Affirmed by published opinion. Chief Judge ERVIN wrote the opinion, in which Judge MURNAGHAN and Senior Judge BUTZNER joined.

OPINION

The Office of Civilian Health and Medical Programs of the Uniformed Services (“CHAMPUS”) challenges the grant of a permanent injunction in favor of one of its beneficiaries, Gail Ann Wilson. Wilson, who has breast cancer, seeks payment for a course of treatment known as high-dose chemotherapy with peripheral stem cell rescue, which she and her doctor claim is necessary to treat her condition. CHAMPUS contends that such treatment is excluded from its program because it is “experimental or investigational” under the applicable federal regulations. Concluding that CHAMPUS’s denial of benefits was arbitrary and capricious, we affirm the district court’s issuance of a permanent injunction, thereby prohibiting CHAMPUS from denying payment for the treatment requested by Wilson.

I.

Gail Wilson suffers from Stage II high-risk inflammatory breast cancer. Based on her doctor’s recommendation, Wilson decided to undergo a series of treatments known as high dose chemotherapy with peripheral stem cell rescue (“HDC/PSCR”) to treat her condition.1 In summary, extremely high doses of chemotherapy are considered more effective in killing cancer cells, and thus can be used to treat breast cancer. Unfortunately, in addition to killing malignant cells, the treatment also kills healthy white blood cells in the patient’s blood stream and bone marrow, leaving her susceptible to deadly infections. To combat this problem, doctors have developed a procedure in which a patient’s “peripheral stem cells” are harvested from her blood prior to the administration of high-dose chemotherapy or radiation. This treatment is called peripheral stem cell rescue (“PSCR”). A similar, alternative procedure known as an autologous bone marrow transplant (“ABMT”) retrieves such cells from the patient’s bone marrow. After the patient’s body is flooded with cancer-killing agents, the healthy cells are reinfused, hopefully sufficient to protect her against disease.

Gail Wilson is the wife of a retired member of the United States Navy. As such, she is a beneficiary of CHAMPUS, a health benefits program that provides medical benefits for dependents of active-duty and retired members of the United States military. See 10 *363U.S.C. §§ 1076-79. Established by Congress and administered by the Secretary of Defense, see 10 U.S.C. §§ 1071, 1072(4), 1073, the program supplements the military’s system of direct care for members of the armed services. Although it resembles insurance, CHAMPUS “is not an insurance program in that it does not involve a contract guaranteeing the indemnification of an insured party against a specified loss in return for a premium paid.” 32 C.F.E. § 199.1(d) (1994).

Pursuant to federal regulations promulgated by the Secretary of Defense, CHAMPUS provides eligible beneficiaries “medically necessary services and supplies required in the diagnosis and treatment of illness and injury.” Id. § 199.4(a)(1). The regulations specify certain “conditions, limitations [and] exclusions,” one of which is the following:

Not in accordance with accepted standards, experimental or investigational. Services and supplies not provided in accordance with accepted professional medical standards; or related to essentially experimental or investigational procedures or treatment regimens.

Id. § 199.4(g)(15). “Experimental” is defined as:

Medical care that essentially is investigatory or an unproven procedure or treatment regimen (usually performed under controlled medicolegal conditions) that does not meet the generally accepted standards of usual professional medical practice in the general medical community.

Id. § 199.2(b). CHAMPUS’s policy manual, promulgated to provide guidance in the implementation of the program, labels some sixty-six specific conditions “experimental or investigational.” Although the list is “for example purposes only and is not to be construed as being all-inclusive,” it does not include HDC, ABMT, or PSCR. Joint Appendix at 424-28.

CHAMPUS specifically provides coverage for chemotherapeutic agents and their administration. The CHAMPUS policy manual states, however, that ABMTs are covered only for certain diseases under specific circumstances, and the list of covered conditions does not include breast cancer. Joint Appendix at 431-32. In addition, the manual indicates that “harvesting of the required stem-cells by apheresis from peripheral blood rather than bone marrow, can be allowed for those beneficiaries for whom it has been established that bone marrow harvesting can not be accomplished due to documented bone marrow involvement” with cancer. Id. at 432.

Dr. David Bogner, CHAMPUS’s Medical Director, received Wilson’s request for coverage of her treatment with HDC/PSCR on July 15, 1994, and denied her request that very day. According to Bogner’s letter of denial, “in the absence of published randomized, prospective trials, CHAMPUS must continue to consider this therapy as investi-gational for the treatment of breast carcinoma.” Id. at 442. Neither Wilson’s physician nor the treatment’s provider would begin HDC/PSCR without an advance commitment from CHAMPUS to cover its cost.

On September 6,1994, Wilson filed a complaint in the United States District Court for the Eastern District of Virginia. She sought an injunction requiring CHAMPUS and the Secretary of Defense to pay for HDC/PSCR, along with declaratory relief, costs, and attorneys fees. On October 24,1994, following expedited proceedings, the district court entered a final judgment permanently enjoining the defendants from denying Wilson coverage for the desired treatment. Id. at 891-905. Reviewing CHAMPUS’s decision under the standard set forth in the Administrative Procedures Act (“APA”), 5 U.S.C. § 706(2)(A), the court found that CHAMPUS and the Secretary had “act[ed] in an arbitrary and capricious manner in denying Plaintiff coverage.” Joint Appendix at 895. The defendants filed timely notice of appeal to this court.

II.

We review the district court’s grant or denial of a permanent injunction for an abuse of discretion. Lone Star Steakhouse & Saloon, Inc. v. Alpha of Virginia, Inc. 43 F.3d 922, 939 (4th Cir. 1995). With respect to injunctive relief, “ ‘[w]hat we mean when we say that a court abused its discretion, is *364merely that we think that [it] made a mistake.’ ” Direx Israel, Ltd. v. Breakthrough Medical Corp., 952 F.2d 802, 814 (4th Cir. 1991). In making that assessment, we review the district court’s factual findings for clear error and its legal conclusions de novo. Lone Star Steakhouse & Saloon, 43 F.3d at 939; North Carolina v. Virginia Beach, 951 F.2d 596, 601 (4th Cir. 1991).

To obtain a permanent injunction, Wilson must have demonstrated that her claim to benefits from CHAMPUS has merit.2 Amoco Production Co. v. Village of Gambell, 480 U.S. 531, 546 n. 12, 107 S.Ct. 1396, 1404 n. 12, 94 L.Ed.2d 542 (1987). To make that showing, she must have established that CHAMPUS’s decision to deny coverage is invalid under the APA. Green Hosp. v. United States, 23 Cl.Ct. 393, 399-400, 403 (1991). Under that statute, we review an agency’s action outside the circumstances of formal proceedings to determine if it was “arbitrary, capricious, and abuse of discretion, or otherwise not in accordance with law.” 5 U.S.C. § 706(2)(A).3 Under this narrow standard, we are “ ‘not empowered to substitute [our] judgment for that of the agency.’ ” Maryland Dep’t of Human Resources v. United States Dep’t of Agric., 976 F.2d 1462, 1475 (4th Cir. 1992) (quoting Citizens to Preserve Overton Park v. Volpe, 401 U.S. 402, 416, 91 S.Ct. 814, 824, 28 L.Ed.2d 136 (1971)). Rather, “we perform ‘only the limited, albeit important, task of reviewing agency action to determine whether the agency conformed with controlling statutes,’ and whether the agency has committed ‘a clear error of judgment.’” Id. (quoting Baltimore Gas & Elec. Co. v. Natural Resources Defense Council, Inc., 462 U.S. 87, 97, 103 S.Ct. 2246, 2252, 76 L.Ed.2d 437 (1983), and Citizens to Preserve Overton Park, 401 U.S. at 416, 91 S.Ct. at 823-24) (citations omitted).

As a benchmark for determining whether CHAMPUS’s refusal to pay was arbitrary and capricious, we interpret the federal regulations excluding coverage for treatments that are “[n]ot in accordance with accepted standards, experimental or investigational,” 32 C.F.R. § 199.4(g)(15). We note again that “experimental” is defined as medical care that “does not meet the generally accepted standards of usual professional medical practice in the general medical community.” Id. § 199.2(b).

CHAMPUS argues on appeal that Dr. Bogner based his decision to deny coverage on several factors. He allegedly relied upon “his continuous review of refereed medical literature on this issue, on discussions with the Office of Technology Assessment at the U.S. Public Health Service concerning this treatment, on the CHAMPUS Policy Manual provision governing high-dose chemotherapy treatments and on various medical and medical technology articles and reports.” Brief for the Federal Appellants at 16.4 After *365closely examining the record in this case, and particularly Dr. Bogner’s letter rejecting coverage, we view CHAMPUS’s denial of Wilson’s request differently. Throughout, there is a disproportionate emphasis on the lack of Phase III clinical trials proving the effectiveness of HDC/PSCR,5 meaning that the efficacy of the treatment had not been measured against conventional therapy given to a control group in a prospective, randomized experiment. In fact, Bogner’s denial letter makes the agency’s decision seem relatively straightforward: “In the absence of published randomized, prospective trials, CHAM-PUS must continue to consider this therapy as investigational for the treatment of breast carcinoma.” Joint Appendix at 442. While the letter mentions consultations with “technology assessment panels” and “oncology consultants,” it intimates that such endeavors were undertaken for the sole purpose of gathering results of Phase III clinical trials, to which CHAMPUS avowedly gave the “greatest weight.” Id. at 443; see also Affidavit of Dr. David F. Bogner, id. at 434-35 (“I determined that the proposed therapies were not CHAMPUS benefits because no Phase III trials have proven HDC/PSCR to be safe, effective and superior to conventional therapies for breast carcinoma_”).

Certainly, such tests are important in determining the validity of medical treatments. Moreover, they may be an important factor in determining whether a particular therapy meets the “generally accepted standards of usual professional medical practice in the general medical community.” Id. § 199.2(b). However, nothing in the Code of Federal Regulations or the CHAMPUS policy manual indicates that published, Phase III clinical trial results are required before a benefit can be provided. Without doubt, a therapy could become standard practice in the medical community before it had been proven more effective than traditional treatments. Cf. Pirozzi v. Blue Cross-Blue Shield, 741 F.Supp. 586, 593 (E.D.Va. 1990) (“Many treatments become accepted without phase III stud-ies_”). Such would be the ease, for instance, where a highly promising treatment for a terminal illness is known to be harmless, but has yet to be proven efficacious in prospective, randomized testing.

In the record before us, there is considerable evidence that Phase III clinical trials are not the critical aspect in determining whether a therapy has become “generally accepted” within the medical community. CHAM-PUS’s expert witness, Dr. Bruce Cheson, who is the Medical Director of the Clinical Trial Division of the National Cancer Institute, acknowledged that the American Society of Clinical Oncology recently published a paper “t[aking] the position that third-party payers should be paying for certain treatments administered during the course of a clinical trial.” Joint Appendix at 332. Dr. Cheson also indicated that a patient “has a very high certainty of getting the best available care if one participates in a clinical trial.” Id. Moreover, he agreed with Wilson’s counsel that, assuming HDC/PSCR was limited to the context of a peer-reviewed clinical trial with which Cheson was familiar, HDC/PSCR “is accepted treatment and it should be mentioned by oncologists as an alternative to their patients.” Id. at 337. Furthermore, at his deposition, even Dr. Bogner recognized the existence of what is commonly known as “home run treatments” — those “that prove themselves so significantly that phase III trials are not necessary.” Id. at 404.

*366In addition to overemphasizing the necessity of Phase III clinical trials, CHAMPUS ignored abundant evidence that HDC/PSCR is gaining widespread acceptance within the medical community. For example, a letter from some of the country’s premier oncologists 6 noted that “[t]he evidence to support the efficacy of these procedures is far in excess of many commonly used and reimbursed medical treatments.” Id. at 266. These physicians also cited the “explosive growth of the use of this procedure in the United States” as “suggest[ing] that a broad consensus of physicians recognize this efficacy.” Id. Even CHAMPUS has conceded that “there is ample scientific background for vigorous clinical investigation in the area of high-dose chemotherapy treatment for breast cancer.” Brief for the Federal Appellants at 16 (citing 12 J. Clinical Oncologf 226, 229 (Jan. 1994), included in Joint Appendix at 614, 618). Moreover, while the omission is not dispositive, the agency failed to list HDC/ PSCR as “experimental or investigational procedure” in its own policy manual. Joint Appendix at 424-48.

Despite this evidence, CHAMPUS relied on an unwritten agency policy mandating Phase III trials before a treatment is provided. In contrast, federal regulations require only that a therapy be generally accepted, see id. § 199.2(b), not that it prospectively be proven to have a statistically significant effect in curing a disease. While the two categories certainly overlap to a substantial degree, they are not co-extensive, and CHAMPUS wrongly ignored the distinctions between them. Effectively, CHAMPUS imposed a requirement beyond those in the applicable regulations by creating an informal, but nonetheless binding, prerequisite that a treatment pass Phase III trials. The agency did so despite regulations mandating that CHAMPUS pay benefits “subject to all applicable definitions, conditions, limitations or exclusions specified.” 32 C.F.R. § 199.4(c)(1). We therefore conclude that CHAMPUS acted arbitrarily and capriciously in denying coverage to Wilson.7 See Bedford County Memorial Hosp. v. Health & Human Services, 769 F.2d 1017, 1022 (4th Cir. 1985) (noting that “[ajgency action is arbitrary and capricious if the agency relies on factors that Congress did not intend for it to consider”).

III.

For the reasons discussed above, we conclude that CHAMPUS’s refusal to pay for Wilson’s HDC/PSCR was arbitrary and capricious and not in accordance with law. Accordingly, we affirm the district court’s judgment granting Wilson a permanent injunction that prohibits CHAMPUS from denying coverage for that treatment.

AFFIRMED.