Reliability Requires Bias Analysis: Seventh Circuit Tightens Rule 702 for Litigation-Generated Epidemiology and Upholds Return-Conditioned “100% Satisfaction” Warranties

Introduction

In Gilbert v. Lands' End, Inc., No. 23-3162 (7th Cir. Oct. 23, 2025), the Seventh Circuit affirmed summary judgment for Lands’ End against two sets of Delta Air Lines employees who alleged harm from new Delta uniforms supplied by Lands’ End. The “Andrews plaintiffs” (605 individuals) brought Wisconsin products liability claims (manufacturing defect, design defect, and failure to warn) for personal injuries allegedly caused by dye and chemical transfer from the garments. The “Gilbert plaintiffs” (five individuals) asserted breach of express warranty claims as intended third-party beneficiaries of the Delta–Lands’ End Uniform Apparel Agreement (UAA), including a “100% satisfaction guaranteed” clause.

After extensive discovery featuring textile, toxicology, epidemiology, and dermatology experts, the district court excluded plaintiffs’ key causation expert, found plaintiffs lacked admissible evidence of defect and causation, and granted summary judgment. On appeal, the Seventh Circuit addressed:

Whether plaintiffs’ experts established a product defect under Wisconsin law;
Whether plaintiffs’ causation expert applied reliable epidemiological methods under Federal Rule of Evidence 702 and Daubert;
Whether Lands’ End breached the UAA’s “100% satisfaction” warranty under Delaware law when plaintiffs did not return the garments.

Summary of the Opinion

The Seventh Circuit affirmed across the board.

No triable defect evidence: Test results showed “crocking” (dye transfer) and the presence of chemicals/metals in some garments, but plaintiffs’ experts did not tie those findings to a manufacturing or design defect. The textile expert conceded the chemicals identified (e.g., fluorine, sodium, magnesium, aluminum, silicon) commonly appear in textiles and were not present at concerning levels, and he identified no industry-wide colorfastness standard violated by the uniforms.
Exclusion of causation expert under Rule 702: The court upheld exclusion of the epidemiology expert (Dr. Freeman). He relied exclusively on plaintiff-only questionnaires created for discovery, failed to identify or control for bias, omitted the required preliminary association analysis before employing Bradford-Hill, and used a “distilled” three-factor Bradford-Hill proxy without justifying how it mapped onto the recognized nine-factor framework. His alternative use of the Naranjo algorithm (an adverse drug reaction tool) likewise lacked a reliable foundation.
Warranty claim fails for noncompliance with reasonable conditions: Applying Delaware law per the UAA’s choice-of-law clause, the court held that the “100% satisfaction” warranty reasonably required return of the garments with a form to obtain a refund or exchange. Because plaintiffs did not return the uniforms, and the record showed Delta employees who returned items received refunds, no breach occurred.

Detailed Analysis

Precedents and Authorities Cited

The panel clarified standards of review and marshaled both state substantive law and federal evidentiary law:

Summary judgment: De novo review under Rule 56, requiring no genuine dispute of material fact and entitlement to judgment as a matter of law (Thompson Corrugated Sys., Inc. v. Engico, S.R.L., 111 F.4th 747 (7th Cir. 2024)).
Expert testimony review: Application of Rule 702 is reviewed de novo; the decision to admit or exclude experts is reviewed for abuse of discretion (Chicago Joe’s Tea Room v. Broadview, 94 F.4th 588 (7th Cir. 2024); Daubert v. Merrell Dow (1993); Naeem v. McKesson (7th Cir. 2006)).
Wisconsin products liability framework: Plaintiffs needed evidence of a defect for all three claims (manufacturing, design, failure to warn) (Godoy v. DuPont, 2009 WI 78), and of causation (Murphy v. Columbus McKinnon, 2022 WI 109; Tanner v. Shoupe, 596 N.W.2d 805 (Wis. Ct. App. 1999)). Complex defect and causation questions require expert assistance (City of Cedarburg Light & Water Comm’n v. Allis-Chalmers, 148 N.W.2d 13 (Wis. 1967)).
Causation standard: The court noted Wisconsin employs a “substantial factor” standard (Burton v. DuPont, 994 F.3d 791 (7th Cir. 2021); Sumnicht v. Toyota, 360 N.W.2d 2 (Wis. 1984); Pieper v. Neuendorf, 274 N.W.2d 674 (Wis. 1979)), though the parties framed causation as general/specific causation and any argument about the proper standard was waived.
Rule 702 methodology versus conclusions: The panel carefully applied Seventh Circuit guidance distinguishing challenges to methodology versus the weight of conclusions (Stollings v. Ryobi, 725 F.3d 753 (7th Cir. 2013); Manpower, Inc. v. Ins. Co. of Pa., 732 F.3d 796 (7th Cir. 2013)) while emphasizing the 2023 amendment to Rule 702 clarifying that sufficiency of basis is an admissibility question.
Epidemiology references: The Federal Judicial Center’s Reference Manual on Scientific Evidence (3d ed.) guided the court’s expectations for bias detection, association measurement, and use of Bradford-Hill. The court contrasted plaintiffs’ approach with the Harvard “Flight Attendant Uniforms” study (McNeely et al., BMC Public Health 2017), which quantified associations and managed reporting bias.
Naranjo algorithm and Milward: Plaintiffs invoked a weight-of-the-evidence approach under Milward v. Acuity (1st Cir. 2011), but the court found Dr. Freeman did not undertake the method’s steps and, in any event, his analysis was undermined by the same data and methodological flaws.
Warranty and UCC: Delaware UCC § 2-316(1) permits reasonable limitations that do not negate an express warranty. The court relied on Bushendorf v. Freightliner, 13 F.3d 1024 (7th Cir. 1993) for the principle that a seller cannot disclaim an express warranty, but can reasonably condition its exercise. Delaware law allows beneficiary rights to be limited by contract (Crispo v. Musk, 304 A.3d 567 (Del. Ch. 2023)). Wisconsin choice-of-law rules enforce contractual clauses absent public-policy concerns (Drinkwater v. American Family (2006); Tanner v. Juniper Realty (7th Cir. 2006)).

Legal Reasoning

Defect Evidence Under Wisconsin Law

Plaintiffs’ defect theory began with mixed lab results:

“Crocking” tests showed dye transfer at scores between 3.5 and 5 out of 5, yet technicians noted garments met “general industry standards” for colorfastness.
X-ray and chemical assays detected various elements and compounds (e.g., fluorine, sodium, magnesium, silicon; PFOA; antimony; chromium; mercury; formaldehyde; bromine). Worn garments showed higher concentrations, possibly due to interactions with deodorants, lotions, and detergents.

The textile expert (Dr. Hauser) opined that improper dye processes caused crocking, but he did not state that the dye itself was hazardous, that chemical transfer occurred at injurious concentrations, or that transfer of identified chemicals routinely accompanies crocking. In deposition, he further undermined the defect claim by acknowledging the chemicals at issue are routinely present in textiles at non-concerning levels and that there is no broad, universal industry colorfastness standard—retailers and consumers negotiate acceptable limits. The toxicology expert (Dr. Apple) described “elevated” or “dangerous” levels in a report but disclaimed knowledge of permissible limits or exposure pathways in deposition. The medical expert (Dr. Scheinman) assumed exposure and opined on symptoms without linking any garment characteristic to a defect.

The panel held this record cannot sustain a products liability defect theory. Presence of chemicals, without expert testimony establishing a departure from specifications or an unreasonably dangerous design, does not create a triable issue. Nor did plaintiffs show that the garments’ dyeing process failed any cognizable industry or contractual standard. In short, the evidence might support property-damage claims based solely on dye transfer, but it did not establish a defect “in a manner resulting in the transfer of toxins” for personal injury claims.

Rule 702 Gatekeeping and Causation Experts

The Seventh Circuit affirmed exclusion of Dr. Freeman’s causation opinions for lack of reliability under Rule 702 and Daubert. The opinion is especially significant for three reasons: the treatment of litigation-generated survey data, the required preliminary association step for Bradford-Hill analyses, and the attempted use of the Naranjo algorithm outside the drug context.

1) Dataset reliability and bias analysis

Dr. Freeman relied exclusively on questionnaires administered to nearly 1,000 Delta employees who were already suing—i.e., a convenience sample of financially interested respondents. The questionnaires were drafted for discovery efficiency, not epidemiological research. Responses contained acknowledged errors and nonsensical entries. Critically, Dr. Freeman did not identify, assess, or attempt to control for obvious sources of bias (e.g., selection bias, reporting bias, recall bias), contrary to routine epidemiological practice as described in the Reference Manual.

The court rejected plaintiffs’ argument that the district court improperly scrutinized “data quality” rather than “methodology.” Here, the absence of bias assessment and error analysis was itself a methodological failure. The panel contrasted Dr. Freeman’s approach with the Harvard flight attendant study, which:

Used a control group;
Quantified associations between uniform exposure and symptoms;
Acknowledged and mitigated reporting bias by embedding uniform-related questions within a broader health survey.

The Seventh Circuit emphasized that, even under pre–2023 Rule 702 precedent (e.g., Manpower; Stollings), failure to apply ordinary, field-specific techniques to known data problems justifies exclusion. The court expressly noted the 2023 amendments to Rule 702, which clarify that “sufficiency of basis” is an admissibility question, but it did not need to rely on the amendment to affirm.

2) Bradford-Hill: prerequisite association and faithful application

Dr. Freeman performed a “distilled” three-factor Bradford-Hill analysis (plausibility, temporality, alternative causes) without first conducting an association analysis—that is, without demonstrating that symptoms occurred among exposed individuals more often than expected by chance (typically by quantifying risk relative to an unexposed control). He also did not compute or report any association metrics (e.g., odds ratios, risk ratios).

Beyond that threshold gap, he did not justify how his three factors mapped onto the recognized nine Bradford-Hill considerations. The literature he cited concerned very different contexts (vehicle crash trauma, arthritis, fibromyalgia, rotator cuff injuries) and did not explain how or why a truncated framework would be valid in a toxic exposure setting. While Bradford-Hill does not impose a rigid “counting” rule, an expert must reliably apply the framework and address the criteria’s core logic. The failure to bridge those gaps warranted exclusion.

3) Naranjo algorithm and Milward’s weight-of-the-evidence

Dr. Freeman alternatively applied a modified Naranjo algorithm—designed to assess causality in adverse drug reactions—to both general and specific causation. Lands’ End argued that the Naranjo tool lacks validation outside pharmacovigilance and that Dr. Freeman’s modifications skewed the scoring. The district court excluded the analysis, noting that Dr. Freeman had not undertaken the steps associated with a Milward-style “weight-of-the-evidence” methodology and that the same faulty questionnaire dependence undermined reliability.

Although the appellate panel observed the district court did not discuss in detail the general causation conclusions under Naranjo, it declined to remand because plaintiffs independently failed on defect and because the same foundational data flaws doomed any causation approach. The court thus left unresolved the broader question of whether and when Naranjo can be adapted beyond the drug context, signaling that any such adaptation must be justified with discipline-specific validation and transparent, bias-conscious application.

Breach of Express Warranty Under Delaware UCC

The UAA’s choice-of-law clause pointed to Delaware law, which Wisconsin courts honor absent public-policy conflicts. Delaware UCC § 2-316(1) allows reasonable contractual limitations that do not negate an express warranty, and Delaware law permits limiting third-party beneficiary rights in accordance with the agreement’s terms.

The UAA’s “100% satisfaction guaranteed” provision promised a refund or exchange “at any time” if an employee was not satisfied, subject to returning the product with a mutually agreed return/exchange form. The court held these were reasonable, commonplace conditions precedent that did not nullify the warranty. The record showed Delta employees who returned items received refunds; plaintiffs did not return their garments and therefore could not establish breach. The fact that the UAA was not initially public to employees did not excuse noncompliance, particularly given counsel’s knowledge of the terms during litigation and Lands’ End’s offer to refund upon return in response to plaintiffs’ demand letter.

Impact and Implications

Gilbert v. Lands’ End has immediate consequences for toxic tort litigation, expert practice under Rule 702, and commercial warranty enforcement.

Rule 702 gatekeeping post-2023 amendment: The Seventh Circuit’s analysis reinforces that courts will treat “sufficiency of basis” and bias control as admissibility questions, not merely jury weight. Experts relying on litigation-generated, plaintiff-only questionnaires must confront selection, recall, and reporting biases, and should validate findings with controls, record review, or independent data where feasible.
Epidemiology in toxic exposure cases: Before invoking Bradford-Hill, experts should quantify association (e.g., odds ratios) using appropriate comparators; if employing a truncated analysis, they must justify how it encompasses Bradford-Hill’s causal logic. Dose-response, consistency, and dechallenge/rechallenge evidence remain powerful. Failure to follow these norms risks wholesale exclusion.
Naranjo algorithm beyond pharmaceuticals: Courts will expect empirical justification for cross-domain use. Absent validation and faithful adherence to a recognized weight-of-the-evidence protocol (e.g., Milward), Naranjo-style scoring is vulnerable to exclusion.
Defect proof in product cases involving chemicals in common consumer goods: Demonstrating mere presence of chemicals or metals is insufficient; plaintiffs need expert evidence that levels deviate from specifications or render products unreasonably dangerous, and that exposure pathways and doses are sufficient to cause the alleged harm.
Warranty drafting and enforcement: Vendors may condition “satisfaction guaranteed” warranties on returns and standardized forms; such conditions are enforceable if reasonable and not negating the warranty. For mass claims, defense counsel should document open refund/exchange pathways. Plaintiffs’ counsel should advise clients to use available return channels or risk forfeiting warranty relief.
Choice of law clauses: The court’s straightforward enforcement of Delaware law underscores the value of clear contract clauses in nationwide programs.
What the case does not resolve: The panel did not decide the precise contours of Wisconsin’s “substantial factor” versus “general/specific” framing because of waiver; it did not determine the viability of Naranjo in non-pharma contexts; and it did not address property-damage class issues or the separate “defect-free” warranty claim, which were not on appeal.

Complex Concepts Simplified

Crocking: Industry term for dye rubbing off a fabric onto another surface (skin, clothing, furniture). Crockfastness tests measure resistance to such transfer.
Colorfastness standards: There is no single universal industry threshold; acceptable limits are often set by contract between buyer and supplier, sometimes by reference to standardized test scores (e.g., 1–5 scales).
Defect types (Wisconsin law):
- Manufacturing defect: the specific product departs from intended design or specs, making it unreasonably dangerous.
- Design defect: the intended design itself is unreasonably dangerous.
- Failure to warn: the product has a dangerous use that requires warnings or instructions, which were insufficient.
“Substantial factor” causation: Under Wisconsin tort law, conduct is a legal cause if it has such an effect in producing the injury that a reasonable person would regard it as a cause, in the ordinary sense. Parties often frame toxic torts as “general” (can this agent cause the harm?) and “specific” (did it cause this plaintiff’s harm?), but Wisconsin doctrine focuses on substantial factor.
Rule 702 and Daubert: Expert testimony must be based on sufficient facts or data; be the product of reliable principles and methods; and reflect a reliable application of those methods to the case facts. Since 2023, Rule 702 clarifies that sufficiency of basis is an admissibility requirement.
Bradford-Hill criteria: A nine-factor framework for assessing causation in epidemiology, including association strength, consistency, temporality, dose-response, plausibility, coherence, specificity, experiment (e.g., symptom resolution when exposure stops), and analogy. Typically preceded by quantifying an association.
Naranjo algorithm: A ten-question scoring tool originally developed to assess whether a drug likely caused an adverse reaction. Use beyond pharmaceuticals requires careful justification and validation.
Weight-of-the-evidence (Milward): A method synthesizing heterogeneous evidence streams. Reliable use requires a transparent protocol: identify the causal hypothesis, gather all relevant evidence, evaluate quality, assess coherence across streams, consider alternatives, and draw reasoned inferences.
Express warranty limitations (UCC § 2-316(1)): A seller may impose reasonable conditions (e.g., return of goods, completion of forms) so long as they do not negate the express warranty itself. Disclaimers cannot erase an express warranty but can regulate how it’s exercised.
Third-party beneficiary limits: Beneficiaries’ rights are defined and limited by the contract creating them; they must comply with conditions precedent.

Conclusion

Gilbert v. Lands’ End is a pointed reminder that federal courts will strictly police the reliability of expert causation testimony, particularly when based on litigation-generated, plaintiff-only surveys. The Seventh Circuit reinforced that epidemiological opinions must grapple with bias, quantify association, and faithfully apply accepted causal frameworks. Attempts to compress or repurpose methodologies—whether by “distilling” Bradford-Hill or importing the Naranjo algorithm—must be justified with discipline-specific rigor and transparent reasoning.

On the contract side, the court confirmed that broad satisfaction guarantees can be conditioned on returns and routine administrative steps without negating the warranty. Beneficiaries who decline to meet those conditions cannot claim breach, especially where the seller demonstrably honors returns.

Going forward, plaintiffs in toxic exposure cases should anticipate heightened Rule 702 scrutiny and design their expert workups accordingly—control groups, quantified associations, dose-response analyses, medical record corroboration, and explicit bias handling will be critical. Defendants, for their part, will find in Gilbert a robust template for challenging unreliable epidemiology and for enforcing reasonable warranty conditions. In both arenas, Gilbert advances methodological discipline and contractual clarity in the Seventh Circuit.