them afterwards, and thus contributed to routine costs. The ancillary costs generated by the patient are reimbursed separately.

The hospitals argued, however, that including all labor/delivery patients in the labor/delivery area at midnight as routine inpatients distorted the hospitals' reimbursement for the care of Medicare patients. Excluding labor/delivery services from the numerator but not the denominator of the fraction in the formula above results in a hospital's receiving less reimbursement for treating Medicare patients than it otherwise would: the fraction is smaller than it would otherwise be, and therefore the fraction multiplied by the number of Medicare patients yields a smaller total than it otherwise would. The hospitals' interest in challenging the regulation is therefore clear.

The claimed "dilution" of Medicare reimbursement--and thus the illegal subsidy of Medicare patients by non-Medicare patients--that results from counting labor/delivery services only in the denominator of the fraction in the formula above occurs with all ancillary services. All patients receiving ancillary care at the census-taking hour are counted as inpatients for the purpose of calculating daily routine costs, under the assumption that patients in ancillary areas at midnight are receiving care ancillary to their receipt of routine services earlier in the day. Ancillary costs are uniformly excluded from the routine-costs total and reimbursed instead in a separate calculation because a separate charge exists for them by definition.

In the case of labor/delivery services, however, we ruled in St. Mary's I that the dilution is a "concrete violation" of 42 U.S.C. Sec. 1395x(v)(1)(A). Because the fundamental assumptions underlying the general reimbursement scheme do not obtain in the case of labor-delivery patients, HHS's reimbursement method forces labor/delivery patients to subsidize the routine costs of other patients. Because labor/delivery patients are disproportionately non-Medicare patients, HHS's reimbursement scheme raises the presumption that non-Medicare patients are forced to subsidize the routine costs of other patients, in violation of 42 U.S.C. Sec. 1395x(v)(1)(A). Because HHS has on remand been unable to rebut that presumption, except by offering evidence in support of post hoc justifications for its treatment of labor/delivery patients that are inconsistent with its other reimbursement policies, we hold its treatment of labor/delivery patients invalid.

The most important difference between the assumptions of the Medicare reimbursement scheme and the reality of rendering labor/delivery services is that labor/delivery patients generally receive ancillary care without receiving any routine care:

[T]he hospitals have demonstrated that the class of patients residing in the labor/delivery area at the census-taking hour have received no routine care, and will not necessarily receive routine care. This is not a case of hypothetical examples at the extremes, therefore, and we need not guess as to whether extreme cases somehow are balancing one another out. We know for a fact that no routine care has been provided to these patients, but routine care costs are being attributed to them.

St. Mary's I, 718 F.2d at 472 (citations omitted). Many women who use the labor/delivery room come there directly from the outside world, whereas patients who receive most other ancillary services have already received routine care. Id. The assumption in the reimbursement scheme that ancillary patients have already received routine care, and thus can fairly be included as inpatients receiving routine care for the purposes of the Medicare reimbursement scheme, therefore does not hold for many labor/delivery patients.2

The court's awareness of, and reliance upon, this general distinction between labor/delivery services and other ancillary services is clear in St. Mary's I not only from our discussion of the issue but from the phrasing of our holding:

If there is no [offset in the form of a disproportionate number of Medicare patients in non-labor/delivery ancillary services] or if the Secretary decides not to contest the issue, then the Secretary may not count labor/delivery patients, who have not previously that day received routine care, among the routine inpatient population for purposes of calculating the average cost per diem for routine services.

....

.... Absent substantial evidence to support [a disproportion of Medicare patients in non-labor/delivery ancillary services], the Secretary is directed to exclude labor/delivery room patients, who have not previously that day received routine services, from the inpatient count used to derive the average cost per diem for general routine services.

718 F.2d at 474 (emphasis added).3 This holding is not, as the government has argued, a general condemnation of Medicare's reimbursement scheme. The court's rationale does not hinge on whether the overall scheme of reimbursement is balanced, and evidence on that broad issue is inappropriate. We held rather that HHS cannot include labor/delivery patients as contributing to routine costs when those patients have not done so, at least when this inclusion would result in subsidization of Medicare patients' routine costs by non-Medicare patients.

The fact that labor/delivery patients are included as routine inpatients though they receive no routine care does not, however, by itself make even a prima facie case that the Medicare reimbursement scheme improperly shifts costs to non-Medicare patients. The relevant statutory provision on cost-shifting requires that

under the methods of determining costs, the necessary costs of efficiently delivering covered services to individuals covered by [Medicare] will not be borne by individuals not so covered, and the costs with respect to individuals not so covered will not be borne by such insurance programs ....

42 U.S.C. Sec. 1395x(v)(1)(A) (1982). Because HHS's reimbursement scheme includes labor/delivery patients in the category of routine inpatients even if the labor/delivery patients have received no routine care, HHS requires labor/delivery patients to subsidize the routine care of other patients. The relevant concern under 42 U.S.C. Sec. 1395x(v)(1)(A), however, is whether HHS's reimbursement scheme requires non-Medicare patients to subsidize the routine care of Medicare patients. If Medicare patients used labor/delivery services in the same proportion that they used other services, then labor/delivery patients would pay a disproportionate share of routine costs, but non-Medicare patients would not be paying a disproportionate share. No violation of 42 U.S.C. Sec. 1395x(v)(1)(A) would occur.

But there was persuasive evidence in St. Mary's I that there is a much smaller proportion of Medicare patients in the labor/delivery area than in other ancillary areas, and thus that the disproportionate burden borne by labor/delivery patients was translated into a disproportionate burden for non-Medicare patients. First, the hospitals presented evidence that a high percentage of those who used labor/delivery services were in the labor/delivery area at midnight.4 Id. at 472. Second, the hospitals presented evidence that those in the labor/delivery area were almost all non-Medicare patients.5 Id. These characteristics of labor/delivery services, taken together with the fact that many labor/delivery patients have received no routine care, imply that non-Medicare patients who have received no routine care will be disproportionately counted at midnight. That disproportionate counting of non-Medicare patients, if uncompensated for by a disproportionate counting of Medicare patients in other ancillary areas, unbalances the compensation of routine costs as between Medicare and non-Medicare patients.

Faced with the arbitrariness of Medicare's decision to include all labor/delivery patients as routine inpatients, and with evidence that labor/delivery patients were disproportionately non-Medicare patients, the court held in St. Mary's I that HHS's policy of including labor/delivery room patients who had received no routine services as routine inpatients was a "concrete violation" of HHS's statutory mandate in 42 U.S.C. Sec. 1395x(v)(1)(A). 718 F.2d at 474.

HHS had in St. Mary's I asserted in its brief, however, that this apparent violation of 42 U.S.C. Sec. 1395x(v)(1)(A) was balanced by the presence at the census-taking hour of patients in non-labor/delivery ancillary areas who had also received no routine services. Brief of Respondents in Nos. 82-1034, et al., at 30. This argument is not meritless on its face. Patients in the emergency room, for example, might be disproportionately Medicare patients, might generally come directly from the outside world to that ancillary-care area, and might remain in the emergency room until the census-taking hour without receiving any routine care. It is conceivable that the disproportionate number of Medicare patients in non-labor/delivery areas who receive no routine care might balance out the non-Medicare patients in the labor/delivery area who receive no routine care. If that were so, the Secretary's policy of counting patients in ancillary-care areas as routine inpatients would not subsidize Medicare patients at the expense of non-Medicare patients. It would be inaccurate with respect to some patients, but not inaccurate with respect to patients' Medicare/non-Medicare affiliation.

The remand instructions in St. Mary's I gave HHS the opportunity to show a disproportionate counting of Medicare patients in non-labor/delivery ancillary areas:

[W]e remand this case for the limited purpose of the PRRB's taking evidence on the question whether the number of Medicare patients found nationally in other ancillary areas at the census hour is sufficient to offset the dilution of Medicare reimbursement created by counting labor/maternity patients in the routine inpatient count.

718 F.2d at 474. HHS has stated repeatedly that it can make no such showing. St. Mary of Nazareth Hospital v. Heckler, 587 F. Supp. at 938, J.A. at 8. Since the prima facie violation of 42 U.S.C. Sec. 1395x(v)(1)(A) remains unrebutted even after HHS has now been expressly informed of the legal basis for its inadequacy, the policy in question becomes conclusively invalid. The District Court was correct to hold that the Secretary could no longer count as routine inpatients labor/delivery patients who had received no routine care during the previous day.

On remand, the government raised an argument not advanced in the briefs in St. Mary's I or in its petition to this court for rehearing. The government now argues that it should be allowed to present evidence on the overall balance of routine -cost reimbursement because that balance is the crux of the issue here. They argue that, if they can show Medicare contributes an amount to hospitals roughly equal to its average routine costs, then the routine-care reimbursement scheme cannot violate 42 U.S.C. Sec. 1395x(v)(1)(A). They propose to show that labor/delivery patients generate higher routine costs, once they do receive routine care, than the average patient. Counting labor/delivery patients as routine inpatients even when they receive no routine services, in other words, makes up for the later, disproportionately high costs labor/delivery patients generate when they do receive routine services.

There is some broad language in portions of St. Mary's I besides the remand order that arguably supports the government's case for a wide-ranging evidentiary inquiry on remand:It is irrational to apportion to labor/maternity patients costs which they have not incurred without either including the costs that they have incurred or demonstrating that the distortion is balanced by some other aspect of the accounting process. Because labor/delivery area costs are not included in the calculation of the average routine cost per diem and because there is no evidence that this imbalance is made up elsewhere, non-Medicare payors are forced to bear some of the costs of the Medicare program.

718 F.2d at 473-74 (emphasis added). The government essentially argues that these general statements of what Medicare has failed to do also define exactly what HHS can do on remand--that is, show that the imbalance is made up somewhere in the necessarily vast expanse of Medicare reimbursement.

Our holding in St. Mary's I, however, was based on the inability of HHS to come up with an affirmative justification for its policy on labor/delivery patients rather than on an overall evaluation of routine-care costs.6 We intended to allow HHS to demonstrate a balancing of the distortion by some other aspect of the accounting process. We intended to allow HHS to show that the subsidy of Medicare patients by non-Medicare patients caused by applying Medicare's general policy to labor/delivery patients was balanced by an opposite and roughly equal subsidy of non-Medicare patients by Medicare patients caused by applying Medicare's general policy to non-labor/delivery services. We allowed this possibility because HHS made this argument in its brief in St. Mary's I, but was denied the opportunity to present evidence to support it before the PRRB owing to the fact that the intermediaries, rather than HHS itself, opposed the hospitals before the PRRB. 718 F.2d at 474. HHS's position on this appeal, however, is that the ancillary-care balancing argument is simply an example of the kind of wide-ranging assertions that this panel permitted the government to attempt to prove before the PRRB.

But this case, from its inception, has never presented the court with an opportunity to review the panoramic equity of the Medicare reimbursement scheme. The issue has always been whether a particular decision by HHS--the decision to count labor/delivery patients who have received no routine care as routine inpatients--is arbitrary, capricious, or otherwise not in accordance with the law. See 5 U.S.C. Sec. 706 (1982). HHS must come up with some affirmative rationale to support that decision. St. Mary's I explored possible rationales at great length but found none satisfactory. 718 F.2d at 466-74. St. Mary's I held that HHS's policy, as applied to labor/delivery services, violated 42 U.S.C. Sec. 1395x(v)(1)(A) unless HHS could show that the application of this same policy to other ancillary patients compensated for the imbalance generated by its application to labor/delivery patients, who are disproportionately non-Medicare patients. HHS has said, however, that it can make no showing regarding ancillary-care utilization by Medicare and non-Medicare patients.

For the court to consider the overall routine-care reimbursement balance, regardless of the rationality of counting labor/delivery patients who have received no routine care as routine inpatients, would be for the court to fail to hold HHS to the deferential but meaningful standards of judicial review of its decisions. The definitions that HHS uses to run the reimbursement scheme must bear some relation to reality. HHS could not attribute to labor/delivery patients all of a hospital's costs incurred in building a new wing for elderly patients and then claim that attribution was rational because the overall amount contributed by non-Medicare patients to all reimbursements was roughly equal to their share of overall costs. Similarly, HHS should not be able to attribute to labor/delivery patients costs of routine care that they never receive simply because HHS asserts that it can show the result is that non-Medicare patients contribute roughly the same amount of money that they generate in routine costs.

HHS's argument is somewhat more specific than this, of course, in that it alleges its policy compensates for higher routine costs generated by labor/delivery patients once they do receive routine care. This is, however, a prohibitively new tack for HHS.7 See Burlington Truck Lines, Inc. v. United States, 371 U.S. 156, 168-69, 83 S. Ct. 239, 245-46, 9 L. Ed. 2d 207 (1962) (agency may not justify its actions by advancing post hoc rationalizations of counsel). The agency did not raise this point in St. Mary's I or in its petition for rehearing. In addition, HHS has long expressly assumed in its own regulations that the average routine costs per day are the same for each patient. See 42 C.F.R. Sec. 405.403(d) (1980), J.A. at 192. And if HHS had come to doubt the validity of its own assumption with respect to a particular sort of care, there was a mechanism available to account for higher routine-care costs: the designation of an area as "special care," see supra p. 1314. HHS did not attempt to employ that mechanism here. Indeed, even when the differential in the costs of routine care is greater than it is here, HHS has refused to employ this mechanism when asked by others to do so. Compare Sun Towers, Inc. v. Schweiker, 694 F.2d 1036 (5th Cir.1983) (averaging upheld despite differential of over $15 per patient-day) with Affidavit of J. Michael Fitzmaurice, J.A. at 127, 128 (differential of $4.31 per patient-day between maternity unit and non-maternity care).

Finally, the government asserts that Walter O. Boswell Memorial Hospital v. Heckler, 749 F.2d 788 (D.C.Cir.1984), implies that its view of the appropriate evidence to be considered on remand should prevail. Appellant's Reply Brief at 9. We must disagree.

Boswell involved a challenge to an HHS rule that changed Medicare's reimbursement of general and administrative (G & A) costs such as record-keeping costs and insurance premiums. Under the old scheme, HHS reimbursed hospitals for the proportion of their G & A costs equal to the hospital's proportion of services provided to Medicare patients, much as HHS reimburses hospitals for routine costs. Under the new scheme, HHS proposed to retain the old method of reimbursement for all G & A costs except malpractice insurance premiums. HHS claimed that the change was necessary to prevent Medicare patients from subsidizing the malpractice insurance premiums of non-Medicare patients.

The Boswell court remanded the case to the District Court on procedural grounds, id. at 792-94, but went on to discuss the proper substantive standards for the District Court to apply on remand. Of most asserted relevance to the case now before us is the standard that HHS must meet to justify subdividing a class of expenses:

For such an action to be consistent with the Medicare Act's prohibition of cost-shifting, the Secretary must have implicitly determined that the old G & A pool, taken as a whole, had come to subsidize non-Medicare patients at the expense of Medicare patients. Without such a determination, there would be no legitimate grounds for removing malpractice premiums from the general G & A pool, since the old G & A pool had long been presumed to balance costs fairly between Medicare and non-Medicare patients. Given no net cost-shifting in the old pool and the removal of Medicare-subsidized malpractice premiums, the new G & A pool would inevitably subsidize Medicare patients at the expense of non-Medicare patients.

Id. at 795. Similarly, the court in Boswell stated:

It would be arbitrary and capricious for HHS to bring varying interpretations of the statute to bear, depending upon whether the result helps or hurts Medicare's balance sheets, without some showing that change in the underlying balance of G & A expenses, taken as a whole, warranted breaking out malpractice premiums from the G & A pool.

Id. at 799 (footnote omitted). HHS argues that, in this case, it is simply asking for the opportunity to show that the general pool (in this case general routine costs rather than G & A costs) is properly balanced. HHS argues that, if it can make that showing, then creating a new category of particular costs (in this case labor/delivery services rather than malpractice premiums), as the District Court mandated here, is prohibited by Boswell.

There is, however, one crucial difference between Boswell and this case. In Boswell, no one disputed that malpractice premiums had initially been properly included in G & A costs. It was clear that both Medicare and non-Medicare patients benefitted from malpractice insurance and generated premium costs. The issue was how properly to apportion those shared costs between Medicare and non-Medicare patients. Here, HHS is attempting to apportion the costs of a service (routine care) to non-Medicare patients when those patients do not benefit from the service at all. We ruled in St. Mary's I that labor/delivery patients, who are predominantly non-Medicare patients, should not be forced to pay such costs unless patients in other ancillary-care areas receiving no routine care also contributed to such costs. The difference in kind between Boswell and this case outweighs the structural similarities between them that arise in any attempt to determine the proper level of aggregation to use in reimbursing hospitals for the costs of treating Medicare patients.

Affirmed.