FDA Draft Guidance: Integrating Randomized Controlled Trials into Routine Clinical Practice

What it is

— Integrating Randomized Controlled Trials for Drug and Biological Products Into Routine Clinical Practice is a draft guidance for industry issued by the U.S. FDA (CDER and CBER) in September 2024 (Docket FDA-2024-D-2052; comment period closed December 2024). It is part of FDA's Real-World Evidence (RWE) program and sits alongside the Agency's framework and the RWD-source guidances. Its purpose is to describe how sponsors can run randomized controlled trials with simplified, streamlined protocols and procedures focused on essential data, embedding the trial in ordinary clinical care and leveraging real-world data infrastructure (EHR, disease and product registries, claims) for recruitment, baseline characterization, endpoint capture, and follow-up. These designs are also called pragmatic trials, point-of-care trials, or large simple trials. The defining feature that separates this guidance from the rest of FDA's RWE suite is that randomization is retained — the real-world component is the data and care setting, not the causal contrast. It is a regulatory framework (what makes such a trial acceptable as substantial evidence), not a reporting checklist and not a risk-of-bias instrument.

When to use

— Reach for this guidance when you are planning or defending a randomized, interventional trial that will be conducted under real-world conditions for an FDA submission: broad eligibility, care delivered by treating clinicians rather than dedicated research staff, objective endpoints (death, hospitalization, major clinical events) ascertained from routinely collected data, and risk-based monitoring. Decision rule for picking the right document: if the study randomizes treatment and uses RWD mainly for data capture and follow-up, this guidance governs; if the study is non-interventional (treatment decisions made in routine care, no randomization, confounding controlled by design and analysis), use `fda-rwe-noninterventional` and `fda-rwd-ehr-claims` instead. PRECIS-2 helps you position a design on the explanatory–pragmatic continuum and CONSORT-Pragmatic governs journal reporting; this FDA guidance addresses regulatory acceptability — they are complementary, not interchangeable. Registry-based randomized trials (RRTs) that randomize within an existing registry fall squarely in scope.

What it requires

— The substantive expectations cluster into: (1) trial design integrity — pre-specified, simplified protocol; preserved randomization and allocation concealment; broadened but defensible eligibility; equipoise. (2) Fitness-for-use of the RWD that supports the trial — the EHR, registry, or claims source used for endpoints, covariates, and follow-up must be relevant and reliable (provenance, completeness, accuracy, traceability), held to the same standard as in FDA's RWD guidances. (3) Endpoint ascertainment from routine data — endpoints must be objective and reliably captured from the data stream; validated computable phenotypes / outcome algorithms, adjudication where needed, and a defensible mortality source. (4) Estimands and intercurrent events — an explicit estimand with treatment-policy (ITT-like) framing for the comparative effect, and pre-specified handling of non-adherence, treatment switching, and crossover that are common when patients are treated in routine care. (5) Retention, missing data, and follow-up — attrition and loss to follow-up minimized and characterized; missing-data mechanisms and analytic handling pre-specified. (6) Human-subject protection and pragmatic conduct — informed consent, safety reporting, and risk-based monitoring adapted to the care setting without compromising participant protection or data integrity.

When NOT to use — limitations and common misapplications

— (a) Do not apply this guidance to non-interventional observational database studies; randomization is its premise, and an observational target-trial emulation needs `fda-rwe-noninterventional` / `fda-rwd-ehr-claims` plus design-based confounding control, not this document. (b) Do not treat it as a reporting checklist or a quality score — it confers no "compliance badge," and a streamlined protocol that ticks its themes is not automatically pragmatic, valid, or generalizable. (c) Do not conflate it with PRECIS-2 (a design-positioning tool) or CONSORT-Pragmatic / SPIRIT (reporting and protocol templates); using a reporting extension where regulatory design guidance is required (or vice versa) is the classic wrong-tool error. (d) "Streamlining" is not license to weaken the parts that carry the evidence: randomization integrity, endpoint ascertainability from routine data, adequate power, and equipoise still bind, and endpoints that cannot be reliably captured from the available RWD (e.g., subjective or imaging-adjudicated outcomes absent in the data) are a fatal design flaw. (e) It does not relax human-subject protections or pharmacovigilance obligations.

How it maps to this catalog

— The randomization premise changes which catalog concepts implement which requirement, and an honest mapping says so. Confounding-control concepts such as `high-dimensional-propensity-score-hdps-rwe` and `active-comparator-new-user` are not the workhorses here — randomization, not propensity adjustment, balances confounders; those concepts belong to the observational alternative, `target-trial-emulation`, which you would reach for only when a randomized pragmatic trial is infeasible. What does implement this guidance: `fit-for-purpose-data-assessment-rwe` (relevance/reliability of the supporting RWD source); `diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe`, `outcome-algorithm-construction-rwe`, and `endpoint-adjudication-chart-review-rwe` (endpoint ascertainment from EHR/claims, with validation); `mortality-source-hierarchy-rwe` (death capture and censoring); `composite-endpoint-construction-rwe` and `estimands-ate-att-intercurrent-events-rwe` (estimand definition and intercurrent-event handling); `attrition-and-loss-to-follow-up-rwe` and `missing-data-pattern-table-rwe` (retention and missing data); `generalizability-transportability-external-validity-rwe` (the representativeness payoff that motivates pragmatic designs); `picots-framework-rwe` and `sample-size-power-precision-rwe` (scoping and adequacy); and `claims-analysis` for the operational mechanics of the supporting data. Applied note for claims/EHR/registry RWE: the trial randomizes, but every endpoint, covariate, and follow-up flag is still read from a real-world feed — so the same phenotype validation, continuous- observability, and mortality-source discipline used in observational studies apply to the trial's data layer, and a weak data source undermines a randomized pragmatic trial just as surely as it undermines an observational one.