Judge JACOBS dissents in a separate opinion.
POOLER, Circuit Judge.Nickolas Zervos suffers from metastatic breast cancer, an essentially incurable condition that often can be held at bay temporarily either by repeated cycles of conventional chemotherapy or by a single cycle of high dosage chemotherapy followed by an autologous blood stem cell transplant (“HDCT”). On his physician’s advice and after being warned that HDCT often involves more — and more severe — side effects than conventional chemotherapy, Zervos elected to use HDCT because he believed it gave him a chance at a longer period of remission. However, Empire HealthChoice, Inc. (“Empire”), the carrier for Zervos’ health care coverage from his employer, Verizon New York, Inc. (“Verizon”), declined to pre-certify Zervos’ eligibility for coverage for HDCT, finding that the treatment was investigational and experimental. As part of its internal appellate process, Empire referred Zervos’ case to an outside reviewer. Although that reviewer found that HDCT was effective for treatment of Zervos’ condition and that the clinical trial in which he would be treated was well-designed, Empire again denied coverage — this time because HDCT had not been proven to be superior to conventional chemotherapy.
Zervos began this litigation in January 2001 and concomitantly requested a preliminary injunction. The district court denied Zervos’ request for injunctive relief and we affirmed. Zervos v. Verizon New York, Inc., 252 F.3d 163 (2d Cir.2001)(‘‘Zervos II”). In affirming, we noted that Zervos might be able to establish that the administrator applied the health insurance plan’s “Experimental/In-vestigational” exclusion erroneously on appeal by defining it in a manner not supported by the language of the plan. Id. at 174. On remand and after a bench trial, the district court found that Empire’s appellate process was arbitrary and capricious and remanded to Empire, directing that a different outside consultant review Zervos’ eligibility. Because the denial of benefits based on the record at the time of the remand was unreasonable, and the remand did not account for the fact that each day was crucial in this case, we find that the remand was an abuse of discretion and reverse and remand for the entry of an injunction requiring coverage.1
*639BACKGROUND
1. Request for Treatment
In February 2000, Zervos discovered a lump in his right breast. A biopsy demonstrated that the lump was malignant, and the subsequent mastectomy at Parkway Hospital indicated that the cancer had spread to twelve out of the twelve lymph nodes examined. After further testing, Zervos learned that the malignancy probably had infiltrated his sternum.2 Based on these findings, Parkway referred Zervos to Columbia Presbyterian Medical Center for more aggressive chemotherapy. On October 13, 2000, Dr. Charles S. Plesdorffer, Zervos’ physician at Columbia Presbyterian, requested insurance approval to proceed with HDCT following a protocol that he attached. Zervos already was receiving standard chemotherapy in the form of Adriamycin and Cytoxan, which he “tolerated well.” He subsequently responded to this conventional chemotherapy and to a course of radiation, and his cancer went into remission. Zervos v. Verizon New York, Inc., No. 01-CIV-685, 2001 WL 253377, *4 (S.D.N.Y. March 14, 2001) (“Zervos I ”).
II. Relevant Policy Provisions
The policy covering Zervos provided coverage for “medically necessary” treatments, defined as treatments that “are required for the necessary treatment of Injury, Illness, or pregnancy, as distinct from those which are unnecessary or Experimental/Investigational.” Under the policy, “Experimental/Investigational” means “services or supplies which are not of proven benefit for the diagnosis or treatment of the Covered Person’s condition, or are not generally recognized by the medical community as effective or appropriate for that condition, as determined by the Claims Administrator.”
Dr. Steven Wolinsky, Empire’s medical policy director, testified that he himself decided whether a treatment was experimental or investigational “based on literature, research and evidence,” or, at his discretion, he referred the decision to Empire’s Technology Assessment Committee (“TAC”). The TAC consists of Empire medical personnel and academic physicians from the New York City or Albany area. In determining whether a procedure is of “proven benefit,” within the meaning of the policy, Wolinsky or the TAC considers “the efficacy [of the treatment] balanced by the toxicity and whether that net positive health effect compares favorably with other standard treatments.” Toxicity, in Wolinsky’s view, includes any side effects of the treatment from the most minor to the most toxic.
The policy also provides for an internal appeal, the resolution of which is committed to the discretion of Empire, as Appeals Administrator. Empire refers these appeals to an external reviewer, usually the Medical Care Ombudsmen Program (“MCOP”), an organization that was set up about five years ago to handle appeals of coverage decisions. MCOP, which has no ties to Empire other than the payment it receives for handling an appeal, refers each case to a doctor who has expertise concerning the insured’s condition. Dr. Wolinsky then poses a series of questions to the external consultant, who answers them based on the patient’s records, his or her own expertise, and the medical literature. Dr. Wolinsky first told the district court that Empire “abide[s] by whatever decision the external reviewer makes.” *640However, at trial, he admitted that Empire accepted only the reviewer’s determination of what the medical literature showed and her “clinical expertise [on] what’s the best treatment for the patient at the given time.”
III. The Evolution of Empire’s HDCT Policy
For several years in the 1990s and based on the available medical research, Empire covered HDCT for patients who, like Zervos, had metastatic breast cancer and had responded to standard chemotherapy. At that time, several Phase II studies, which are summarized in Edward A. Stadtmauer, M.D., et al., Conventional-Dose Chemotherapy Compared with High-Dose Chemotherapy Plus Autolo-gous Hematopoietic Stem-Cell Transplantation for Metastatic Breast Cancer, 342 New Eng. J. Med.1069 (2000),3 and one Phase III randomized study — the “Bezwo-da” study — supported the conclusion that HDCT provided an overall survival benefit for patients with metastatic breast cancer.4 However, in May 1999, Dr. Wolinsky became aware of abstracts that had been presented at an American Society of Clinical Oncology (“ASCO”) meeting. These abstracts discussed controlled, randomized Phase III studies and in Dr. Wolinsky’s view did not show an improvement in result for HDCT. Dr. Wolinsky therefore convened a meeting of the TAC in July 1999. At that meeting, Empire decided to continue coverage based in part on a presentation from Zervos’ treating physician, Dr. Hesdorffer.
Later in 1999 evidence of fraud discredited the Bezwoda study. In addition, the principal study showing “little or no benefit” from HDCT, the Stadtmauer study, had gained greater acceptance in the oncology community. This new information caused Wolinsky to again convene the TAC. At a meeting in March 2000, Dr. Diana Lake of Sloan Kettering Memorial Hospital presented the results of three randomized Phase III studies of the use of HDCT in Stage IV breast cancer patients: the Philadelphia trial involved in the Stadt-mauer article, a French trial, and a study at Duke University. The French trial showed a three-year overall survival rate of 59% in the HDCT group versus 28% in the control group. The Duke study, on the other hand, showed a three-year survival rate of 41% in the HDCT patients versus 55% in the control group. The Stadtmauer study, which considered the largest population of patients, showed no significant difference in overall survival and mortality between the control group and the HDCT group. Based on Lake’s presentation, the TAC determined to initially deny coverage for HDCT for metastatic breast cancer.
Dr. Wolinsky also testified that at the time of the March 2000 TAC decision, he was aware of ASCO’s position that because there was a “lack of persuasive data dem*641onstrating superior effectiveness” of HDCT, that treatment “should only be performed in the context of a high quality clinical trial.” He denied that cost played any role in determining whether a treatment was of proven benefit or considered effective or appropriate by the medical community.
Since the March 2000 TAC meeting, Empire has denied three of the four requests it received for coverage of HDCT. It also initially denied coverage to the fourth applicant, a woman whose cancer was not as advanced as Zervos’ cancer, but reversed its position after receiving a letter from an outside reviewer stating that HDCT was as effective for this woman’s condition as conventional chemotherapy and was the best possible treatment for her because it would be administered in a well designed and randomized clinical trial. The appellate decision to grant coverage was made by Empire’s Dr. Downs, not Dr. Wolinsky.
IV. Empire’s Action on Zervos’ Request
Empire denied Dr. Hesdorffer’s request for coverage for Zervos on October 18, 2000, because “Autologous Peripheral Stem Cell Harvest/Transplant is Experimental and Investigational in Breast Cancer Stage IV.” Zervos’ physician appealed, and Empire sent Zervos’ records along with a list of questions to Dr. Thomas R. Spitzer, director of the bone marrow transplant program and deputy chief of the hematology-oncology unit at Massa-chussetts General Hospital. Dr. Spitzer indicated, based on available data, that HDCT would not improve Zervos’ health outcome.5 However, he also said that “[h]igh dose chemotherapy and autologous stem cell transplantation appears to be as effective as prolonged combination chemotherapy for metastatic breast cancer. A possible benefit of high dose chemotherapy is the avoidance of a protracted course of continuous combination chemotherapy.” Finally, Dr. Spitzer indicated that Dr. Hesdorffer’s study was well designed.
In reviewing Dr. Spitzer’s answers, Dr. Wolinsky ignored Dr. Spitzer’s assertion that HDCT was as beneficial as any other treatment because, in Dr. Wolinsky’s view, Dr. Spitzer’s other answers demonstrated that HDCT had not been proven to be beneficial. Dr. Wolinsky assumed that Dr. Spitzer had overlooked the toxicity portion of a proven benefit analysis. On October 24, 2000, Dr. Wolinsky sent Zervos a notice reaffirming the initial denial and stating:
“There are no convincing data to show that high dose chemotherapy and Auto-logous Peripheral Stem Cell Transplant is superior to conventional chemotherapy alone for males with metastatic breast cancer.” The Autologous Peripheral Stem Cell Harvest and Transplant remains denied as experimental and in-vestigational.
After learning of Empire’s denial, Zer-vos received radiation treatments in an effort to stabilize his condition. As a result of those treatments and conventional chemotherapy, Zervos’ cancer was in remission as of the end of February 2001, *642and he was eligible to receive HDCT in Hesdorffer’s study.
V. Initial District Court Proceedings
On January 29, 2001, Zervos filed a complaint against Empire and Verizon6 in the United States District Court for the Southern District of New York and moved for a preliminary injunction directing Empire to cover the cost of HDCT. Zervos claimed that Empire discriminated against him based on his disability and gender and that its denial of coverage was arbitrary and capricious and thus violated the Employee Retirement Income Security Act (“ERISA”), 29 U.S.C. § 1001, et seq. Three days after Zervos began his lawsuit, Empire — at the request of co-defendant Verizon — obtained a second medical opinion from an MCOP outside reviewer, Dr. Raymond B. Weiss. Weiss said that “[tjhere is no published evidence that indicates [HDCT] for men with metastatic breast cancer is any more beneficial than any other form of chemotherapy” and that Dr. Hesdorffer’s study was not well designed.
The district court denied Zervos’ request for a preliminary injunction, finding that his evidence failed to demonstrate irreparable injury because he did not show that HDCT would improve his likelihood of survival and that defendants reasonably concluded HDCT was experimental and investigational. Zervos I, 2001 WL 253377, at *8-10.
VI. The First Appeal
We affirmed the district court’s order. We agreed that Zervos failed to establish that he had a likelihood of success on any of his claims including the ERISA claim, at least insofar as he premised that claim on the administrative record presented to the district court. Zervos II, 252 F.3d at 173. We noted, however, that
Our decision ... does not necessarily mean that Zervos may not be able to develop a successful federal claim.... Zervos may be able to establish that, under the plan, the focus of an “arbitrary and capricious” inquiry should not be on Empire’s initial decision to deny coverage for HDCT — that is, the decision that Empire made after the TAC meeting. Rather, Zervos may be able to show (1) that judicial review here should focus on the final, definitive decision to deny coverage (which was all but made by an external reviewer); (2) that in this case that decision was the one communicated to Zervos by letter dated October 24, 2000; and (3) that that decision was “arbitrary and capricious” because it was based on a material misunderstanding of what is meant under the plan by the term “Experimental/Investigational.” Compare The Plan at 6 (defining “Experimental/Investigational”) as “services ... which are not of proven benefit for the ... treatment of the Covered Person’s condition, or are not generally recognized by the medical community as effective or appropriate for that condition” with Letter from Empire to Zervos (Oct. 24, 2000) (“We are unable to authorize [HDCT] for the following reason: ‘There are no convincing data to show that [HDCT] is superior to conventional chemotherapy alone for males with metastatic breast cancer.’ [HDCT] remains denied as experimental and inves-tigational.”).
Id. at 174-75 (footnote omitted). We also cautioned that “[i]n this case — whatever it *643may ultimately require — -justice can tolerate no substantial delays.” Id. at 174.
During the pendency of the appeal, Zer-vos unfortunately suffered a recurrence of his metastatic disease, making him ineligible to receive HDCT. A new three-month course of chemotherapy did not reduce the size of the metastatic lesions, which now were in his lung. However, another experimental course of chemotherapy begun in May 2001 substantially improved Zer-vos’ condition. Zervos had significant complications from this treatment, including damage to the right lung.
VII. Proceedings on Remand
On July 30, 2001, two days after testing that established Zervos’ renewed eligibility to participate in the Hesdorffer protocol, Zervos renewed his request for a preliminary injunction and asked to combine that application with a trial on the merits. At a conference on August 13, 2001, the district court denied Zervos’ request for an accelerated trial because defendants had made a summary judgment motion. The court heard oral argument on the summary judgment motion on September 6, 2001. On October 1, 2001, the first date of the scheduled trial, Judge Daniels granted Empire’s motion for summary judgment dismissing Zervos’ ADA claim but refused to dismiss his ERISA claim.
During the five-day trial, the district court heard testimony from the key players in the formation of Empire’s policy on HDCT and from the doctors who played a role in reviewing Zervos’ initial claim and internal appeal. In addition, Dr. Weiss and several expert witnesses offered their opinions. At the conclusion of the trial, on October 5, 2001, the district court stated that it intended to issue a decision early the following week. However, on October 11, 2001, Judge Daniels’ chambers called Zervos’ attorney and asked whether Zer-vos could wait for a decision until the following week. The attorney replied that based on Zervos’ physician’s advice and the strong likelihood of an appeal to this court by the non-prevailing party, Zervos could not wait for a decision beyond October 16, 2001. On October 16, chambers staff advised Zervos that the court had not yet reached a decision and asked how much longer Zervos would be eligible to receive HDCT. The following day Zervos submitted a declaration from Dr. Hesdorf-fer who stated that Zervos remained eligible for HDCT but that within the past month his cancer markers had increased slightly, indicating that the disease was about to “bounce back.” Dr. Hesdorffer also said:
[Ujnless the Treatment is administered to Mr. Zervos by the end of October, there is a grave risk that his metastatic disease will return, as it did in March of this year, and he will again become ineligible to receive the Treatment. This time, however, because tumor cells develop resistance to chemotherapy with each course of therapy, the chance of finding another chemo-agent that will yield another response sufficient to make him eligible for the Treatment a third time is remote at best.... [Without [HDCT], Mr. Zervos’ treatment options going forward are fairly limited, and his long-term prognosis, based just on continuing with his current treatment, would be severely limited because there is a limit to the amount of his current treatment that he can ultimately tolerate.
On October 22, 2001, the district court issued two memorandum opinions and orders: the first gave the court’s reasons for dismissing the ADA claim, and the second addressed the ERISA claim without specifically addressing Zervos’ renewed request for a preliminary injunction. In his *644second decision, Judge Daniels found that (1) Empire’s decision to stop covering HDCT was not arbitrary and capricious; (2) Empire’s initial decision to deny coverage for Zervos’ HDCT was not arbitrary and capricious; but (3) Empire’s appellate process was conducted in an arbitrary and capricious manner because the questions asked did not directly relate to policy terms and were confusing, Wolinsky weighed the answers to the same question by different reviewers inconsistently, and Empire sometimes, but not always, covered Experimental/Investigational treatment. Zervos v. Verizon New York, Inc., No. 01-CIV-685, 2001 WL 1262941, at *7-15 (S.D.N.Y. Oct.22, 2001) (“Zervos III”). The court remanded the case to Empire for “an immediate expedited review within 48 hours of plaintiffs submission of current relevant information.” Id. at * 15. It directed that Empire ask a new MCOP reviewer to answer questions “that elicit a clear and unequivocal response regarding whether the treatment is of proven benefit or generally recognized by the medical community as effective or appropriate.” Id. at *15. Judge Daniels further ordered that the new reviewer “be clearly and directly asked whether there is any other medical or clinical reason that the patient should receive the treatment despite its experimental/investigational nature.” Id. The court neither explicitly retained jurisdiction nor explicitly directed the entry of judgment.
VIII. The Second Appeal
Upon receipt of the district court’s opinion and order, Zervos immediately filed a notice of appeal and a motion requesting that we consider that portion of the appeal that contested the appropriateness of the remand on an expedited basis and issue a decision no later than October 31, 2001. The motion was granted, and Chief Judge Walker referred the appeal to this panel. Empire cross-appealed from the district court order to the extent that it found Empire’s appellate process to have been arbitrary and capricious. We heard oral argument on October 29, 2001, after which the panel directed the parties to appear before a CAMP mediator. The settlement discussion failed. On October 30, 2001— with Judge Jacobs dissenting — we issued an order reversing the district court’s order and directing the entry of an injunction requiring that Empire cover the cost of HDCT for Zervos.7 Judge Daniels promptly entered an order in compliance with this directive.
On October 31, 2001, Empire moved for a stay and for recall of the mandate, and Zervos informed the court that Empire declined to obey Judge Daniels’ order. We unanimously denied Empire’s motion.
Zervos asserts that we have jurisdiction of his appeal pursuant to 28 U.S.C. § 1292(a)(1) because the district court, in effect, denied his request for immediate injunctive relief. Empire argues that we lack appellate jurisdiction because the district court did not make a final determination on the request for injunctive relief. On the merits, Zervos argues that the district court abused its discretion in remanding the case to a plan administrator instead of entering an injunction ordering treatment.
DISCUSSION
I. Jurisdiction
Section 1292(a)(1) gives the courts of appeals jurisdiction of appeals from *645“[ijnterlocutory orders of the district courts ... granting, continuing, modifying, refusing or dissolving injunctions, or refusing to dissolve or modify injunctions.” 28 U.S.C. § 1292(a)(1). A party may appeal from an order that does not explicitly deny an injunction if (1) it has the practical effect of denying injunctive relief and (2) it has a “serious, perhaps irreparable, consequence.” Carson v. Am. Brands, Inc., 450 U.S. 79, 83-84, 101 S.Ct. 993, 67 L.Ed.2d 59 (1981) (internal quotation marks omitted). In Carson, the Supreme Court held that the court of appeals had jurisdiction over an appeal from a district court order declining to approve a settlement agreement because the order to proceed to trial deprived the plaintiffs of the advantages the settlement agreement conferred, which included immediate injunctive relief. Id. at 87-89, 101 S.Ct. 993. The Court found that these combined circumstances “constitute[d] serious, perhaps irreparable consequences that petitioners can effectually challenge only by an immediate appeal.” Id. at 90, 101 S.Ct. 993 (internal quotation marks omitted). We also have found that we had jurisdiction over an appeal from an interlocutory order that dismissed without leave to amend particular counts on which the plaintiff sought injunctive relief because the denial of injunctive relief pending appeal from a final judgment was “likely to have a serious and irreparable impact upon [the plaintiffs’] ability to compete.” Volvo N. Am. Corp. v. Men’s Int'l Prof'l Tennis Council, 839 F.2d 69, 76 (2d Cir.1988).
In this case, we find that the practical effect of the district court’s refusal to order immediate injunctive relief, despite (1) an established violation of ERISA; (2) the fact that the district court issued its order at a time when the plaintiffs cancer could recur at any day; and (3) our earlier recognition that time was of the essence, was an effective denial of the plaintiffs request for injunctive relief. Time has always been a critical factor in this case. As early as June of last year when the plaintiff had not established his ERISA claim, we instructed that justice would not tolerate any further “substantial delays.” Then in October 2001, Dr. Hesdorffer generally predicted that the plaintiff might become ineligible for HDCT treatment if it was not administered approximately by the end of October., Thus, the record shows that when the district court entered its order towards the end of October, any further delays would increase the risk that the plaintiff would become ineligible for the relief sought.
Of course the presence of time pressure alone is insufficient to establish that the district court’s remand was an effective denial of the plaintiffs request for relief. For example, even in cases where time is of the essence, a remand is probably not a final determination when the district court is unable to determine whether there has been a violation of ERISA because of an incomplete record. But in this case, the record was sufficiently complete for the district court to find that the appeals process was arbitrary and capricious. As we discuss below at III, the record also supported coverage for Zervos. Thus, the main purpose the remand served was to give the defendants an opportunity to remedy their defective appeals process. Even assuming that the defendants ordinarily should be allowed this opportunity, here the remand essentially decided that the plaintiff, who had established the predicate condition for relief, and whose cancer could recur at any time, would not receive the treatment for another week or two. We hold that because Zervos established that (1) the appeals process was arbitrary and capricious both procedurally and, as we discuss below, substantively, and (2) his request for relief could be rendered moot *646before he received relief, the delay triggered by the remand effectively denied injunctive relief.
Moreover, plaintiff established that this effective denial had a “serious, perhaps irreparable, consequence.” The denial is serious because each delay meant that there was a risk that Zervos would become ineligible for HDCT. And it is “perhaps irreparable” because the Hesdorffer declaration established- that Zervos probably has run out of conventional treatment options based on the increase in his cancer markers, the toxicity of his current treatment, and the increasing resistance of his cancer to conventional treatments. At this juncture, HDCT may well be the only effective or appropriate treatment available to Zervos.
Consequently, we find that we have jurisdiction over this appeal. We stress, however, that this holding is limited to the exigent and unique circumstances of this case. We do not reach the issue of whether a remand to a plan administrator is a final judgment within the meaning of 28 U.S.C. § 1291,8 and we do not suggest that a district court order granting a remand rather than the preliminary injunctive relief that a party requested ordinarily would be appealable immediately pursuant to Section 1292(a).
II. Standard of Review and Definition of the Record
The parties agree that Empire had discretion to interpret the provisions of the plan it administered. Therefore, reviewing courts can disturb Empire’s interpretations and actions only if they are arbitrary and capricious. Miller v. United Welfare Fund, 72 F.3d 1066, 1070 (2d Cir.1995). A decision is “arbitrary and capricious [if it] is without reason, unsupported by substantial evidence or erroneous as a matter of law.” Id. (internal quotation marks omitted). Because this standard is highly deferential, we have said that “the trial court should not conduct a de novo hearing on a rejected applicant’s eligibility for benefits.” Miles v. New York State Teamsters Conference Pension and Retirement Fund Employee Pension Benefit Plan, 698 F.2d 593, 599 (2d Cir.1983). We also have implied that the district court should not accept information from the insurer that was not part of the original record unless the insured is instrumental in causing the information to be added to the record. See Zuckerbrod v. Phoenix Mut. Life Ins. Co., 78 F.3d 46, 50 n. 2 (2d Cir.1996). Even where the district court exercises de novo review of the plan administrator’s determination, the district court “ought not” to accept additional evidence absent “good cause.” DeFelice v. American Int’l Life Assurance Co. of NY, 112 F.3d 61, 66 (2d Cir.1997).
We review de novo the district court’s assessment of the plan administrator’s decision. Miller, 72 F.3d at 1070.
Zervos argues that the district court should have conducted, and we must conduct, a review limited to the administrative record and thus not consider the testimony of Empire’s experts and of Dr. Weiss. In light of Miles, Zuckerbrod, and DeFelice, we agree that our review must be limited to evidence that was before Empire when it (1) determined in March 2000 that it would deny coverage for HDCT, (2) denied Zervos coverage on October 18, 2000, or (3) rejected Zervos’ appeal on October 24, 2000. Although the incompleteness of an administrative record *647might — under certain circumstances — constitute good cause for hearing additional evidence, the record before Empire at the time it made the October 24, 2000, decision was not incomplete. Rather, Empire followed the steps in handling Zervos’ application and appeal that it followed for any other insured up to and including its solicitation of Dr. Spitzer’s opinion. The additional information that Empire obtained after Zervos filed this lawsuit appears to be aimed at bolstering its legal position and not at providing fuller review of Zer-vos’ claim. Therefore, we limit our review of Empire’s decision to the administrative record as it is described by the testimony of Dr. Wolinsky and by the minutes of the TAC meeting. This record includes the written submissions of Drs. Spitzer and Hesdorffer made prior to October 24, 2000.
III. The Merits
For purposes of this appeal, neither party questions either the TAC’s initial decision to deny coverage for HDCT to insureds with metastatic breast cancer or Wolinsky’s October 18, 2000, initial denial of coverage to Zervos. Therefore, we have no cause to review these determinations, and, in any case, little basis to question them. The studies before the TAC reasonably could be read to establish that HDCT did not provide a survival benefit over long term conventional chemotherapy for patients with metastatic breast cancer. These studies also demonstrated that although HDCT conveyed approximately the same survival benefit as very long term conventional chemotherapy, it had more side effects and more serious side effects. The evidence that the TAC considered, and particularly ASCO’s view that HDCT is appropriate in the context of a well designed clinical trial, also supports Zervos’ contention that HDCT can be appropriate for individual patients in the context of a well designed clinical trial. The TAC took all of this evidence into account by determining to initially deny coverage but to allow coverage if an external reviewer found HDCT was appropriate for an individual patient.
Thus, we focus our review on (1) the correctness of the district court’s conclusion that Empire’s appellate process functioned in an arbitrary and capricious manner and (2) the appropriateness of remand rather than injunctive relief as a remedy.
We agree that Empire’s appellate process was arbitrary and capricious, primarily because Dr. Wolinsky required that a treatment be superior to another existing treatment in order to avoid exclusion under the policy’s experimental/investigational language while the language itself requires only that the treatment be effective — not more effective than alternatives — and considered to be appropriate by the relevant medical community. Thus, contrary to basic ERISA principles, Dr. Wolinsky in effect added additional language to the policy. See, e.g., Gallo v. Madera, 136 F.3d 826, 330 (2d Cir.1998) (“Even when trustees of a pension plan are entitled to deference in interpreting the terms of the plan, deference cannot be so broad as to permit them to graft additional requirements onto unambiguous plan definitions.”). Dr. Wolinsky also acted in an arbitrary and capricious manner by refusing to consider Dr. Spitzer’s opinion that HDCT would benefit Zervos as much as long-term conventional chemotherapy. Dr. Wolinsky testified that the question which produced Dr. Spitzer’s favorable response was one of the three which ordinarily “carried] the most weight.” Because Dr. Wolinsky’s basis for rejecting Dr. Spitzer’s answer was his unfounded assumption that Dr. Spitzer had not considered side effects, it was completely lacking in reason.
*648Having adopted the district court’s view that Empire’s internal appellate process was arbitrary and capricious at least as applied to Zervos, we review the district court’s choice to remand rather than grant injunctive relief for abuse of discretion. See Kinek v. Paramount Comm., Inc., 22 F.3d 503, 508 (2d Cir.1994); see also Halpin v. W.W. Grainger, Inc., 962 F.2d 685, 697 (7th Cir.1992) (reviewing district court’s choice of remedy for an ERISA violation for abuse of discretion). We find that remand in this case was an abuse of discretion.
First, a remand of an ERISA action seeking benefits is inappropriate “where the difficulty is not that the administrative record was incomplete but that a denial of benefits based on the record was unreasonable.” Zuckerbrod, 78 F.3d at 51 n. 4. On October 24, 2000, when Dr. Wolinsky denied Zervos’ appeal, the evidence in the record did not support a denial. The materials that the TAC considered established that (1) HDCT and conventional chemotherapy produced essentially the same survival benefit; (2) HDCT had greater but shorter side effects than repeated cycles of conventional chemotherapy; and (3) HDCT can be an appropriate therapy only if it is administered in a high-quality clinical trial. Although Zervos did not establish that the trial in which he would participate was of high quality when he made his initial application, the Spitzer letter unequivocally stated that the trial was of high quality and that the treatment would be effective and appropriate for Zervos. Moreover, Dr. Wolinsky’s assumption that Dr. Spitzer failed to consider side effects when he said HDCT was as effective as conventional chemotherapy was completely lacking in reason because Dr. Spitzer specifically referred to the lesser duration of side effects with HDCT. Because the record was complete and it supported coverage for Zervos’ treatment, we find that the substance of the plan administrator’s appellate decision also was arbitrary and capricious.
In addition, the remand order was arbitrary and capricious because it did not adequately address the time constraints Zervos faced. Any delay necessitated by a remand could effectively deny justice in this case, as we observed over half a year earlier even before the plaintiff had established a violation of ERISA. Because there was sufficient evidence to conclude that the denial of benefits was unreasonable and the record establishes that each day mattered, the district court’s remand, which served primarily to give the defendants an opportunity to retool a defective system, was an abuse of discretion. Here the administrative record did not contain substantial evidence supporting a denial of benefits and in fact could only be read to support granting coverage. Therefore, remand was an inappropriate remedy, and the district court should have reversed the determination to deny benefits and directed Empire to provide the requested coverage.
CONCLUSION
Because (1) Empire’s argument that the district court’s conclusion that its appellate process was arbitrary and capricious lacks merit; (2) Empire’s decision to deny benefits was also substantively erroneous; and (3) seen in the light of the exigency of Zervos’ situation, the flaws in the appellate process, and the lack of support in the record for Empire’s decision, the remand order was an abuse of discretion, the order of the district court is reversed. By our order of October 30, 2001, we have directed the district court to order Empire to provide the requested coverage.
. We entered an order to that effect on October 30, 2001, with an opinion to follow.
. Because Zervos’ cancer had spread to the sternum, doctors identified it as Stage IV or metastatic. Stage IV cancer has a less favorable prognosis than other more localized stages.
. The Phase II trials "consistently reported high overall rates of response (combined complete and partial responses), ranging from 73 to 100 percent. Despite a median survival of only 10 to 24 months, 7 to 18 percent of patients in these studies remained free of progressive disease for up to 5 years after the treatment.” Id. at 1069.
. Tests evaluating medical treatment are categorized as Phase I, Phase II, or Phase III studies. Phase I studies evaluate the potential toxicity of a recommended treatment. Phase II studies are non-randomized studies that evaluate the proposed treatment against the results historically obtained for proven treatments. Although helpful in establishing the efficacy of a treatment, Phase II trials may be affected by the nature of the patients who participate in them. Randomized Phase III studies present the best evidence of whether a treatment is effective.
. Dr. Spitzer noted that the Bezwoda study included only a small number of patients and the conventional chemotherapy regimes were not equal, and that Bezwoda had been proven guilty of scientific misconduct. He also stated that the Stadtmauer study had been criticized "for the high number of patients who were lost to follow-up and the 'nonstandard' (extended duration) nature of the chemotherapy control group," but that the oncology community nevertheless had widely accepted the results "and many have advocated abandonment of [HDCT] as 'standard of care' for metastatic breast cancer.”
. Zervos also sued United Healthcare Co., Inc., but the parties later stipulated to dismiss the complaint against United Healthcare.
. In connection with the appeal, Empire submitted the opinion of a third MCOP doctor, which it obtained after the district court's remand order. This doctor strongly recommended that Zervos not receive HDCT.
. This issue is unresolved in this circuit. Crocco v. Xerox Corp., 137 F.3d 105, 108 (2d Cir.1998).