Framework for FDA's Real-World Evidence Program

What it is

— The Framework for FDA's Real-World Evidence Program is a strategic, programmatic document published by the U.S. Food and Drug Administration (Center for Drug Evaluation and Research and Center for Biologics Evaluation and Research) in December 2018, in response to Section 3022 of the 21st Century Cures Act. It articulates the principles FDA will use when assessing whether real-world data (RWD — data on health status and care delivery routinely collected from sources such as claims, EHRs, product and disease registries, and digital health technologies) can generate real-world evidence (RWE — clinical evidence about a product's use and potential benefits or risks derived from analysis of RWD) of sufficient quality to support a new indication for an approved drug or to satisfy post-approval study requirements. The Framework is organized around three evaluation pillars that the agency applies to any proposed RWE submission: (1) whether the RWD are fit for use — relevance and reliability of the data; (2) whether the study design used to generate RWE provides adequate scientific evidence to answer the regulatory question; and (3) whether the study conduct meets FDA regulatory requirements (e.g., monitoring, data integrity). It is maintained by FDA, which has since operationalized the Framework through a series of more specific RWE guidances (2021–2024). The Framework is the umbrella; the operational expectations live in the sibling guidances.

When to use

— Consult the Framework when scoping a regulatory RWE strategy for an approved drug or biologic: deciding whether a non-interventional study, a hybrid/pragmatic design, an externally controlled trial, or a registry-based study could plausibly support a label expansion or fulfill a post-marketing requirement, and when preparing for a sponsor-FDA meeting on RWE. It is the right reference for the programmatic, decision-grade question — "will FDA consider this kind of RWE credible for this kind of decision?" — and for orienting a team to FDA's three-pillar evaluation logic before any protocol is drafted. Decision rule for which document applies: use THIS Framework for high-level strategy and the agency's evaluation lens; switch to the operational FDA guidances for design and submission detail — `fda-rwe-noninterventional` (non-interventional study design and analysis expectations), `fda-rwd-ehr-claims` (assessing EHR and medical-claims data for fitness-for-use), and `fda-rwe-devices` (RWE for medical devices, a separate CDRH pathway). For EU/EMA strategy use the ENCePP and GVP Module VIII references; for HTA/payer dossiers, RWE strategy is governed by HTA-body methods guidance (e.g., NICE RWE framework), not by this FDA document. For protocol templating and reporting — the steps the Framework expects but does not itself specify — use HARPER/StaRT-RWE (protocol) and STROBE/RECORD-PE (reporting).

What it requires

— The Framework does not impose a numbered checklist; it sets evaluation expectations across its three pillars, each of which maps onto concrete RWD methods. Data relevance and reliability (fit-for-use): the data must capture the exposures, outcomes, and key covariates needed for the question, with adequate accuracy, completeness, provenance, and quality controls — including validated phenotype/outcome algorithms and a documented data-curation/linkage trail. Study design rigor: FDA explicitly endorses applying clinical-trial design principles to observational data — pre-specification, a clear causal estimand, an appropriate comparator, correct time-zero/index alignment to avoid immortal-time and selection bias, and rigorous confounding control. Study conduct: transparency, pre-registration where applicable, data integrity, and pre-specified sensitivity and quantitative bias analyses so that the evidence is reproducible and defensible. In practice this means a submission must document data-source fitness, phenotype validation (PPV/sensitivity), exposure and outcome definitions, estimands and intercurrent-event handling, attrition, and bias analyses — the substantive domains a reviewer will probe.

When NOT to use — limitations and common misapplications

— The single most common error is treating the Framework as a checklist to complete. It is a strategic/programmatic document, not a reporting checklist, not a critical-appraisal or risk-of-bias instrument, and not a quality score; "satisfying the Framework" is not a meaningful claim, and an analyst who needs item-level reporting structure should reach for STaRT-RWE, HARPER, RECORD-PE, or STROBE instead. A second error is using the 2018 Framework where the operational FDA guidance is required — it is the umbrella, and the design- and submission-level expectations live in the 2021–2024 guidances (`fda-rwe-noninterventional`, `fda-rwd-ehr-claims`, `fda-rwe-devices`); citing the Framework when a reviewer expects the specific guidance is the "wrong-document" trap. The Framework also does not, by itself, make an observational study causal: invoking it does not substitute for the actual design and analytic work (active comparator, time-zero alignment, confounding control, sensitivity analysis) that earns a causal interpretation. It is U.S.-FDA and drug/biologic specific — it does not govern EU/EMA submissions, HTA decisions, or (directly) devices — and it addresses effectiveness for approved products, not initial approval or safety surveillance per se. Finally, it is a framework, not a standard: it signals what FDA values but defers the methodological specifics to other guidances and to the scientific literature, so it should never be the sole cited authority for a design choice.

How it maps to this catalog

— The three pillars map directly onto implementing concepts in this repository. Data fit-for-use is implemented by `fit-for-purpose-data-assessment-rwe` (relevance and reliability assessment), `claims-analysis` (claims structure and limitations), `medicare-ffs-ma-commercial-claims-differences-rwe` (payer-driven data-capture differences), and the phenotype concepts `diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe`, `ehr-phenotyping-algorithms-rwe`, and `claims-outcome-algorithm-ppv-sensitivity-rwe` (algorithm development and PPV/sensitivity validation). Study design rigor is implemented by `target-trial-emulation` (pre-specify the hypothetical trial before emulating it), `active-comparator-new-user` (comparator choice and incident-user restriction), `time-zero-index-date-alignment-rwe` and `immortal-time-bias-handling` (correct index alignment), `estimands-ate-att-intercurrent-events-rwe` (causal estimand and intercurrent events), and `high-dimensional-propensity-score-hdps-rwe` and `propensity-score-methods-psm-iptw` (confounding control). Study conduct, transparency, and robustness is implemented by `study-protocol-or-sap-elements`, `attrition-and-loss-to-follow-up-rwe` and `database-feasibility-attrition-funnel-rwe` (attrition reporting), `e-value-sensitivity-analysis` and `quantitative-bias-analysis-toolkit-rwe` (quantitative bias analysis), and `regulatory-readiness-rwe` (assembling the submission package). A reviewer evaluating a claims- or EHR-based submission under this Framework will, in effect, walk the chain from `fit-for-purpose-data-assessment-rwe` and `diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe` (is the data fit and are the phenotypes validated?) through `active-comparator-new-user` and `time-zero-index-date-alignment-rwe` (is the design trial-emulating and free of immortal time?) to `e-value-sensitivity-analysis` (how robust is the result to unmeasured confounding?). Treat the Framework as the lens and these concepts as the lenses' implementation.