Causal Inference
for Regulatory-Grade Evidence
Overview
Sponsors increasingly need observational analyses that hold up under the same scrutiny as randomized trials — for external control arms, label expansions, post-authorization safety studies, and HTA submissions.
Our consultants contributed to the development of the target trial emulation framework, including the TARGET reporting statement, the structural classification of bias, g-methods, and have engaged directly with FDA and EMA on the methodological evaluation of RWE. We work with sponsor teams to specify defensible estimands, design analyses that target them, and document the assumptions on which causal interpretation rests.
Target trial Specification and Protocol Development
Most of the inferential damage in observational pharmacoepidemiology is done before any data are analyzed. We work with sponsors to articulate the causal question in target-trial form and to specify the estimand attributes that the analysis will target.
Typical deliverables:
​
-
Target trial protocol and its observational emulation
-
Explicit specification of time zero and alignment of eligibility, treatment assignment, and start of follow-up
-
Design alternatives when time zero specification is problematic
Target Trial Emulation
We design and execute target trial emulations using the analytic toolkit appropriate to the question selected on the basis of the causal contrast and the feasible assumptions according to the available data.
​
Applications include:
​
-
Comparison of sustained treatment strategies were time-varying confounders are affected by prior treatment and thus conventional adjustment methods do not work
-
Clone-censor-weight designs when, due to the observational nature of the data, the research question implies that patients cannot be classified into treatment strategies at the time when eligibility is applied (e.g., because the comparison is about different durations of treatment or because the use of grace periods)
-
Emulations supporting marketing authorization, label expansion, and HTA evidence packages
​​
External Control Arms and Hybrid Designs
External controls for clinical trials drawn from registries, claims, EHR, or completed trials are increasingly used to support marketing authorization. We support sponsors with:
-
Fit-for-purpose assessment of candidate data sources
-
Quantitative evaluation of overlap and effective sample size
-
Transportability analyses when source and target differ on effect modifiers
-
Quantitative bias analysis for unmeasured confounding and outcome misclassification
-
Review of recent regulatory precedents involving external controls and the methodological grounds on which they were accepted or contested
Regulatory Interaction Support
Our consultants have served as Special Government Employees at FDA, presented to FDA advisory committees, and contributed to regulatory-grade pharmacoepidemiology programs including the SIGMA consortium and EMA-aligned PASS networks.
We support the methodological dimensions of regulatory interactions where causal inference is at issue — FDA Type B and Type C meetings, EMA scientific advice and qualification procedures, and HTA engagements — including:
​
-
Methodological strategy for RWE submissions, positioned relative to current FDA RWE guidance and EMA reflection papers
-
Preparation of briefing documents with explicit articulation of the estimand and identifying assumptions
-
Responses to agency questions
-
Advisory committee preparation that anticipates common methodological objections
-
Post-meeting protocol amendments that address agency feedback
Independent Methodological Review
We review internal and vendor-led studies prior to database lock, submission, or publication. Reviews focus on:
​
-
Alignment between the stated estimand and the analysis that targets it
-
Credibility of the identifying assumptions (e.g., no-unmeasured-confounding, consistency)
-
Handling of post-baseline confounding and interpretation of the resulting estimate
-
Use of appropriate statistical methodology
Common findings include:
​
-
Conditioning on post-baseline variables leading to selection bias
-
Misguided per-protocol analyses
-
ITT estimates reported as if they answered a per-protocol question
Response strategies for agency, peer-review, and HTA critiques are also supported.
​
​​​
​
Training
We deliver technical training on causal inference for time-varying treatments, target trial emulation, estimand specification, and the diagnosis and remediation of design flaws in observational pharmacoepidemiology. Formats are calibrated to prior background in causal inference methodology.
