FDA RWE for Medical Devices

What it is

— Use of Real-World Evidence to Support Regulatory Decision-Making for Medical Devices is an FDA guidance for industry and FDA staff, issued by the Center for Devices and Radiological Health (CDRH) with the Center for Biologics Evaluation and Research (CBER). The first edition was finalized in August 2017; a revised final version was issued in December 2025 (Federal Register notice of availability, 18 December 2025) pursuant to section 3629 of the Food and Drug Omnibus Reform Act of 2022 (FDORA), which directed FDA to update its real-world data (RWD) / real-world evidence (RWE) device guidance. It is a regulatory decision framework, not a reporting checklist or a risk-of-bias instrument: it describes how FDA evaluates whether RWD are of sufficient relevance and reliability to generate RWE that can support a device regulatory decision, and it is maintained by FDA (CDRH).

When to use

— Use this guidance whenever a device sponsor proposes to generate or submit RWE to support a CDRH/CBER regulatory action: a marketing submission (510(k), De Novo, PMA, PMA supplement, HDE), an expanded or modified indication, label changes, a condition-of-approval post-approval study (PAS), a section 522 postmarket surveillance order, active safety surveillance / signal evaluation through programs such as NEST, or construction of an external/historical control for a single-arm device study. Decision rule: apply this device guidance — rather than the drug/biologic-oriented `fda-rwe-framework` and its companions (`fda-rwe-noninterventional`, `fda-rwd-ehr-claims`) — whenever the regulated product is a medical device or device-led combination product reviewed by CDRH/CBER. Device-specific realities (UDI- rather than NDC-based exposure ascertainment, operator/site learning-curve effects, iterative design changes within a product family, and registry-centric data) make the device guidance controlling. Pair it with study-conduct and reporting tools (HARPER/STaRT-RWE for protocols, STROBE/RECORD-PE for reporting) — those are complementary, not substitutes.

What it requires

— The substantive backbone is fit-for-purpose assessment of RWD against the specific regulatory question. Relevance: the data must capture the device exposure (unique device identifier or device/procedure codes), the target population and indication, and the outcomes at adequate granularity, with sufficient follow-up and sample size. Reliability: data accrual (provenance, completeness, timeliness, representativeness) and data quality assurance/control (accuracy, conformance, transformation/transcription integrity, auditability). Beyond data fitness, the guidance expects the methodological discipline of a credible non-randomized study: a pre-specified protocol and statistical analysis plan; transparent, validated operational definitions for device exposure and outcome phenotypes; correct time-zero/ index-date alignment to avoid immortal-time and other time-related biases; an explicit estimand with pre-stated handling of intercurrent events; rigorous confounding control (active comparator, propensity-score and high-dimensional methods) given the absence of randomization; rigor for external/historical controls where used; accounting for attrition and missing data; and sensitivity / quantitative bias analyses sized to the decision's risk. It also addresses curating and linking device registries, the use of data collected under Emergency Use Authorization, and early engagement with FDA (pre-submission/Q-Submission) to align on data and design before lock.

When NOT to use — limitations and common misapplications

— (1) It is not a reporting checklist or quality score; you cannot "complete" it the way you tick STROBE or RECORD items — it demands substantive evidence of relevance, reliability, and design validity. (2) It does not lower the evidentiary bar: clean, high-quality data do not make an observational comparison causal; confounding, selection, and time-related biases must still be designed out. (3) Wrong-document error: applying it to a drug/biologic (use the `fda-rwe-framework` family) or applying a drug-oriented RWE guidance to a device. (4) Treating registry participation as automatic fitness-for-use — many device registries lack the linkage, comparator, or outcome ascertainment needed for the question at hand. (5) Assuming RWE substitutes for a trial when the question (novel device, no adequate comparator, high residual-bias risk) genuinely requires a randomized design or a single-arm study with a rigorously justified external control. (6) Checklist-as-theater: asserting "fit-for-purpose" without the phenotype validation, balance diagnostics, and bias analysis that substantiate it.

How it maps to this catalog

— Each requirement is implemented by a concept entry here. `fit-for-purpose-data-assessment-rwe` operationalizes the relevance/reliability core. `diagnosis-phenotype-algorithm-1ip-2op-time-window-rwe` and `claims-analysis` implement device-exposure and outcome operational definitions and their validation (UDI, procedure/CPT/ ICD-10-PCS coding). `time-zero-index-date-alignment-rwe` implements index/time-zero discipline. `active-comparator-new-user` and `high-dimensional-propensity-score-hdps-rwe` implement non-randomized confounding control, while `target-trial-emulation` supplies the overarching design discipline. `estimands-ate-att-intercurrent-events-rwe` implements estimand and intercurrent-event specification. `attrition-and-loss-to-follow-up-rwe` implements attrition and missing-data accounting; `e-value-sensitivity-analysis` and `quantitative-bias-analysis-toolkit-rwe` implement sensitivity and quantitative bias analysis. `single-arm-external-control` and `rare-disease-external-controls-rwe` implement the external/historical control arms that are common in device evaluation; `generalizability-transportability-external-validity-rwe` and `regulatory-readiness-rwe` support transportability and submission readiness. Applied note for registry/claims/EHR device RWE: unlike drugs, device exposure has no NDC — anchor exposure on the UDI, device-specific procedure codes, and curated device registries (e.g., cardiovascular implant registries), reconcile operator/site learning-curve effects and iterative device versions across a product family, and link registry data to claims and a death index to complete follow-up and capture mortality.