Defendant-Appellant-Cross-Appellee Liberty Life Assurance Company of Boston ("Liberty") appeals the district court's denial of its motion for summary judgment and that court's grant of summary judgment in favor of plaintiff-appellee-cross-appellant Linda Ellis ("Ellis"). The district court concluded that no genuine issue of material fact existed as to Ellis's claim under the Employee Retirement Income Security Act ("ERISA"), 29 U.S.C. § 1001, et seq., and that she was entitled to summary judgment. The court ultimately ruled that Liberty in its role as plan fiduciary abused its discretion when it terminated Ellis's long-term disability ("LTD") benefits because substantial evidence did not demonstrate a change in her medical condition after Liberty initially determined that Ellis qualified for LTD benefits. The district court dismissed Ellis's state-law claims, however, holding that they are preempted by ERISA; and Ellis cross-appeals that ruling. For the reasons that follow, we (1) affirm the district court's dismissal of Ellis's state-law claims, (2) reverse the district court's grant of summary judgment and award of costs and fees in favor of Ellis, and (3) grant summary judgment in favor of Liberty, rendering a take-nothing judgment on Ellis's ERISA claim.

Liberty is a nationwide insurance carrier that issued a disability insurance policy ("the Policy") to Chase Manhattan Bank ("Chase") in January 1997. The Policy, which is an integral part of an employee welfare benefits plan governed by ERISA, provides LTD benefits to eligible Chase employees.

The Policy specifies that LTD benefits are payable for the first 24 months of disability to a covered employee who is "unable to perform all of the material and substantial duties of his occupation on an Active Employment basis because of an Injury or Sickness."1 The Policy further provides that after 24 months, LTD benefits continue to be payable if the disabled employee "is unable to perform with reasonable continuity, all of the material and substantial duties of his own or any other occupation for which he is or becomes reasonably fitted by training, education, experience, age and physical and mental capacity."2 As the plan fiduciary, Liberty is expressly vested with discretionary authority to make all coverage, eligibility, and interpretation decisions with regard to the Policy: "Liberty shall possess the authority, in its sole discretion, to construe the terms of this policy and to determine benefit eligibility hereunder."3

In 1997, Chase hired Ellis as a mortgage loan officer. Ellis worked at Chase until 1999, when she applied for short-term disability ("STD") benefits — under a different Liberty policy — because she could no longer perform her job duties as a loan officer. Although the exact nature of Ellis's medical condition remains somewhat unclear from the evidence in the record on appeal, her medical records indicate that she might suffer from fibromyalgia, a rheumatic syndrome that causes pain in muscles, tendons, and fibrous and other connective tissues. Liberty reviewed Ellis's STD claim, approved it, and started paying her STD benefits in January 2000.4

When Ellis's STD benefits expired later that year, her claim automatically converted into one for LTD benefits under the Policy. Liberty then began to investigate whether Ellis's claim fell within the Policy's definition of LTD. In June 2000, Liberty informed Ellis by letter that it had reviewed her file and determined that she was eligible for LTD benefits. Liberty also informed Ellis that it would periodically require updated medical information "to support total disability as defined by the Policy." Liberty continued its investigation, and, in light of additional medical evidence that it subsequently gathered, Liberty determined that Ellis was not eligible for LTD benefits. In December, Liberty wrote to Ellis:

While it is apparent you were ill and met the criteria for your policy's definition of disability initially, based on the medical information received, you no longer meet your Long Term Disability Policy's definition of disability. Therefore, we must close your claim for benefits, effective December 31, 2000.

The following month, Ellis administratively appealed Liberty's decision to terminate her LTD benefits. Ellis submitted further medical information to Liberty, which forwarded her file to its Managed Disability Services Unit ("MDSU"). The MDSU concluded that no objective medical findings existed that would render Ellis "disabled" within the contemplation of the Policy. Liberty then affirmed its decision to terminate Ellis's LTD benefits. (Liberty has made no effort, however, to recoup the LTD benefits previously paid to Ellis.)

In October, Ellis sued Liberty in Harris County, Texas. Ellis asserted Texas statutory and common law claims for violations of the state insurance code, breach of contract, and breach of the duty of good faith and fair dealing. Liberty timely removed the suit to the district court pursuant to 28 U.S.C. § 1441(b) on the basis of ERISA preemption.

The following fall, after the close of discovery, Liberty filed a motion for summary judgment seeking dismissal of Ellis's state-law claims. In response, Ellis filed a cross-motion for summary judgment and sought to amend her complaint to state an ERISA claim. The district court granted Ellis leave to amend her complaint, and Liberty filed a supplemental motion for summary judgment to dismiss her ERISA claim.

The district court eventually denied Liberty's motion for summary judgment and granted summary judgment to Ellis on her ERISA claim. The court dismissed Ellis's state-law claims, however, holding that they were preempted by ERISA. The district court subsequently issued a supplemental memorandum and order clarifying its award of attorneys' fees and prejudgment interest to Ellis, ultimately entering final judgment in favor of Ellis.

Two days later, Ellis filed a motion to alter or amend the judgment on the amount of damages, attorneys' fees, and prejudgment interest. The district court granted the motion in part, increasing the quantum of Ellis's future disability benefits and clarifying the rate of prejudgment interest. Liberty timely filed its notice of appeal.5

Liberty first argues that the district court erred when it granted Ellis leave to amend her complaint to state an ERISA claim. We review a district court's decision to grant leave to amend a complaint for abuse of discretion.6 Federal Rule of Civil Procedure 15 states that leave to amend pleadings "shall be freely given when justice so requires."7 In determining whether to grant leave, a district court may consider such factors as (1) undue delay; (2) bad faith; (3) dilatory motive on the part of the movant; (4) repeated failure to cure deficiencies by any previously allowed amendment; (5) undue prejudice to the opposing party; and (6) futility of amendment.8 Although the district court assigned no reasons on the record for granting Ellis leave to amend her complaint, we are satisfied that it did not abuse its discretion when it did so.

Although Liberty argues that Ellis's amendment demonstrates undue delay, bad faith, and dilatory motive, we find no evidence in the record to support such an argument. Liberty's strongest argument concerns the potential prejudice that it may have suffered as a result of Ellis's filing of her amendment so late in the proceedings in district court. We reject this argument. Liberty removed Ellis's state-court suit on the basis of ERISA preemption. Ultimately, and as Liberty argued in its Notice of Removal, the district court concluded that ERISA preempted all of Ellis's state-law claims.9 We have previously held that "ERISA's preemptive and civil enforcement provisions operate to `recharacterize' such claims into actions arising under federal law."10 Thus, for removal purposes, ERISA's preemptive power recharacterized Ellis's state-law breach of contract claim as a claim arising under federal law, specifically ERISA. Liberty might not have known with certainty that Ellis's breach of contract claim would be recharacterized as an ERISA claim and that Liberty would ultimately have to litigate such a claim. Having removed on the basis of ERISA preemption, however, Liberty cannot now be heard to complain about the district court's grant of leave for Ellis to amend her complaint to include an ERISA claim. There was no prejudice to Liberty, and the district court did not abuse its discretion when it granted Ellis leave to amend her complaint to state an ERISA claim.

We review a district court's grant of summary judgment de novo.11 "Whether the district court employed the appropriate standard in reviewing an eligibility determination made by an ERISA plan administrator is a question of law."12 We thus review this decision de novo.13 When the ERISA plan vests the fiduciary with discretionary authority to determine eligibility for benefits under the plan or to interpret the plan's provisions, "our standard of review is abuse of discretion."14 As the Policy vests Liberty, as plan fiduciary, with the "sole discretion" to construe the terms of and to award benefits under the Policy, we review Liberty's interpretation of the Policy for abuse of discretion.15

We have previously explained in detail the appropriate two-step process to review a plan fiduciary's interpretation of its plan:

First, a court must determine the legally correct interpretation of the plan. If the administrator did not give the plan the legally correct interpretation, the court must then determine whether the administrator's decision was an abuse of discretion. In answering the first question, i.e., whether the administrator's interpretation of the plan was legally correct, a court must consider:

(1) whether the administrator has given the plan a uniform construction,

(2) whether the interpretation is consistent with a fair reading of the plan, and

(3) any unanticipated costs resulting from different interpretations of the plan.16

If we determine that the fiduciary's interpretation of the plan was legally correct, the inquiry is over, pretermitting any need to consider whether a legally incorrect interpretation of the fiduciary was not an abuse of discretion.17

We have also held that when a complaining participant or beneficiary shows that the plan fiduciary has a conflict of interest, we apply a sliding scale to the Wildbur standard: "The greater the evidence of conflict on the part of the administrator, the less deferential our abuse of discretion standard will be."18 "The degree to which a court must abrogate its deference to the administrator depends on the extent to which the challenging party has succeeded in substantiating its claims that there is a conflict."19 In its Objections and Responses to Plaintiff's Request for Admissions, Liberty acknowledges that it has a financial interest in the dollar value of the claims that are paid under the Policy. This is enough to satisfy us that a legal conflict of interest exists here. Accordingly, we apply the sliding-scale standard of review articulated in Vega to Liberty's interpretation of its Policy provision.

As noted above, the LTD Policy provides that benefits are initially payable only to an employee who is "unable to perform all of the material and substantial duties of his occupation on an Active Employment basis because of an Injury or Sickness."20 The district court concluded that, under this language, Ellis would be eligible to receive LTD benefits if she "could not perform any one of the material duties of her occupation."21

The district court erred when it interpreted the phrase "unable to perform all" — the language in the policy — as synonymous with "unable to perform any one." We interpret "unable to perform all" as synonymous with "not able to perform every." In other words, "unable" is synonymous with "not able," and "all" is synonymous with "every." Applying the Wildbur methodology, we hold that Liberty gave a legally correct interpretation to this provision of the plan.

The first Wildbur factor — whether the fiduciary has given the plan a uniform construction — weighs in favor of Liberty's interpretation. The district court mistakenly relied solely on the deposition testimony of Liberty's litigation manager, Paula McGee, as support for crediting Ellis's proffered interpretation that she is entitled to LTD benefits if she is unable to perform "any one" of the material and substantial duties of her occupation. McGee testified:

Q. Under that definition, if Ms. Ellis could not perform one of the material duties of her occupation, she would be disabled?

A. Yes.

In a subsequent affidavit, however, McGee explained that counsel's question at the deposition confused her and that the company had consistently interpreted "Disability" to mean a person who is unable — not able — to perform all — each and every one — of the material and substantial duties of her own occupation:

Liberty has consistently interpreted the Policy. Specifically, when evidence reveals that during the first 24 months of disability, an employee is capable of performing the material and substantial duties of her own occupation, the Company has denied benefits. In my five years of employment with Liberty, I cannot recall an instance where this Policy provision was interpreted differently.

McGee's post-deposition affidavit is buttressed by the testimony of Liberty's disability claims consultant and its appeals consultant. Both testified in depositions that Liberty decided to terminate Ellis's benefits by virtue of its interpretation that a disabled person under the LTD Policy is a person who is not able to perform every material and substantial duty of her occupation. All this tips the scale in favor of Liberty on the first Wildbur factor.

The next Wildbur factor — whether Liberty's interpretation is "consistent with a fair reading of the plan" — also supports Liberty's interpretation. For Ellis to qualify for LTD benefits under the Policy, Liberty determined that she had to show that she could not perform "each" of the material and substantial duties of her occupation; in other words, "each and every duty" or "every single duty." This is consistent with a fair reading of the plain wording of the plan. There is no dispute that the Policy language requires that Ellis be unable to perform all of the material and substantial duties of her occupation to receive LTD benefits. We conclude that in the context of the Policy as a whole, a fair reading of the term "unable to perform all" is that Ellis is not disabled for purposes of LTD if she can perform "at least one" of the material and substantial duties of her occupation. Ellis's proffered interpretation, that she is disabled if she cannot perform one ("any one") of the material and substantial duties of her occupation — i.e., "unable to perform all" means "not able to perform any one" — cannot be squared with the Policy's language.

Our conclusion that Liberty's interpretation is legally correct is strengthened by consideration of the third Wildbur factor — whether a different interpretation of the plan would result in unanticipated costs to the plan. A comparison of the Policy provisions that define "Disability" and "Partial Disability" in pari materia leads inescapably to the conclusion that adoption of Ellis's proffered interpretation would lead to Liberty's incurring of unanticipated costs. Section 4 of the Policy defines Partial Disability:

"Partial Disability" or "Partially Disabled" means as a result of the Injury or Sickness, the Covered Person is:

1. able to perform one or more, but not all, of the material and substantial duties of his own or any other occupation on an Active Employment or a part-time basis ...22

Liberty reasons with irrefutable logic that if we were to credit Ellis's interpretation of "Disability," the definitions of "Disability" and "Partial Disability" would conflate these separate categories into one, i.e., there would be no difference between the eligibility prerequisites for total disability and those for partial disability. It follows that if that were the case, Liberty would be required to provide both LTD and partial disability benefits to a covered employee if he could not perform "any one" of the material and substantial duties of his occupation, a patently absurd result. If the definition of long term disability were interpreted to mean "unable to perform just one," as Ellis urges, "unable to perform all" in the definition of Disability would be synonymous with "unable to perform one or more" in the definition of Partial Disability. That simply cannot be: Such a reading would render partial disability's phrase "but not all" meaningless surplusage, not to mention putting it in direct conflict with Ellis's proffered interpretation of "all" in the phrase "unable to perform all" in the definition of Disability. Obviously, this cannot be the intended result under the Policy and — just as obviously — unanticipated costs would be incurred by Liberty.

Ellis attempts to counter by asserting that Liberty's interpretation is legally incorrect because "[u]nder this contorted interpretation, virtually no person could ever satisfy the definition of `Disability.'" Ellis offers the example of a secretary who is rendered paraplegic, contending that this employee would not be disabled under Liberty's interpretation if she could sit at her desk in a wheelchair and answer a speaker phone. Ellis's argument ignores, however, the two adjectives that modify "duties" — "material and substantial." Merely because a disabled employee can perform a minor, collateral duty of his job, e.g., answering a speaker phone, would not justify the plan fiduciary's considering such an employee ineligible for benefits under Liberty's interpretation of the LTD Policy. In such a situation, the disabled employee would be disabled under Liberty's interpretation, despite his ability to perform minor duties, as long as he could not — was "unable to" — perform any of the material and substantial duties of his occupation. We conclude that Ellis would have to demonstrate that she cannot perform "every single" or "each and every" "material and substantial duty of her occupation" — which she could not prove — to obtain LTD benefits. Liberty gave a legally correct interpretation of the plan provision in question.23

Among the rest of Ellis's arguments, we perceive that two merit brief consideration. The first concerns the burden of proof under ERISA. Ellis insists that substantial record evidence supports her claim of total disability, in light of which Liberty abused its discretion when it determined that she was not disabled. This argument misapprehends the burden of proof under ERISA. The law requires only that substantial evidence support a plan fiduciary's decisions, including those to deny or to terminate benefits, not that substantial evidence (or, for that matter, even a preponderance) exists to support the employee's claim of disability.24 Substantial evidence is "more than a scintilla, less than a preponderance, and is such relevant evidence as a reasonable mind might accept as adequate to support a conclusion."25 We are aware of no law that requires a district court to rule in favor of an ERISA plaintiff merely because he has supported his claim with substantial evidence, or even with a preponderance. If the plan fiduciary's decision is supported by substantial evidence and is not arbitrary and capricious, it must prevail.

The second argument that we address is more problematic, as it tangentially concerns the degree or level of proof that is needed to sustain a plan fiduciary's interpretation of its policy provision. The crux of the dispute here is whether — as Ellis contends and the district court ruled — a plan fiduciary's decision to terminate LTD benefits once it has initially agreed to provide them must be supported by evidence that a substantial change in the covered employee's medical condition occurred after the initial grant of benefits. The parties dispute whether a higher standard of proof is required to sustain a plan fiduciary's decision to terminate benefits once granted than is needed to sustain a plan fiduciary's initial denial of benefits. Ellis argues that because Liberty initially determined that she qualified for LTD benefits, it abused its discretion when it terminated her LTD benefits months later, without medical evidence reflecting that a substantial change in her condition had occurred in the interim. The district court accepted Ellis's evidentiary dichotomy and ruled that the absence of credible, substantial, or positive evidence in the record to demonstrate that Ellis had become medically ineligible for permanent disability benefits after having been found eligible initially precluded Liberty from terminating the benefits previously granted.26

We disagree with the district court's view of the applicable law. We have found no statutory, regulatory, or jurisprudential authority — and neither Ellis nor the district court has cited any to us — that would heighten the level of the proof needed for a plan fiduciary to determine entitlement or non-entitlement to LTD benefits simply because the fiduciary previously had approved entitlement and paid benefits to the employee in question. The district court committed legal error when it concluded that, once the fiduciary approves entitlement to LTD benefits, subsequent termination of those benefits would have to be supported by substantial evidence of a change in the employee's condition. We have never articulated such an evidentiary distinction or imposed such a requirement on the plan fiduciary: All that ERISA requires is that substantial evidence support a plan fiduciary's benefits decision — whether it be to deny benefits initially or to terminate benefits previously granted — when, as here, the plan fiduciary is vested with the discretion to determine, inter alia, both initial and continued eligibility for benefits. In the investigation that continued following its initial grant of LTD benefits to Ellis, Liberty acquired subsequent medical evidence that supported termination of her LTD benefits months after it had approved Ellis's entitlement to them on the basis of the evidence that it had before it at that time.

We hold that when a plan fiduciary initially determines that a covered employee is eligible for benefits and later determines that the employee is not, or has ceased to be, eligible for those benefits by virtue of additional medical information received, the plan fiduciary is not required to obtain proof that a substantial change in the LTD recipient's medical condition occurred after the initial determination of eligibility. Indeed, evidence could exist — as it did here — at the time that the plan fiduciary initially granted benefits that demonstrates that the ERISA plaintiff is not totally disabled. In addition, a plan fiduciary could receive evidence that an ERISA plaintiff is not totally disabled months after it has made the initial grant of benefits. A contrary holding would basically prohibit a plan fiduciary from ever terminating benefits if it later discovered evidence that the ERISA plaintiff was not disabled at the time of the initial grant of benefits.27 More importantly to plan participants and beneficiaries, such a rule would have a chilling effect on the promptness of granting initial benefits in the first place. This we are unwilling to do. A plan fiduciary that has granted plan benefits to a participant or beneficiary is not estopped from terminating those benefits merely because there is no evidence that a substantial change in the covered employee's medical condition occurred after the original grant of benefits.

As we reverse the district court's grant of summary judgment in favor of Ellis, we vacate the award of costs and attorneys' fees to Ellis.

Ellis cross-appeals the district court's ruling that ERISA preempts her state-law claims. She sued Liberty for violations of Texas Insurance Code ("TIC") articles 21.21 and 21.55 and for breaches of the common law duty of good faith and fair dealing. TIC article 21.21 prohibits unfair competition and unfair practices by insurance companies and subjects them to civil liability for violations.28 TIC article 21.55 subjects insurance companies to civil liability if they unfairly and untimely process and treat a claim.29 With respect to Ellis's common-law claim, the courts of Texas have held that "[a] cause of action for breach of the duty of good faith and fair dealing is stated when it is alleged that there is no reasonable basis for denial of a claim or delay in payment or a failure on the part of the insurer to determine whether there is any reasonable basis for the denial or delay."30 Ellis argues that the United States Supreme Court's holding in Kentucky Association of Health Plans, Inc. v. Miller31 brings her claims under ERISA's savings clause.32 We review ERISA preemption of state law de novo.33

Under conflict preemption,34 ERISA preempts state laws "insofar as they may now or hereafter relate to any employee benefit plan."35 As an exception, however, ERISA's so-called savings clause allows state laws "which regulate insurance, banking, or securities" to survive ERISA preemption.36 In Miller, the Supreme Court simplified the test for ERISA conflict preemption when it made a clean break with the McCarran-Ferguson factors that it traditionally applied to determine whether a state statute regulated insurance and thus survived preemption under ERISA's saving clause.37 After Miller, for a state law to be deemed a "law... which regulates insurance" under Section 1144(b)(2)(A) and thus be exempt from traditional ERISA preemption, such law must (1) be directed toward entities engaged in insurance, and (2) substantially affect the risk pooling arrangement between the insurer and the insured.38

On the one occasion that we considered Miller's change to ERISA preemption, we observed that "[t]he only pertinent difference between the Miller analysis and the previous test is that in place of the second Miller inquiry, the previous test asked whether the statute in question `transfers or spreads the risk from the insured to the insurer.'"39 Also, prior to Miller, we held that ERISA preempts TIC articles 21.2140 and 21.5541 well as the Texas common law duties of good faith and fair dealing.42 Thus, we need only answer whether the "simplified" Miller analysis affects our prior holdings. We conclude that it does not with respect to Ellis's state-law claims. Thus, ERISA preempts Ellis's common law claim for breach of the duties of good faith and fair dealing. Specifically, this common law doctrine fails the first prong of the Miller analysis because it is not directed toward entities engaged in insurance.43

Regarding Ellis's two statutory claims, Liberty contends that TIC articles 21.21 and 21.55 likewise fail the first prong of the Miller analysis because they address the misconduct of insurers and thus do not regulate insurance. This argument misses the mark. As the Supreme Court noted in Rush Prudential v. Moran, to determine whether a law "regulates insurance," "we start with a `common-sense view of the matter,' ... under which `a law must not just have an impact on the insurance industry, but must be specifically directed toward that industry.'"44 In Moran, the Supreme Court went on to hold that "when insurers are regulated with respect to their insurance practices, the state law survives ERISA."45 In Sharpless, we analyzed the effect of the Miller decision on our ERISA preemption analysis, noting that Miller had not changed this factor.46 Indeed, whereas the McCarran-Ferguson analysis required that we determine whether the practice "is limited to entities within the insurance industry,"47 the Miller analysis merely requires that we determine whether the "statute is specifically directed towards entities engaged in insurance."48 TIC articles 21.21 and 21.55 are undeniably directed toward entities engaged in insurance, as they regulate any possible unfair practices and expose the insurer to civil liability for violations of the statutes. We are satisfied therefore that articles 21.21 and 21.55 satisfy the first prong of the Miller analysis.

To survive ERISA preemption, however, TIC articles 21.21 and 21.55 must also satisfy Miller's second prong; they must "substantially affect the risk pooling arrangement between the insurer and the insured."49 We hold that these two articles do not.

In Miller, the Supreme Court explained that, to affect the risk-pooling arrangement, a statute must "alter the scope of permissible bargains between insurers and insureds" and thus substantially affect the risk-pooling "arrangements that insurers may offer."50 TIC articles 21.21 and 21.55 are remedial in nature — they provide remedies "to which the insured may turn when injured by the bad faith of the insurer."51 Being remedial, these two articles cannot possibly affect the bargain that an insurer makes with its insured ab initio. They provide only that "whatever the bargain struck," the insured may recover additional damages if thereafter the insurer acts in bad faith or unfairly.52 As TIC articles 21.21 and 21.55 provide remedies above and beyond those provided in ERISA, they are remedial in nature and do not affect the risk — here, the covered employee's disability — for which the insured contracted with the insurer.53

Within the insurance industry, "risk" signifies "the risk of occurrence or injury or loss for which the insurer contractually agrees to compensate the insured."54 Here, the risk pooled is that of long-term disability, and this risk is reflected in the terms of the Policy itself. The remedies that TIC articles 21.21 and 21.55 provide for unfair practices and bad faith are not risks identified in the Policy as a risk of loss that Liberty agrees to bear for Chase or for Ellis and other similarly situated Chase employees. As a basic tenet of insurance law, the insurance policy "defines the scope of risk assumed by the insurer from the insured."55 Here, Liberty did not contract with Chase or Ellis to assume the risk of any unfair practices or bad faith violations. As TIC articles 21.21 and 21.55 fail Miller's second prong, Ellis's claims grounded in violations of those articles are preempted by ERISA.

We affirm the district court's dismissal of Ellis's state-law claims and that court's grant of leave to Ellis to amend her complaint. We reverse the district court's grant of summary judgment and award of costs and fees in favor of Ellis, and we grant summary judgment in favor of Liberty, rendering a take-nothing judgment against Ellis on her ERISA claim.

AFFIRMED IN PART, REVERSED AND RENDERED IN PART.

I respectfully dissent.

As an initial matter, I disagree with the majority's conclusion that "unable to perform all of the material and substantial duties of his occupation" can only mean unable to perform "each and every one" of the material and substantial duties of an occupation, so that if an employee can perform even one material and substantial duty of his or her occupation, the employee is not disabled. Although this is a reasonable interpretation of the policy language, the policy language is ambiguous and susceptible to more than one reasonable interpretation. I would interpret the policy provision differently. If there are ten material and substantial duties of an occupation and the employee can perform only six of those ten duties, then the employee is by definition "unable to perform all of the material and substantial duties" of the occupation. That too is a reasonable interpretation. The ambiguity should be construed against the insurer. See Wegner v. Standard Ins. Co., 129 F.3d 814, 818 (5th Cir.1997). But to construe the policy differently than the administrator would create an internal inconsistency between the policy provisions for total disability and partial disability. Consequently, though I disagree with the majority's interpretation of the policy, this issue is not outcome determinative.

However, I do respectfully dissent from the majority's conclusion that when a plan administrator initially determines that a covered employee is eligible for benefits and later determines that the employee is not eligible for those benefits, the plan administrator may terminate benefits without demonstrating that its initial decision was erroneous, or without substantial evidence of a change in the claimant's medical condition.

Because this case involves an insurer who is also the plan administrator, producing a direct conflict of interest on the part of the administrator, we apply a sliding-scale standard of deference to the administrator's decision. See Vega v. Nat'l Life Insur. Servs., 188 F.3d 287, 294-99 (5th Cir.1999). Under such circumstances, this court still applies the abuse of discretion standard, "but gives less deference to the administrator in proportion to the administrator's apparent conflict." Id. at 296. Where there is a conflicted administrator, "the administrator has a financial incentive to deny the claim and often can find a reason to do so." Id. The court must "focus on whether the record adequately supports the administrator's decision." Id. at 298. "[W]e are less likely to make forgiving inferences when confronted with a record that arguably does not support the administrator's decision." Id. at 299.

At the summary judgment stage, it is the movant who has the burden of showing that there is no genuine issue of material fact. The parties agree that this claim involves a policy of insurance issued by Liberty. Liberty is the administrator. All reasonable inferences are to be drawn in favor of Ellis. If Liberty pays a claim it is not unreasonable to infer that every penny of the claim comes directly out of Liberty's coffers. Any argument that the administrator did not have a direct and total, or almost total conflict of interest in this situation is to ignore reality. The fox guarding the chicken house is not entitled to great deference.1 Thus, in analyzing the administrator's decision to terminate benefits, I would give little deference to the administrator's exercise of discretion.

It is undisputed that the administrator initially determined that Ellis was entitled to disability benefits based on the medical evidence, and later reaffirmed that fact in the letter of termination. I would hold as a matter of law, that once the administrator determined Ellis was entitled to disability benefits, a subsequent termination of those benefits would be an "arbitrary and capricious" decision by the administrator, and hence an abuse of discretion, unless there is substantial evidence to support either (1) that the initial decision to grant benefits was incorrect; or (2) that there has been a change in condition that would justify the termination of benefits. Once the administrator has exercised its discretion and determined that a claimant is entitled to benefits, a later decision to terminate those vested disability benefits without justification is by definition arbitrary and capricious and an abuse of discretion. See Meditrust Financial Servs. v. Sterling Chemicals, 168 F.3d 211, 215 (5th Cir.1999) (holding that administrator's decision is arbitrary if made without rational connection between known facts and the decision or between found facts and the evidence).

In the termination letter of December 22, 2000, the administrator acknowledged that "it is apparent" that Ellis "met the criteria for [her] policy's definition of disability initially." Thus, in the opinion of the administrator the initial decision to grant disability benefits was correct. The administrator made no effort to show that the initial decision was wrong, but to the contrary, reaffirmed the initial determination. The question then becomes whether there was a change in condition that would justify the later termination of benefits. The majority opinion fails to answer this question.

In the termination letter, the administrator listed (with little explanation of its relevance) the medical evidence used in support of the decision to terminate benefits. The district court analyzed this evidence and found that it did not support the administrator's conclusion to terminate benefits. I agree. I would affirm the district court for basically the same reasons given in the district court's opinion. The record does not adequately support the administrator's decision to terminate benefits because the administrator admitted that the plaintiff was not initially disabled and because there was no substantial evidence of a subsequent change in condition. Thus, the administrator abused its discretion.

The majority contends that the dissent argues that a plan administrator cannot reverse an initial erroneous decision to pay benefits. Contrary to the majority's characterization, I certainly would agree that an insurer can correct an erroneous initial decision to pay benefits, but only if there is substantial evidence to support that decision. But in this case the administrator, Liberty, acknowledged, even in its letter of termination, that the initial determination to award benefits was appropriate.

For the above reasons, I respectfully dissent.