The Paraquat Playbook: Why Written Inclusion Criteria Save Your Case (In re Paraquat)

By John Mahoney · Founder, MedLegal AI · May 2, 2026

In April 2024, Chief Judge Nancy Rosenstengel of the Southern District of Illinois excluded Dr. Martin Wells, the plaintiffs' general-causation epidemiologist in the Paraquat MDL. Wells had built a meta-analysis from a starting pool of 36 studies. He had narrowed that pool to seven. Of those seven, one study contributed roughly 74% of the analytic weight — even though, by Wells's own stated criteria, that study should have been excluded because it included residential exposures.

The court's central problem with Wells was not the statistics. It was that the inclusion and exclusion criteria for the meta-analysis had never been reduced to writing before the rebuttal report and were not objectively replicable. Method by reverse engineering.

Four bellwether plaintiffs lost on summary judgment when Wells went down. He was the only general-causation expert.

If your case rides on epidemiology — toxic tort, pharmaceutical, occupational exposure — Paraquat is the case that explains why the order of methodology steps matters as much as the methodology itself.

What Happened in In re Paraquat

The plaintiffs alleged that occupational paraquat exposure caused Parkinson's disease. The general-causation theory required an epidemiologic showing that the exposure causes the disease at a population level. Wells, a credentialed epidemiologist, was retained to perform that showing.

His methodology was a meta-analysis. He started with a pool of 36 epidemiology studies bearing on paraquat and Parkinson's. He narrowed the pool to seven studies for inclusion in the meta-analysis. He pooled effect sizes and reported a result.

The defense moved to exclude. Three findings drove the court's decision:

1. No written inclusion/exclusion criteria pre-analysis. The criteria Wells used to narrow 36 studies to seven were never reduced to writing before the rebuttal report. The court found that without pre-specified criteria, the selection process was not objectively replicable — a different epidemiologist with the same starting pool could not be expected to arrive at the same seven studies.
2. Internal inconsistency in the criteria as articulated. When Wells did articulate criteria — eventually — one of his stated criteria was that residential-exposure studies should be excluded. But one of the seven studies he included did include residential exposures, and that study contributed roughly 74% of the analytic weight in the pooled result. Either the criterion was not really applied, or the result was driven by a study that should not have been there. Either reading was fatal.
3. Methodology, not credentials. Wells was a credentialed epidemiologist. The exclusion was not about him personally. It was about the methodology as applied — and the post-2023 amendment to FRE 702 specifically directs courts to evaluate whether the expert "reliably applied" the methodology, not just whether the methodology is in principle valid.

Summary judgment for the four bellwether plaintiffs followed. With no general-causation expert, the cases ended.

Why It Matters for Plaintiff Attorneys

The Paraquat exclusion is a methodology problem, not a science problem. Meta-analysis is a perfectly valid methodology for general-causation epidemiology. Many of the cases that have survived Daubert in toxic-tort contexts have involved meta-analyses. The difference is procedural — whether the inclusion criteria were pre-specified, written down, replicable, and consistently applied.

This is the same underlying problem that drove the Acetaminophen MDL's collapse, in a different methodological form. In Acetaminophen, the experts cherry-picked studies in support of their Bradford Hill weighting. In Paraquat, the expert effectively cherry-picked studies in his meta-analytic pool. Different methods, identical defect: the science was reverse-engineered from the desired result.

The post-2023 FRE 702 amendment — and the Fourth Circuit's Sardis v. Overhead Door framework that the amendment formalized — directs courts to treat reliability as admissibility, not weight. A meta-analysis with retroactively-articulated criteria is not "reliably applied" in the amendment's sense. The proponent has not carried the preponderance burden of admissibility, even if the underlying science is sound.

For plaintiff attorneys, the implication is procedural. Your epidemiology expert should be able to hand the court a written, time-stamped protocol that pre-dates the literature search. That protocol should specify: the population of studies to be searched, the search terms, the databases, the inclusion criteria, the exclusion criteria, the planned statistical methodology, and the planned sensitivity analyses. The protocol should be dated. It should not be revised after the studies are pulled.

Most plaintiff-side meta-analyses do not work this way. The expert is hired, given a research question, and produces a meta-analysis as part of the report. The criteria — to the extent they exist — are inferred from the report. That sequence is exactly what the Paraquat court found problematic, and it is exactly what the defense bar is now attacking.

How to Avoke This in Your Next Case

Four concrete steps:

1. Require a pre-specified protocol before the literature search begins. Before your expert pulls a single study, the engagement should produce a written protocol with: research question, search strategy, databases, inclusion criteria, exclusion criteria, planned statistical analyses, and planned sensitivity analyses. The protocol should be signed and dated. Save it. Produce it in discovery.

2. Run a sensitivity analysis on the dominant study. If one study contributes more than half the weight in the meta-analysis — as happened in Paraquat — the report needs a sensitivity analysis showing what the result would be without that study. If the result reverses, that has to be addressed openly. Hiding that fact is the kind of methodology problem the defense bar will surface in deposition.

3. Document every excluded study by name and reason. A complete enumeration of studies considered, with a documented reason for each exclusion mapped to the pre-specified criteria, is the single most defensible procedural artifact. It makes the methodology objectively replicable. It defeats the central Paraquat objection.

4. Plan for the second-bite scenario. The Acetaminophen MDL gave plaintiffs a second bite with replacement experts after the initial exclusion. The replacement experts went the same way. Plan from the start as if your meta-analysis will be challenged — because in any high-stakes mass tort it will be. Build the protocol once, build it to survive scrutiny, and use the same protocol if you need to retain a replacement expert.

A broader point: meta-analytic methodology is a process. The process leaves artifacts. The artifacts — the protocol, the search log, the inclusion/exclusion record, the sensitivity analyses — are what the gatekeeper actually evaluates. An expert who has done the work but did not generate the artifacts is, after 2023, indistinguishable from an expert who did not do the work.

How a Daubert Workup Tool Can Help

MedLegal AI's Daubert workup tool ships a meta-analysis methodology audit module designed for exactly the Paraquat failure mode. The tool requires the expert to submit (a) the date the inclusion/exclusion criteria were finalized, (b) the criteria themselves in writing, (c) a complete enumeration of studies considered, and (d) a documented reason for each exclusion. The tool flags any meta-analysis where the criteria appear to have been written after the studies were selected, and runs a sensitivity check on the dominant study. The point is to prepare smarter for Daubert challenges — to give your epidemiology expert a procedural audit trail that mirrors what the gatekeeping order will look for. You can run a free Daubert workup on your expert here.

Bottom Line

The Paraquat exclusion is not a critique of meta-analysis. It is a critique of meta-analysis without procedural rigor. The cases that survive Daubert in 2026 will be the ones whose epidemiology workup produces an audit trail — pre-specified protocol, dated search log, complete exclusion record, sensitivity analyses on the dominant study — that the gatekeeper can replicate. That audit trail is cheap to build at the start of the engagement. It is impossible to build retroactively. Plan accordingly.

Run a free Daubert workup on your expert.

MedLegal AI is software, not a law firm. We do not provide legal advice. All AI-generated outputs require independent review by a licensed attorney.

← All posts · About John Mahoney · Free Daubert tool · Start a 14-day free trial