What the Studies Actually Say: Scientific Literature and Expert Reliability under Daubert

This is the third part in a “Law Meets Science” series on how the scientific record affects expert evidence in litigation. Part 1 examined the implications of unpublished and selectively reported research. Part 2 focused on the potential impact of evolving scientific knowledge on the reliability of expert opinions over time.  Now, in Part 3, we examine what experts can reliably conclude from the published science and the implications for admissibility in the courtroom.

Epidemiologic studies typically evaluate associations between exposures and outcomes, often along with statistical uncertainty and potential sources of bias. In litigation, experts may rely on those studies as part of the basis for opinions about causation, even when the study authors themselves do not conclude a causal relationship. Determining whether an observed association supports a causal conclusion, however, generally requires consideration of the broader body of evidence.  The distinction between what the scientific literature reports and the causal conclusions an expert draws from that literature can have significant consequences under Federal Rule of Evidence 702 and Daubert, as the acetaminophen multidistrict litigation illustrates.

The Southern District of New York’s December 2023 ruling in In re Acetaminophen – ASD-ADHD Products Liability Litigation excluded all five plaintiffs' general causation experts, who offered opinions on whether prenatal acetaminophen exposure can cause autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD). The court deemed each of these experts to be "eminently qualified" but concluded that none had applied their methods reliably to the question of causation. Four themes emerge from the district court’s ruling:

• Association does not equal causation. A positive association may occur due to chance, bias, or confounding rather than being causal. The district court noted that one of the plaintiffs’ own experts characterized the association between prenatal acetaminophen exposure and ADHD as modest.
• Potential confounding requires careful consideration. The court identified two major potential confounders: confounding by indication and genetic confounding. Confounding by indication may occur when the reason for taking acetaminophen, such as fever or infection, is itself associated with ASD or ADHD. Genetic confounding may occur when genetic factors are associated with both acetaminophen use during pregnancy and the likelihood of ASD or ADHD in offspring. More broadly, familial confounding may arise from genetic or shared environmental factors. The court cited a sibling-control study (Gustavson et al., 2021) that compared ADHD risk among siblings who differed in prenatal acetaminophen exposure. In that study, the association between acetaminophen and ADHD weakened to non-significance in the sibling-control analysis, leading the authors to conclude that it “may at least partly be due to familial confounding.”
• Accepted methods must be applied reliably. The district court scrutinized how the experts applied the Bradford Hill criteria, including whether the experts explained their conclusions and how the factors informed their overall causal assessment.
• Medical and regulatory authorities had not reached the same causal conclusions. No medical organization or regulatory body, including the Food and Drug Administration, had concluded that prenatal acetaminophen exposure causes ASD or ADHD.

Applying these principles, the district court excluded each of the five experts. In each case, the problem was how the expert applied an accepted methodology, not the methodology itself:  

1. The epidemiologist performed a single Bradford Hill analysis that treated ADHD and ASD together and combined studies measuring symptoms with studies measuring clinical diagnoses. The district court concluded that this combined approach obscured weaknesses in the evidence, particularly the more limited research on ASD. It also found that he did not reliably address inconsistent study results or genetic confounding. 
2. The psychiatrist likewise supported analyzing ADHD and ASD together and relied on the epidemiologist’s assessment of the epidemiologic evidence, so the district court found his analysis suffered from similar defects. 
3. The teratologist did not explain how he weighed the Bradford Hill factors in reaching his overall causal conclusion, and the district court found that he did not adequately account for inconsistent findings and study limitations. 
4. The professor of clinical pharmacy opined that risk increased after 28 cumulative days of acetaminophen use but did not adequately explain how the studies supported that specific threshold. 
5. The toxicologist relied on preclinical evidence to support biological plausibility, but the district court found that he did not reliably account for the heterogeneous and inconsistent results in that research.

Because these five exclusions left plaintiffs without admissible general causation evidence, the district court granted summary judgment for the defendants.

But gatekeeping has limits. In July 2026, the United States Court of Appeals for the Second Circuit vacated the district court’s judgments and remanded. Significantly, the Second Circuit held that the district court exceeded its discretion in excluding three of the five experts: the epidemiologist, psychiatrist, and toxicologist. It upheld the exclusions of the teratologist and professor of clinical pharmacy. The Second Circuit emphasized that it was deciding only the admissibility of the expert opinions, not whether acetaminophen actually causes ASD or ADHD. As to the three reinstated experts, the appellate court concluded that the district court went beyond assessing reliability and resolved scientific disagreements that should not have been decided at the admissibility stage. The distinction is important: Rule 702 and Daubert require an expert’s methodology and its application to be reliable, but they do not require the expert’s scientific conclusion to be correct. Where qualified experts using reliable methods can legitimately reach different scientific conclusions, resolving that disagreement is generally for the jury rather than the court at the admissibility stage.

Evaluating an expert's reliance on scientific literature requires looking beyond the citations to the studies themselves: what question was studied, how exposure and outcomes were measured, what alternative explanations remain, what limitations the authors acknowledged, and whether the expert's conclusion accurately reflects the evidence. Courts, in turn, must test the reliability of that analysis without resolving genuine scientific disputes themselves. That is where law truly meets science.

Together, the three installments of this series highlight different dimensions of expert reliability: whether the scientific record is complete, how it evolves over time, and whether an expert's conclusions reliably reflect what the evidence supports.

If you have any questions about the issues raised in this series, please contact the authors or the Womble Bond Dickinson attorneys with whom you normally work.