2017 could be big for litigation over “Big Data” applications in insurance. This past year saw the filing of several “price optimization” class actions, and claims against a fraud detection tool, similar to the ones used by insurers, survived an initial round of motions. Zynda v. Arwood, 175 F.Supp.3d 791 (E.D. Mich. 2016). But even as the industry braces for a wave of suits over cutting edge models and analytics, battles over less exotic systems are still being fought. Last month, in Johnson v. GEICO Cas. Co., No. 16-1132 (3rd Cir. Nov. 29, 2016), the U.S. Court of Appeals for the Third Circuit addressed several key theories and arguments that have been used against some early applications of machine learning—tools that identify excessive medical charges or unnecessary procedures. While ruling in favor of the insurer, the court left the status of those theories and arguments essentially unresolved. Because the same contentions are sure to re-surface when plaintiffs go after tools that rely on telematics, clickstreams and social media, Johnson provides a useful window into lawsuits yet to come.
In August 2004 – long before Big Data went viral – Sharon Anderson was involved in a motor vehicle accident and was treated for headaches, neck pain and back pain. Nearly a year later, in June 2005, she complained of similar symptoms during a visit to her physician; he prescribed 12 sessions of physical therapy, which Ms. Anderson received over the next four months. Both the doctor and the physical therapist submitted bills for their services directly to Ms. Anderson’s automobile insurer, seeking reimbursement under her policy’s Personal Injury Protection (“PIP”) feature. PIP coverage is mandatory in Delaware, where Ms. Anderson resides, and it requires insurers to pay certain “reasonable and necessary expenses” for medical services arising out of a covered accident. 21 Del. C. § 2118(a)(2).
In the course of processing these bills, the insurer had them reviewed by an automated system. One component of that system is a database, to which multiple insurers contribute information about the millions of bills they receive for healthcare services and equipment. By accessing the database, the system can compare the prices on a given bill with the prices for the same services that are charged by other providers in the same geographic area.
The system then performs a machine learning function known as “classification”: prices that exceed the 80th percentile in that area (i.e., prices that are higher than the ones charged by 80% of the area’s providers) are classified as unreasonably high. (In many jurisdictions, they are said to exceed the “usual, customary and reasonable,” or “UCR” rate.) Like most carriers, Ms. Anderson’s insurer responded to such bills by making a “UCR reduction”—paying only the 80th-percentile amount. That practice (the “geographic reduction rule”) resulted in a $31 reduction of the bill for Ms. Anderson’s June 2005 doctor visit.
The bill review system also implemented computerized “rules” to identify services that are unlikely to have been medically necessary. These rules (known as a “knowledge base”) were designed to reflect reimbursement policies which the insurer had formulated independently of the bill review system. (Applications that apply rules in this way are called “expert systems”; they were used widely in the 1970s and ’80s.) One such rule (the “passive modality rule”) directed the system to flag certain procedures, if they were performed more than eight weeks after the date of the accident. It caused Ms. Anderson’s insurer to deny payment for several of her physical therapy sessions, on the ground that “physical therapy modalities … provide no therapeutic benefit during the chronic period [and] are not reimbursable.” The insurer provided a similar explanation for denying additional bills for “stimulation” and “hot/cold pack treatment.”
Challenging Technology With Class Consciousness
In 2006, Ms. Anderson brought an action over her medical bills in a Delaware state court, and the insurer removed it to the District of Delaware. (The case was called Johnson v. GEICO, because Ms. Anderson had a co-plaintiff, Mr. Johnson, who dismissed his claims in 2015.) Ms. Anderson’s case was one of dozens of challenges to similar bill-review systems that were brought in the first decade of this century.
The complaint in Johnson alleged breach of contract—a claim Ms. Anderson could win by proving (i) that the charges submitted by her medical providers had been reasonable and necessary to treat injuries caused by her covered accident, and (ii) that her insurer had failed to pay them. The total amount of the alleged underpayments was slightly more than $1,000. Ms. Anderson, however, also asserted claims on behalf of three putative classes, including classes consisting of all other policyholders affected by either the geographic reduction rule or the passive modality rule. Where the certification of a class is a real possibility, the value of the case increases exponentially, and the plaintiff has considerable leverage in forcing a settlement.
Establishing that every member of a putative class received medically necessary services at a reasonable price would be cumbersome and expensive. Moreover, if questions about individual bills “predominate” over common questions, they can also prevent a court from certifying a class (a problem that is discussed below). Ms. Anderson’s complaint, like those in many similar cases, said little about the reasonableness of the providers’ bills, but focused instead on the proposition that the insurer’s bill review system was inherently unreasonable—making it inevitable that the insurer would “systematically and arbitrarily” reduce or deny payments that its policies required. Relatedly, the plaintiff contended that the computerized system was inconsistent with policyholders’ reasonable expectations about how their claims would be processed.
1. Breach of Contract
One possible way to establish that a system is inherently unreasonable is to contend that it uses flawed data, or that the data is in some way mishandled. In the past, allegations about systemic defects have been made against bill review systems of the type at issue in Johnson—especially those that used the database formerly known as “Ingenix,” which was itself the target of several class actions, as well as a highly-publicized investigation by New York’s Attorney General. See, e.g., M.W. Widoff, P.C. v. Encompass Ins. Co. of Am., No. 10 C 8159 (N.D. Ill. March 2, 1012) (alleging “selective data contribution,” “data scrubbing” and “flawed algorithms”); McGovern Physical Therapy Assocs. v. Metropolitan Pro. & Cas. Ins. Co., 802 F.Supp.2d 306 (D. Mass. 2011) (alleging system was “flawed, biased and unreliable”).
Ms. Anderson, however, chose a different approach. Instead of claiming her insurer’s system processed data incorrectly (a charge that might be difficult and expensive to prove), Ms. Anderson argued that the system was theoretically flawed: because it is possible for even a very high price to be reasonable in some circumstances, and for a contraindicated treatment to be medically necessary in some cases, Ms. Anderson—like most other plaintiffs in her position—asserted that her insurer’s system was necessarily overinclusive. See, e.g., Halvorson v. Auto-Owners Ins. Co., 85 Fed.R.Serv.3d 1491 (8th Cir. 2013) (alleging insurer “routinely caused nonpayment of reasonable medical expenses,” because reduction to 80th percentile is “not a method to identify and exclude outliers or unreasonable bills”); State Farm Mut. Auto. Ins. Co. v. Reyher, 266 P.3d 383 (Colo. 2012) (alleging insurer “systematically failed to pay … reasonable medical bills because it … us[ed] a database incapable of determining reasonableness”).
Ms. Anderson contended, for example, that the geographic reduction rule was “not based on scientific principles designed to ensure that reasonable payments are made,” because “[m]edical expenses are not unreasonably excessive simply because they exceed what is usually charged in the locality.” She maintained that a legitimate determination of reasonableness requires consideration of such individualized issues as whether the provider had “greater-than-usual expertise,” or whether he used “state-of-the-art (and hence more costly) equipment.” She similarly asserted that the passive modality rule was improper, because it caused the insurer to deny payment “without obtaining any independent medical or expert opinion” about the necessity of each individual treatment. Determinations based on medical generalizations, she contended, are “without any credible medical basis.”
One potential problem with this approach is that the automated system might be only one of several components of the bill review process. Consequently, the plaintiff must also assert (or, at least, suggest) that the insurer’s system actually precluded consideration of the unique circumstances of each case. The success or failure of that assertion is often important to the outcome of the case; in some instances, it has been dispositive. See, e.g., In re Farmers Med-Pay Litigation, 229 P.3d 551 (Okla. Ct. App. 2010) (certifying class on basis of allegation that insurer “had essentially abandoned an individualized approach to assessment of med-pay claims”); State Farm Mut. Auto. Ins. Co. v. Reyher, 266 P.3d 383 (Colo. 2012) (reversing class certification, based on testimony at certification hearing that system included an appeal process, and that each bill was “independently reviewed by the assigned adjuster”).
The evidence on this point is often equivocal. In Johnson, Ms. Anderson alleged that the passive modality rule was applied “without the possibility of human review”; but she also alleged that the insurer had written to her physical therapist at the time it denied payment, inviting the provider to “[s]ubmit medical records … includ[ing] … findings to indicate the appropriate use of the physical modality.” In Strawn v. Farmers Ins. Co., 350 Or. 336 (2011), cert den, 132 S.Ct. 1142 (2012), which affirmed a jury award in a class action against an insurer, the Oregon Supreme Court concluded that “the ‘recommendation’ [of the automated system] was, as a practical matter, the final determination of reasonableness,” because claims adjusters were graded on their adherence to those recommendations. On the other hand, in Halvorson, supra, the Eighth Circuit held that individual issues predominated over common questions, despite the plaintiffs’ assertions that adjusters “typically accept[ed] the recommendation of the [automated] bill reviewer.”
2. Bad Faith
Another way to keep the focus off of individual bills is to assert a cause of action for which failure to pay a “reasonable” amount is not an element. For example, in Bemis v. Safeco Ins. Co. of Am., 407 Ill.App.3d 1164 (2011), the plaintiffs maintained, as a matter of Illinois law, that “Medpay [i.e., medical payments] coverage language compels [the insurer] to pay the medical provider’s actual charges, absent fraud or bad faith on the provider’s part.” In other words, a system that reduced charges on the ground that they were unreasonably high, but without a finding of fraud or bad faith, would straightforwardly violate the insurer’s contract. (An Illinois Appellate Court initially accepted this argument, but reversed itself two years later.) In Ivanov v. Farmers Insurance Company of Oregon, 344 Or. 421, 185 P. 3d 417 (2008), the plaintiffs contended that Oregon’s PIP statute prohibits an insurer from denying any claim for lack of medical necessity, unless it first conducts an independent medical examination.
In Johnson, the plaintiff tried a slightly different angle, contending that the common law of Delaware made an insurer liable for breaching the duty of good faith it owes to each policyholder, even if the insurer did not breach its contract by reducing the policyholder’s payments. Thus, she maintained, the claim could be asserted on behalf of a class, even if some members of the class might not recover damages. The plaintiff argued further that this same rule applied to claims for “bad faith breach of contract.” Like the breach of contract claim, these legal theories ultimately depended on the factual proposition that the automated bill review system was inherently unreasonable—so much so, that the insurer could not have adopted it in good faith.
Ms. Anderson also made an argument based on Delaware’s Unfair Trade Practices Act, 18 Del. C. § 2304, which defines certain acts as “unfair claim settlement practices,” if they are performed “with such frequency as to indicate a general business practice.” Ms. Anderson claimed the statute’s requirements are implied by law into every Delaware policy. In other words, she asserted that insurers owe a duty to each individual policyholder—even those who are unaffected by a particular practice—to refrain from engaging in unfair practices systematically.
Plaintiffs challenging automated bill review systems have often tried to assert claims for fraud—both common law fraud and alleged violation of state consumer fraud statutes. In most cases, these claims are a variation of the breach of contract claim: they assert that the insurer misleadingly promised (in the policy itself) to pay “reasonable” expenses, knowing that it was employing a system that would cause “reasonable” bills to go unpaid. E.g., Strawn, supra (“[D]efendants offered their policies … representing that they would pay all reasonable medical expenses. They intentionally failed to explain … that, in general, they would not reimburse for any excess above a charge based on the 80th percentile”); M.W. Widoff, supra (alleging violation of Illinois Consumer Fraud Act, based on “[d]efendants’ concealment of their intent not to honor promises to pay reasonable expenses”).
In Johnson, Ms. Anderson alleged that her insurer’s policies contained implied representations that it “would neither delay nor deny payment … without reasonable justification,” and that it “would deal fairly with its insureds.” She alleged that the insurer “knew” these representations were “false”—because it employed a bill review system that it expected to delay or deny payment unreasonably in at least some cases.
Another proposition was at least implicit in the claims advanced in Johnson and Strawn: the suggestion that consumers enter into insurance contracts with expectations about the methods their insurer will or will not use to process claims. This argument posits that a reasonable consumer might choose not to purchase a policy from an insurer that uses an automated system to review medical bills. See, e.g., Ross-Randolph v. Allstate Ins. Co., No. DKC 99–3344 (D. Md. May 11, 2001) (plaintiffs alleged they were induced to maintain PIP coverage by representations that insurer would interpret policies “in accordance with Maryland law, recognizing Plaintiffs’ reasonable expectations”).
In some cases, courts have rejected theories like those advanced by Ms. Anderson in rulings on dispositive motions before trial. More commonly, the critical struggle is over class certification. Because the amount at issue for each individual claim is often minimal, the value of a case depends on maintaining at least a realistic possibility that a class will be certified.
Federal Rule 23(b)(3) (as well as many state rules that are modeled on it) permits class certification where “questions of law or fact common to class members predominate over any questions affecting only individual members.” There is no hard-and-fast rule for determining predominance in federal court, but it generally requires a showing that the class “will prevail or fail in unison” with respect to an essential element of the plaintiff’s claim. Amgen Inc. v. Connecticut Retirement Plans and Trust Funds, 133 S.Ct. 1184 (2013). Some states have developed clearer standards; in Illinois, for example, a plaintiff must show that success on her individual claim “will establish a right of recovery in other class members.” Avery v. State Farm Mut. Auto. Ins. Co., 216 Ill.2d 100 (2005).
In cases like Johnson, insurers have argued that plaintiffs may not assert breach of contract claims on behalf of a class, because the claim requires a showing that each claimant was charged a “reasonable” amount for “necessary” services; even if the named plaintiff’s bills are reasonable, that fact cannot establish reasonableness with respect to any other member of the proposed class. This argument has succeeded in Illinois. Bemis, supra; Shipley v. St. Paul Fire & Marine Ins. Co., No. 5-10-0619 (Ill. App. Ct. June 22, 2012). But the Supreme Court of Oregon rejected the argument in Strawn, supra, based on a questionable finding that the Oregon PIP statute creates a presumption that bills are reasonable and necessary. (A detailed discussion of the Strawn decision is available here.)
In In re Farmers Med-Pay Litigation, supra, an Oklahoma appellate court evaded the argument, ruling that it could not investigate the merits of an allegation that adjusters lacked discretion to override the automated system. That strategy probably will not work in federal court; in Wal-Mart Stores, Inc. v. Dukes, 131 S.Ct. 2541 (2011), the Supreme Court noted that “[t]he necessity of touching aspects of the merits in order to resolve preliminary matters … is a familiar feature of litigation.”
Johnson Changes Noting
In Johnson, Ms. Anderson pursued each of these arguments on a long and tortuous journey.
1. Certifying Classes
Three years after the case was filed, the district court found that a Delaware plaintiff who sues for breach of contract, based on a denial of PIP benefits, “bears the burden of proving the claimed expenses are both reasonable and necessary.” On that basis, it held (as the Illinois court had in Bemis, supra) that individual breach of contract issues predominated over common ones, and it declined to certify a class in connection with that claim. Johnson v. GEICO, 673 F.Supp.2d 255 (D. Del. 2009).
It also held, however, that the insurer’s conduct “may still amount to a breach of the duty of good faith and fair dealing[,] even if no underlying breach of contract ultimately occurs.” As for “bad faith breach of contract,” the court followed the approach of the Oklahoma court in In re Farmers Med-Pay:
[I]f, as Plaintiffs allege, Defendants reduce bills based solely on an 80th percentile cap, and deny bills based solely on treatment dates and [procedure]] codes which reflect passive modality claims, such conduct applies on a class-wide basis. Thus, the Court concludes that commonality is present, and that any individualized issues are not predominant.
The court therefore certified a “geographic reduction class”—consisting of policyholders who received services for which the insurer paid less than the billed amount under the geographic reduction rule—and a “passive modality class”—whose members received passive modality services for which the insurer refused to pay. Ms. Johnson represented both classes in asserting claims for breach of the covenant of good faith and for “bad faith breach of contract.” Id.
2. De-Certifying Classes
That was in 2009. In the intervening years, the insurer uncovered evidence that the symptoms for which Ms. Anderson was treated in 2005 might have been caused by a pre-existing condition, unrelated to her 2004 automobile accident—in which case the insurer would have no PIP liability at all. In 2014, a new judge found this evidence fatal to the claim for “bad faith breach of contract,” because he appeared to disagree with his predecessor about the law governing that claim. The court stated:
[Where] there is no breach of contract, there can … be no bad faith breach of contract.
Johnson v. GEICO, No. 06–408 (D. Del. June 16, 2014).
The court was also unconvinced, as a matter of fact, that the insurer’s bill review system “frustrated the overarching purpose of the insurance contacts”—a position that knocked out the claim for breach of the covenant of good faith. Finally, the court rejected the claim that the requirements of Delaware’s Unfair Trade Practices Act could be implied into Ms. Anderson’s policy, because it found no “evidence that the parties would have agreed to include the clause if they had negotiated the issue.” Id.
On the basis of these findings, the district court awarded summary judgment to the insurer. Id. The following year, for the same reasons, it de-certified the classes asserting claims for breach of covenant and bad faith breach of contract. Johnson v. GEICO, 310 F.R.D. 246 (D. Del. 2015).
3. A Tepid Affirmance
On appeal, the Third Circuit confirmed those conclusions, but it did so on the basis of narrow and somewhat obscure reasoning. The Court of Appeals affirmed the awards of summary judgment, on the ground that Ms. Anderson’s treatments fell outside the scope of her PIP coverage, because she could not demonstrate that her accident had caused the relevant injuries. The court thereby avoided ruling on the validity of Ms. Anderson’s underlying legal theories.
With respect to Ms. Anderson’s claim that requirements of the Unfair Trade Practices Act are implied by law into Delaware insurance policies, the court was similarly non-committal. In a footnote, it found that the plaintiff had “not offered any evidence of the parties’ intent at the time of contracting.” That finding was grounds for affirming the district court, but it left open the general question of whether consumers have specific expectations about the mechanics of claims handling.
But the court’s work was not over: the issues presented by the award of summary judgment were also relevant to the district court’s de-certification of the geographic reduction and passive modality classes. The Third Circuit affirmed that decision, and, in doing so, it appeared to find that “reasonableness” and “necessity” are essential elements of the two causes of action those classes were asserting. But the court’s actual language made it difficult to draw that conclusion with any confidence.
To begin with, the opinion stated that “individual proof of reasonableness, necessity, or causation would be required for a class member to prevail under the certified causes of action.” If, as this statement suggests, the claims could be established by a showing of “causation” only, then a class action remains a viable option for future cases, because an insurer that pays even a reduced amount on a PIP claim has already conceded that the underlying injury relates to a covered accident.
The court further muddied the waters with another statement. In many cases, the possibility that each class member might separately have to prove the amount of her damages is not an impediment to class certification; courts often permit individual “mini-trials” on damages after liability has been established on a class-wide basis. But the Third Circuit concluded its ruling by stating that damage calculations—as well as deference to the district court—were additional grounds for the decision:
Given those individualized inquiries, the individualized issues relating to the calculation of damages, and the fact that “[t]he trial court … possesses broad discretion …,” we cannot find that the District Court abused its discretion when it concluded that … predominance … was no longer satisfied … .
That statement leaves open the question of whether predominance would have been satisfied if damages were uniform.
To top things off, the opinion was also marked, “NOT PRECEDENTIAL,” pursuant to Third Circuit Internal Operating Procedures 5.3 and 5.7, meaning that the Court does not consider the case to be “binding precedent.”
What Will The Future Bring?
If there’s a lesson in Johnson, it’s that the complete arsenal of theories and arguments employed against an earlier generation of automated tools will be available and brought to bear against the fruits of the “Big Data” revolution. Policyholders will contend that new benchmarking and auditing tools violate their insurance contracts, by interfering with the prompt and appropriate resolution of their claims. They will assert that underwriting, fraud detection and activity optimization products cause insurers to act “unreasonably,” either because the tools rely on allegedly spurious correlations, or because they employ machine learning in a way that makes the insurer unaware of the ultimate basis for its own decisions. They will maintain that various statutes and state causes of action impose liability for presumed or speculative harm. And they will use these claims to maximize the exposure of defendant insurers.
Next year’s claims will also involve issues that Johnson did not present—such as disparate impact and the duties owed to first-party insureds during settlement negotiations. But those are lessons for another day.