Why FDA issued the guidance
FDA's 2023 draft guidance, 'Considerations for the Design and Conduct of Externally Controlled Trials for Drug and Biological Products,' codifies what the Agency has been signaling at advisory committees for years: external control arms (ECAs) can support substantial evidence of effectiveness, but only when sponsors can defend the comparability of the external cohort, the relevance and reliability of the source data, and the analytic plan that links the two.
The bar is highest when the control is fully external (no concurrent randomized arm) and the endpoint is subject to ascertainment or measurement bias. ECAs are most defensible in single-arm trials of rare or serious diseases where randomization is impractical and the natural history is well characterized.
Fit-for-use data assessment
Before any propensity score is fit, the data source must be assessed for relevance (does it capture the target population, treatments, and outcomes the trial measures?) and reliability (are the data complete, accurate, and provenance-documented?).
We build a data quality memo that maps every protocol-defined variable to its source field, documents missingness patterns, and pre-specifies the imputation or sensitivity analyses that will address them. This memo becomes the backbone of the Type C briefing book.
Propensity scoring choices that hold up
Variable selection should be driven by a documented causal framework — typically a directed acyclic graph (DAG) — not by stepwise statistical selection. Include confounders of treatment and outcome; exclude instruments and colliders.
Pre-specify the matching or weighting estimator (1:1 nearest neighbor with caliper, IPTW with stabilized weights, overlap weights), the caliper width, and the diagnostics that will declare balance achieved (standardized mean differences <0.1 on all covariates).
Pre-specification choices that survive AdComm
Lock the protocol, statistical analysis plan, and data quality memo before unblinding the external cohort to treatment assignment in the trial arm. Document the lock in a date-stamped, version-controlled package.
Plan for tipping-point and E-value sensitivity analyses up front. Advisory committees consistently ask how robust the result is to an unmeasured confounder — answer that question in the primary submission, not during the meeting.
Key takeaways
- Treat the data quality memo as a primary regulatory deliverable, not an appendix.
- Pre-specify propensity methodology, balance diagnostics, and sensitivity analyses before the external cohort is touched.
- Bring tipping-point and E-value analyses to the first FDA meeting — don't wait to be asked.
Want this brief as a PDF, or to discuss it with our team?
We share the full deck and supporting references on request.