This case presents several questions concerning the interpretation and application of an employee health insurance plan, an area litigated with increasing frequency under the Employee Retirement Income Security of Act of 1974 (ERISA), 29 U.S.C. Secs. 1001-1461 (1988). Belden & Blake Corporation appeals the district court's final judgment directing it to pay for employee Richard Heasley's liver/pancreas transplant as a treatment for pancreatic cancer which spread to both lobes of his liver. After a bench trial, the court held the transplant was not an "experimental procedure" excluded from coverage under Belden & Blake's self-funded medical insurance plan. We will vacate and remand.I.

Richard Heasley, now deceased, was an engineer employed by Belden & Blake Corp., an oil and gas producing company incorporated in Ohio and operating in Ohio, Pennsylvania, and New York.1 In September 1991, Heasley was diagnosed as suffering from Zollinger-Ellison's Syndrome, a condition in which a pancreatic tumor produces a hormone called gastrin. Known as a gastrinoma, the condition is a type of neuroendocrine tumor. Although gastrinomas typically grow slowly and remain confined to the pancreas, any spreading tends to occur in the liver.

The treatment of a neuroendocrine tumor depends upon its size, growth rate, and the severity of its symptoms. Smaller tumors that are not growing can be treated with drug therapy, followed by surgical removal if necessary. Where, as in Heasley's case, the tumor spread to both lobes of the liver, surgical removal was not possible without destroying the liver. Before surgeons performed liver transplants, the only treatments available for these metastasized tumors were chemotherapy and drug therapy.

After diagnosing the tumor, Heasley's doctors at the University of Pittsburgh Medical Center (the "University of Pittsburgh") recommended a liver/pancreas transplant. This procedure entails replacement of the diseased liver, removal of the tumor and, if necessary, replacement of the pancreas. Heasley's doctors believed a transplant would be his best hope for a cure in light of the tumor's rapid growth and the fact it had not spread beyond his pancreas and liver. They also believed his relatively young age (42 years old) and otherwise excellent health made him a good candidate. The doctors expected the transplant would give Heasley a higher quality of life for a longer period than chemotherapy. In the meantime, they reduced the size of the tumor through chemotherapy. But after responding favorably to the first six treatments, Heasley showed no improvement from the seventh and eighth treatments. This led the doctors to worry Heasley would no longer respond to chemotherapy, a further reason for recommending the transplant.

As a Belden & Blake employee, Heasley contributed to and received coverage from the company's self-funded group medical plan.2 The Plan is managed by AultCare, a Preferred Provider Organization affiliated with Aultman Hospital in Canton, Ohio, which administers claims and has significant influence over coverage determinations. The Plan covers organ transplants subject to its general exclusions, one of which bars coverage for "experimental procedures."

The University of Pittsburgh requires prospective transplant patients to make full payment for the procedure prior to surgery. Following diagnosis of his gastrinoma in September 1991, Heasley requested authorization for the procedure from Belden & Blake. On October 18, on a form labeled "AultCare Predetermination," AultCare Medical Director Gregory Haban advised Heasley the transplant would not be covered because it was "[c]onsidered experimental."

An exchange of correspondence among Heasley, Belden & Blake, and AultCare followed. In a letter from his counsel to Belden & Blake on March 12, 1992, Heasley requested a Summary Plan Description and Plan Document to determine whether the denial of coverage was proper. Belden & Blake's Human Resources Manager James Ewing responded, enclosing a health insurance policy issued by McKinley Life Insurance Company that contained the "experimental procedures" exclusion. Ewing advised Heasley that Belden & Blake had converted the policy to a self-insured plan in June 1990. Noting the policy contained no provision to appeal the denial of coverage, Heasley's counsel asked Ewing to whom he could direct questions about an appeal. Ewing directed him to AultCare. Responding to Heasley's counsel on March 23, AultCare Vice President Richard Haines confirmed the denial was based on "[t]he determination of AultCare and Belden & Blake ... that a liver transplant for treatment of Mr. Heasley's diagnosis is experimental." When Heasley requested further information explaining the denial of coverage, AultCare offered to review its determination, and asked Heasley to answer certain questions about the transplant. Aided by his physicians, Heasley prepared written responses and sent AultCare copies of pertinent medical journal articles. Six weeks later, on July 15, AultCare Medical Director Haban advised Heasley that after reconsideration, it had decided to deny coverage on the same basis.

On August 4, Heasley filed suit against Belden & Blake in the Western District of Pennsylvania, invoking ERISA's civil actions provision, 29 U.S.C. Sec. 1132(a)(1).3 In his complaint, Heasley sought a declaration that his proposed liver/pancreas treatment was not an "experimental procedure" within the meaning of the Plan, and that Belden & Blake had breached its fiduciary duty and its contractual obligations under the Plan. Heasley requested an order directing Belden & Blake to authorize the transplant, civil penalties, attorney's fees, and costs. Belden & Blake moved to dismiss the complaint for failure to join AultCare as a necessary party, and the court denied the motion. The court then held three days of hearings, receiving expert testimony from both sides.

Following the hearings, the court found the transplant was not an "experimental procedure" under the Plan and that Belden & Blake had breached its fiduciary duty by denying coverage. It directed Belden & Blake to pay the University of Pittsburgh the full cost of the transplant. The court certified its order as a final judgment under Fed.R.Civ.P. 54(b). Belden & Blake appealed. In the meantime, Belden & Blake paid a total of $279,000--$208,000 to the University of Pittsburgh and $71,000 to Heasley's doctors--and Heasley received his transplant. Shortly after we heard oral argument, Heasley died from post-operative complications associated with his transplant.4

On appeal, Belden & Blake challenges the district court's choice of the de novo standard of review over its denial of coverage, its interpretation of the "experimental procedure" exclusion in the Plan, and its conclusion that Heasley's liver transplant fell outside the exclusion. We have jurisdiction under 28 U.S.C. Sec. 1291 (1988).

Different standards of review govern our consideration of the district court's rulings. The court's choice of the de novo standard and its definition of the term "experimental procedure" were mixed questions of law and fact, involving the the application of legal precepts, Gregoire v. Centennial School Dist., 907 F.2d 1366, 1370 (3d Cir.), cert. denied, 498 U.S. 899, 111 S. Ct. 253, 112 L. Ed. 2d 211 (1990), and the interpretation of contractual language to determine the parties' intent, PaineWebber Inc. v. Hartmann, 921 F.2d 507, 511 (3d Cir.1990). We review the legal aspect of these determinations under a plenary standard and their factual aspects for clear error. Ram Constr. Co. v. Amer. States Ins. Co., 749 F.2d 1049, 1053 (3d Cir.1984). Our review over the district court's ruling that Heasley's transplant was not an "experimental procedure"--under the court's or Belden & Blake's definition of that term--is more limited. Because the court based this ruling on its resolution of conflicting expert testimony and documentary evidence, we review for clear error. Holder v. Prudential Ins. Co. of America, 951 F.2d 89, 91 (5th Cir.1992) (applying clear error standard to district court's ruling that high-dose chemotherapy was "experimental" treatment for breast cancer); Farley v. Benefit Trust Life Ins. Co., 979 F.2d 653, 661 (8th Cir.1992) (same standard applied to ruling that high-dose chemotherapy for melanoma was "experimental").

We first consider whether the district court chose the proper standard of judicial review over Belden & Blake's denial of benefits. Under Firestone Tire & Rubber Co. v. Bruch, 489 U.S. 101, 109 S. Ct. 948, 103 L. Ed. 2d 80 (1989), a court must review de novo a company's denial of benefits "unless the benefit plan gives the administrator or fiduciary discretionary authority to determine eligibility for benefits or to construe the terms of the plan," in which case we review a benefits denial under the narrower arbitrary and capricious standard. Id. at 115, 109 S.Ct. at 956-57. Selection of an appropriate standard of judicial review therefore turns on the terms of the plan. Id. at 111, 109 S.Ct. at 954-55.

In Luby v. Teamsters Health, Welfare & Pension Trust Funds, we held a plan's grant of discretion can either be express or implied. 944 F.2d 1176, 1180 (3d Cir.1991). We ruled a general grant of administrative power to the plan trustees did not imply authority to decide specific disputes between beneficiary claimants. We also determined an express grant of discretion in one specific area of a plan undermined a claim that the administrator possessed implied discretion in another area of the plan. An express grant in one area, we reasoned, demonstrated that had the drafters intended to grant discretion in another area, "they knew how to say so and would have expressly done so." Id. at 1181.

In contending the Plan accorded AultCare discretion to determine whether Heasley's transplant was experimental, Belden & Blake relies on language in the Utilization Management Benefit section of the Plan which provides:

We [AultCare] will evaluate the proposed admission for certification of medical necessity and appropriateness under the terms of the Master Group Policy and send a letter documenting the recommendation to you....

App. 100. Purporting to act under this grant of authority, AultCare sent a letter to Heasley on March 23, 1992, advising him the plan would not approve his transplant because it was experimental.

The district court held the Plan neither expressly nor impliedly granted AultCare discretion to interpret the terms of the Plan or to make eligibility determinations. Applying Luby, the court ruled the Plan's grant of discretion regarding "medical necessity and appropriateness," coupled with the absence of a grant of discretion over the "experimental procedure" exclusion, meant Belden & Blake did not intend to grant AultCare discretion over this exclusion.

Holding that a plan must express a "clear and unequivocal" intention to grant discretion, the court ruled the Plan fell considerably short of this standard. Because Firestone requires de novo review "unless the benefit plan gives the administrator or fiduciary discretionary authority," 489 U.S. at 115, 109 S.Ct. at 956, the court first looked to see whom the Plan designated as an administrator or fiduciary. Finding the Plan made no such designation, the court applied ERISA's default rule, under which Belden & Blake, the Plan sponsor, is deemed the Plan administrator. See 29 U.S.C. Sec. 1002(16)(A)(ii); Kendal Corp. v. Inter-County Hospitalization Plan, 771 F. Supp. 681, 684 (E.D.Pa.1991).

The court next considered whether the Utilization Management Benefit section of the Plan could be read to express Belden & Blake's clear and unequivocal intention to confer upon AultCare discretion to construe and apply the Plan's "experimental procedure" exclusion. Rejecting this claim, the court reasoned the language of the provision--authorizing AultCare to "evaluate" a proposed procedure and then "send" the participant a "letter documenting the recommendation"--did not signify AultCare had authority to make final coverage determinations. At most, the court continued, the language granted AultCare discretion to determine whether a proposed procedure was "medically necessary and appropriate," not to determine whether a procedure was "experimental."

On appeal, Belden & Blake contends the discretion granted to AultCare applies to coverage determinations generally and that, in any event, the experimental procedure exclusion is part of the inquiry into medical necessity and appropriateness. In assessing these contentions, Firestone requires that we look to the terms of the Plan. The Plan does not define the phrase "experimental procedures," nor does it reference that phrase in its definition of "medically necessary," which includes services and procedures "considered necessary to the amelioration of sickness or injury by generally accepted standards of medical practice in the local community." App. 89.

The text of the Plan suggests contradictory inferences. The provision on which Belden & Blake relies grants AultCare discretion to "evaluate the proposed admission for certification of medical necessity and appropriateness under the terms of the Master Group Policy." Because the "experimental procedure" exclusion is among those terms, the provision could be read to confer upon AultCare discretion over that exclusion. But other sections of the Plan militate against such a reading. The Plan contains separate exclusions for care, services, and supplies "which are not deemed Medically Necessary" and for "experimental procedures." App. 93. There would be no need for separate exclusions were the term "experimental procedures" subsumed in the concept of "medical necessity."

The parties' course of dealing is ambiguous as to whether either party had discretion.5 It appears Belden & Blake deferred to AultCare's medical expertise in making coverage determinations. Yet coverage determinations are economic as well as medical, and there is evidence Belden & Blake retained discretion over certain issues. For example, the record contains letters in which Belden & Blake advised AultCare of changes in coverage. One letter authorized a lifetime maximum benefit of $25,000 for AIDS, another letter limited coverage for dependents to those under age 19 unless they are full-time students, and a third letter authorized nicotine patches as a "covered medical expense."6 AultCare's letters to Heasley were unclear on whether it or Belden & Blake had discretion. Furthermore, Heasley testified that Belden & Blake's president told him Belden & Blake would make the final coverage determination.

In contending that whether a procedure is "experimental" is part of the medical necessity inquiry, Belden & Blake also relies on the testimony of its expert witness on health insurance, Dr. Grant Lawless of Blue Cross of Western Pennsylvania. But like the text of the Plan, Dr. Lawless' testimony on the overlap between medical necessity and experimental procedures is ambiguous. He initially testified that a procedure "considered experimental automatically falls under those broad categories of medically necessary and appropriate." App. 538. When pressed by the court, he indicated the two inquiries were discrete, stating: "if an area of medicine is considered experimental, ... it must be looked at [for] medical appropriateness and necessity, just as any other area of medicine would need to be looked at for medical appropriateness and necessity." App. 539-40. For these reasons, we find the textual and testimonial evidence is ambiguous as to whether the Plan granted discretion.

Both parties maintain our treatment of implied discretion in Luby supports their respective constructions of the Plan. But in light of the textual and testimonial evidence supporting both parties' constructions, we believe Luby does not resolve the standard of review inquiry. Accordingly, we consider whether the district court correctly determined a plan must contain a "clear and unequivocal" statement of discretion to trigger arbitrary and capricious review.

Belden & Blake maintains the district court's insistence that a grant of discretion be "clear and unequivocal" conflicts with Firestone. We recognize that under Firestone, the question whether discretion is granted turns on the Plan, 489 U.S. at 115, 109 S. Ct. at 956-57, and Courts of Appeals applying Firestone have looked to plan language to determine whether discretion is granted. Ziaee v. Vest, 916 F.2d 1204, 1207 (7th Cir.1990), cert. denied, --- U.S. ----, 111 S. Ct. 1581, 113 L. Ed. 2d 646 (1991); Madden v. ITT Long Term Disability Plan, 914 F.2d 1279, 1283 (9th Cir.1990), cert. denied, 498 U.S. 1087, 111 S. Ct. 964, 112 L. Ed. 2d 1051 (1991); Moon v. American Home Assur. Co., 888 F.2d 86, 88 (11th Cir.1989); de Nobel v. Vitro Corp., 885 F.2d 1180, 1187 (4th Cir.1989); Baxter v. Lynn, 886 F.2d 182, 187-88 (8th Cir.1989).

In attempting to determine what type of plan language confers discretion under Firestone, several courts have required a clear statement of discretion. de Nobel, 885 F.2d at 1187 (grant of discretion "must appear on the face of the plan"); Adams v. Blue Cross/Blue Shield of Maryland, Inc., 757 F. Supp. 661, 666 (D.Md.1991) ("benefit plan language must demonstrate on its face a clear and unequivocal intent to vest the plan administrator with discretionary authority"); Martin v. Masco Indus. Employees' Benefit Plan, 747 F. Supp. 1150, 1152 (W.D.Pa.1990) (same); Graham v. Federal Exp. Corp., 725 F. Supp. 429, 433 (W.D.Ark.1989) (same).

Yet in construing a Plan to determine whether it grants discretion, Firestone directs us to consider "the provisions of the [plan] as interpreted in light of all the circumstances and such other evidence of the intention of the [plan's creator] with respect to the [plan] as is not inadmissible." 489 U.S. at 112, 109 S.Ct. at 955. Obviously, if a plan contains a clear statement of discretion, it warrants arbitrary and capricious review under Firestone. But requiring a clear statement appears to truncate the Firestone inquiry into "all the circumstances and other evidence of ... intention." Under the clear statement approach, there is no need to consider the parties' intent; if the grant of discretion is not clearly stated, that is the end of the inquiry.

This is not to say Firestone always or even often requires a full-fledged factual examination of the circumstances under which a plan was created. In most cases, the language of the plan alone will determine whether the plan grants discretion. Thus, in concluding discretion could be implied as well as express in Luby, we recognized Firestone did not condition a finding of discretion on a particular verbal formula but did not depart from Firestone's primary focus on plan language. Indeed, we found no discretion was implied solely by construing the language of the plan. Id. at 1181.7 But where, as here, both the Plan and the evidence are ambiguous regarding discretion, neither Firestone nor Luby compels a particular result, and we must seek another approach for determining the standard of review.

Other Courts of Appeals have resolved ambiguities in plan exclusions through application of the rule of "contra proferentem." See Phillips v. Lincoln Nat. Life Ins. Co., 978 F.2d 302, 311-12 (7th Cir.1992); Glocker v. W.R. Grace & Co., 974 F.2d 540, 544 (4th Cir.1992); Masella v. Blue Cross & Blue Shield of Conn., 936 F.2d 98, 107 (2d Cir.1991); Kunin v. Benefit Trust Life Ins. Co., 910 F.2d 534, 540 (9th Cir.), cert. denied, 498 U.S. 1013, 111 S. Ct. 581, 112 L. Ed. 2d 587 (1990). Under that rule, "if, after applying the normal principles of contractual construction, '[an] insurance contract is fairly susceptible of two different interpretations, ... the interpretation that is most favorable to the insured will be adopted.' " Kunin, 910 F.2d at 539 (quoting A. Windt, Insurance Claims and Disputes Sec. 6.02, at 281-82 (2d ed. 1988)).

Although we declined to decide whether to apply contra proferentem in an ERISA case involving a severance pay agreement, Taylor v. Continental Group Change in Control Severance Pay Plan, 933 F.2d 1227, 1233 (3d Cir.1991), we expressly reserved the question whether this precept was appropriate in an ERISA insurance case, "where the principle ... is often strictly applied." Id.; see also Eley v. Boeing Co., 945 F.2d 276, 279-80 (9th Cir.1991) ("[a]lthough contra proferentem is strictly applied in the interpretation of insurance contracts, it is not automatically or universally applied to other contracts").

The principle of contra proferentem derives from the recognition that:

[i]nsurance policies are almost always drafted by specialists employed by the insurer. In light of the drafters' expertise and experience, the insurer should be expected to set forth any limitations on its liability clearly enough for a common layperson to understand; if it fails to do this, it should not be allowed to take advantage of the very ambiguities that it could have prevented with greater diligence. Moreover, once the policy language has been drafted, it is not usually subject to amendment by the insured, even if he sees an ambiguity; an insurer's practice of forcing the insured to guess and hope regarding the scope of coverage requires that any doubts be resolved in favor of the party who has been placed in such a predicament.

Kunin, 910 F.2d at 540.

This reasoning is especially convincing where, as here, an insured employee seeks contractual benefits under ERISA, a statute designed to protect the interests of such employees. Adoption of contra proferentem as a federal common law rule in ERISA insurance cases makes sense because "to do otherwise 'would require us to ... afford less protection to employees and their beneficiaries than they enjoyed before ERISA was enacted,' a result that would be at odds with the congressional purposes of promoting the interests of employees and beneficiaries and protecting contractually defined benefits." Masella, 936 F.2d at 107 (quoting Firestone, 489 U.S. at 113-14).8

Furthermore, incorporation of contra proferentem into ERISA as a federal common law rule is consistent with Firestone's statement that a court must construe a plan "without deferring to either party's interpretation." 489 U.S. at 112, 109 S. Ct. at 955; see Phillips, 978 F.2d at 311, Kunin, 910 F.2d at 541. To hold otherwise would be to ignore that a presumption and deference are distinct constructional tools. As the Ninth Circuit aptly put it, "we might say that a baseball umpire should not defer to a baserunner's claim that he is safe; but this proposition is fully consistent with the separate rule that ties go to the runner." Kunin, 910 F.2d at 540.

Adoption of contra proferentem does not violate ERISA's pre-emption clause. That clause provides: ERISA "supersede[s] any and all State laws insofar as they ... relate to any employee benefit plan," 29 U.S.C. Sec. 1144(a), except, inter alia, state laws which "regulate[ ] insurance, banking, or securities." id. Sec. 1144(b)(2)(A). Where, as here, a particular state law rule is consistent with the purposes of ERISA, we find it appropriate to adopt that rule as a matter of federal common law without regard to whether it falls within the insurance savings clause. See Kunin, 910 F.2d at 540 (finding contra proferentem applies in an ERISA insurance case "regardless of whether it applies as a matter of uniform federal law or because federal law incorporates state law on this point").9

We recognize other Courts of Appeals have applied contra proferentem to construe particular terms in insurance contracts subject to ERISA rather than to determine whether a plan grants discretion. Masella, 936 F.2d at 107 (construing the term "dental"); Phillips, 978 F.2d at 311 ("mental illness"); Kunin, 910 F.2d at 541 (same). But under Firestone, a court's choice of the standard of review is itself a question of contract construction. Accordingly, we believe the reasoning of these cases applies in this context as well. Application of contra proferentem in this case requires us to find the Plan did not grant Belden & Blake or AultCare discretion to determine which procedures are experimental because the Plan and the evidence are ambiguous. For this reason, we affirm the district court's choice of the de novo standard in reviewing Belden & Blake's denial of benefits.10

We now consider whether the court properly applied that standard in concluding the denial of benefits was erroneous. In assessing Belden & Blake's denial of benefits, we must define the phrase "experimental procedure." The district court interpreted "procedure" to refer generally to the type of operation involved, here a liver transplant. Noting the parties agreed a liver transplant operation itself is not experimental, the court held the denial of coverage improper.11

Belden & Blake contends the term "experimental procedure" must be defined in terms of its particular application, i.e., whether it is an experimental procedure for the type of tumor involved here. We agree. Experts for both sides focused on the efficacy of a transplant for Heasley's particular condition rather than as a general matter. The federal Medicare guidelines take a similar approach, providing that a liver transplant is experimental for certain illnesses and tumors and not for others. 56 Fed.Reg. 15006 (April 12, 1991) (approving liver transplants as not experimental for certain specified conditions). So does the applicable case law. See DiDomenico v. Employers Co-op Indus. Trust, 676 F. Supp. 903, 908 (N.D.Ind.1987) (liver transplants are not experimental for adults as well as for children); Pirozzi v. Blue Cross-Blue Shield of Virginia, 741 F. Supp. 586 (E.D.Va.1990) (high dose chemotherapy with autologous bone marrow transplant is not an experimental treatment for Stage IV metastasized breast cancer).

The proper classification of Heasley's tumor to determine whether the transplant was experimental is more difficult. Heasley's gastrinoma, a neuroendocrine tumor, is part of a larger category of metastatic tumors that have spread to the liver. Choosing the proper category of tumors is significant because it affects the efficacy of the procedure and because it determines the number of transplants that have been performed for the tumor, two important criteria for determining whether the procedure is experimental.

The parties' views on the relevant family of tumors have evolved over the course of this dispute. AultCare and Belden & Blake initially denied coverage based on the federal Medicare guidelines, which focus on the largest of these three categories, providing no coverage for liver transplants as treatments for "adult patients with primary or metastatic malignancies of the liver." 56 Fed.Reg. 15006. App. 370 (testimony of Dr. Rovner).

AultCare and Belden & Blake shifted their focus in reconsidering Heasley's request for coverage, after Heasley's doctors submitted letters and medical journal articles contending that neuroendocrine tumors are a particular subset of metastatic malignancies in the liver for which transplants have proved effective and are no longer experimental. One article states: "Generally, the results of liver transplantation for metastatic liver disease have not been favorable. One exception has been the unique group of neuroendocrine tumors, the slow growth of which allows liver transplantation to effectively palliate and control symptoms." Angel E. Alsina, et al., "Liver Transplantation for Metastatic Neuroendocrine Tumor," J. Clin. Gastroenter. 1990; 12(5):533. App. 943.

In the district court, the parties also focused on the category of neuroendocrine tumors. For example, Dr. Maureen Martin of the University of Pittsburgh, a member of Heasley's transplant team, testified that "[t]he issue of other metastatic tumors to the liver is totally separate from what we consider for neuroendocrine tumors." App. 207. Similarly, Dr. Marks, Belden & Blake's oncologist, focused on whether a transplant was appropriate for this group of tumors. App. 463, 470. In analyzing whether the transplant was an "experimental procedure" under Belden & Blake's definition of that term, the district court believed neuroendocrine tumors comprised the relevant category.

In its brief on appeal, Belden & Blake suggests variously that we consider the general category of neuroendocrine tumors and the specific category of gastrinomas. This distinction is significant because gastrinomas comprise only a small percentage of the number of neuroendocrine tumors for which transplants have been performed, and because the number of transplants performed bears on whether the procedure is experimental. But Belden & Blake offers no medical rationale for differentiating between gastrinomas and neuroendocrine tumors. Nor has it called our attention to any expert testimony or other record evidence on this point. The absence of such evidence suggests gastrinomas share the salient characteristics of other neuroendocrine tumors: a slow growth rate, a tendency to spread only to the liver and not beyond, and a tendency not to recur in the liver following a transplant. Therefore we conclude, subject to the introduction of any new evidence on remand, that the relevant inquiry is whether a liver transplant is experimental for neuroendocrine tumors generally.

In order to determine whether liver transplants for neuroendocrine tumors are covered under the Belden & Blake Plan, we now consider the meaning of the term "experimental." Courts construing the term in similar health insurance plans have found it resistant to precise definition. As the Court of Appeals for the Ninth Circuit has stated, "[i]n the context of modern medicine, the term 'experimental' seems clearly ambiguous on its face." Johnson v. Dist. 2 Marine Eng. Beneficial Ass'n, 857 F.2d 514, 516 (9th Cir.1988); see also Dahl-Eimers v. Mutual of Omaha Life Ins. Co., 986 F.2d 1379, 1382-83 (11th Cir.1993) (same); Pirozzi, 741 F. Supp. at 589 (same).

The concept can be captured by a verbal formula, but only at a high level of generality. The definitions offered by the experts in this case are illustrative. For example, the University of Pittsburgh transplant surgeon Dr. Martin defined as experimental "a procedure that has not been tested before, so that the outcome is unknown." App. 224. Similarly, AultCare physician Dr. Rovner testified a procedure remains experimental "until you can show that it will ... cure patients and prolong their lives in a predictable fashion." App. 411. These definitions, although easy to state, are hard to apply. This problem is exacerbated because, as experts for both parties agreed, it is difficult to identify precisely when a procedure ceases to be experimental and becomes accepted. App. 569 (testimony of Dr. Lawless); App. 584 (testimony of Dr. Carr).

Insurers and plan administrators have employed several approaches to cure the inherent ambiguity of the term "experimental." Some plans enumerate particular conditions for which a given procedure is experimental or simply write into the plan specific exclusions for these conditions without reference to their experimental nature. See Wilson v. Group Hosp. & Medical Serv., Inc., 791 F. Supp. 309, 311 (D.D.C.1992) (adopting the latter approach). This is also one of the approaches taken in the Medicare guidelines. 56 Fed.Reg. 15006 (1991).

Other insurers and employers resolve the definitional problem by specifying who will decide the meaning of the term "experimental." Often, such plans incorporate by reference classifications adopted by one or more independent authoritative medical bodies. For example, the plan in Fuja v. Benefit Trust Life Ins. Co., 809 F. Supp. 1333, 1336 (N.D.Ill.1992), excluded coverage for procedures "deemed to be experimental, educational or investigational in nature by an[ ] appropriate technological assessment body established by any state or federal government." See also Zuckerberg v. Blue Cross & Blue Shield, 108 A.D.2d 56, 62, 487 N.Y.S.2d 595, 600 (2d Dep't 1985) (plan provided "a procedure will be covered if an appropriate governmental agency Federal or New York State, recognizes it as sufficiently effective to justify any risks that may be involved"), aff'd 67 N.Y.2d 688, 499 N.Y.S.2d 920, 490 N.E.2d 839 (1986). As the Court of Appeals for the Eleventh Circuit observed in Dahl-Eimers, courts have "found th[e] ambiguity cured when the term 'experimental' is qualified by specifying who will determine" its meaning and application. 986 F.2d at 1383.12

As a corollary, courts have refused to rely exclusively on particular third-party classifications where the plan has not explicitly referenced them in defining its experimental procedure exclusion. See Bucci v. Blue Cross-Blue Shield of Connecticut, 764 F. Supp. 728, 730-31 (D.Conn.1991) (rejecting reliance on defense expert's use of five-factor Technical Evaluation Criteria because they were not incorporated in plan); Pirozzi, 741 F. Supp. at 590-91 (same); see also DiDomenico, 676 F. Supp. at 906 (granting preliminary injunction to reject denial of liver transplant although the procedure was not approved by the Medicare guidelines where guidelines were not part of the plan). This rule guards against the risk that a plan defendant will opportunistically invoke a self-serving third-party classification in order to avoid liability. See Bucci, 764 F. Supp. at 733 ("[w]hat defendant has attempted to do is to deny benefits on the ground that the procedure is not accepted when measured against a standard which is not defined in the plan ...").

The Belden & Blake Plan neither enumerates experimental procedures nor defines that term by incorporating the classification schemes employed by other medical bodies. Important consequences flow from these omissions. We cannot uphold Belden & Blake's initial justification for its denial of benefits--that liver transplants for neuroendocrine tumors are not yet approved by the federal Medicare guidelines--because the plan neither incorporates nor otherwise references the guidelines.13 Therefore we must turn to other sources to define the term "experimental" in the Belden & Blake Plan. As have other courts faced with this problem, e.g., Pirozzi, 741 F. Supp. at 590, we look to the testimony and the medical literature submitted by the expert witnesses.

Employing this approach, the district court found Heasley's proposed transplant was not experimental for several reasons. Noting neither liver transplants generally nor surgical removal of neuroendocrine tumors are considered experimental, the court reasoned that a liver transplant was the next logical step in the treatment of these tumors. That relatively few patients (50 in its estimation) have received transplants for neuroendocrine tumors did not, in the court's view, establish the procedure was experimental in light of the rarity of the condition, the recent availability of the surgery, and its demonstrated success. The court also relied on data showing only 5 percent of liver transplant patients at the University of Pittsburgh died of complications associated with the surgery. Finally, it credited the views of Heasley's experts, who testified the procedure was not experimental, and bolstered its conclusion by referring to medical journal articles introduced by these experts.

Several aspects of this analysis are problematic. First, we are uncertain whether the court should have relied on data showing the post-operative mortality rate for the procedure at the University of Pittsburgh was only 5 percent. The undisputed testimony reveals this was the mortality rate for all liver transplants at the University of Pittsburgh, App. 591, whereas the post-operative mortality rate for patients receiving transplants for neuroendocrine tumors was much higher. The data supplied by Heasley's doctors to AultCare revealed that four of the eight University of Pittsburgh patients with neuroendocrine tumors receiving transplants died of post-operative complications. App. 939. The article offered by Heasley's experts entitled "Liver Transplantation for Metastatic Neuroendocrine Tumor," co-authored by Dr. Alsina, indicated that 4 of 14 patients studied, or 29 percent, died of post-operative complications. App. 946. We note that neuroendocrine tumors are exceptionally rare, comprising only 14 of the 5,000 liver transplants performed at the University of Pittsburgh. App. 587. On remand, the district court should consider whether the higher post-operative mortality rate for neuroendocrine tumor patients receiving transplants has a medical explanation or whether it simply reflects a statistical anomaly resulting from the small number of such patients who have been transplanted.

Second, we are not certain the record supports the district court's conclusion that over 50 liver transplants have been performed for neuroendocrine tumors worldwide. The record provides solid support for about half that number. Dr. Alsina's article reports on 14 such patients, including five at the University of Pittsburgh. Yet Dr. Carr testified that 14 transplants had been performed at the University of Pittsburgh by the time of the district court hearing, and counsel advised us at argument that a fifteenth such transplant has since been performed, bringing the total to 25.14 The larger estimates came from two of plaintiff's experts. Dr. Shaw testified he estimated between 30 and 60 such transplants had been performed worldwide, App. 681, and Dr. Martin estimated the number was between 40 and 55, of which 30 to 40 had been reported in the literature. App. 218. Yet neither witness cited a source for these estimates, which are hotly disputed by Belden & Blake.

Finally, the district court offered no rationale for accepting the testimony of Heasley's experts rather than that of Belden & Blake's experts. The parties dispute which of these groups of experts were better informed. Only Heasley's experts were transplant surgeons with personal experience with performing the contested procedure. Yet Belden & Blake maintains these witnesses were biased in favor of surgery and contends the district court should also have relied on the testimony of the oncologists for each side. Furthermore, only Belden & Blake offered an expert in the field of health insurance. Although resolution of conflicting testimony is within the province of the district court, where, as here, apparently credible and informed experts testified for both sides, we believe some explanation of its apparent choice to rely entirely on Heasley's experts was warranted.

For these reasons, we are unable to determine whether the district court's conclusion that the transplant was not an experimental procedure was clearly erroneous. Accordingly, we believe the best course is to vacate the district court's order and remand for further proceedings.

As did the district court and the experts who appeared before it, we believe the term "experimental" basically hinges on the safety and effectiveness of the procedure in question, as demonstrated by its use on an appreciable number of patients. Yet given the increasing amount of litigation over what constitutes an "experimental procedure" and the fact-intensive nature of this determination, we believe a more systematic approach is warranted than simply applying a general definition.15 Absent any analytic framework for construing the term "experimental," a court is forced to choose between the testimony of two sets of experts who often seek "a floating standard which can rise or fall in any fact situation [and which is], not reviewable against identifiable criteria." Bucci, 764 F. Supp. at 733. Yet bright-line rules risk over- or under-inclusiveness in such a fact-bound area. For these reasons, we believe it is better to set forth a non-exclusive list of factors to determine whether a procedure is experimental.

The first factor is the judgment of other insurers and medical bodies. Heasley's doctors offered undisputed evidence that several commercial and government insurers have covered the transplant. App. 940. Because Belden & Blake conceded these insurers' policies contain an experimental procedure exclusion, their approval of coverage provides support for a finding that the procedure is not experimental. Similarly, to the extent other authoritative medical bodies have considered the procedure, this evidence is probative.16

A second factor is the amount of experience with the procedure. Often this is simply a function of the number of times the procedure has been performed, but it may be more complicated. Belden & Blake contends only transplants in the United States should be considered in arriving at a total figure. We disagree. First, had Belden & Blake wished to limit the analysis this way, it could have so defined "experimental" in the plan. Compare Fuja, 809 F. Supp. at 1336 (defining medically necessary as "given in accordance with generally accepted principles of medical practice in the U.S."); cf. DiDomenico, 676 F. Supp. at 906 (refusing to find dispositive Medicare guidelines not included in Plan). Moreover, we note European data and studies were cited on several occasions in the liver transplant study conducted by the federal Office of Health Technology Assessment, App. 1031, 1034, 1038, which formed the basis for the Medicare guidelines on which Belden & Blake relies. See Rollo v. Blue Cross/Blue Shield of N.J., 1990 WL 312647 at * 6 (D.N.J. Mar. 22, 1990) (relying on English studies in concluding autologous bone marrow transplant was not experimental for Wilms' tumor). Consideration of European cases was especially appropriate in this case. Two of the three largest liver transplant centers are in London, England and Hannover, Germany, and the surgeons there were trained by Dr. Starzl of the University of Pittsburgh, the surgeon who pioneered the liver transplant. App. 584.

The number of transplants performed may not always indicate a procedure is experimental. For example, AultCare physician Dr. Rovner refused to say how many transplants it would take before the procedure ceased to be experimental. App. 422. This has proven a familiar pattern with defense experts, and courts faced with such testimony have appropriately discounted the significance of this factor. Thus, as the district court stated in Bucci, "[t]o permit defendant to base a denial on the lack of cases which it regards as sufficient to prove the efficacy of the treatment, when the number it would accept as sufficient is not ascertainable from the language of the policy nor testified to, would permit an arbitrary and capricious denial." 764 F. Supp. at 732; Pirozzi, 741 F. Supp. at 590 (expressing skepticism with defense expert testimony that the number of high-density chemotherapy treatments was "still too small" to determine whether treatment was experimental for breast cancer because expert "was unable to suggest a number of patients that would provide sufficient data").

A third factor is the demonstrated effectiveness of the procedure. Several indicia bear on this inquiry, including the long-term survival rate associated with the procedure,17 the likelihood of recurrence of the cancer, and the post-operative mortality rate. Because effectiveness is a relative inquiry, data on these rates must be compared to data associated with no treatment or with alternative treatments. Pirozzi, 741 F. Supp. at 590-91 (comparative data is one of five technology assessment criteria used by Blue Cross and Blue Shield Association).

Measures of effectiveness should be compared not only to other treatments but also over time. As the Court of Appeals for the Fifth Circuit stated, "it is the nature of medical research that what may one day be experimental may the next be state of the art treatment." Holder v. Prudential Ins. Co. of America, 951 F.2d 89, 91 (5th Cir.1992). Indeed, Heasley's experts testified that, as the University of Pittsburgh developed more experience with transplants for neuroendocrine tumors, success rates improved. App. 594 (Dr. Carr).18

In certain cases, there will be differences in the quality of the evidence regarding a procedure's effectiveness. Belden & Blake emphasizes no randomized double-blind studies have been done with respect to liver transplants for neuroendocrine tumors. Such studies track two groups of patients with the same condition, with only one group being given the disputed treatment and the other serving as the "control" group. App. 450 (Dr. Marks). Pirozzi, 741 F. Supp. at 591. Belden & Blake contrasts these studies with the medical journal articles offered by Heasley in the district court, which contain only anecdotal information regarding transplants for neuroendocrine tumors.

In evaluating this critique, we note initially, as at least one other court has observed, that "nothing in the plan requires that a treatment be the subject of completed [double-blind] studies to escape the experimental treatment exclusion." Pirozzi, 741 F. Supp. at 591. More importantly, such studies represent only one, albeit a generally reliable, means for comparing two treatments. The significance of their absence therefore depends on the availability and quality of other comparative data, some of which is contained in the record.19 Often, peer-reviewed medical journal articles, or, in certain cases, expert testimony, will provide helpful sources of this data as well.

Part of the comparative analysis involves the attractiveness of the various treatment alternatives for the particular patient. Thus, in concluding an autologous bone marrow transplant with high-dose chemotherapy was not experimental for a patient with multiple myeloma, one court noted "it was simply the only appropriate treatment available to treat plaintiff's condition" in light of expert testimony that the plaintiff would receive no more benefit from standard chemotherapy. Dozsa v. Crum & Forster Insurance Co., 716 F. Supp. 131, 138. Indeed, one reason Heasley's doctors recommended a transplant was their belief he was no longer responding to chemotherapy. Also, there was testimony that the patients with neuroendocrine tumors who were transplanted at the University of Pittsburgh had no realistic survival options other than the transplant given the extent of their metastasization. App. 708 (Dr. Shaw). Additionally, Heasley's experts testified that transplants are recommended only for the patients with the worst tumors, and many patients with the most advanced tumors come to the University of Pittsburgh, a leading liver transplant center. App. 708 (Dr. Shaw); App. 592-93 (Dr. Carr). If true, these facts would tend to make statistics regarding effectiveness less favorable, and, to the extent they suggest no other treatment option is viable, would help place in perspective data regarding the effectiveness of the transplant.

In addition, there will always be particular facts in the case that may affect application of the factors we have outlined. In this case the rarity of neuroendocrine tumors makes it difficult to collect statistically significant amounts of data, and to conduct double-blind studies. We leave to the district court the proper inference, if any, to draw from the rarity of the disease in assessing whether the transplant was an experimental procedure.

On remand, both parties may submit testimony concerning application of the term "experimental procedure" to Heasley's transplant in light of the factors and concerns we have identified. But neither party may supplement the record with data generated since the district court's initial ruling. For example, evidence of liver transplants performed for neuroendocrine tumors or studies published during the pendency of this appeal would be inadmissible. This distinction makes sense in part because Heasley received his transplant and is no longer living. Were Heasley alive and awaiting a decision on payment or treatment, more recent evidence would be admissible.

We recognize that any determination of whether a particular procedure is experimental will necessarily turn on the facts of the particular case. Nor are we unaware of the context in which these cases arise. The archetypal case presents a gravely ill patient requesting an injunction after being denied benefits. Faced with the exigencies of such a claim and a wealth of conflicting and complex expert testimony, the trial court has a daunting task.

For the foregoing reasons, we will vacate the district court's order and remand for further proceedings consistent with this opinion.