Third Circuit Requires Causation-Grade Statistical Proof (Not Mere Correlation) to Establish Predominance in Pharmaceutical-Fraud RICO TPP Classes; Confirms Issue-Limiting Authority in Rule 23(f) Appeals

1. Introduction

This precedential decision arises from long-running multidistrict litigation over Avandia, a Type II diabetes medication developed and marketed by GlaxoSmithKline LLC (“GSK”). Third-party payors (“TPPs”)—health plans and similar entities that reimbursed prescriptions—sought Rule 23(b)(3) certification of a nationwide class, alleging that GSK’s misrepresentations about Avandia’s cardiovascular risks and benefits violated RICO, 18 U.S.C. § 1962(c), and caused economic injury.

The Plans’ central liability theory was a “quantity-effect” chain: (1) GSK’s fraud influenced physicians’ prescribing decisions; (2) more Avandia prescriptions were written than would have been absent the fraud; and (3) TPPs reimbursed those additional, higher-cost prescriptions instead of cheaper alternatives. The appeal, accepted under Rule 23(f), presented two principal certification issues: (i) ascertainability and (ii) predominance as to causation (including the role of “reliance” in the causal chain). The Court also requested briefing on a threshold appellate-procedure issue: whether the Third Circuit may limit the scope of review in a Rule 23(f) appeal to selected questions.

2. Summary of the Opinion

3. Analysis

3.1. Precedents Cited

A. Appellate procedure: limiting issues in Rule 23(f) appeals

• Microsoft Corp. v. Baker, 582 U.S. 23 (2017): Used to frame Rule 23(f) as granting “unfettered discretion,” akin to § 1292(b), and as a discretionary interlocutory vehicle with fewer “limiting requirements.”
• Newton v. Merrill Lynch, Pierce, Fenner & Smith, Inc., 259 F.3d 154 (3d Cir. 2001): Cited for the proposition that Rule 23(f) models § 1292(b).
• Yamaha Motor Corp., U.S.A. v. Calhoun, 516 U.S. 199 (1996): Establishes that, in § 1292(b), “the order” is appealable (not the certified question), and the Court reads Yamaha’s permissive framing as consistent with appellate discretion to choose which issues to decide.
• Lundeen v. 10 W. Ferry St. Operations LLC, Barbato v. Greystone All., LLC, Gruber v. Price Waterhouse, Miller v. Bolger, In re Data Access Sys. Sec. Litig. (en banc), and related Third Circuit § 1292(b) authorities: Used to show the Third Circuit’s long practice of treating interlocutory review as discretionary in scope, guided by “prudential considerations,” and sometimes refusing to consider certain asserted grounds for reversal.

Influence on the holding: By equating Rule 23(f) discretion to (and even broader than) § 1292(b) discretion, the Court announced an institutional rule for future class-certification appeals: a motions/merits panel may accept an appeal yet confine review to particular certification issues, subject to mandatory consideration of subject-matter jurisdiction.

B. Rule 23(b)(3) certification framework in the Third Circuit

• In re Hydrogen Peroxide Antitrust Litig., 552 F.3d 305 (3d Cir. 2008): The anchor for “rigorous analysis,” preponderance burden, and the duty to resolve disputes relevant to class certification.
• Ferreras v. Am. Airlines, 946 F.3d 178 (3d Cir. 2019): Reinforces that courts must resolve “every dispute that is relevant to class certification” before granting certification.
• Tyson Foods, Inc. v. Bouaphakeo, 577 U.S. 442 (2016): Provides a functional test for statistical/representative evidence in predominance—permissible when it would be admissible in individual actions and when defenses are common.
• Comcast Corp. v. Behrend, 569 U.S. 27 (2013): Used to distinguish “data” from “what the data prove,” emphasizing that courts must scrutinize whether a model actually matches the theory of liability (and, here, whether the evidence speaks to the fraudulent conduct at issue).
• Amchem Prods., Inc. v. Windsor, 521 U.S. 591 (1997) and Gen. Tel. Co. of Sw. v. Falcon, 457 U.S. 147 (1982): Cited to underscore that predominance is not presumed even in fraud-like categories; Rule 23 elements must be proved.

Influence on the holding: These cases supplied two critical constraints. First, certification cannot rest on assumptions about proof at trial; the court must test the evidence. Second, where plaintiffs propose class-wide proof (including statistical proof), the court must confirm the proof actually bears on the disputed element—here, but-for causation via prescriber reliance.

C. Ascertainability doctrine

• In re Niaspan Antitrust Litig., 67 F.4th 118 (3d Cir. 2023); Hargrove v. Sleepy's LLC, 974 F.3d 467 (3d Cir. 2020): State the two-part ascertainability test (objective criteria; reliable and feasible membership mechanism).
• Byrd v. Aaron's Inc., 784 F.3d 154 (3d Cir. 2015); City Select Auto Sales Inc. v. BMW Bank of N. Am. Inc., 867 F.3d 434 (3d Cir. 2017); Kelly v. RealPage Inc., 47 F.4th 202 (3d Cir. 2022): Support using records plus affidavits and “yes-or-no” record review without devolving into “mini-trials.”

Influence on the holding: The Court treated the Plans’ proposed combination of purchase/reimbursement records and corroborated affidavits as squarely within established Third Circuit practice.

D. RICO causation and reliance as a link in the causal chain

• Sergeants Benevolent Ass'n Health & Welfare Fund v. Sanofi-Aventis U.S. LLP, 806 F.3d 71 (2d Cir. 2015): Central to the Court’s framing that “reliance” is often not an element but is commonly a necessary causal link; also central to the correlation-versus-causation critique and to the idea that sufficiently robust aggregate proof may establish class-wide causation.
• Harnish v. Widener Univ. Sch. of L., 833 F.3d 298 (3d Cir. 2016) and In re Cmty. Bank of N. Va. Mortg. Lending Pracs. Litig., 795 F.3d 380 (3d Cir. 2015): Used to reconcile when courts may (and may not) infer reliance from a common scheme—particularly distinguishing one-dimensional decisions (e.g., paying an allegedly inflated bill) from multidimensional medical prescribing decisions.
• In re Warfarin Sodium Antitrust Litig., 391 F.3d 516 (3d Cir. 2004): Demonstrates that where reliance is irrelevant to liability, common scheme evidence can support predominance; contrasted with Avandia where reliance is the causal hinge.
• United States v. Mosley, 454 F.3d 249 (3d Cir. 2006): Supplies a but-for causation definition and reinforces the need to test alternative causal hypotheses.

E. Statistical proof of causation in pharmaceutical-fraud settings (inter-circuit alignment)

• In re Neurontin Mktg. & Sales Pracs. Litig. ("Neurontin I"), 712 F.3d 21 (1st Cir. 2013); In re Neurontin Mktg. & Sales Pracs. Litig. ("Neurontin II"), 712 F.3d 51 (1st Cir. 2013); In re Neurontin Mktg. & Sales Pracs. Litig. ("Neurontin III"), 712 F.3d 60 (1st Cir. 2013): Treated regression analysis as a recognized method capable of proving but-for causation in analogous RICO pharmaceutical marketing cases.
• In re Celexa & Lexapro Mktg. & Sales Pracs. Litig., 915 F.3d 1 (1st Cir. 2019): Reinforced that statistical evidence quantifying prescriptions attributable to fraud (and the probability that a TPP paid for at least one such prescription) can support predominance.
• UFCW Local 1776 v. Eli Lilly & Co. ("Zyprexa"), 620 F.3d 121 (2d Cir. 2010): A cautionary comparator where evidence of prescription decline following risk disclosure was insufficient to prove class-wide causation.
• Painters & Allied Trades Dist. Council 82 Health Care Fund v. Takeda Pharm. Co., 2025 WL 1683472 (9th Cir. June 16, 2025): Approved certification where regression modeling controlled for confounders and tied disclosure failures to prescription declines.

Influence on the holding: The Third Circuit expressly “join[ed]” the First, Second (in suggestion and reasoning), and Ninth Circuits on a core principle: TPPs may rely on statistical evidence to establish prescriber-reliance causation, but the evidence must be “sufficiently rigorous” to show causation, not mere temporal correlation after a disclosure event.

F. Gatekeeping and methodological rigor

• Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993): The Court linked acceptable class-wide statistical causation proof to reliability standards that “pass muster” under Daubert.
• In re Paoli R.R. Yard PCB Litig., 916 F.2d 829 (3d Cir. 1990): Used earlier in the opinion for a definition of “meta-analysis,” but also emblematic of the Third Circuit’s attention to scientific methodology in litigation.

3.2. Legal Reasoning

A. The Court’s two separate doctrinal moves

The opinion makes two doctrinally significant moves that operate at different levels of the case:

• Procedural (appellate) rule: Because Rule 23(f) is modeled on § 1292(b) and confers at least as much discretion, the Third Circuit may accept a 23(f) appeal yet limit review to selected issues.
• Substantive certification rule: In a RICO pharmaceutical-fraud quantity-effect case, causation depends on prescriber reliance as a link; because reliance is typically individualized, plaintiffs must show that the element is susceptible to class-wide proof. Statistical evidence can do so, but only if it actually establishes causal attribution (not merely that a disclosure was followed by a prescription decline).

B. Why “common scheme” evidence did not carry predominance here

The District Court treated evidence of a uniform deceptive scheme (marketing materials, employee testimony, and the like) as supporting an inference of “class-wide reliance.” The Third Circuit rejected that approach as overbroad for two related reasons.

• No “fraud shortcut” to Rule 23(b)(3): Relying on In re Hydrogen Peroxide Antitrust Litig. and Amchem Prods., Inc. v. Windsor, the Court insisted on “actual, not presumed” compliance with Rule 23.
• Prescribing is not a one-dimensional decision: The Court distinguished settings like In re Cmty. Bank of N. Va. Mortg. Lending Pracs. Litig.—where paying an allegedly illegitimate charge may allow a streamlined reliance inference—from clinical prescribing decisions that vary across providers and patients and are influenced by multiple factors.

In short, “scheme” evidence may show wrongful conduct, but in this case it did not itself bridge the key causal gap: whether and how much that conduct changed prescribing behavior.

C. Correlation is not enough: the “Nissen drop” and the natural-experiment framing

The Plans leaned heavily on the market-wide decline in Avandia prescriptions after publication of the Nissen study and argued that earlier disclosure (e.g., ICT-37) would have caused a similar decline. The Court treated this as the same logical deficiency identified in UFCW Local 1776 v. Eli Lilly & Co. ("Zyprexa") and Sergeants Benevolent Ass'n Health & Welfare Fund v. Sanofi-Aventis U.S. LLP: a temporal association—disclosure then decline—does not, without controlling for confounders, establish that the disclosure (and thus the earlier nondisclosure/fraud) caused the prior prescription volume.

The Court did not reject “natural experiment” logic in principle; it rejected unsupported natural-experiment claims resting on correlation rather than an analysis that tests competing causal explanations. That is why it repeatedly referenced regression analysis as an exemplar: regression’s value is not “math for its own sake,” but its capacity to control for other variables and quantify causal effects.

D. The clarified evidentiary pathway for predominance on causation

The Court’s clarified standard can be stated as follows:

• TPP plaintiffs in pharmaceutical-fraud RICO cases may prove but-for causation (via prescriber reliance) with class-wide statistical evidence.
• But that evidence must be rigorous enough to support causal attribution—i.e., to “distinguish correlation from causation” by isolating the “causal significance” of the fraud-related variable and addressing alternative hypotheses.
• A regression analysis is a paradigmatic way to do this; the Court also allowed for “comparably robust” quantitative methods.
• Such statistical evidence is typically strengthened (and sometimes operationalized for class-wide inference) by circumstantial and direct evidence—consistent with Neurontin I, In re Celexa & Lexapro Mktg. & Sales Pracs. Litig., and Painters & Allied Trades Dist. Council 82 Health Care Fund v. Takeda Pharm. Co.

On the present record, that pathway was not satisfied because the Plans’ principal regression evidence (Dr. Rosenthal’s) was excluded under Daubert and not appealed, and the remaining statistical evidence did not isolate the alleged fraud’s causal effects.

E. Why vacatur and remand (not outright reversal) mattered

Two features drove the remand posture:

• Unresolved certification disputes: Under Ferreras v. Am. Airlines and In re Hydrogen Peroxide Antitrust Litig., the District Court’s failure to grapple with key challenges to the statistical evidence required vacatur.
• Possibility of adequate proof: The Court emphasized that the Plans argued (late) that Dr. McGuire’s work could potentially address causation and noted the District Court had reserved judgment on that question—leaving room for further fact-finding and, potentially, reopened discovery under Rule 16(b) for “good cause.”

3.3. Impact

A. On Third Circuit class certification in pharmaceutical-fraud RICO cases

The decision sets a demanding—but not categorically prohibitive—template for TPP RICO quantity-effect classes: plaintiffs must be prepared to offer (and defend under Daubert) econometric or similarly rigorous quantitative proof that the challenged misconduct caused incremental prescriptions. A “before/after disclosure” graph, without more, is unlikely to carry predominance where multiple confounders could explain prescribing trends.

B. On expert strategy and Daubert posture

The Plans’ experience illustrates a practical lesson: when a case’s predominance theory depends on statistical causation proof, exclusion of the core model can be certification-dispositive. Future plaintiffs in the Third Circuit will likely invest earlier in model design (variable selection, confounder controls, robustness checks) to withstand Daubert scrutiny—particularly given the opinion’s emphasis on false positives and “results-oriented” modeling flagged by the District Court below.

C. On the “inference of reliance” doctrine

By narrowing when courts may infer reliance from a common scheme, the Court limits the reach of “class-wide reliance” rhetoric in contexts where the relevant decision is not one-dimensional. That doctrinal tightening may influence not only pharmaceutical-fraud RICO suits, but also other fraud-based class actions where individualized decision-making is complex.

D. On Rule 23(f) appellate practice

The Rule 23(f) holding is likely to affect how litigants draft petitions and how motions panels frame granted appeals: parties should expect that the Third Circuit may accept interlocutory review to address a discrete legal question (e.g., a predominance standard) without necessarily reaching every certification issue.

E. The concurrence’s caution (what the opinion does—and does not—require)

Judge Shwartz concurred in part and in the judgment, agreeing remand was required but expressing concern about any implication that regression is “almost always” necessary. The majority formally stated that causation-grade statistical proof may take the form of regression analysis “or any comparably robust statistical method,” leaving doctrinal space for other rigorous causal-inference approaches. The tension signals that, on remand and in future cases, litigants may contest not merely whether statistics are needed, but what kinds of non-regression causal proof can satisfy predominance when combined with strong circumstantial evidence.

4. Complex Concepts Simplified

• But-for causation: An act is a but-for cause if the injury would not have happened without it (here: would the Plans have paid for the same number of Avandia prescriptions absent the fraud?).
• Reliance as a causal link (in RICO): Even if “reliance” is not an element of mail/wire fraud, the plaintiff often must show someone relied on the misrepresentation to connect the fraud to the plaintiff’s loss (here: physicians’ prescribing decisions).
• Correlation vs. causation: Two events moving together (prescriptions fell after Nissen) does not prove one caused the other; other factors may have changed at the same time.
• Regression analysis: A statistical method that estimates the relationship between an outcome (prescription volume) and potential causes (marketing, disclosures, time trends), while “controlling for” other variables—used to isolate causal effects when designed and validated appropriately.
• Natural experiment: Using a real-world event as if it were an experiment. The Court’s point: calling something a natural experiment does not eliminate the need to prove the event caused the observed change.
• Daubert: The evidentiary gatekeeping standard for expert testimony; unreliable methods (or models producing “false positives”) can be excluded, which can undermine class-wide proof.
• Predominance (Rule 23(b)(3)): Common issues must outweigh individualized ones; if proving an essential element requires individualized proof member-by-member, certification is inappropriate.
• Ascertainability: The class must be defined objectively and members must be identifiable through feasible, reliable methods (records plus corroborated affidavits, not mere self-identification).

5. Conclusion

In re: Avandia Marketing makes two durable contributions to Third Circuit law. First, it confirms that the Third Circuit may limit the scope of issues reviewed in a Rule 23(f) appeal, aligning Rule 23(f) practice with (and even more discretionary than) § 1292(b). Second—and more consequentially for pharmaceutical-fraud RICO litigation—it adopts a causation-focused predominance standard: TPP plaintiffs may use class-wide statistical evidence to prove prescriber-reliance but-for causation, yet the proof must be robust enough to establish causation rather than correlation by isolating the fraud’s effect and testing alternative explanations. Because the Plans’ remaining evidence largely showed only a post-disclosure correlation and the District Court did not resolve key disputes, the Third Circuit vacated certification and remanded for further proceedings under the clarified standard.