What JAMA’s Causal Language Guidance Actually Says About Difference-in-Differences Studies

On August 13, 2026, JAMA published a methodological guidance piece titled “Causal Language for Studies Using Difference-in-Differences Analyses,” addressed to the clinical research community at large. It is not a regulatory ruling, and it carries no enforcement weight. That is precisely why it deserves closer reading than the headlines will give it. Press coverage of methodological guidance tends to flatten nuance into a single takeaway — “use causal language carefully” — and stop there. But the document itself is more specific, more operationally consequential, and more uncomfortable for sponsors than that framing suggests. The piece engages directly with the difference-in-differences (DiD) design, one of the most commonly invoked quasi-experimental methods in real-world evidence packages submitted to FDA, and it draws lines that the field has been avoiding drawing for years.

“Many clinical investigations are intended to evaluate or demonstrate causal relationships (eg, the effect of a treatment or other intervention on patient outcomes), whether using information from well-designed randomized clinical trials, leveraging observational data with suitable methods, or combining the 2 approaches. Evaluating and communicating the degree to which a given study supports a causal conclusion can be complex, typically involving both clinical and methodological considerations, and using appropriate language to describe the relevant considerations — causal language — is critical.”

This opening framing does something quiet but significant: it places observational methods and RCT methods on a continuum of causal evidence rather than treating them as categorically separate epistemologies. The phrase “with suitable methods” is doing real work here. Suitability is not assumed; it must be demonstrated. Under FDA’s March 2024 draft guidance on non-interventional studies, this same logic governs whether an observational submission can support an effectiveness claim — the agency is explicit that reliability and relevance of the data source must be established before causal framing becomes defensible. JAMA is aligning its editorial standards with the regulatory direction of travel.

Where DiD Studies Earn Their Claims

The document then moves quickly from philosophy to mechanics, and this is where its forensic value begins. The guidance does not condemn DiD designs. Sponsors who read this piece as a warning against quasi-experimental methods will misread it. What JAMA is drawing a line around is the gap between what a DiD design can establish and the language authors routinely use to describe it.

“The difference-in-differences design relies on the parallel trends assumption — the assumption that, in the absence of the intervention, the outcomes in the treatment and comparison groups would have followed the same trajectory over time. When this assumption holds, the design can support causal conclusions. When it does not hold, the design estimates a quantity that conflates the effect of the intervention with pre-existing differences in trends between groups.”

The parallel trends assumption is the load-bearing wall of every DiD analysis, and the guidance names it explicitly because most published DiD studies treat it as a formality rather than an empirical question. This section maps directly onto ICH E9(R1)’s framework for estimands and sensitivity analyses: the assumption is not decoration, it is a condition that must be interrogated, tested where possible, and transparently reported. A DiD study that asserts causal effects without documenting parallel trends testing has not earned its causal language, regardless of how impressive the point estimate looks.

The implication for regulatory submissions is direct. FDA’s 2024 draft guidance on non-interventional studies calls for sponsors to pre-specify their analytical assumptions and demonstrate their robustness. An RWE package built on a DiD design that does not include a pre-intervention trend analysis and a sensitivity test for assumption violation will face questions in review that the guidance literature is now making harder to deflect.

“Authors of studies using difference-in-differences analyses should explicitly state whether and how the parallel trends assumption was assessed, report the results of any such assessment, and qualify their causal language accordingly. When the assumption cannot be assessed or its plausibility is uncertain, language should reflect that uncertainty rather than assert causal effects.”

The phrase “qualify their causal language accordingly” is the operational core of the entire document. JAMA is not asking authors to abandon words like “effect” or “caused.” It is asking those words to be conditioned on demonstrated assumption validity. The standard here is closer to what a statistician would call “honest uncertainty quantification” than it is to the binary “causal vs. associational” framing that most methods sections currently use. Reviewers at FDA, reading a DiD-based RWE submission, will now have a published JAMA standard to cite when asking whether the parallel trends assumption was adequately interrogated.

The Language Gap Between Journals and Regulators

There is a counterintuitive dynamic embedded in this guidance that the clinical research community has not yet fully absorbed. The common assumption is that regulatory language is more conservative than academic language — that journals push further toward causal claims while agencies pull back toward association. The JAMA piece inverts that assumption in a specific and important way.

“The use of causal language in observational research has increased substantially in recent years, paralleling the development of causal inference methods and frameworks. However, the use of causal language does not itself establish a causal relationship; it reflects the authors’ judgment that the study design and analytic approach are sufficient to support such an interpretation. Readers, editors, peer reviewers, and downstream users of research — including policymakers and clinicians — may not always have the methodological expertise to evaluate that judgment independently.”

This passage is doing something genuinely important: it locates the risk not in bad-faith misrepresentation but in the asymmetry of expertise between authors and readers. A principal investigator who uses “DiD-estimated causal effect” in a manuscript abstract may be technically accurate within the assumptions of the design, but a payer medical director reading that abstract to inform a formulary decision does not have the tools to evaluate whether those assumptions held. The 2020 retraction of the hydroxychloroquine observational study from The Lancet — which used registry data to draw conclusions about treatment effects without adequate assumption testing — demonstrated precisely this failure mode at scale and at speed. The damage from that paper was not confined to its authors; it cascaded through clinical practice guidelines, regulatory emergency authorizations, and public trust.

FDA’s own guidance on evidence-based review for health claims makes the same point from the regulatory direction: observational studies generally cannot be used to rule out alternative explanations for observed associations, and the agency treats intervention studies as the evidentiary standard against which observational submissions are compared. The JAMA guidance is now giving that regulatory posture a peer-reviewed methodological foundation.

“We recommend that authors using difference-in-differences designs adopt language that is transparent about the assumptions underlying causal interpretation, including explicit statements such as ‘under the parallel trends assumption’ when presenting effect estimates, and that they avoid language implying certainty of causal effect when such certainty is not warranted by the design or the results of assumption assessments.”

The phrase “under the parallel trends assumption” as a parenthetical qualifier in an abstract or results section represents a small typographic change with real downstream consequences. It signals to an FDA reviewer, a payer analyst, or a meta-analyst that the causal claim is conditional, not absolute. It is the methodological equivalent of a confidence interval: it does not weaken the finding, it contextualizes it honestly. Sponsors building RWE packages for label expansion or post-marketing commitment fulfillment should treat this language convention as a compliance expectation, not an optional stylistic choice, given where FDA’s draft guidance on non-interventional studies is pointing.

What This Means Before the Next Submission

Read end-to-end, the JAMA guidance is not a methodological attack on DiD designs. It is a quality standard for how those designs communicate their own limitations. The document fits a pattern visible across the last 18 months of methodological literature: a convergence between academic journals and regulatory agencies around the idea that causal inference language must be earned through transparent assumption documentation, not asserted through design labeling. A 2023 NEJM perspective on causal inference from observational data made the same argument at the public health level. FDA’s March 2024 draft guidance operationalized it at the submission level. JAMA has now anchored it at the publication level. The three documents are converging on a single standard, and sponsors working with DiD-based RWE need to treat them as a unified set of expectations rather than separate conversations happening in parallel silos.

The piece also reveals something about where peer review is heading. JAMA’s explicit recommendation that authors include assumption assessment results in the main text — not supplementary materials — signals an editorial shift. Assumption tests buried in an appendix have historically allowed the causal language to travel through citations without the caveats. That gap is closing. Endpoints News noted in its coverage of the FDA’s March 2024 draft guidance that the agency’s 8-page document was designed to provide clarity on when drug sponsors can use observational studies to demonstrate effectiveness — and “clarity” in that context means documented conditions, not implied adequacy. JAMA is enforcing the same standard one manuscript at a time.

For sponsors with DiD-based analyses in active FDA submissions or in preparation, the immediate action is a document-level audit: locate every instance of causal language in the study report, clinical summary, or integrated analysis, and map it to an explicit parallel trends assessment with documented results. Where that assessment was not conducted or cannot be reconstructed, the language must be downgraded to associational framing. This is not a concession of weakness — a well-documented DiD study with honest causal language qualifications will withstand FDA review more durably than one that asserts causal effects and cannot defend the assumption architecture when reviewers probe it.

The next regulatory document to watch in this sequence is FDA’s final guidance on non-interventional studies, which will convert the March 2024 draft’s recommendations into binding expectations. When that document issues, the JAMA causal language standard will shift from best practice to the baseline against which submissions are evaluated.

References

1. JAMA — “Causal Language for Studies Using Difference-in-Differences Analyses”
2. Federal Register — “Real-World Evidence: Considerations Regarding Non-Interventional Studies for Drug and Biological Products” (FDA Draft Guidance, March 2024)
3. MDedge — “Lancet, NEJM Retract Studies on Hydroxychloroquine and COVID-19”
4. FDA — “Guidance for Industry: Evidence-Based Review System for the Scientific Evaluation of Health Claims”
5. PMC / NEJM — “Methods of Public Health Research: Strengthening Causal Inference from Observational Data”
6. Endpoints News — “When Can Observational Studies Support a Drug’s Effectiveness? FDA Explains in New Draft Guidance”