Real-World Evidence in FDA Prescribing Information Labeling

Real-world evidence in FDA prescribing information: when it makes the label, and when it does not

Real-world data (RWD) and real-world evidence (RWE) increasingly appear in FDA review packages, but the higher bar is whether they reach the approved prescribing information (PI) itself. This overview identifies specific FDA-approved drugs whose labeling incorporates RWD/RWE, contrasts them with cases where FDA reviewers concluded the same class of data could not be included, and summarizes the fit-for-use principles that drive those decisions.

Background: what FDA guidance permits

FDA guidance states RWD/RWE can support regulatory decisions, including labeling changes, when the data are relevant, reliable, and fit for the specific question. For drugs and biologics, RWE may support approval of a new indication for an already approved drug, satisfy post-approval study requirements, and support labeling changes such as adding or modifying an indication, changing dose or route, expanding to a new population, or adding comparative-effectiveness or safety information ³³⁴⁸. RWE may also serve as confirmatory evidence in some circumstances ³²⁴⁶. Critically, the guidance does not create a blanket approval or pass/fail scoring: acceptability depends on the product, the indication, the analyses, and the reliability and relevance of the data source ³⁴⁴⁰⁴⁴. Recurring limitations FDA cites include accuracy and completeness of capture, traceability, missing data, selection bias, heterogeneity across sources, and coding differences ^37394346.

Examples where RWD/RWE appears in the approved prescribing information

Ibrance (palbociclib): real-world data for male patients

The Ibrance label addresses use in men, a population not studied in the pivotal randomized trials, using post-marketing and real-world sources. The PI states: "Based on limited data from postmarketing reports and electronic health records, the safety profile for men treated with IBRANCE is consistent with the safety profile in women treated with IBRANCE" ¹³⁰. The named data source is "postmarketing reports and electronic health records," and the same language is carried through later label revisions ^{134137145146147}. This is a widely cited example of EHR-derived RWD supporting a labeled statement about a new population.

Prograf (tacrolimus): registry data for lung transplantation

The Prograf PI supports lung transplant use with SRTR registry data rather than a randomized controlled trial. The label describes "a non-interventional (observational) study using data from the U.S. Scientific Registry of Transplant Recipients (SRTR)," analyzing outcomes by discharge immunosuppression regimen in primary lung transplant recipients alive at discharge and transplanted between 1999 and 2017 ¹⁴⁸. For the pediatric population specifically, the label notes support from "the experience in the U.S. Scientific Registry of Transplant Recipients (SRTR) including 450 pediatric patients receiving tacrolimus immediate-release products in combination with mycophenolate mofetil and 72 pediatric patients ... in combination with azathioprine between 1999-2017" ^149150151, with one-year graft survival estimates reported by subgroup ¹⁵⁴¹⁵⁵.

Fabrazyme (agalsidase beta): matched external control in Clinical Studies

The Fabrazyme PI explicitly states the evidence base includes "four clinical studies ... and one matched analysis based on data from observational studies" ¹⁶³. Study 5 is described as "a long-term, observational study assessing the rate of decline in renal function (eGFR slope) in 122 patients," with treated patients matched 1:1 to "a historical cohort of untreated patients with Fabry disease" on age, sex, disease subtype, and baseline eGFR; the label reports an eGFR slope of -1.5 mL/min/1.73 m2/year for treated versus -3.2 for untreated (difference 1.7; 95% CI 0.5, 3.0) ¹⁶⁸. The label separately references a Fabry patient registry for long-term monitoring ¹⁶⁸¹⁶⁹.

Skyclarys (omaveloxolone): natural history comparison with an explicit caution

The Skyclarys PI includes a post hoc, propensity-matched external control drawn from a natural history study: "In a post hoc, propensity-matched analysis, lower mFARS scores were observed in patients treated with SKYCLARYS after 3 years relative to a matched set of untreated patients from a natural history study. These exploratory analyses should be interpreted cautiously given the limitations of data collected outside of a controlled study, which may be subject to confounding" ¹⁷⁷. The cautionary framing, describing the analysis as exploratory and confounding-prone, is characteristic of how FDA labels supportive (rather than pivotal) RWE, and the text persists across label versions ^178179180.

RWE as an external control supporting approval (review-level, not always in the label)

Several rare-disease and oncology approvals relied on real-world external controls in the review, though the extent to which that language reaches the PI varies:

• Brineura (cerliponase alfa): a non-treatment natural history cohort (Study 901, registry-based) served as the external comparator for single-arm treatment studies 201/202, assessing the CLN2 Motor domain over 48, 72, and 96 weeks; FDA noted the comparator relied largely on retrospective chart review and patient interviews ⁶⁴¹³.
• Voxzogo (vosoritide): FDA reviewed single-arm long-term extension data against a real-world natural history control (primarily the AchNH registry) with age/sex matching to compare height change over 5 years ⁵¹¹¹⁵. Notably, the reviewed natural history comparison did not surface in the label text examined, which describes the randomized, placebo-controlled Study 1 and its open-label extension instead ^108115120. This illustrates the gap between review-level RWE and label content.
• Duvyzat (givinostat): natural history external controls were used in an integrated analysis of long-term efficacy in Duchenne muscular dystrophy; the sponsor submitted RWE datasets and code, and FDA treated the natural history analysis as exploratory but part of the confirmatory-evidence picture ⁸⁴⁹⁷.
• TALVEY (talquetamab): a propensity-weighted comparison of single-arm MonumenTAL-1 against individual patient data from prospective non-interventional studies (LocoMMotion, MoMMent) served as a real-world external control in relapsed/refractory multiple myeloma ².
• Paxlovid (nirmatrelvir/ritonavir): FDA reviewed real-world studies using national EHRs, administrative claims, and integrated healthcare data (including the U.S. Veterans Health Administration and datasets from Quebec, Israel, and Hong Kong) associated with reduced hospitalization/worsening outcomes ⁹⁸¹⁰⁰. However, the label text examined presents the randomized EPIC-HR results and exploratory rebound analyses, not separate real-world effectiveness results ²⁰⁵²⁰⁷, again underscoring that review-level RWE does not automatically become label content.

A useful contrast is IWILFIN (eflornithine): FDA reviewed an external control derived from a prior clinical trial (Study ANBL0032), but the review explicitly notes this external control was not considered real-world data because it came from a previously conducted clinical trial ¹⁶. External control does not equal RWE.

Cases where FDA said RWD/RWE could not be included

The mirror image, where reviewers concluded observational or real-world data had limitations that precluded a labeling claim or an effectiveness conclusion, is equally instructive:

• Chantix (varenicline): FDA concluded the observational studies were "of insufficient quality" to rule in or rule out increased risk of suicide, non-fatal self-harm, or neuropsychiatric hospitalizations, citing outcome-validity concerns, use of bupropion as comparator, channeling bias, residual confounding (including with nicotine replacement and withdrawal), limited statistical power, and under-ascertainment of outcomes; FDA therefore recommended not including the findings in labeling ⁶⁴⁷².
• Rozlytrek (entrectinib): FDA judged a real-world crizotinib comparator arm unlikely to be generalizable to the ROS1-positive NSCLC population (low real-world ROS1 testing rates, estimated 15%-30% sensitivity) and not sufficiently comparable to the trial population; specific problems included differential eligibility, baseline imbalance/selection bias, small sample size, treatment beyond progression complicating time-to-discontinuation, missing radiographic imaging in the EMR limiting PFS, and potential OS bias ⁵⁵.
• Siliq (brodalumab): for suicidal ideation and behavior, FDA determined the available data were insufficient and that existing pharmacovigilance and pharmacoepidemiology methods would not be adequate to assess the risk post-market, so no observational data collection was recommended given the limitations of such data for suicidal outcomes ⁵⁷⁷⁰⁷³.
• Aqneursa (levacetylleucine): the submitted RWD comprised three published case series of off-label Tanganil use with unclear, possibly overlapping patient counts; FDA characterized them as early proof-of-concept only, not supportive data, with limitations precluding acceptability as confirmatory evidence on their own ⁵⁸.
• Forzinity (elamipretide): FDA found expanded-access case reports "not complete or comprehensive enough," not systematically collected, and limited (three described cases, one with a different disease), and rejected the natural history / left-ventricular-volume analyses because FDA had not agreed that LV volume is a reliable biomarker predicting clinical benefit, with unknown overlap between natural-history and trial populations ⁵⁴⁹.

Why RWE is accepted in some labels and rejected in others

The pattern across these examples tracks FDA's stated fit-for-use framework. RWD/RWE has reached labeling where it addresses a defined, answerable question with a credible comparator and transparent limitations: a safety-consistency statement for an unstudied population (Ibrance) ¹³⁰, a registry-based effectiveness/graft-survival description for a transplant setting where randomization is impractical (Prograf) ¹⁴⁸¹⁴⁹, and a matched observational renal-outcome analysis in a rare disease (Fabrazyme) ¹⁶⁸. Where it is supportive rather than pivotal, the label carries explicit cautionary language about confounding and non-controlled data (Skyclarys) ¹⁷⁷.

Conversely, FDA declined RWD/RWE where the data could not support the causal claim at issue: unreliable or biased outcome ascertainment and confounding for sensitive safety signals (Chantix, Siliq) ⁶⁴⁵⁷, a non-generalizable and non-comparable external comparator (Rozlytrek) ⁵⁵, sparse or overlapping case series offered as effectiveness evidence (Aqneursa) ⁵⁸, and unsystematic expanded-access reports paired with an unvalidated biomarker (Forzinity) ⁵⁴. This is consistent with FDA guidance that acceptability turns on relevance, reliability, and fit for the specific decision, with no blanket approval and no change to existing evidentiary standards ³⁴⁴⁵⁴⁶.

Limitations of this review

This is an illustrative, not exhaustive, survey. It is drawn from Drugs@FDA labels and reviews plus FDA's RWE guidance; it does not capture every product whose labeling references registry, EHR, claims, or natural history data, and terminology for "real-world" evidence is inconsistent across reviews, so keyword-based discovery can miss cases. Where a product appears only in review text, that is flagged, because review-level use of RWE does not establish label inclusion. A regulatory team planning a specific submission should confirm the current label wording for any product cited here and discuss proposed RWE use with the relevant FDA review division, as the guidance advises ⁴⁷⁴⁸.